Portal Factory · the board

What did they do while I was asleep?

What happened

The last brief — 16 days old, in 2 lines

Staff messages and group announcements plan · October 22 · lead

0 commits, last 24 hours — 3319 on every branch all time · 317 unpushed

Who did what

Claude account 1 is working now.

Who	Entries on this board	Changes saved since the switch
Unknown	82	0
Claude account 1	347	858
Claude account 2	23	659
An unregistered Claude account	0	2
Unknown	1	0
Account not recorded	9	0
Commits with no stamp	0	19

Claude account 1 did all the work up to October 6 at 1:31 AM, when Alyssa switched to a second account. Since then, every saved change carries the name of the account that made it, and every entry below says whose it is.

What changed — 462 entries over 40 days

Every change any session made, newest first. Today is open; each earlier day is a fold, and none of them opens itself.

Tuesday, November 3 2 changes

8:54 PM8:54 PM ET: Question bank: Riley’s trim is LIVE on MK11 (since 8:42 PM ET today), on Alyssa’s word “please just make edits for Riley.” Module quizzes went from 15 quizzes and 105 questions to 4 quizzes and 21 questions, in the order Riley listed. The final exam was not touched (28 items). Nothing was deleted: 94 questions sit in the bank’s archive and can be restored. This was a content change, so the live release stays 125. A report went to Dave by email. Nothing paid. Not done: nobody has signed in on the live site to look yet, and Riley has notes to review (the term length answer disagrees with the lesson).Unknown
5:52 PM5:52 PM ET: RELEASE 125 IS LIVE (since 5:49 PM ET today) on MK11, on Alyssa’s go (“Yes, release the fix”). Question bank: an emptied quiz can no longer lock learners out. Before, if every question in a module quiz was retired, the later modules, the final exam and the certificate stayed locked for learners; now an empty quiz blocks nothing. No question changed on the live portal, and no quiz there is empty today, so nobody will see a difference yet. The fix was proven on a copy of the live portal and by tests, not watched working live, and it has not been walked signed in on the live site. Riley’s trim of the module quizzes (15 quizzes down to 4) has been rehearsed on a copy and is NOT live; it waits on Riley’s answers. Nothing paid (a hosting release, no model calls). Going back to release 124 is possible and changes no data.Unknown

Monday, October 26 — 2 changes

12:59 PM12:59 PM ET: Alyssa chose to post nothing to the staff What’s New board for release 122 (her item 212, about 12:2x PM ET): the release 117 back-arrow post already covers it. That question is closed, so it is off the needs-you list. The one call left on this release is her go or hold on the skip-link fix (register row 11.28).Unknown
12:16 PM12:16 PM ET: RELEASE 122 IS LIVE (since 12:13 PM ET today) on MK11, on Alyssa’s go (her item 211). It is release 121 plus the back-arrow history keys. What changed for a person: if you hold the browser’s Back button to jump several pages at once, or use the skip link and then press Back, the page’s back arrow now names the right page every time, and from there pressing the browser’s Back twice takes you home. The style files are identical to release 121 (nothing changed in light or dark), and there is no database change. Checked: a restore point first; the release steward said DEPLOY OK before the deploy; the health check answers healthy on all three addresses; both reviewers cleared it (accessibility officer: PASS WITH NOTES, design critic: SHIP). Alyssa accepted one rare case as how it behaves live: a single Forward after a Back can name the page two back. One older skip-link problem is recorded for its own small release. Release 120 (larger text on the step-9 pages) and release 121 (avatars on the manager side) went live overnight. A signed-in walk of the new release on the live site is still to come. Nothing paid (hosting only). Going back to release 121 is possible and changes no data. Update 12:59 PM ET: she chose no What’s New post for this release (the release 117 post covers it). Still open for Alyssa: a go or hold on the skip-link fix.Unknown

Sunday, October 25 — 12 changes

11:10 PM11:10 PM ET: RELEASE 121 IS LIVE (since 11:06 PM ET Oct 25) on MK11, on Alyssa’s go. Avatars a person draws in their Learner view now carry over to the manager side: the manager’s own avatar in the masthead, and each learner’s avatar on the manager home, roster, learner page and messages (initials where none is drawn). Her words: “All avatars should also carry over to the manager side.” A signed-in walker saw her avatar in the masthead live, in light and dark. Nothing paid. Open and not new: the Stripe webhook errors with no Stripe keys set, and the roster header seat count (1 of 25) differs from Home (2 of 25).Unknown
10:26 PM10:26 PM ET: RELEASE 120 IS LIVE (since 10:18 PM ET Oct 25), on Alyssa’s go, and plan step 9 (design pass) is marked done. Release 120 carries the larger-text fixes the design reviewer (SHIP) and accessibility officer (PASS, not a certification) signed off and Alyssa approved: dark sample-bill box on the exam, roster scrolling inside its card, exam start page, home card and sidebar at larger text, the lesson page at larger text with the Read-aloud bar kept, and the square light-mode logo plate. A signed-in walker passed six live checks. Not verified live: the “All announcements” link, the real sample-bill picture on the live exam, signed-in phone widths. Still open, not blockers: two narrow-column spots at the largest text, four parked brand-audit items waiting on her yes or no, the Read-aloud bar appearing late. Nothing paid; nothing posted to What’s New (that waits on her pick).Unknown
10:08 PM10:08 PM ET: Alyssa posted the release 117 item to the staff What’s New board herself, at 10:06 PM ET Oct 25, for MK11 only. Title: “Back arrows follow your browser’s Back button.” It says, in two sentences, that the back arrow at the top of a page now returns you to the page you really came from, even after you press your browser’s Back button or swipe back on your phone, and that it never sends you around in a circle. The Back-button fix is release 117 (live since 5:18 PM ET). Production is on release 119 (Super Admin “View as”, live since 6:30 PM ET), which sits on release 118 (Grid Map’s light mode, since 5:37 PM ET); Fly confirms release 119 is the newest. Not read back from production by this session; the posting tool’s own read-back showed the row. The What’s New question for release 117 is closed, and the only item left there is the go or hold on the next small release (register rows 11.24 and 11.25). Nothing paid.Unknown
9:29 PM9:29 PM ET: Step 9 (design pass) has its sign-off. The design reviewer said SHIP and the accessibility officer said PASS (not a certification) on the final fixes, after five review rounds today; Alyssa reviewed the screenshots and approved them. Fixed on the working copy only: the dark sample-bill box on the exam, the roster scrolling inside its card, the “All announcements” link contrast, the exam start page and sidebar at larger text, and the lesson page at larger text with the Read-aloud bar kept. NOT LIVE: production is still release 116, and a release needs Alyssa’s separate go. Step 9 stays “In progress” until then. Nothing paid; nothing posted to What’s New.Unknown
6:36 PM6:36 PM ET: Super Admin “View as” is LIVE for every portal (Fly release 119, since 6:30 PM ET Oct 25), on Alyssa’s go. A Super Admin now sees “View as” beside their name on every manager page and can look at any portal as an Admin, a Manager, a Mentor or a Learner. It is view-only (looking changes nothing), every switch is written to the audit record, and sign-out ends the view. A client’s own staff never receive the control. It is Alyssa’s pick, concept 3, “View As, Under Your Name”, and it sits on top of release 118 (Grid Map’s light mode). Checked: all the safety and quality checks passed on a fresh database; the release steward said DEPLOY OK before and “live and clean” after; the health check answers healthy three times; the style files compared with release 118 show 0 changed (the new control’s rules were added; dark mode is untouched). Two tries failed on the way, plainly: the first build was refused because a piece of the control was reaching into protected sign-in code from the browser side (fixed, and a test now pins it), and then Fly’s own builder ran out of memory, so nothing went live from either; on Alyssa’s “Continue” it was built on this Mac and deployed from here. Not yet seen by a person: a signed-in walk on the live site as her Super Admin (nobody holds that password here), and nobody has used the menu on production yet. Nothing paid: no model runs, only design mock-ups and code. Going back to release 118 is possible and changes no data.Unknown
6:27 PM6:27 PM ET: Alyssa answered the What’s New question for release 118 (Grid Map’s light mode): “Do not post anything to the What’s New board for this.” Closed as answered: no post was made, and nothing was posted. The manual’s Grid Map pictures stay as they are.Unknown
5:41 PM5:41 PM ET: release 118 is LIVE for MK11 (since 5:37 PM ET Oct 25, on Alyssa’s go): the arcade game Grid Map now has a light mode. She asked for three concepts and picked A, “ink on paper” (white board, navy active word, gold ink for finished letters and the lit run, no glows); B and C are kept as a record. It sits on top of release 117’s back-arrow fixes. Checked: restore point first, 1 of 1 health checks, all three portal addresses healthy, style files against release 117 with 0 removed, 0 changed, 18 added (all under light-only Grid Map rules; dark-mode-only declarations 114 before and 114 after, so dark mode is untouched), steward DEPLOY OK. Along the way: the checks stopped at two failed tries (database connections used up by other sessions’ servers, then a Walk Map guard that needed narrowing), and the first deploy try at 5:24 PM failed during the build on a style-compressor crash, so nothing went live from it; fixed, with a test that refuses the pattern. The signed-in live look at Grid Map in light and dark is still owed: the walker could not run it at about 5:39 PM ET because the Chrome connection on this Mac is down, and no page was opened. Correction to the 5:23 PM entry: the release 117 live walk at 5:21 PM could not run (no browser connection), so its two back-arrow checks are not yet confirmed on the live site. Answered at 6:26 PM ET (see the entry above this one): no What’s New post. Nothing paid.Unknown
5:22 PM5:23 PM ET: release 117 is LIVE for MK11 (since 5:18 PM ET Oct 25, on Alyssa’s go). The browser’s Back button and a phone’s back swipe now name the true previous page; no loop with scripts off; the Team strip keeps a focused item in view when the window narrows. Re-run on a quiet machine first: 33 of 33 passed. Checked: restore point first, 1 of 1 health checks, all three portal addresses healthy, style files 0 removed, 0 changed, 0 added against release 116 (dark mode untouched), accessibility officer PASS WITH NOTES, steward DEPLOY OK. Not yet on this board: the live walk of the two back-arrow checks (it could not run at 5:21 PM, the Chrome connection was down) and the steward’s after-deploy re-read. Two rare long-press-Back cases are open for the next small release at her word. Open for her: which release 117 changes to post to What’s New. Nothing paid.Unknown
2:18 PM2:18 PM ET: the staff What’s New board on MK11 has a new post, “Question bank”, on Alyssa’s word (posted 2:18 PM ET Oct 25). One sentence: Admins can now change every module quiz question, final exam question and flash card from one Question bank page in Manager view, and set a time limit for the quizzes and the final exam. Nothing paid.Unknown
2:11 PM2:11 PM ET: release 116 is LIVE for MK11 (since 2:08 PM ET Oct 25, on Alyssa's go). On top of the question bank (release 115), a timed, recorded final exam's start page now names the exact clock time its clock starts if the learner has not shared their screen by then, in her wording; it closes the one Important finding of the final review. Checked: restore point first, 1 of 1 health checks, all three portal addresses healthy, style files compared with release 115 with 0 removed, 0 changed and 0 added (dark mode untouched), accessibility officer PASS WITH NOTES, steward DEPLOY OK. The new words are not visible to anyone yet: nothing is timed on the live portal until an Admin sets an exam time limit. Not walked by hand for that reason. Nothing paid. Still hers: the manual edition with the question-bank section, and the stage 2 and stage 3 questions. Next release in line: the back-arrow fix as release 117.Account not recorded
1:38 PM1:38 PM ET: Alyssa told us we got our first paying client (Oct 25, her words: "We got our first paying client."). Step 13, First paying client, is marked done on her word. Not yet checked against Stripe or our records: no client or amount is written anywhere we can see, so the board names none.Unknown
1:28 PM1:30 PM ET: the MK11 question bank is LIVE again, as release 115 (since 1:08 PM ET Oct 25, on Alyssa's go). Admins can edit quiz questions, exam items, flash cards and time limits. It came from the clean line: release 112 plus the bank, none of the eight 2.0 fixes she held (that hold still stands). Checked: 1 of 1 health checks, all three portal addresses healthy, style files 0 removed and 0 changed against the previous release (dark mode untouched), and a live walk at desktop width in light and dark passed. Phone width and the learner quiz and exam pages were not walked. Nothing paid.Unknown

Saturday, October 24 — 8 changes

4:27 PM4:27 PM ET: ROLLED BACK. On Alyssa's word ("Roll v113 back to v112 now") production went back to the exact release 112 build at 4:25 PM ET (Fly number 114). Checked: 1 of 1 health checks, all three portal addresses answer healthy, the sign-in page serves exactly release 112's style files. The question bank (live 3:59 to 4:25 PM ET as release 113) and the eight held 2.0 fixes are off the live portal; production data was untouched while they were live (no question edits, no audit entries). The question bank returns through a clean, fully checked release line on her go. Nothing paid.Unknown
4:10 PM4:10 PM ET: the MK11 question bank is LIVE as release 113 (3:59 PM ET, on Alyssa's go): Admins can edit quiz questions, exam items, flash cards and time limits. Every check passed and the live walk passed at desktop width in light and dark. But release 113 also carries eight 2.0 fixes Alyssa held on Oct 22, which the lead found after the deploy and should have caught before. A rollback to 112 was tried and refused by the session's permissions, so it is her call: roll back to 112 or keep 113. A clean line without the held fixes is being tested and deploys only on her word. Nothing paid.Unknown
3:55 PM3:55 PM ET: release 112 is checked live. The style files match release 111 with nothing removed or changed (53 new rules for the Team strip), the three live addresses answer healthy, and a live walk as Alyssa's own views passed 7 of 8 checks (phone width was not walkable). One known flaw is confirmed live: after the browser's Back button from the accessibility statement, the settings arrow can name the wrong page. The accessibility records now read live in release 112. The follow-up fix for that flaw stopped at two failed tries; the cause is found and it waits on Alyssa's word. Nothing paid.Unknown
1:21 PM1:21 PM ET: release 112 is LIVE for MK11: back arrows on the learner and staff pages, and the Team menu as the Desk Strip (her pick, item 190), on top of release 111, on her word to release as soon as finished. No database change. Fly shows release 112 complete. The checks were recorded later (see the 3:55 PM entry).Unknown
12:53 PM12:53 PM ET: Back arrows and the Team menu (Desk Strip, her pick, item 190): built and reviewed in the other working session, now cut as a release line on top of what is live. On her go (items 191 to 195) the back-arrow fixes were made after the tests were shown failing first; the working session stopped twice at its two-strike limit (once on the profile Back case, which turned out to be a test racing the page load, and once on the release gate's seed data) and, at last report, was stopped at the second on the release gate (a gate-database seed problem) and waiting to be unblocked. UPDATE 3:55 PM: it went live at 1:21 PM ET as release 112 on top of 111, and her word "release as soon as finished" (item 197) covers it once the gates pass. Nothing paid.Unknown
12:50 PM12:50 PM ET: release 111 is LIVE: the two name fixes (her item 197b, a one-time exception to the hold; her words: "please release the updates live as soon as finished"). It is release 110 plus: Jackson Armstrong's briefing no longer says "Jackson Reeves", and the interview candidate Maya Okafor is now Stephanie Okafor (her live record renamed in place, her past calls kept). No database structure change, no style change, nothing paid. Restore point made at 12:43 PM; going back to release 110 is one command. A live walk as her learner view passed (Stephanie Okafor with her photo, the other three candidates, no "Maya"; 24 customers, Jackson Armstrong under Bear). The back-arrows release goes on top as 112, so the held 2.0 and exam-fixes release is now 113 or later; her hold (item 169) stands for everything else. Still open: manual page 111 still says Maya (being retaken).Unknown
12:31 PM12:31 PM ET: the MK11 user manual is now the BOLT edition (her item 196: "Yes, it's the manual still saying the old DISC names. Get the BOLT edition of the manual onto the site."). Edition r7c (149 pages, dated October 24, with the interview-practice steps and no DISC wording) replaced the old 145-page edition on MK11; the upload was checked and logged. Its last fixes were checked by measurement, not re-read by the three reviewers. One page (111) still says "Maya" and will be retaken. Nothing paid.Unknown
11:52 AM11:52 AM ET: the staff What's New board on MK11 has a new post, "Interview practice for hiring managers" (her item 194, posted after a dry run and her "Yes, post ... as written"). Role play now has a Candidates side where you can practice interviewing four job candidates, and afterward the coach checks off what you covered. It sits beside the BOLT post; posts are never edited in place. Read back correct.Unknown

Friday, October 23 — 9 changes

6:02 PM6:02 PM ET: release 110 is LIVE: MK11 now has interview practice for hiring managers (her items 186 and 187, a second one-time exception to the hold). It is release 109 (BOLT) plus the interview role plays and one database change, 0083 (the live database went from 80 to 81 changes, read back correct). Gates green, the production check said go, and a restore point was made at 5:48 PM; it went live at 5:52 PM and the style files were compared with release 109: nothing removed or changed, so dark mode is untouched. After she saw the dry run, the four candidates (Maya Okafor, Devin Castillo, Andre Whitfield, Cody Brennan) were added to MK11 at 5:59 PM. The Role play page has a Customers | Candidates switch; the Candidates side is shown only to learner accounts linked to a staff account. A live walk passed in light and dark (Customers side unchanged with 24 customers; Candidates side shows the four, no customer on it); phone width was not re-checked live. One interview call with 0 turns, opened during the walk, was hers (she confirmed at about 6:05 PM ET and said to leave it). Going back to release 109 is one command. Naming: Fly gave this release the number 110, so the held 2.0 and exam-fixes release is now 111 or later. Her hold on everything else (item 169) stands. Nothing paid for the release. Nothing posted to the staff What's New board; that waits on her pick.Unknown
5:18 PM5:15 PM ET: release 109 is LIVE: MK11's role plays now speak BOLT (her item 184, a one-time exception to the hold). It is release 108 plus only the two BOLT changes; nothing from the Apple Mac rebuild ("2.0") went live, and there is no database, add-on or style change. Gates green, a read-only database check found nothing wrong, a restore point was made at 5:11 PM, and the live and sign-in pages answer. After she saw the dry run, MK11's setting was switched to BOLT (read back matching, eight other settings unchanged). A learner walk in light and dark passed: Bear, Owl, Lamb and Tigger, 6 customers each, no DISC wording; every other portal stays on DISC. Going back to release 108 is one command. Naming: Fly gave THIS release the number 109, so the held 2.0 and exam-fixes release is now the NEXT release, 111 or later (release 110 went to the interview role plays). Her hold on everything else (item 169) stands. Nothing paid. Nothing posted to the staff What's New board; that waits on her pick.Unknown
3:04 PM3:04 PM ET: the Apple Mac rebuild ("2.0") has passed all ten checks and is with Alyssa to test. At about 2:50 PM ET Alyssa ruled that the browser page walks count as the same result on both chips (the goblin game's motion test failed the same way on both, 28 versus 27, a random reading), and asked for one re-run of the focus-ring check. The re-run read the same focus ring on both chips; the overnight difference was a one-time timing catch, not a chip difference. All ten checks are done. Her test guide passed the accuracy checker (it was sent back twice, fixed, and passed on the third read) and was handed to her at about 3:02 PM ET. Next: she walks the Apple version with the guide, then picks a switch-over day. Unchanged: the live portal is still release 108, release 109 is still on hold (her item 169), the Mac switch-over day is not picked, and nothing was sent live.Unknown
12:29 PM12:30 PM ET: Library round 2 is FINAL. Both of Alyssa's picks are done (items 181 and 182). The Share panel title breaks a word only when it cannot fit, and the hyphen is now proven in Apple's browser engine (WebKit, installed on her yes). Section names in the Sections panel now show whole at phone width with large text: the field grows with the name, two lines at least, tested with an 80-character name; desktop and normal-size text are unchanged. The design critic and the accessibility officer passed it and the final checks are green. It is with the release session, not live; her deploy word is next (database change 0081 first). The name-field and WebKit questions are closed and removed from Needs you; what remains is optional leftover look items. Nothing paid.Unknown
5:05 AM5:05 AM ET: the Apple Mac rebuild ("2.0") now NEEDS YOU. The overnight browser page walks on the Apple and Intel chips finished at 4:58 AM: 77 of the 78 walks gave identical results on both chips, and 389 tests passed on each. The 78th, the goblin game's motion test, failed on BOTH chips with a random reading (28 on Apple, 27 on Intel). The goblins move at random by design, and the check's own rule counts any difference as a miss, so this was the second miss on this topic and the work stopped for Alyssa's ruling. One accessibility check also read a button's focus ring in a different color on each chip (both colors are used on the page); it is probably a timing matter rather than the chip, and one targeted re-run would show. Besides the five passed checks, the plan was corrected, the proof scripts were saved and her test guide is ready. Her ruling on the walks, then her own walk of the Apple version with the guide. Unchanged: the live portal is still release 108, release 109 is still on hold (her item 169), the Mac switch-over day is still not picked, and nothing was sent live.Unknown
1:22 AM1:22 AM ET: overnight checks on the Apple Mac rebuild ("2.0"). Alyssa's own words, about 11:39 PM ET on Oct 22: "A: finished means the Apple version passes every proof and is handed to me to test; a later switch-over day makes it official. Start the ten checks you can do without me, with nothing sent live." Five of the ten checks have passed so far: (1) the leak-scanner test, where both the Apple and the Intel scanner catch a planted fake key and stay quiet on a clean folder; (2) the strict side-by-side test of the Intel and Apple chips, identical on every check (same results, same failing tests, same counts, and the Apple side ran faster); (3) one real PDF through the virus scan and text reading, which passed on the second run after a fix to the check's own rule (the first run failed because of how the check counted a reused connection, not because of the PDF); (4) Alyssa's golden 0.1 video rebuilt on the Apple tools, which matches the original frame for frame, with the same length, slide changes, captions and loudness (nothing bought or attached; it is used only as the yardstick, and no Series 65 content changed); (5) a build-and-backup rehearsal with nothing sent, which made a backup point labeled as a rehearsal and built the release image on this Mac only. Also done: the plan documents were corrected, the proof scripts were saved, and her test guide was drafted. Still running: the browser page walks on both chips, about 4 hours. Unchanged: the live portal is still release 108, release 109 is still on hold (her item 169), the Mac switch-over day is still not picked, and nothing was sent live.Unknown
1:11 AM1:10 AM ET: Library round 2, Alyssa's pick on the Share panel title (item 181) is done: a word in the panel title now breaks only when it cannot fit, with a hyphen on real phones; desktop and normal-size text unchanged. The release candidate moved again to include it and the checks are green. The Sections panel name fields were measured, not changed: at phone width with large text there is room for about 140 pixels of text against names needing 188 to 245, so she chooses a two-line field (recommended) or smaller text. Test browsers cannot show the hyphen; installing Playwright's WebKit (about 90 MB) would, and waits on her yes. Still with the release session, not live; her deploy word is next (0081 first). Nothing paid.Unknown
12:19 AM12:20 AM ET: Admin unlock (Alyssa's item 166) moved from waiting on her to built, approved to build, handed over, and still NOT live. Alyssa gave her written OK (item 167) for the unlock table's database change (0082), its undo script and its isolation test, and for three accessibility fixes the design critic named. All were done: the table is created with row rules on and forced, and the app can read, add and relock rows but never delete them; the isolation test passes 6 of 6; the notice now takes keyboard focus after a press, the section clears the sticky header, and the gold look shows the right stripe for 'already open'. The design critic passed round 3. Every check is green on a copy of the database (leak test 388 of 388; only the six known, unrelated failures; secret scan clean). The branch is the release candidate after Library round 2 with the release session (0081, then 0082). Not merged, 0082 not on production, nothing deployed: those wait on her deploy word in the release session. Nothing paid.Unknown
12:18 AM12:18 AM ET: Alyssa's pick on the Library's sideways scroll is DONE (item 180). At phone width with large text, the Library pages no longer scroll sideways (358 down to 320 on five surfaces, light and dark); nothing changed at normal text size; before-and-after pictures went to her. Design critic PASS, accessibility officer release PASS, final checks green. Library round 2 is final and with the release session as the release candidate; her deploy word is next, with database change 0081 first. Not live. The 'sideways scroll' ask is closed. New small picks for her (none being built): the Share panel title breaking a name mid-word, the Sections name fields cutting names off, the upload button's wrap, and three older learner-page flaws at phone width (coin icon over the logo, a stray dot, a missing space). Unchanged: her demo learner, Dave's address, the exam recorder's three follow-ups, the PR #1 merge call, and the Admin unlock OK (0082; its design review passed and its checks are running). Nothing paid. The live portal is still release 108.Unknown

Thursday, October 22 — 30 changes

11:53 PM11:54 PM ET: Two new MK11 lessons are LIVE on the MK11 portal (her item 179, an exception to the 2.0 hold for this write only): The Four Impulse Factors and BOLT: Four Kinds of Customer, her exact approved text, with MK11's Systems Recap and her training notes as text-only sources. Read back independently: both present, citations all resolve, and they open in her learner view. The exam readiness review is done (14 of 28 questions covered by the old lessons, pass 23; 11 questions for MK11 in a new Google Doc, not shared). The games and practice audit is sent back with six proposals, none applied, waiting on her. Library round 2 and the exam recording upload fix remain release candidates, not deployed, waiting on her deploy word. Admin unlock still waits on her OK. Nothing paid. The live portal is still release 108.Unknown
11:45 PM11:45 PM ET: Alyssa approved all sixteen Library round 2 changes as pictured (item 178). The phone-width walk passed, the automatic accessibility scan found 0 problems in 56 runs, and the accessibility officer gave a release pass with both conditions met. The branch was merged under the evening's reader and statement fixes, with checks green on the merged result. It was handed to the release session as a release candidate. NOT deployed: the database change (0081) goes first, then her separate deploy word. Now waiting on her: the deploy word when the release session asks, and a small pick on an older sideways-scroll flaw at 150% text size; the earlier items are unchanged (her demo learner, Dave's address, the exam recorder's three follow-ups, the merge call on PR #1, and the Admin unlock OK and pictures). The 'approve the 16 changes and phone-width capture' ask is closed. Nothing paid was spent. The live portal is still release 108.Unknown
10:51 PM10:52 PM ET: Library round 2 is built and reviewed, and NOT merged or live (Alyssa's items 163 and 165). An upload now lands as a draft that only Admins can open, with Publish and Discard; Admins can make labelled sections (a Sections panel, section pickers, grouped lists). The database change (0081) was placed on her OK; checks on a copy of the database are green; the design critic passed all 16 items on the third pass; the accessibility officer passed it for showing her and lists release conditions (the phone-width walk, a fuller scan, merge order). Pictures went to her at 10:50 PM ET. Waits on Alyssa: approve the 16 visible changes and the critic's seven taste notes, and say whether the phone-width capture runs (it stopped at two strikes). Also waiting on her, unchanged: Admin unlock (built, needs her OK for 0082 and a look at the pictures), her demo learner, Dave's Super Admin address, the exam recorder's three follow-ups, and the merge call on PR #1. The staff What's New board is DONE (ten posts at 9:07 PM ET, item 170) and is no longer pending. Nothing paid was spent. The live portal is still release 108.Unknown
10:24 PM10:25 PM ET: Admin unlock (Alyssa's item 166) is finished for now and NOT live. Built and green on a copy of the database: every module, quiz and exam opens for an Admin's own learner view, and an Admin can unlock or relock one module for one learner from the learner's page. Waits on Alyssa: her written OK for the database change (0082) and its isolation test, and her look at the pictures. The design critic sent the control back twice on small accessibility points, so per her loop rule it stopped there for her word. The MK11 hub session merges it after Library round 2 once she says yes. Spend: none (no paid calls; the build used no model). The live portal is still release 108.Unknown
9:38 PM9:38 PM ET: the list of what remains before the Apple Mac rebuild ("2.0") counts as finished is written: 34 requirements proven done, 42 still open. Alyssa has not yet said what "finished" means: (A) the Apple version passes every proof and goes to her to test, with a later switch-over day making it official, or (B) the switch-over is done too. The switch-over day is not scheduled and stays paused. Release 109 stays on hold and the live portal is still release 108. Correction to the 9:03 PM note: two safety-guard changes for this Mac's command guard (approved by Alyssa this morning) are on the working branch and GitHub; none of the release 109 fixes is.Unknown
9:03 PM9:03 PM ET: release 109 is ON HOLD, on Alyssa's word. Alyssa's own words, about 8:58 PM ET (her item 169): "Hold the v109 deploy. Nothing goes live until the 2.0 build is completely finished. Keep the release branch as it is." This replaces her 8:52 PM go for tonight (item 167). Release 109 is the short-screen side menu and banner fixes, the Library reader fixes and the approved accessibility statement wording, built as part of the Apple Mac rebuild ("2.0"). It did not go live tonight. The live portal is still release 108 and nothing on it changed: no restore point, no look at the live database and no deploy ran (the release was checked on this Mac only and stopped after the secret scan). None of the release 109 fixes is merged into the main working branch or on GitHub. (Two safety-guard changes for this Mac's command guard, which Alyssa approved this morning, are on the working branch and GitHub; they change the Mac's protections, not the portal.) Next for this plate: a checked list of what remains before 2.0 counts as finished, going to Alyssa. Any later deploy needs a new written go from her.Unknown
8:33 PM8:33 PM ET: the 13 dead Library rows are removed, and all 14 documents were proven present first (Alyssa's item 164). This evening, each live document's stored PDF matched the original file and its page pictures, page text and original were in storage (14 of 14). The 13 retired rows from the failed morning try were then set to deleted by their exact ids, one audit entry each. Read back on production: 14 active, 13 deleted, 0 retired; the staff list and the learner rule both show 14. That decision is off the waiting list. New and waiting on Alyssa, not started: two Library requests from this evening (item 163), an Admin preview of the whole document before publishing and Admin-made labelled sections, to be built as one round with one database change once she picks. Everything else is unchanged: her demo learner, Dave's Super Admin address, the What's New picks (nothing posted), the exam recorder's three follow-ups and the signed-in walk. Release 108 is live; PR #1 is green and the merge call is hers.Unknown
8:08 PM8:08 PM ET: the MK11 Library is loaded. 14 documents are live on production, all Admins only: the 13 Leadership Documents plus a new one Alyssa asked for this evening, TM/STM Fronted Pay, labelled Onboarding documents. Read back on production: 14 active rows, 14 audit entries, and the temporary loader login dropped. The first try failed at 7:57 AM at the storage step; the same storage keys passed a full test at 4:21 PM with no change made, so that question is closed and no new keys were needed. Riley widens each document from its Share panel. Waiting on Alyssa: removing the 13 retired rows the failed morning try left behind (by exact id, on her word), her demo learner, Dave's Super Admin address, her What's New picks (nothing posted), the exam recorder's three follow-ups, and a signed-in walk of the S3 desk. Release 108 (group announcements, database change 0080) is deployed and live since 3:01 PM. Also resolved: the laptop's file-storage key question (the keys on this Mac passed a full read and write test at 4:21 PM and loaded the 14 documents at 8:08 PM; nothing further to do) and the question of pushing the repository (pushed this morning; PR #1's checks are green, and the merge call is Alyssa's, asked by the CI session).Unknown
4:19 PMPHP security check (php2shell): nothing of ours runs PHP, so there is nothing to patch. Two critical PHP flaws were disclosed on Oct 21. We checked every site, server, container, project and repository we could reach from here, and none uses PHP. The only PHP site nearby belongs to a client and is on their own host, not ours; a heads-up note is drafted for Alyssa to send, and nothing has been sent. A few older hosting accounts and one old laptop could not be reached from this Mac and wait on Alyssa's own check, so the sweep is finished for everything we could see and still open for those. Status: done. Cost: nothing on the paid model service (all of the work was Claude Code session usage, no paid pipeline run). The full written report is in the project's reviews folder.Unknown
3:03 PM3:01 PM ET: release 108 is live for MK11: group announcements. A Mentor posts to their own group or everyone; a Manager or Admin to everyone or one Mentor's group; learners see everyone-posts plus their own group's. Alyssa said yes (items 156 and 157); a restore point was made first, database change 0080 was applied to production (history 79 to 80), the production steward passed all six checks, and health answers 200 on MK11, Series 65 and the app's own address. Item 151: the Messages half needed no build, since Admins already see every thread. Still waiting on Alyssa: her signed-in walk (not possible from this Mac), her What's New picks (the board is empty; nothing posted), Dave's Super Admin sign-in address, and whether to add an 'Answered by' column to Messages (a proposal only).Unknown
2:38 PMAbout 2:40 PM ET: boards corrected for the afternoon. Release 107 has been live since 2:35 PM (release 106 plus the S3 exam-proctor change); Riley Gibbs is MK11's one top Admin (done this morning, the board still said none was named); the thirteen Library documents are waiting on her word about storage keys and removing 13 retired rows; her demo learner, Dave's Super Admin address, the What's New picks and the exam recorder's follow-ups wait on her. Still in progress elsewhere: the pull request's checks, and the next release line with group announcements (merge approved, not deployed). One fix to the board's own history: it said the repository had never been pushed; the pull-request branch was pushed this morning.Unknown
2:35 PM2:35 PM ET: release 107 is live for MK11. It is release 106 plus the S3 exam-proctor change: a Super Admin's test exam sitting is now recorded and flagged like a learner's, with playback on the S3 desk only (never a client Admin). Alyssa said yes at 2:28 PM; a restore point was made first; health answers 200 on all three addresses and the new S3 testing page and Library send a signed-out visitor to sign-in. Practice-copy checks were green apart from the eight known failures. Signed-in walk still waits on her.Unknown
11:25 AMAbout 11:25 AM ET, on Alyssa's word: Riley Gibbs was named MK11's one top Admin on production and read back (one top Admin, one audit entry). No MK11 Admin can step Riley down. Dave's test announcement was also removed by its exact id (the first try rolled back and wrote nothing; the second removed it and logged it).Unknown
7:43 AMAbout 7:43 AM ET: the boards now show the morning release as live. Release 106 has been serving MK11 since 7:38 AM, health answers 200 on both portals, and the new pages guard themselves for signed-out visitors. Not done and waiting on Alyssa: her own signed-in walk, a decision on how to load the thirteen Library documents (a production login is needed that cannot bypass the safety rules; nothing was written), Riley's email for the top Admin, the 'Show link' choice, and her What's New picks. Nothing was paid and nothing was posted to the staff What's New board.Unknown
7:42 AMAbout 7:42 AM ET: two steps of the morning runbook stopped. The signed-in walk was not done (this Mac's Chrome has no saved MK11 login, and the walker will not type a password); the document load was blocked before writing anything, because the loader needs a production login that cannot bypass the safety rules. Three options are written for Alyssa; the recommended one is a short-lived dedicated loader login, on her OK.Account not recorded
7:38 AM7:38 AM ET: release 106 is live for MK11 (the Library, the Structure view with What's New, and Admin seats). Fly's remote builder ran out of memory and killed the build first, so release 105 kept serving the whole time; the same image was built on this Mac and deployed from here. The health check passes and the signed-out pages answer 200 or send the visitor to sign-in. A restore point was made before the deploy so release 105 can be put back.Account not recorded
7:31 AMAbout 7:31 AM ET, on Alyssa's written OK (7:05 AM): the three database changes were applied to the live database (76 entries became 79) after the production steward's GO and a restore point. One first try was refused and rolled back with nothing written, because the hosted database now opens each session with no default schema; the apply script now sets one for itself. Read back: the seven new tables have their safety rules on and forced, the app can add, read and change but never delete in them, and no top Admin exists yet.Account not recorded
7:25 AMAbout 7:25 AM ET: the secret scan found two of the night's capture files with one-time sign-in links from the Mac's scratch copy (not production). They are redacted, the capture helper now masks recorded text, and the older findings are recorded with the reason. The repository was never pushed, so nothing left this machine.Account not recorded
3:57 AMAbout 3:57 AM ET: the user manual (Edition 2) is at edition 6e, built clean (145 pages in the full book), and the three reviewers are confirming it now. The round-6c reviews had sent it back for a build bug (a retaken picture could print its old pixels; two did) plus a short list. Round 6d fixed the builder, so a picture is remade when its screenshot is newer and the build refuses a stale one, and fixed the list. The round-6d checks found every text fix held and all 199 printed pictures true to their screenshots, with small leftovers (one old Mentor-menu picture, some pictures to refit, one false sentence about the report's cards, and the Learners count, which is 7 active plus 6 deactivated people); all are applied in 6e. The sign-in lesson stays as corrected (a rename, not a lock). The Library release is unchanged since 3:22 AM: gates green, accessibility officer passed, design critic shipped. Nothing is deployed (production is still release 105), nothing was paid tonight, and nothing was posted to the staff What's New board.Unknown
3:22 AMAbout 3:22 AM ET: the Library is finished and checked, still not deployed. The accessibility officer's sixth round passed (the accessibility statement is true for MK11 with the final build) and the design critic's fifth round shipped the reader on the live practice page in both themes. Three small fixes tonight made each page's text keep its lists and lines right: Word bullets no longer run together, nested points stay under their parent, and spreadsheet pages keep no indents with wrapped cells rejoined. Final checks all green on the last build (typecheck, lint, secret scan, 369 of 369 leak checks, only the eight known unit failures). The accessibility register was updated to match. The reviewers' failed sign-ins tonight (about half an hour) came from the Library test accounts being renamed on the practice copy for the manual's pictures without telling them, not from a login lock; final checks now also run on a cloned database as a precaution.Account not recorded
3:20 AMAbout 3:20 AM ET: the user manual (Edition 2) was rebuilt after all three round-6b reviews sent it back. Twenty-five pictures were retaken on the release copy, every must-fix was reworded, and two new build checks were added (a picture's crop must hold what it outlines; a number in a picture's description must be in the picture). It builds clean at 145 pages in the full book. The design, plain-English and accuracy re-reviews started at 3:20 AM and are running. Nothing was paid tonight and nothing is deployed.Account not recorded
2:10 AM2:10 AM ET, Oct 22: the night run so far. Nothing is deployed; production is still release 105 (checked against the host's release list). One morning release is assembled and tested on a practice copy of the database: the Library, the Structure view with the What's New board, and Admin seats. Reviews are in for Library security (GO), Library accessibility and design (fixes made, round-3 re-reviews still running) and Admin seats security (GO with fixes, made). The user manual is being rebuilt with the new sections. In the morning, nothing ships without Alyssa's written OK for the protected areas, her deploy yes, the name of MK11's top Admin (Riley), and her What's New picks. Nothing was paid tonight.Unknown
2:00 AMAbout 2:00 AM ET: the second round of Library re-reviews came back. Accessibility: pass with fixes, but the text under the page pictures still lost its structure for spreadsheets and nested lists, and a change to the top bar had moved the exam review page, which Alyssa's Oct 17 rule says stays untouched. Design: sent back, chiefly because the share panel slid under the top bar once the cards ended. Third-round fixes were made: page text now keeps Word's real indents and spreadsheet rows, the panel stays pinned past the last card, the exam page keeps its old position, the reader's links are bigger, and cards reflow at 150% text. A browser test that timed out once under load passed alone. The third reviews are running.Unknown
1:45 AMAbout 1:45 AM ET: Admin seats passed its security review with fixes, and they are made. The serious finding: a portal that already exists, such as MK11, has no top Admin yet, so any Admin could have stepped Riley down. That is now refused, and a new step to name the top Admin (a Super Admin can also pick one) is mandatory in the morning. A top Admin can no longer be re-invited or crowned except through the two approved routes. The Library design critic's first round (send back on six pages) and the accessibility officer's must-fixes were also fixed by this time.Unknown
1:19 AMAbout 1:15 AM ET: the morning release was assembled. The Library, the Structure view with What's New, and Admin seats are on one line on top of release 105, as three database changes (numbers 0077, 0078 and 0079). The practice copy for the checks is separate from any live data. The Library security re-review came back GO; two small follow-ups were fixed at once.Unknown
1:00 AMAbout 1:00 AM ET: Alyssa asked that, once the Library is finished, work start back on the rest of the MK11 to-do list and continue through the night until everything is built and ready for a morning deploy. Series 65 and the Mac rebuild stay paused, and nothing is deployed tonight.Unknown
12:56 AMAbout 12:56 AM ET: the Library fix round was built, not deployed. Security fixes (an upload that will not render no longer leaves a half-made record; the undo script restores the learner-erase step), accessibility fixes (the words of each page shown under its picture, the hint color, 44-pixel buttons), and Alyssa's Gold Shelf picks for the reader, the staff Library, upload, the document page and the reading report. The thirteen documents were re-converted on the new Mac mini and loaded on the practice copy. All 337 leak checks pass.Unknown
12:50 AMAbout 12:50 AM ET: Alyssa ruled that every Library document starts Admins only until Riley or another MK11 Admin opens it to other levels, and that Admins must be able to delete a document they no longer want. Both are built: new uploads and the thirteen documents default to Admins only, and delete removes the files from storage and the document from every list, while the record of what happened stays.Unknown
12:25 AMAbout 12:25 AM ET: Alyssa lifted the pause for the Library only: fix the review items and apply her Gold Shelf picks, with nothing deployed yet.Unknown
12:02 AMAbout 12:02 AM ET: the new Mac mini was proven like-for-like. The Library branch's checks give the same results here as on the laptop (330 of 330 leak checks, and the same known eight unit failures), and the browser tools were restored. No machine problem remains; clearing out the old Docker and installer copies waits for Alyssa to name it.Unknown

Wednesday, October 21 — 3 changes

4:06 PM4:30 PM hard stop, Oct 21. The Library (Leadership Documents) is built on its own branch and is NOT live; release 105 stays live. Tests pass (330 of 330 leak tests, the same 8 known unit failures as release 104). Security review: GO with two Medium fixes. Accessibility review: refuse until two fixes land. Design: concept A approved; B and C blocked after two send-backs for Alyssa. No fixes were made after the stop. Waiting on Alyssa: concept pick, who sees each document, protected-area OK, then deploy OK. Nothing was paid for today.Account not recorded
2:55 PMRelease 105 is live for MK11 as of 2:47 PM Oct 21, after Alyssa's yes. Admins can remove a learner and restore them from Recently Removed for 10 days. Admins, Managers and Mentors get a Manual page with a flip-through book preview. Admins can give coins to one learner. Super Admins get an internal testing mode that clients never see. Two database changes were applied to the live database first and read back. A restore point was made, so release 104 can be put back. The first deploy try timed out on the laptop's phone hotspot (live stayed up on release 104); the second went through, and both portals' health checks answered 200 afterward. Nothing was paid for today. Alyssa also lifted the pause on Riley's items: Structure and What's New and Admin seats are next, then the Library concepts and round 6 of the manual. Still waiting on Alyssa: her What's New picks.Account not recorded
11:12 AMNew session started at 11:12 AM Oct 21 on Alyssa's second account. A restore point was made, the live site is still release 104, and the MK11 and Series 65 health checks both answered OK at 11:13 AM. The laptop disk is at about 5.5 GB free.Unknown

Tuesday, October 20 — 6 changes

5:50 PMAlyssa paused everything at about 5:50 PM Oct 20. The four helpers stopped, their unfinished work was saved, and nothing was merged or deployed afterward. MK11 and the release line stay paused until she says otherwise.Unknown
12:43 AMThe team chart's data file was damaged when two updates overlapped, and that stopped the boards from publishing from 10:38 PM Oct 19 to about 12:45 AM Oct 20. The damaged tail was cut off with every agent's entry kept, the original was saved, and the boards are publishing again.Unknown
12:37 AMAlyssa approved the What's New posts for MK11 (invite emails, Reset password, no time limit, new quiz look, and Remove learners once it is live). Nothing is posted yet because the board itself is still being built.Unknown
12:26 AMThe staff Manual page is built on its own branch: Admins, Managers and Mentors will get a Manual entry that serves their portal's current PDF, never shown to learners. It waits for Alyssa's picture and the manual update.Unknown
12:19 AMAlyssa's test seat was switched off so it no longer shows anywhere. It was deactivated earlier and never deleted.Unknown
12:08 AMRelease 104 is live: internal staff access for S3 Elite AI. Alyssa approved the release first. A restore point was made, the database change was applied, the site was deployed and its health check passed. The one-time set-up commands for Alyssa and Dave are still to be run.Unknown

Monday, October 19 — 11 changes

11:57 PMRemove learner is built: an Admin removes a learner with a confirm step, the learner sits in Recently Removed for 10 days and can be restored, then is erased for good. Alyssa approved the pictures. It is being merged on top of release 104 and will need her release OK.Unknown
10:55 PMRelease 103 is live: the portal now sends invite emails itself, and each learner row has a Reset password button that sends the same email. The sending domain was verified first. Security and production checks passed. Riley's mail never arrived before because the portal had never been able to send any.Unknown
10:15 PMRiley's two learners were reactivated and emailed their invites. Alyssa's own test seat then confirmed the new email worked.Unknown
5:50 PMThe laptop's disk filled up again at about 5:40 PM and stopped the spell check release checks. Four finished working copies were removed with Alyssa's OK and Docker was restarted by her; the glossary trial that was running when Docker went down was interrupted.Unknown
5:28 PMAlyssa answered 'continue' on Super Admin and the glossary, and approved the spell check release. The spell check release candidate was put together and is not yet deployed.Unknown
11:10 AMSpell check passed its design and security re-checks and Alyssa approved it.Unknown
11:00 AMAlyssa ruled: focus on MK11 only. Series 65 work is paused until she says otherwise.Unknown
10:30 AMGlossary and plain summaries are built and checked. A paid trial on the laptop (about $0.25, cap $4, an estimate from our ledger, not the Console) is running; the summaries are held for Alyssa.Unknown
10:27 AMRelease 102 is live: the manager's no-time-limit setting, the Raised Cards quiz page, the single accessibility statement dated Oct 19 and the phone-wrap fix. The first push was cancelled by a network drop and the old version stayed up; the retry finished cleanly. MK11 and Series 65 home, sign-in and accessibility checks all pass.Unknown
10:20 AMThe MK11 Learner Guide (73 pages, unreviewed edition) was emailed to Dave from Alyssa's own email. The Manager, Mentor and Admin guides and the reviews are still to come.Unknown
5:00 AMOvernight builds finished but none went live on their own: Raised Cards, the single accessibility statement and the no-time-limit setting passed their checks; spell check, Super Admin and the glossary stopped at two strikes or waited on Alyssa, and the Help pictures were sent back twice.Unknown

Sunday, October 18 — 8 changes

11:20 PMAlyssa paused all work so her laptop could rest. Nothing was mid-write to the live site. Overnight the laptop disk was nearly full; idle work folders were moved to the external drive and free space recovered.Unknown
11:05 PMRelease 101 is live: the MK11 final exam, the four access levels (Admin, Manager, Mentor, Learner), bigger tap targets, six live fixes, voice typing and the smoother role-play voice. Security checks passed first and the home, sign-in and accessibility pages answered on both portals.Unknown
10:25 PMMK11 quizzes now need 80% to pass (15 quizzes live; the exam was already 80%). A learner under 80% is offered Try again.Unknown
8:43 PMWriting run on the laptop: 8 of 9 new MK11 lessons written and fact-checked, 1 refused, 2 quizzes in progress; about $2.02 by our ledger (an estimate). Not live. The MK11 question Doc gained a Part 8 for the missing sample bills.Unknown
8:43 PMTonight's release is assembled: bigger tap targets, six live fixes, voice typing and the smoother role-play voice. Typecheck, lint, security tests and secret scan are clean; 7 full-suite failures were diagnosed as already present before tonight's work.Unknown
8:43 PMFinal exam, access levels and the no-time-limit setting are in build. The exam mirrors MK11's own online exam question for question; Admins review and release every grade. The exam build needed two strikes on a real defect and was resumed on Alyssa's 'continue'.Unknown
8:43 PMMK11 quiz corrections are live: 13 questions rewritten and 8 retired, read back 21 of 21. The three select-all questions became one-answer questions as Alyssa chose.Unknown
8:43 PMMK11 lesson corrections are live: 21 lessons, 184 fixes, read back from the live portal 21 of 21 at about 9:00 PM. An independent accuracy check had passed every fix first.Unknown

Thursday, October 15 — 3 changes

11:16 PMRelease 99 is live. It carries the MK11 final-exam code (per-portal attempt cap, answer key held until a pass, exam take and result pages) and a corrected accessibility statement. All release checks passed, the security-test suite passed, and both MK11 and Series 65 answered correctly after deploy. The MK11 exam has no questions yet; it waits on MK11's file, and both portals stay at 2 attempts until it is loaded.Claude account 2
11:16 PMChecked against reality: the missions and the second voice-limit fix were already live, so nothing was applied. The Harassment term does not show an internal code to learners on the live lessons (it is stripped at render); a small extra safeguard is committed and will ship with the next release.Claude account 2
11:16 PMFive accessibility design sets went through the critic and all five were sent back. Voice and spelling and Help and tour are blocked for Alyssa after three send-backs; glossary is blocked after two; no time limit is back in a fix round after her 'continue'; targets and line length is in a fix round. A separate accessibility check also found the Series 65 statement was missing its audio-description line; that is fixed and live.Claude account 2

Wednesday, October 14 — 14 changes

3:09 PMRelease 93 is live: each portal's accessibility statement now lists every open problem the October 14 accessibility review found, instead of saying there are none. Seven lines are true on every portal, plus lines for flashcards, role play, games, slide downloads and uncaptioned videos, shown only where the portal has the feature. Each line comes off when its fix ships. A line about the games over-lists rather than under-lists on Series 65, whose arcade may not have all the same defects. Live read-back: MK11 and Series 65 both list their problems and neither says there are none. 24 unit tests fail in the clean release copy; they fail the same way without this change, so they predate it, and they need their own diagnosis.Claude account 1
3:08 PMThe accessibility review is finished. Three reviewers covered 42 pages in 15 configurations (630 page loads), keyboard and screen-reader behavior, and media, documents and legal. The verdict is not top tier yet. Five Level A or AA failures were confirmed by eye: text-field borders too faint (1.05 to 2.48 against 3 needed), the game banner at 1.58, generic page titles, arcade links told apart by color alone, and text clipping at narrow widths with large spacing. It also found a 15-minute session redirect, a timer that could submit 'Sign out', one-press graded submit and unannounced tutor answers. Alyssa's decisions: list the known problems now (done in release 93); draw the nine visible fixes as pictures first; build the non-visual fixes; add a written transcript under every video. The report is a Google Doc for Alyssa.Claude account 1
3:07 PMThe sign-out fix is stopped under the two-strike rule and waits on Alyssa. Part one is committed: signing out now ends the session even when another browser tab had renewed it a moment earlier (before, the browser could stay signed in after pressing Sign out). The second round's failed-network test failed twice, because the portal resets the form after the action and a learner's choices are lost after a failed submit.Claude account 1
11:41 AMRelease 92 is live: the accessibility statement is now true for each portal. The review found the shared 'no known problems' text false on Series 65: its 22 lesson videos have no captions or transcript, and its Messages link leads to a page that is switched off there. The statement now reads each portal's own features when the page is shown: it names the captions gap where a portal has lesson videos, offers Messages only where it is on, and lists the owner's email so a signed-out reader can report a problem. Checked live on both portals.Claude account 1
10:46 AMRelease 91 is live: everything Alyssa approved after reviewing the pictures this morning. Games: the four accessibility limitations are closed and the Grid Map tablet layout she approved is in. The saved-settings bug is fixed (a browser add-on could make the portal forget a learner's text size and theme). Learner pages S3, S4, S5 and P6 and the whole manager set are in the new look, with the skyline banner on every manager tab. Five accessibility features are live: a focus ring, more contrast, reading comfort, calm mode with a motion setting, and 'Read practice questions aloud'. Automatic reads use the device voice; the studio voice is used on Listen. Series 65 also got the arcade keyboard fixes. Checks after release: both sign-in pages, the health check and the logo route answered correctly, and the statement shows the new settings. Not shipped: the final-exam tools, the paused quiz timer change, and the five features for other portals.Claude account 1
9:47 AMAlyssa's decisions recorded overnight. Manager pages: keep Home and Learners as approved, keep the headlines, restyle the bodies, access box at the bottom, solid red confirm button, no seats eyebrow, the typo fix for all clients, and no Message button for a deactivated learner. Final exam: 80% to pass, no time limit, unlimited tries, the answer key held until a pass, opens after all lessons, questions shuffled.Claude account 1
9:47 AMOvernight queue finished for Alyssa's morning review (she asked at about 2:15 AM to get everything in the queue ready). Nothing went live. Two changes are committed and ready to ship on her OK: the games release (Grid Map tablet layout she approved, four accessibility fixes in all eight games, statement dated Oct 14) and a fix for a live bug where a browser add-on makes the portal forget a learner's saved text size and theme. Built and design-critic passed but waiting for her picture review: the manager pages with her A to H answers, learner pages S3, S4, S5 and P6, the manager top bar and My Journey at large text, and five accessibility features. The final-exam tools are built and wait for MK11's questions.Claude account 1
9:47 AMTwo role-play findings. The wait from when a learner stops talking to when the customer's voice starts is about 5 seconds against a 1.5 second target; two no-cost fixes could bring it near 2. Separately, the cost tracker does not count one kind of AI charge (cache writes), and one model's price on file may be out of date, so role-play cost is slightly under-counted. No harm done; both wait for Alyssa.Claude account 1
9:47 AMThe overnight run's incidents are all handled. A designer's working copy accidentally copied a 10 GB backup folder and filled the laptop's disk at about 1:12 AM, which stopped Docker and the local database. With Alyssa's approval, four leftover working copies and the trash were cleared (their unsaved changes were saved first). The laptop then restarted once, and the database and local site were brought back and checked with a real query. At about 9:42 AM the local site ran out of memory after eight hours of checks and was restarted. The live portal was never affected. About 20 GB of disk is free, so it needs watching.Claude account 1
1:15 AMConsole balance read: $46.67, down $3.37 from $50.04 (last read 2026-10-04). Auto-reload still off.Unknown
1:10 AMRelease 90 is live: every fix from the October 13 security review is now running for clients. In the portal itself, a sign-in link can no longer be bent to send a signed-in person to another website; a deactivated learner can no longer switch themselves back on with an old set-password link; the sign-in lockout no longer treats the whole internet as one person (before, thirty wrong passwords from anywhere could lock every learner and manager out for fifteen minutes); a logo file can no longer carry a script; and a branding upload is no longer accepted before the permission check. Three new factory-wide guards mean every new portal inherits this: a test that fails if anyone writes a second copy of the safe-destination check, a deploy gate that refuses to release while any leaked key is still live, and a check that refuses to connect a client address whose setup would put all its visitors in one lockout bucket. After the release, sign-in pages on the live addresses, the health check and the logo route answered correctly, and a deliberately broken login pass was turned away cleanly. The security work ran no paid model calls; the only cost was hosting releases.Claude account 1
12:44 AMThe last leaked credential is now dead. The master password for the database (the one that can see every client's data) was reset at the database host with Alyssa signed in herself. The new one went from the host's copy button straight into a locked file and never entered the working session; the old one is now refused, the new one was proven to connect, and the laptop was switched over with a backup first. The live portal was never affected, because it connects with a different, weaker login that was never in the leak. With this, every credential from the September 13 leak is replaced and dead: the two outside vendor keys (September 14), the three session and signing keys (October 13), the file-storage keys (October 14, old key revoked) and this database password (October 14). The release check that guards this now passes.Claude account 1
12:40 AMFile-storage keys replaced and the old ones revoked. Alyssa signed in to Cloudflare and a new key, limited to the portal's two storage areas, was created. Its values moved from Cloudflare's copy button straight onto the live portal without ever appearing in the working session (release 89, no downtime). Shortly after midnight the old, leaked key was deleted, and Cloudflare confirmed it. Logos and uploads load normally on the new key.Claude account 1
12:09 AMThe security review's fixes were written, attacked and tightened before release. Alyssa asked for an honest review of the live MK11 portal, then for the fixes, then for the same protections to be built into the factory so every new portal gets them. The review rated the portal 6 out of 10 overall (the application code itself 8 out of 10), held down by leaked keys that had never been replaced. Every fix was first shown failing in a test, then fixed. Twelve independent reviewers then tried to break the fixes and found several holes, including a release check that would have gone green with the database master password still live; all were closed before release. Two sign-in test suites that failed when run side by side were traced to two tests deleting each other's lockout counters, not to the new code, and now delete only their own.Claude account 1

Tuesday, October 13 — 14 changes

11:46 PMThree of the leaked values were replaced on the live portal (release 88): the one that signs users' login sessions, the one that scrambles visitor addresses before they are stored, and the signing key for learner access passes. Nobody was signed out, because the longer-lived passes are kept in the database. A script made the new values, proved the new signing key loads, signs and verifies before anything restarted, and printed none of them. The portal answered normally ten seconds after the restart.Claude account 1
10:46 PMThree quiz content fixes are live on MK11. Alyssa ruled that LOA means Letter of Authorization, that the question naming 'Abby Riley' should read 'Abby or Riley', and the company-values question was fixed. While checking, we found MK11's own documents disagree: the standards document opens with one list of values and the About page gives another. The question now says which list it draws from, so it no longer contradicts the values lesson. Six questions were corrected on the laptop copy; the three that exist on the live portal were corrected there too and read back equal. The three 'Abby or Riley' questions exist only on the laptop copy. Each change is recorded in the audit trail.Claude account 1
10:05 PMRelease 87 is live: the audit fixes Alyssa approved. Learners get a slimmer phone menu; there is one sign-in page and managers land in Manager view automatically; the sign-in photo is sharp on phones; the quiz review answer sheet keeps one width; My Journey no longer scrolls sideways at large text sizes; there is one Export CSV button and no empty band above the seat count; the browser tab shows the client's own mark; 'Listen to the key points' uses line icons instead of emoji; unfinished progress segments show in light mode; and the Field cheat sheet is hidden until MK11's real documents arrive. American spelling was also fixed across the portal's screens and in 90 stored lesson and quiz lines on the live portal, each recorded in the audit trail. All new styling applies only to the Gold look; Series 65 is unchanged. A security review of the one-sign-in change found one medium problem, fixed before release.Claude account 1
9:51 PMMK11's Gold portal now lives at its house-standard '.app' address, and the old '.com' address forwards to it, keeping the page the visitor asked for (including set-password links). Alyssa ruled earlier that every portal's address should end in .app; this is the first. The move needed a new secure certificate, a record at the domain registrar (checked with Alyssa first), a forwarding rule in Cloudflare, and the portal re-bound to the new address with an audit entry.Claude account 1
7:55 PMRelease 86 is live: the first brand-audit fixes. The course outline page was retired on the Gold portal (Alyssa: 'if it serves no real purpose'); manager and learner now see the same lesson, quiz and topic counts; the picked role-play difficulty is easier to see in light mode; the client's manager no longer sees Alyssa's own email, name or sign-in time; announcement fields are visible in dark mode; photo labels are readable; and one lesson's definition wording was fixed on both copies. The coaching and 'topics checked' wording changes apply on every portal. Signed-in live pages were not walked (no live password was used); the proof is the laptop walk at the same code.Claude account 1
7:25 PMBrand audit finished (progress step 9 stays In progress). A brand reviewer and a design reviewer went through 92 screenshots of 23 MK11 Gold pages in light and dark, desktop and phone. Verdict: very good, one MK11 product. They found 7 bugs and 26 suggestions. Under Alyssa's new rule that no design changes happen without her approval first, nothing was changed; the whole list waits on her.Claude account 1
7:03 PMRelease 85 is live: the Learner/Manager switch no longer disappears after about 15 minutes. Alyssa also set a new rule: no design change of any kind until she has approved or rejected the suggestion.Claude account 1
2:37 PMRelease 84 is live: a stray yellow line down the manager pages is gone. It appeared because a manager now holds a learner session too, and the hidden learner sidebar was drawing its edge.Claude account 1
1:45 PMRelease 83 is live. Managers can now switch between Learner and Manager view with one sign-in, from either login page. The Learner view takes no seat, is left out of reports, and ends when the manager is deactivated. Dave is linked on production. A security review FAILED the first version (one manager could have reached another's account) and PASSED the fix before anything went live. Role play on Hard now hides each customer's personality type.Claude account 1
12:51 PMRelease 82 is live: the new Cleveland sign (Diner Arch, an original open-licensed lettering), the Spotlight manager look, the Quiz Ledger code (switched off), and 24 role-play customers on the live Gold portal, which had been showing none.Claude account 1
12:18 PMFair answer order: MK11's quiz answers were almost all option A (about 94 in 100 on the live portal) because nothing in the factory ever shuffled them. They are now shuffled on the live and laptop copies with every answer proven unchanged, and the factory shuffles every new question.Claude account 1
11:49 AMMK11's own shareable web address is live for the Gold portal, with a valid certificate. Dave was set up as a full admin with a one-time sign-up link (no generated password) and emailed it from Alyssa's account. Alyssa's own owner-demo login was created the same way.Claude account 1
7:19 AMThe MK11 Gold portal went onto production (release 79) and was published after a sample-bill module problem was fixed. A new database table for learner portraits was added first, with its security rules, and the dark-mode check showed no existing portal's look changed.Claude account 1
4:00 AMAlyssa picked Spotlight as the new manager look and Diner Arch as the Cleveland sign. Learners can now buy up to 2 extra portrait draws a month for 25 coins each.Claude account 1

Monday, October 12 — 2 changes

7:30 PMMK11 Gold gained, for Alyssa to walk locally: learner profile with AI portraits (her first avatar worked), the Podium leaderboard under Where you stand, a Library of placeholder documents, a shop with eight color palettes, the Customer Wall role-play page, and a section quiz that opens right after a section's last lesson.Claude account 1
12:30 PMThe Ask tutor is live on the working MK11 portal (release 77) after fixes to how it finds answers: 15 of 15 test questions pass. MK11 role play now has a $70 monthly team cap, and low-allowance warnings plus an optional per-learner monthly allowance exist for learners and managers.Claude account 1

Sunday, October 11 — 1 change

6:00 PMSeries 65 Module 3 is complete: all 6 videos approved and live. Module 4 has lessons 3, 4 and 5 approved and live; lessons 1 and 2 were in fix rounds. A full backup was made and its restore drill passed after a missing security-test registration was added.Claude account 1

Saturday, October 10 — 2 changes

12:50 PMSpend and tools: Module 3 pictures cost about $3.43 on Oct 10 (about $18.17 of the $35 image budget so far, by our ledger). Narration moved to the Pro plan with 624,220 characters, resetting Nov 10; lessons 1 and 3 used about 13,100.Claude account 2
12:48 PMUpdated the boards to today. Module 3 videos are in production: all six storyboards signed, all pictures made, lessons 1 and 3 narrated and in final review, lessons 2, 4, 5 and 6 in design review. Nothing from Module 3 is live. Alyssa's standing rule: videos run continuously through Module 3 and then Module 4 and on, pausing only for credits; each finished video still waits for her approval.Claude account 2

Wednesday, October 7 — 1 change

5:30 PMModule 2 is complete: all four videos approved by Alyssa and live on the Series 65 portal. The last four Apply missions are live too: 17 are released, and every Module 1 and Module 2 lesson has one.Claude account 2

Tuesday, October 6 — 19 changes

11:20 PMStarted all four Module 2 videos (Measuring Return and Risk) at Alyssa’s request. Each lesson gets its own fresh analogy, Module 2’s own narrator voice, and the quality bar she set today. Four builders are drafting the scripts and slides; nothing has been spent yet.Claude account 2
11:04 PMChecked the live Series 65 site: both welcome videos and all seven Module 1 videos are live, nine in all. The two thought to be missing were already there. Module 2 videos have not been started.Claude account 2
3:21 PMModule 1 is complete: Alyssa approved lesson 6 and it is now live on the Series 65 portal. All seven Module 1 lessons have an approved video; lesson 1 may still need copying to the live site, which waits on her yes.Claude account 2
3:01 PMLesson 6 (Where Companies File, and How They Count), the last video in Module 1, passed every review and was sent to Alyssa. Once she approves it, Module 1 has a video for all seven lessons.Claude account 2
6:55 PMLesson 6's finished video passed its picture-matches-voice and teaching reviews and moved into final inspection, the last step before Alyssa sees it.Claude account 2
6:40 PMLesson 7 (Four Ratios That Size Up a Company) went live on the Series 65 portal after Alyssa approved it, replacing an old sample video that predated this month's quality rules.Claude account 2
6:35 PMToday's narration purchases: lesson 7 bought about 5,700 characters of voice; lesson 6 bought about 6,400 characters plus a small top-up after an ear check caught a date read the British way, fixed by respelling just that line rather than re-buying the whole lesson.Claude account 2
1:29 PMModule 1 lessons 2 and 5 are live on the Series 65 portal after Alyssa approved them. Lesson 7 (Four Ratios That Size Up a Company) was finished to the new standard and sent to her. Lesson 6 has its voice and is having its slides rebuilt. Her quality notes became video rule 42 and the new reference videos.Claude account 2
11:44 AMLessons 2 and 5 of Module 1 were remade the way Alyssa approved and sent to her: lesson 2 with its arrows, bridges and icons redrawn, lesson 5 with every slide laid out again on lesson 3’s structure. Both reused their existing voice, so nothing was spent, and both passed every review and the final inspection. They wait on her verdict.Claude account 2
3:11 AMAlyssa reviewed lessons 2 and 5: the layout of lesson 2 is good but its drawn arrows, bridges and small icons are crude; lesson 5 has the same icon problem and its slides are too busy and unorganized. The cause was traced (fixed-angle arrowheads, mis-sized bridge geometry, and for lesson 5 nine rounds of additions instead of a re-layout). A new standing rule says drawn graphics must match the artwork. Video work is on hold for the night at her request.Claude account 2
2:49 AMModule 1 videos resumed at Alyssa’s request (quizzes and missions stay paused). Lesson 2 was sent to her for approval. Lesson 5 was assembled from voice clips already paid for, so no new voice was bought, and is in its final reviews. Lessons 6 and 7 are being reviewed and fixed before their voice is bought.Claude account 2
2:50 AMThe board now names the Claude account behind every entry, shows which account is working, and counts each account’s work under “Who did what” on What happened.Claude account 2
2:45 AMFixed a board bug that silently erased new entries: two rebuild steps read the board, waited on the database, then wrote their old copy back over anything added meanwhile. They now change only their own numbers.Claude account 2
2:40 AMFound why this board stopped updating on October 4 at 7:40 pm: a new team member (the mission writer) was added without a box on the team chart, and the board refuses to publish an incomplete chart, so 1,373 publish attempts failed. The box is added.Claude account 2
2:21 AMFound a gap: the secret scan does not run by itself when a change is saved, though the rules say it should. Ran it by hand on today’s changes and found nothing. Adding it waits on Alyssa’s yes.Claude account 2
2:20 AMEvery sign-in checked and working (Fly, GitHub, Docker and the laptop database, the backup drive). Restore point made. The live site is healthy: version 76 running and its health check answering.Claude account 2
2:19 AMClaude account 2 started. The desktop app still reported the first account, so Alyssa confirmed the switch and this session was registered as account 2; its first saved change is stamped account 2.Claude account 2
1:31 AMEvery saved change now carries an ID and the name of the Claude account that made it. A dated restore point (the code, its history and the laptop database, copied to the backup drive) can be made at any time, and going back never erases anything.Claude account 1
1:05 AMHandoff to a second Claude account: every unfinished piece of work was saved, and the pick-up notes were rewritten so a fresh account can resume where this one stopped. Typing "start" now resumes the work.Claude account 1

Sunday, October 4 — 2 changes

7:55 PMAlyssa paused video making, quiz writing and mission writing until the weekly usage limit resets. Everything in progress on those three stopped where it was.Claude account 1
12:35 PMCorrection: the quiz 'hold' code reached the laptop portal this morning inside the Apply-missions merge, so unchecked quiz questions are already hidden there. The last four quiz tooling changes were merged just now. Video 6.1 'What Every Client Must Be Told' is live, which makes 20 videos. The 10 failing test files fail the same way on the code from before the merges, so the merges broke nothing.Claude account 1

Saturday, October 3 — 42 changes

8:35 PMThree videos stopped at two strikes tonight, each with its fix already written and $0 more spent: lesson 2.2 (a missing progress strip the last edit accidentally deleted, plus three small picture edits), lesson 0.2's full rebuild (one icon suggests the exam is taken online; two words read as unclear), and lesson 2.3 (one unsourced sentence, an answer shown before the question is asked, a term used before it's defined). All three wait on Alyssa's "continue."Claude account 1
8:17 PMThe missions trial (5 lessons, up to $3 approved) ran and failed: 4 of 5 drafts were refused by the fact-checker and 1 passed with fixes — unsupported claims, and two mission items contradicting each other. Cost about $1.18 at the bill ($1.02 on the ledger) of the $3 allowed. Nothing was released; the other 73 lessons were never started, waiting on Alyssa's ruling.Claude account 1
8:17 PMDeploy v73 went out: the owner-only module unlock (her account, once marked, sees every module open; every other learner is unaffected). Security re-check: pass, no new findings. It changes nothing for anyone yet — her own production learner account is still unmarked, and marking it resets that account's password and signs it out by design, so she runs that step herself.Claude account 1
8:05 PMDeploy v72 went out: the approved games (Treat Drop with her crab pick, Walk Map) plus the Apply-missions code. All gates green (typecheck, lint, secret scan, the 133-test leak suite); the live site answers and Series 65 sign-in works. On Alyssa's "for my account unlock everything," her production learner account was given 1,000,000 coins, verified on the live account.Claude account 1
7:55 PMSeries 65's live portal had "Leaderboard" and "Your team" switched off (the earlier removal had only reached the laptop copy). Checked and confirmed off on the real site.Claude account 1
7:35 PMLesson text fixes applied to the laptop copy only: lesson 2.2 lost a sentence about the exam format that no source backs, and the "Four Yields" lesson's ranking of bond returns was corrected to what its source actually says. Both changes are audited. Production text is unchanged, waiting on Alyssa.Claude account 1
7:30 PMThe old lesson 0.2 video (the pre-rulebook clip that had stayed live by mistake) was taken off the laptop portal on Alyssa's "take the old clip down now, rebuild next." Production still has the old clip; taking it down there needs a production access step only Alyssa can run.Claude account 1
5:17 PMAlyssa read the Anthropic Console again this evening: $50.04 left on the account, unchanged from this morning's reading.Claude account 1
5:10 PMSeries 65 go-live moved forward: Alyssa ran all 67 production migrations herself, the tenant-isolation security gate passes on production, and the domain record and certificate for the Series 65 address have been issued. A deploy to production is running now. Actually promoting Series 65 onto it is a separate, still-to-come step.Claude account 1
4:30 PMA quiz rewrite pass was released: 146 questions were rewritten and 130 released to the course. Lesson quizzes ready to use rose from 8 to 21 of 76 lessons; the Final Boss question pool rose from 73 to 117.Claude account 1
4:00 PMLesson 0.2 ("What Happens on Exam Day") had its body updated with the enrollment and exam-day facts it was missing, approved by Alyssa and applied. Its video (2.2) is now in progress: 2 new pictures made, $0.75 spent of a $1 cap.Claude account 1
3:30 PMTwo of Sadie's new games — Sniff Match and Light the Lighthouse (the beach-and-jetty option Alyssa picked) — are built and merged into the laptop portal.Claude account 1
2:30 PMVideo takedown and rebuild: on Alyssa's "take them all down now," 19 videos that had not been approved were removed from the portal. The lead then rebuilt lesson 2.1's video from scratch; after several rounds of review Alyssa approved the finished cut and it is now attached on the laptop portal. Her notes from this round became four new rules in the video rulebook and new automatic checks that catch the same mistakes next time.Claude account 1
12:10 PMFixed a broken team-chart data file — it had a duplicate, truncated tail after its closing brace, so no agent's status could be written and this catch-up itself failed on its first try — rewritten to the valid record it actually held; nothing was lost.Claude account 1
12:05 PMSadie's new "look A, a light touch" art — critic-approved on every pose and game, checked against 88 before/after pictures with nothing unwanted changed — is being merged into the laptop portal now, on Alyssa's "Yes, merge it." One stylesheet conflict was resolved in the merge; the test gates are being confirmed next.Claude account 1
11:35 AMMerging turned up one real bug, fixed on the spot: the new test for "can one client see another's learner settings" was using a generic test value that a brand-new database rule rejected before the real check ever ran, so it was failing for the wrong reason. Fixed to use a value each column actually allows; every other column still fails loudly if it leaks. The full leak-proofing suite now passes 133 of 133. The slower full test suite, and a side-by-side comparison against the code from before today's merges, are still running.Claude account 1
11:33 AMCreative Visual Mode's code (the illustrated-lesson on/off switch, the diagram engine, the new per-learner settings table) merged into the laptop portal, right after the missions code so both of today's database changes stay in one sequence. MK11 confirmed untouched: 0 markers, 0 changed dark-mode rules, 420 lesson pages byte-identical.Claude account 1
11:30 AMThe Apply missions code (the hands-on exercise after each exam section, its database table, its scoring) merged into the laptop portal. The 4 pilot missions are now part of the real build, not a side branch.Claude account 1
11:30 AMAlyssa read the Anthropic Console herself this morning: $50.04 left on the account. The money page now carries that reading instead of the stale one.Claude account 1
11:00 AMHer "Continue all of them" round on the videos finished: 2 of the 8 modules-3-4 videos that were waiting (lesson 3.3, and separately lesson 6.5) passed every final review and are now live. The other 7 (3.1, 3.2, 3.4, 4.2, 4.3, 4.4, 4.5) hit a NEW problem on re-review — mostly a word the voice says differently than the slide, or a reviewer catching something the first fix missed — and are stopped again under the two-strikes rule. Each needs a fresh look from Alyssa, not just another "continue."Claude account 1
10:00 AMA content draft for lesson 0.2 ("What Happens on Exam Day") is written and passes every free check: it adds the enrollment and exam-day facts the study guide has (Form U4 or self-scheduling, the 120-day window, closed book, pass/fail, section scores on a fail) that the lesson itself was missing, which is why no video could be made for it last night. It is waiting for Alyssa's review before anything is applied, alongside three small word-choice fixes the education reviewer wants in the part of the lesson already approved.Claude account 1
9:34 AMTwo more videos stopped overnight for Alyssa: 6.5 "Performance Fees and Soft Dollars" is fully built and only needs her look; 6.6 "Holding and Handling a Client's Money" needs one more review round. Also flagged for whoever owns these lessons' wording: 6.5 says missing a condition "becomes a violation" (stronger than the rule says), and 6.4 says "the adviser's firm" where the actual rule says "the adviser or its employees."Claude account 1
9:25 AMMorning brief sent: 17 new lecture videos are live (lesson 0.1; all 7 of Module 2; 4.1 and 4.6; all 5 of Module 5; 6.3 and 6.4), each one passed the fact-check, the visual match, the final inspection and the education review on its exact file. 13 more are written and checked, stopped at the two-strikes rule with a ready, gate-passing fix, waiting only on her "continue." Recommended first job of the day: merge the quiz branch, with every check run first — the laptop portal does not yet hide unchecked quiz questions (see below).Claude account 1
8:30 AMOvernight video run, one narrator voice per module (Sarah, Eric, Bella, Brian, Matilda, Chris): 17 lessons now have a real, fully-checked video. The most common reason a video stopped: a short on-screen label that dropped a limiting word the lesson itself uses ("non-issuer", "existing", "registered"). Lesson 0.2 could not be attempted at all — the lesson's own text doesn't carry the facts a video would need to say, see the content-draft entry above.Claude account 1
7:30 AMOvernight quiz run: 941 questions written, 622 passed the automatic checks, and a releaser fact-checked and education-checked all of those overnight. Result: 256 released live on the laptop portal (183 lesson-bank questions, 73 Final Boss pool questions). Only 8 of 77 lessons reached the 5 questions their quiz needs to turn on. 209 are held for a wording fix (almost all the same habit: a wrong-answer explanation brings in a finance fact the course doesn't teach — the correct answers themselves are nearly always right); 16 are refused outright. IMPORTANT, unresolved: the laptop portal's code doesn't have the "hold" feature yet — that's on a separate, still-unmerged branch — so right now all 941 questions, including the 209 that need fixing and the 16 that are flatly wrong, are visible in practice, module checks and the arcade. Nobody walked the portal overnight, but this is live exposure until that branch merges.Claude account 1
6:30 AMThe Apply-missions writer (for the other 73 lessons) was built and tested, and its test failed twice: once on an output cut-off (fixed by raising the cap), then on the fact-checker refusing one sample for repeating the lesson's own unsupported claim. Its full run is priced at about $23 at the bill, over Alyssa's $15 cap, so it was not run and nothing was written. Alyssa has since approved an improved version (v2) up to $15, once it's rebuilt and re-tested.Claude account 1
6:00 AMCreative Visual Mode pilot: 11 fact-checked diagrams built across modules 1 and 2, in a neutral placeholder style, with the on/off switch working end to end. Paused for Alyssa: which of 5 picture styles to use, where the switch sits (top row or side card), and its wording ("Diagrams" or "Creative visuals"). MK11 confirmed unchanged.Claude account 1
4:49 AMA fresh backup went to the G-DRIVE SSD (bundle 560 MB, database dump 8.5 MB, 2.1 GB of work files, all checksum-verified). Every structural restore check passed (RLS forced, 184 policies, 143 grants, 0 app-owned tables). The one check that failed twice was the leak-proofing suite in the restore drill — not an actual leak, but because the shared local database was carrying tables from two still-unmerged branches at once, and no single branch's setup registered all of them. Fixed by this morning's merge order (missions, then Creative Visual Mode).Claude account 1
4:45 AMAlyssa authorized the overnight run: narrate the remaining videos one voice per module; release quiz questions locally as they pass their checks; run the Apply-missions writer only if it prices at $15 or under; back up to the G-DRIVE SSD.Claude account 1
4:33 AMAlyssa approved the reference video: "You did it! YES!!! This is an example of what acceptable looks like." Every slide and video agent was retrained on this exact file before any further narration.Claude account 1
3:05 AMAlyssa watched the fully reviewed video 0.1 and said the slides did not match the narration. Measured: slides change on time, but each slide showed everything at once and held 17 to 33 seconds while the voice walked through it, so a viewer could not tell what was being talked about. Every reviewer had checked whether the spoken words were on the slide, never when. Her ruling: educational value first, fewer slides. A one-minute Part 1 sample is being rebuilt so each piece appears as it is spoken, with a learner test that must pass before she sees it. No new narration is being bought.Claude account 1
3:00 AMThe board now shows a heads-up line at the top whenever its written notes fall behind, names which parts, and flags any brief missing from the board. It was built after Alyssa found the board had gone stale while its timestamp kept updating.Claude account 1
2:49 AMThis catch-up: the board's recent-changes list had stopped moving yesterday evening even though the page kept re-publishing every few minutes, so it looked current when it wasn't. Everything below was missing and has now been added, checked against the actual files and commits, not against anyone's memory of the night.Claude account 1
2:44 AMSeries 65 quiz pilot: a real test batch for one module asked for 42 questions and held 33 for review (24 for the lesson quiz, 9 for a reserve pool). Cost about $0.75 on the factory's own books, about $0.87 on the real bill. The fact-check on those 33 just finished: 25 passed clean, 7 need small wording fixes, 1 is refused — every issue was a true fact that wasn't actually taught in that lesson, never an invented fact.Claude account 1
2:41 AMResearch on how bond yields rank against each other (premium, par and discount bonds) was pulled from FINRA, SEC and MSRB for the yields lesson and its Apply mission. A separate independent auditor rejected it twice: the first time it caught a sentence credited to FINRA that is not on FINRA's page; the second time, wrong statements in a new explanation of when the ranking flips. The math was right both times. Under the two-strikes rule it stopped for Alyssa, who chose to narrow it to the standard ranking with worked examples and audit it once more. Nothing has gone into the lesson.Claude account 1
2:39 AMLesson 0.1's video, review copy r6, is ready: slightly different running pictures each time (so the same image doesn't repeat for the learner), slide 6 rebalanced per Alyssa's note, a pronunciation fix for the exam's name, and every quality check passed — facts, visuals, slide design, final inspection, and the "can a new student follow this" check. 6 minutes 29 seconds, 18 slides. Sent to Alyssa for her look; NOT attached to the live lesson.Claude account 1
2:37 AMCreative Visual Mode (an optional illustrated version of a lesson): the on/off switch was built, left off until a client asks for it, and a safety check now proves a lesson's original words are never changed when the illustrated version is added. Next: five picture styles for the first lesson, for Alyssa to pick from — not drawn yet.Claude account 1
2:31 AMAlyssa ruled on the quiz test's one failed check: "Count it as a pass." The rule now counts a question set as failed only if it's still broken after the automatic repair step catches it — no quality standard was lowered, only how a successful repair is counted.Claude account 1
2:30 AMSadie's look: Alyssa picked option "A, a light touch" from three variants after rejecting a more realistic style and then calling the next version "too cute and a little girly." That look is now on every pose and every game, checked against 88 before-and-after pictures with zero unwanted changes. Not yet merged into the main build.Claude account 1
2:24 AMApply missions (the hands-on exercise after each exam section, modeled on the SIE study guide): Alyssa approved the design and ruled that a learner must finish a mission to clear a lesson. The build started — the database table, the scoring rules and the learner-facing screen are built; 3 of 4 pilot missions are written and have passed the fact-checker and content reviewer, the 4th is waiting on the bond-yield research above.Claude account 1
1:53 AMFive new Sadie-themed game designs were delivered for Alyssa's pick, each a dog-themed skin for an existing game: Biscuit Bake, Sniff Match, Walk Map, Fetch!, and Bone Yard. Design only — no code, no database, nothing spent.Claude account 1
12:05 AMAlyssa said no to the video swap from earlier tonight — verbatim, "I am upset that you swapped the new video in without my approval." The new video was removed and the original restored exactly as it was, verified by playing both the old and new web addresses through to the end. Standing rule going forward: never attach or replace a lesson's video without Alyssa's approval of that exact file first; new videos go to a review copy she can watch, never straight into the lesson.Claude account 1

Friday, October 2 — 36 changes

11:37 PMLesson 0.1's video was narrated and attached to the local test lesson, replacing the old clip, without first showing Alyssa the exact file. (Reversed overnight — see above.)Claude account 1
10:54 PMSeven narrator voices were sampled for the one-voice-per-module rule (Sarah, already approved, plus six candidates). Waiting on Alyssa's picks.Claude account 1
10:33 PMLesson 0.1 was renamed "Passing the Series 65" (the course stayed published the whole time) so its video's narration could use the lesson's real title.Claude account 1
7:35 PMThe step-by-step formula for lecture videos was written up (how many slides, how long each runs, what goes on screen versus in the narration) and the earlier video samples were audited against it. Draft, waiting on Alyssa's approval.Claude account 1
7:30 PMThe team chart grew: clicking any team member now opens their own page with work history and who they collaborate with; helper roles are now shown as regular employees rather than hidden; and a new "worked with" report shows who paired with whom.Claude account 1
7:15 PMA new quality role, the video inspector, was added as the final check before any video can be attached: it verifies facts and arithmetic, checks the voice matches the picture on every slide, and confirms the exact file being attached is the one that was reviewed.Claude account 1
7:00 PMMK11's light-mode sidebar: Alyssa picked option 3 of 3 — a white sidebar with the original MK11 logo set on a navy plate. Dark mode is untouched.Claude account 1
6:47 PMMK11 Gold Standard home page: the light-mode gold text was darkened so the phase numbers and lesson counts are easy enough to read (meets the standard accessibility contrast ratio). Dark mode untouched.Claude account 1
6:41 PMMK11's lesson page: the "Reading Room" design Alyssa picked from five options was built into the MK11 test clone, in both light and dark mode. The live MK11 site is untouched.Claude account 1
6:19 PMSadie's game art started its own round of improvements: a baseline set of 72 pictures of her in the three Series 65 games (desktop and phone, light and dark) was captured first, so every change can be measured against it. In progress.Claude account 1
6:05 PMVideo work resumed this afternoon after Alyssa's pause: a sample is being built first for her approval. Along the way, her review rules became built-in checks rather than just instructions: video length shown as minutes and seconds, a check that refuses any line drawn to empty space (every connector must land on a real picture), and the closed 'Sources' list under every player.Claude account 1
5:44 PMFive MK11 lesson-page designs were delivered for Alyssa's pick (Gold Header, Reading Room, Rule Book, Split Cover, Bold Chapters), built from a real lesson page so she could compare them side by side. She picked Reading Room; it is being built into the test clone now, in both light and dark. The live MK11 site is untouched.Claude account 1
5:20 PMA slide redesign for the Module 1 ratios lesson was approved as the new quality reference, in Alyssa's words: 'This is the quality I expect.' Nine checkable rules were written from it for every future slide to follow.Claude account 1
3:03 PMThe MK11 'Gold Standard' home page was built on the test clone and shown to Alyssa on a live link. Two known issues remain, both still to fix: a footer color that is hard to read in light mode, and a banner that pushes part of the page below the fold.Claude account 1
2:35 PMAlyssa picked 'Gold Standard' for the MK11 home page, one of five designs shown. A fresh practice copy of MK11 was made for it; the live MK11 site and the separate working-draft copy are both untouched.Claude account 1
2:00 PMAlyssa asked for a bolder MK11 look and five different styles to choose from, using her own logo and a Cleveland skyline photo. Five home page designs were delivered: Cleveland Night, Gold Standard, Executive Light, Command Center, and Momentum.Claude account 1
1:41 PMAlyssa's notes on the model bake-off, from a plain-language comparison document built for her: she liked one model's card design best, but not its square boxes (a tell that gives away AI-made work), and the other model's color tint. The games hub did not work well in either version; she asked for Sadie's game art to be improved.Claude account 1
1:17 PMNew safety rules went live: file deletes and risky stop commands are now blocked and redirected to a safe holding folder instead of happening right away, and the board's own update programs are protected from being stopped by accident. A step-by-step picture comparison of the two bake-off portals was also built for Alyssa in her Google Drive, since the changes were hard to see without one.Claude account 1
12:55 PMAlyssa paused all video work to react to the early samples: the clips felt like fragments. New rule going forward: one short video per lesson, made just for that lesson, each with its own opening, closing, and sources list, never cut from a bigger video. The two 'Before You Begin' clips will be remade this way.Claude account 1
12:41 PMDELIVERED: the Series 65 visual bake-off comparison. Two versions of Alyssa's portal (the home page, the lesson's Watch and Study steps, and the games hub) were built blind by two teams on two different AI models from one brief. 32 before-and-after pictures are on a private page, and both versions are running on this computer with a fresh test login. She is reviewing now; her pick comes first, then the reveal of which model built which.Claude account 1
12:30 PMThe test run of the new 'no em dashes' rule in the writing instructions did not pass. The new checker works (it caught lessons the model still wrote with the banned pattern), but it throws out quiz questions instead of sending them to be repaired, so question banks came out too small. Per the house rule, the instruction change stays parked until a repair step exists. Spent $7.68 of the $12 approved for this work (ledger).Claude account 1
12:19 PMAlyssa asked for every effort on the bake-off so she could review it. The MK11 brand-revamp directions were paused (the work is saved) and resume on the other AI model after her pick.Claude account 1
12:03 PMFINISHED: the em dash clean-up of Mike's Series 65 lessons, applied on this computer after Alyssa approved the sample. 77 lessons rewritten, punctuation only; every change is recorded in the audit trail and can be undone from a saved copy of the old text. Em dashes in the course went from 779 to 21. The 21 left sit inside quoted official titles (kept exactly as published, her ruling) or in sentences the fact guard refused to change, which kept their original wording. The main client's MK11 lessons were checked and are unchanged. Nothing is deployed; the local portal shows the new text now.Claude account 1
11:25 AMFINISHED: the autopilot-nights rulebook section and its helper scripts (a pre-flight checklist, a decision log and a night queue), with tests; both reviews passed. It sits on its own branch, ready to merge after the design work lands, once the full test suite is re-run on a quiet machine.Claude account 1
11:20 AMAlyssa asked for the Series 65 'Apply' step, modeled on the hands-on missions in the SIE demo: the step plus four pilot missions (one per exam section), a flat +25 XP once, Series 65 only. It starts after the bake-off and the MK11 directions land. Improvement ideas are put to her as questions.Claude account 1
11:12 AMFINISHED: the two Module 1 video facts are sourced and audited: the IRS lets a business combine accounting methods, and Social Security's yearly raise follows the CPI-W price index (confirmed on the live Social Security page in a real browser). Alyssa chose the short wording for slide 28 (consistency only); no dollar threshold and no 'by law' wording will be narrated.Claude account 1
10:58 AMNew rule from Alyssa: every video names its sources. Not on screen: a closed 'Sources' drop-down under each video player lists the client documents it came from, and a check refuses a video whose list is missing or does not match its fact-check record.Claude account 1
10:29 AMA copy of the Drive link's token had been printed into a local log during tonight's audit; the copy was removed. The owner's ruling: no new token needed, keep the existing one where it belongs.Claude account 1
10:16 AMResumed after the usage-limit stop (hit about 3:50am ET; Alyssa's go-ahead at 6:45am ET: "continue" and "build all of it"). Since then: the overnight read-only quality-and-production audit finished (114 verified findings, most important ones first); the MK11 brand-revamp practice clones exist and both design directions are being built; the Series 65 visual bake-off teams are building; the stray-AI-wording check is fixed and wired into the automatic checks; the autopilot rulebook section is written. Needs Alyssa: revoke and reconnect the Google Drive link for the backup tool (a token was printed into a local log, never shown on any board); say go on re-uploading the live MK11 logo and hero, which are currently missing on the live MK11 Full portal; and pick the bake-off version and the MK11 direction once the pictures are ready. Nothing from tonight is deployed to production.Claude account 1
10:12 AMThe overnight quality and cost audit finished: 114 findings confirmed after double-checking (43 others ruled out or unconfirmed), split into 10 high, 49 medium, 55 low priority. Top items: a security key accidentally printed into a log file (fixed), a two-strike safety check silently broken since mid-September, a production monitor that had read green for 16 days while actually broken, and the live MK11 site showing no logo or hero image (fix queued for after the design work).Claude account 1
2:52 AMTonight's two visual projects are now under way: the Series 65 look bake-off (two teams building the same pages in different visual styles, to be judged side by side without knowing which team made which) and the MK11 brand revamp (two practice-only clone copies of MK11's look, tried out on copies of the portal — the real MK11 site was checked and confirmed untouched, including its dark mode). The read-only quality audit, the wording cleanup work and the autopilot-nights rules continue alongside them. Nothing new needs Alyssa's decision right now — her morning picks will be waiting in the brief.Claude account 1
2:35 AMFINISHED: the Series 65 portal-walk fixes — her four tour items from walking the portal herself, the shop's "House colors" spelling fix, and the accessibility statement — passed their fourth and final design review tonight and were merged into tonight's working branch. The automatic checks run on the merge all came back clean: the type checker, the secrets scanner, and the tenant-isolation security test. The full automatic test suite was still finishing as of this update. Nothing has reached the production site — this merge lives on tonight's branch only.Claude account 1
2:26 AMAlyssa said go on the factory upgrade proposal and the night run (Queue 12) started: "finish everything, even into the morning." A fresh pre-edit backup finished and passed a real restore test, saved to the G-DRIVE SSD, before any code changes began. Alyssa added $10 to the Anthropic account and read the balance herself right after: $69.83. In progress, in order: the Series 65 portal-walk merge, the no-AI-tells wording cleanup, the autopilot-nights rules, a read-only quality and production audit, the Series 65 visual bake-off, the MK11 brand revamp clone (demo tenant only), and Creative Visual Mode.Claude account 1
1:36 AMA written proposal for upgrading how the factory works is finished, with Alyssa's own decisions recorded on each of six parts: removing stray AI writing habits (like em dashes) from anything a learner reads, letting safe overnight work continue on a short list of low-stakes items with a morning review, a read-only quality audit, a visual style bake-off that keeps today's layout and navigation and only changes the look, a new optional 'Creative Visual Mode' that turns a lesson into an illustrated page, and general housekeeping. Nothing is being built from it yet — by her own rule, all code changes stay paused until she has read it and given the go-ahead.Claude account 1
1:10 AMSeries 65's portal-tour fixes went through a second design review tonight. Every fix held up except one shop item's name. Alyssa ruled on it herself: "House colours" becomes "House colors" on Series 65, and MK11's stray "No pressure," becomes "No pressure, friend." Neither change has been made yet — nothing is merged, and no code changes are happening until she has read the factory upgrade proposal above.Claude account 1
1:05 AMA full backup of tonight's work and the database finished and passed a real restore test, saved to the G-DRIVE SSD.Claude account 1

Thursday, October 1 — 17 changes

11:40 PMSadie's games for Series 65 are now merged into the main line, after the design critic's remaining items were fixed: the game hub shows one Sadie, the losing screen reads "Sadie needs a nap — try again," the new teal button color is in place, and the main client's site was checked and found byte-for-byte unchanged. This is local only — nothing here is live on the real site yet. Still open before Series 65's arcade can ship: the energy-sales wording on five of the eight hub cards, one busy text field, how big Sadie's side-eye look should be on the small game pieces, Treat Drop still needing its quiz questions, and a gold text color that needs a contrast check. Separately: the automatic board updater's error log, which had grown very large after crash-looping since yesterday's restart, was cleared by Alyssa herself; the fix made to the updater's startup file so it stops crash-looping is still waiting on her word to keep it or undo it.Claude account 1
10:48 PMBOARD-KEEPER: the boards had not updated since about 9:48 AM yesterday — the automatic updater's process did not survive the computer's restart, and nobody had re-started it since. Found and restarted (along with its backup copy, which had a separate problem: it was trying to run from the wrong starting folder and had been failing silently in the background ever since, each failure writing another error line to a file that grew past 57 megabytes). Both are confirmed running again. Caught up everything below.Claude account 1
10:34 PMSadie's games came back from a final review with one thing still missing: no picture shows the new game-hub layout with Sadie's face on it the way it's supposed to (the camera tool scrolled past it by mistake). Everything else from this round of notes checked out — the happier poses, the frog face, the button color, and the main client's site unaffected. Not merged yet; being re-shot now.Claude account 1
6:23 PMSadie's games updated with Alyssa's own picks: the hub layout she chose, the new losing message ('Sadie needs a nap — try again'), and the new teal button color for Series 65 only (the main client's site is untouched, checked byte-for-byte). Sent on for a final review.Claude account 1
5:40 PMAlyssa lifted the games' two-strikes stop herself and gave the picks above directly, plus confirmed she'd rather keep the current game sounds than switch to a licensed sound pack.Claude account 1
5:10 PMTried to send the MK11 invites tonight: no real accounts were created. The one-time safety override needed to create them was blocked by the safety system on its own, not worked around. Separately, the email tool used to draft the invites turned out to be signed into a colleague's mailbox instead of Alyssa's, so the drafts are sitting there unsent rather than in her own inbox.Claude account 1
5:02 PMSeries 65's first module video is now loaded on this computer only, cut into its two lesson clips, and attached to the practice portal. Checked for real: skipping to the middle, start and end of a clip all served the correct piece of video, byte-for-byte, straight from storage — not a shortcut or a guess. Across the whole 14-module video plan: 1 finished, 1 nearly finished, 12 not started yet; of the 77 total lesson clips planned, 2 are done.Claude account 1
4:30 PMModule 2's video slides are finished (one small wording fix applied without touching the one fact-check had refused to approve). Narration — the step that actually costs money — is being held back for two facts the client's own expert still needs to confirm.Claude account 1
9:47 AMThe computer's hiccup this morning cost nothing real: everything had already been saved and committed beforehand. The database and the local server were brought back after a restart, and the practice portal's course (77 approved lessons) and client list (29) were both confirmed intact.Claude account 1
7:35 AMBefore this morning's restart: finished the next round of Sadie's games (a visual seam fixed, a game piece resized) and committed everything so nothing would be at risk going into the restart.Claude account 1
2:11 AMOvernight spend: about $1.56 on Google image art for a video's slides (under its cap); nothing spent generating lessons, tonight, on this computer or on the live site. The Anthropic account balance read $59.83 by hand at 2:37 AM UTC on October 1, right after Alyssa added $15.Claude account 1
2:09 AMA full backup of the work and the database finished overnight and was verified with a real test restore, which passed. No passwords or secret keys went into it.Claude account 1
2:07 AMMK11's shop item names (rewritten to match MK11's own colors) failed a second review — wrong color names and descriptions promising visuals MK11's shop doesn't draw — and stay held for Alyssa to see in person.Claude account 1
2:05 AMSadie's practice-arcade games stay blocked after a second failed review (a visual flaw on one game's losing screen); still waiting on Alyssa's pick of game-menu layout before the fix lands.Claude account 1
2:03 AMSeries 65 Module 2's video reached a finished 35-slide deck with new painted artwork; a fact-check came back clean and a second design review passed 33 of 35 slides. Narration (the paid step) is still on hold for a few wording fixes and two open questions for the client's own expert.Claude account 1
2:01 AMSix reviewed fixes merged overnight, each cleared by a security review first: more happy, fewer worried-looking poses for Sadie on Series 65; Sadie's artwork locked so only the Series 65 address can see it; a lesson video's fact-check now checks against its own lesson group instead of a neighboring one; a real video player for lesson clips that only works for the right learner on the right tenant; two fixed security tests; and MK11's flipped flashcard now shows the term above the definition again.Claude account 1
1:59 AMGoing live stays held for Alyssa's own word tomorrow: production is missing three recent database updates, the pending deploy is about 182 code changes ahead of what's live (all of which would reach MK11 the same day), Series 65's web address still needs to move to the right place, and the owner's own demo login isn't set up on this computer yet.Claude account 1

Wednesday, September 30 — 1 change

2:00 AMConsole balance re-read: $45.43 (owner hand reading), down $9.10 from $54.53 read earlier (2026-09-29T14:06:51Z). Our own ledger ESTIMATE for the same period is about $11.41 at list / $13.29 at the bill (tutor ~$7.20, author ~$2.70, a stronger tutor model test ~$0.97, grounding ~$0.54) — higher than the $9.10 the Console shows having moved, most likely because the Console has not posted all of it yet; the two figures are not expected to reconcile exactly. Also missing from that estimate: a cancelled batch that finished 11 of its 13 items before it was stopped, which Anthropic bills at batch price (roughly $1–$2) but whose rows in our ledger still show as 'running' at $0.Claude account 1

Tuesday, September 29 — 2 changes

10:06 AMConsole balance re-read: $54.53 (owner hand reading), down $2.91 from $57.44 read earlier (08:00:54Z). Our own ledger ESTIMATE for tonight's work since 05:13:46Z is about $4.87 at the bill (Wave 1 author ~$3.39, grounding ~$0.23, assess eval ~$1.09, an unrecorded eval batch ~$0.16) — the Console usually lags, so part of that may not have posted yet; the two figures are not expected to reconcile exactly. Remaining of the $25 overnight envelope: about $20.13 by the ledger.Claude account 1
1:15 AMConsole balance re-read: $57.44 (owner-reported), down from $58.17 read earlier tonight. Today's two paid Series 65 re-write runs cost about $1.17 at the bill against the owner's $3 cap (Run A ~$0.97, Run B ~$0.20) — that's more than the $0.73 the Console shows having moved since $58.17; shown as two separate figures, not reconciled, since one is the Console and the other is our ledger estimate. No ElevenLabs spend today. A new $25 cap was approved tonight for the Series 65 base portal (remaining lessons + quiz banks); nothing spent against it yet.Claude account 1

Monday, September 28 — 2 changes

8:19 PMCONSOLE READ ATTEMPTED but failed: Chrome extension not connected — cannot access the Console to read the current balance. The previous reading ($62.18) remains unchanged, and its age should be noted before approving any paid run.Claude account 1
5:05 AMConsole balance: $62.18 (was $62.55, $0.37). Auto-reload OFF.Claude account 1

Sunday, September 27 — 1 change

6:24 PMBOARD-KEEPER: fixed three wrong statements found by the accuracy cross-check. The CFO desk note and two money notes were quoting a Sept 15 balance ($43.24) or mis-converting the Sept 21 reading's UTC timestamp into ET (typed as 5:01 AM, really 1:01 AM) — both now say the one true reading, $62.55 from Sept 21, 1:01 AM ET, and its real age. The 'spent this week' tile was showing $0.00 because the Anthropic bill's day-by-day data stops at Sept 12 and the 7-day window found no days to sum — that read as a confirmed zero when it was really no reading at all; it now says so plainly and shows our own ledger figure instead, labelled as the ledger. The demo date (Sept 24) has passed; the board no longer shows a negative day count, it says the date passed and it's waiting on Alyssa.Claude account 1

Friday, September 25 — 1 change

2:15 PMBOARD-KEEPER INCIDENT: this board's play-by-play had nothing in it for the whole back half of Monday (Sept 24) — the practice-coach feature going live on production, the new step-by-step lesson quiz page, two practice-arcade games getting their own daylight colors and going live, and six rounds of review on the neighborhood map — even though the work session closed cleanly at 3:09 PM ET that day. The close correctly turned the session light off, but nobody wrote up what had happened in the same pass, and no one checked the board again until now, about 23 hours later. Filled in below.Claude account 1

Thursday, September 24 — 7 changes

4:45 PMBOARD-KEEPER INCIDENT (from yesterday): the automatic updater that republishes this board every few minutes did not restart itself after the overnight laptop restart, so the live board sat still from about 10:33 AM to 12:45 PM ET while real work continued underneath it. Caught and restarted by hand at 12:45 PM.Claude account 1
1:05 PMThe neighborhood map's newest round (its 16th) is now the version shown live, replacing round 14: one building per lot instead of stacked shapes, a previously-barred lot opened up, lit single-column building fronts, and small yards added around the cottages. A further round of small fixes (a few houses that still read as too wide) is planned next.Claude account 1
1:05 PMTwo of the practice-arcade games now have their own daylight colors live on the real site — the Cleveland skyline game and the driving game — chosen by Alyssa herself, color by color, using a picker tool built so she could see each choice on the real game instead of guessing from a swatch. Six more games are still night-only and waiting their turn.Claude account 1
10:31 AMBuilt and reviewed the first six pieces of a new, more guided lesson page — a progress header, a reading step, and a five-question quiz per lesson — kept OFF for every real client while it's being proven out. A follow-up polish round on the quiz (hiding the old 'mark complete' button once the new page is on, and not letting a learner win by always picking the first answer) hit two separate technical dead ends in a row and stopped there per the two-strikes rule; nothing broken, nothing lost, waiting on the word to try again.Claude account 1
4:14 AMThe laptop's disk filled up again partway through the night and paused three pieces of work (the new lesson quiz page, part of the map redesign, and a design review) until it was restarted with more free space; all three resumed within the hour with nothing lost.Claude account 1
1:15 AMTurned the practice-coach feature on for MK11 specifically, after Alyssa played four real rounds of the practice arcade herself on the live production site to prove the scoring records correctly.Claude account 1
12:52 AMThe practice-coach feature — spaced flashcards, a ten-question practice quiz, and the practice arcade all reporting into one 'where a learner is weak' picture — went live on the real production site.Claude account 1

Monday, September 21 — 7 changes

9:53 AMBOARD-KEEPER INCIDENT: the live board pages have been stuck since about 12:58 AM last night, even though the board's own numbers looked like they were still being updated. The mechanism: a new teammate (the education critic) joined the team chart with no seat drawn for it in the chart's own layout, so every automatic rebuild since 1:04 AM has stopped partway through and never reached the step that publishes to the public pages — roughly 90 attempts in a row, all failed the same way. This is a fix to the board's own display code, which is outside what this pass is allowed to touch; it's reported here rather than patched. Everything in this file itself is now checked and correct as of this morning's read; it will reach the live pages the moment someone adds that one seat.Claude account 1
8:31 AMThe first section video with a finished voice track is done: Phase 3 of the 60-lesson MK11 course, about 3.5 minutes, narrated. It's the only one of six that both quality checks (facts against the client's documents, and teaching quality) approved outright — the other five are back with specific, named fixes before a character of narration gets bought for them, since narration spends a monthly allowance that doesn't come back until mid-October.Claude account 1
8:30 AMOwner ruling: ship tonight's arcade work with two known-red visual checks. Both checks were built around an older car shape; the shape the owner actually approved measures outside their rule on purpose, not by accident, and neither the designer nor the build loop widened the check to force a pass — that call was left to her, and she made it. A separate, real color-contrast gap the same check run turned up (a new dot color nothing was measuring) was fixed first and shipped clean.Claude account 1
4:30 AMProduction database change: the table that lets a human operator vouch for a promoted course now exists live, locked down the same way every client table is (row-level security, append-only). The very first real attempt to publish the bigger 60-lesson MK11 course through it was correctly refused, for two honest reasons: one lesson with no content is now formally excluded with a reason on record, and two whole modules (the sample-bill walkthrough and the commission tables) turned out to contain nothing a learner could actually open — a content gap, not a system fault.Claude account 1
3:00 AMA new automatic check now catches when the course tests a word it never explains to the learner. Tuned against the real course until it stopped crying wolf (started at 5 false alarms out of 12, ended at zero), it found seven real terms MK11's new hires are expected to know but are never taught — some of them on the trainer's own checklist.Claude account 1
2:30 AMAll six MK11 section recap videos were fact-checked against the client's own documents, line by line — about 160 separate claims traced. Five passed clean; one round of six specific issues came back, including a schedule detail with two source documents that genuinely disagree and a claim that invented a reason for something that had a different, real explanation. None of it reached a customer, because nothing had been narrated yet.Claude account 1
12:00 AMAll six MK11 section recap videos were also reviewed for teaching quality by a brand-new checker built and proven overnight (tested against five real lessons first, including one a person had already judged good, before it was trusted with anything). Every one of the six came back with a specific, fixable note; none were vague. The same review confirmed the bigger 60-lesson course covers roughly three times the ground the current live course does.Claude account 1

Sunday, September 20 — 3 changes

10:30 PMYesterday's portal problem turned out to be one missing thing, not three: the live MK11 client portal had zero source documents behind it in production, so the built-in Q&A returned nothing (even for the company's own name), the Resources page was blank, and every citation in every lesson dangled. Fixed by copying the client's real documents into production (4 documents, 601 blocks) and confirmed in a real signed-in browser session, not just a database count: Ask now answers, Resources now shows real content, and lesson citations now show real quoted source text.Claude account 1
4:13 PMConsole balance: $62.55 (unchanged). Auto-reload OFF.Claude account 1
12:13 PMConsole balance: $62.55 (was $62.59, −$0.04). Auto-reload OFF.Claude account 1

Saturday, September 19 — 17 changes

6:12 PMAlyssa asked tonight to have the COO shown on the team chart as "Lisa 2.0" going forward. Noted here so it isn't lost — putting the new name on screen means changing one line of the board's own display code (a label, not the underlying role), which this pass does not touch since it only edits content and data. Queued for the next session that can edit the board's code.Claude account 1
6:11 PMBoard-keeper pass, requested directly ("update the dashboard"): the play-by-play had nothing in it for the whole of today's session (about 9.5 hours) even though a full day of real work happened — added the entries below to close that gap. Also found: BUILD_LOG has not been updated since this morning either, so everything from midday onward exists only in working files and review notes, not in the project's own written history — that is a gap for whoever writes it next, not invented here.Claude account 1
6:10 PMINCIDENT, corrected: the 'hours today' figure (4.3 hours) only counts committed work, and almost none of today's afternoon and evening design work has been committed yet — so a long, real working day reads on that tile as a short one. The session itself has been running continuously since about 8:25 AM, and the first commit of the day was just after midnight. The board's weekly and total hour figures already use a better clock (the actual working-session length, not just commits) and are not affected; the daily calendar square for today is the one place still blind to uncommitted work. Flagging rather than faking a number: roughly thirty files are modified and several new files added, none of it committed, which is itself worth knowing about, not just the hours math.Claude account 1
6:09 PMTwo new practice-arcade games were designed today but not yet built: a driving-themed one and a troubleshooting one. Nothing to play yet; this is planning work only.Claude account 1
6:08 PMStill open in the practice arcade: one game's focus still doesn't jump in automatically when it opens (five of six now do), one game's touch targets are too small on a phone, and the pause screen isn't announced properly to a screen reader. None of these block play.Claude account 1
6:07 PMA round of real accessibility defects in the practice arcade were found and fixed today, each one measured before and after: every game boundary was too low-contrast to see clearly (now fixed); one game could trap a keyboard-only learner inside it with no way out (fixed — Escape now releases it); one game's fireworks animated forever even after pressing Pause (fixed to stop for good); every game's end-of-round "review what I missed" link led nowhere (fixed); the difficulty setting was nearly invisible on a phone screen (fixed); choosing a difficulty inside a game used to silently get forgotten by the hub (fixed); and the marketing banner from the hub used to stay on screen inside every game, covering play (fixed).Claude account 1
6:06 PMAdded a plain-language accessibility statement, linked from the bottom of every portal page, that honestly says the portal partially meets the standard accessibility bar and names the specific gaps still open, instead of overclaiming.Claude account 1
6:05 PMAdded sound effects to the practice arcade — six short tones built entirely in the browser, no sound files. Off by default, with a plain on/off switch on the arcade's home screen.Claude account 1
6:04 PMThe electricity-matching game's spark effect was tuned by hand today and now has a proper payoff: a lightning strike, three separate spark bursts on their own timing, and the matched cards flash gold.Claude account 1
6:03 PMCell Drop's falling pieces reached their fourth and likely final look today — solid, readable shapes with shading instead of outlines. The meter piece went through its own final round this evening (a thicker, more visible hand and a distracting glare removed) and passed every automatic check.Claude account 1
6:02 PMThe car-matching game's fleet is fully settled after several rounds today: the two cars still in question are done (one a classic roadster, one a long-nosed sport coupe, both picked from option sheets), and all six cars are now approved.Claude account 1
6:01 PMAlyssa gave a flat "Perfect" tonight to the Cleveland skyline sign after six rounds of revisions. The final version is built from a vector graphic she supplied herself, redrawn as a clean on/off neon sign, with a city skyline that has warm window variation, real depth, a river carrying the sign's own blue, and animated fireworks.Claude account 1
6:00 PMNothing from today's design and accessibility work is committed yet — it exists only in the working files on this laptop. That is a real exposure (it could be lost) as well as a reason tonight's hours undercount; committing it is the lead's next step, not this board's.Claude account 1
1:35 AMAlyssa asked for three more things tonight, none started yet: a short recap video at the end of each of MK11's six phases, a fix for the handful of places where light-mode text is hard to read, and a redesign of the practice-arcade games around an energy theme instead of outer space.Claude account 1
12:51 AMBuilt the testing set the 'Ask a question' help feature has to pass before it can go live for learners — twelve real questions, including ones designed to catch the assistant inventing a fact, answering about a different company, or falling for a hidden instruction to leak its own setup. Along the way found and fixed two separate reasons the feature could not have worked at all yet, on top of the one already known about. It is not turned on for learners yet, and the part of tonight's check that needs the database could not be fully confirmed — the database is down.Claude account 1
12:42 AMThe practice arcade turned out not to be missing after all — it was one setting switched off for this client, which looked exactly like a broken page. Switched on, with a plain record kept of exactly what was changed and why.Claude account 1
12:18 AMYou can now click any word in a lesson while it's being read aloud and the narration jumps straight to that spot, instead of starting over from the top. The key-takeaways summary at the end of a lesson now uses the same studio voice as the rest of the lesson, instead of the computer's built-in robot voice.Claude account 1

Friday, September 18 — 24 changes

11:44 PMWrote down the pattern behind about twenty separate slide fixes as three simple rules, so the next slide gets it right the first time instead of being fixed one problem at a time.Claude account 1
11:43 PMSlide headlines were carrying about five times the punctuation of real professional headlines, measured against 764 published examples. Six new automatic checks now catch semicolons, ALL CAPS, exclamation points and more before a slide ever ships. Also added small icons and solid panels behind on-screen text so words never sit directly on top of busy artwork.Claude account 1
10:33 PMAdded new automatic checks so a slide can't repeat out loud what's already written on it, and every slide has to stay on screen long enough to actually be read — both rules come from real research on how people learn from slides, not a guess.Claude account 1
10:25 PMWrote three more teammates tonight: a copywriter, a copy critic, and a fact-checker whose job is to trace every claim on a slide back to the client's own documents — and it can refuse to let a video ship if a claim doesn't check out. Design and writing are now reviewed separately, since research commissioned tonight found they fail in different ways and one reviewer looking at both catches neither.Claude account 1
10:20 PMAlyssa reviewed the first sample and said slides looked empty, unbalanced, and the artwork didn't match style. Rebuilt the whole slide layout into two clear halves (words on one side, picture on the other), added artwork to all ten slides, and added two automatic checks that refuse bad AI artwork before a person ever sees it — proven by testing them against the exact two images Alyssa had rejected.Claude account 1
8:50 PMBuilt a real video from an actual MK11 lesson — "Who Sells It, and Who Delivers It" — 10 slides, 3 minutes 22 seconds, with a spoken narrator (the Female 1 voice Alyssa picked). This is the first video in the project with a voice on it. Two new AI teammates (a slide designer, and a critic that reviews the actual rendered pictures rather than the plan) went to work and found eight real problems, including one it could measure directly.Claude account 1
8:06 PMWorked out how to let the AI image tool paint the illustrations while we keep full control of every word and every fact that appears on a slide, instead of letting the image tool improvise text.Claude account 1
6:17 PMThe first lecture video was built end to end for the SIE sample course — slides made and assembled into a real, watchable video file. No voice yet; silent.Claude account 1
6:10 PMNow building a lecture-video maker for the SIE course — turning one written lesson into slides and narration — using the real prompt library just found (see the correction below).Claude account 1
6:04 PMThe six module narrator voices are confirmed by Alyssa, and the per-video character budget is corrected to match.Claude account 1
5:40 PMCORRECTION: an earlier research finding said a person had built the slide deck for the SIE lecture videos. That was wrong — the tool generated its own slides straight from the study-guide text, nobody built a deck by hand. Also corrected in the same pass: the estimated cost per video was overstated by about half.Claude account 1
5:30 PMBuilt the Portal Scope Summary the survey has promised clients since September 5th: one page showing the setup a client chose, an estimated size, a timeline, and price lines. The prices are left blank on purpose — Alyssa sets those by hand, a page never should.Claude account 1
4:46 PMThe intake survey — previously one long page of 35+ fields — is now ten short screens that save as you go, so a call that gets cut short costs nothing. Twenty questions an earlier rewrite had dropped are back in, including the one Alyssa's own notes call the number one cause of rework.Claude account 1
4:34 PMThe intake survey is now the first step of building a portal — no documents can be uploaded for a new client until it's finished. MK11 and the pilot are exempted, since they already hold documents; without that, today's change would have locked out our only two real clients. Also found and fixed: ten security tests on the document upload page had quietly been failing for a while, believed harmless — they were not being run at all. They now run for real, and all twenty-five pass.Claude account 1
4:27 PMEvery feature a portal can have is now marked as either included in the base price or sold as an add-on: 15 base, 28 add-ons. Four features that already existed but had never been priced, listed or able to be turned off were added to the list — including the announcements tool, which every learner could already see with nothing charging for it.Claude account 1
12:39 PMFixed a client-facing listing that described the practice-conversation coach as 'coming soon' — it has been live for twelve days.Claude account 1
12:13 PMCleared old downloaded files and traced the full picture of what is filling the laptop's disk: mostly real working files, apps and caches, nothing mysterious left. About 12GB of breathing room now, up from under 6GB — still worth freeing more when there's time, but no longer an unknown.Claude account 1
1:53 AMConsole balance: $62.59 (was $43.24, +$19.35). Auto-reload OFF.Claude account 1
1:38 AMThe spend ceiling you approve is now actually enforced, not just recorded: a course-generation run stops itself before starting the next paid stage once spending reaches the figure you approved, instead of only writing the number down. It can still go over by at most one stage's cost, because a batch already sent for processing is already paid for whether it finishes or not — the button says this plainly rather than promising more than it does.Claude account 1
1:22 AMWrote three things for you before calling it a night: a client-ready list of the MK11 training material still missing (with an offer to mark two topics as trainer-led instead of written lessons), a full breakdown of the Step 5 situation so you can decide with real numbers, and a one-page plan for Sunday's domain session with the one open call flagged.Claude account 1
1:21 AMThe 'Generate course' button now shows a real estimated price before you click — based on our own actual billing history, not a guess — and requires an approved spending ceiling before it will run at all. Back-tested against four real past runs; worst miss was under 5%.Claude account 1
12:45 AMINCIDENT: the laptop's disk filled completely overnight, which force-killed Docker and took the local database down for about 40 minutes and corrupted one of its indexes. No data was lost — a two-second repair fixed the index once the disk had room again — but anything that ran in that window was reading a database that could have thrown an error on any lookup. The disk is still at 98-99% full; this is the top risk left standing tonight.Claude account 1
12:27 AMBuilt a way to price a course-generation run from what we were really billed for past runs, rather than guessing. Checked it against four real completed runs before trusting it anywhere — worst error under 5%.Claude account 1
12:11 AMFound and cleared nine stuck generation runs that were silently blocking BOTH real clients (MK11 and the Modern Sales pilot) from generating anything — the oldest was two weeks old, all of them $0.00 with nothing actually in progress. Fixed the underlying rule so a dead run can no longer lock the system this way. Also: the Publish button now tells you how many blockers there are before you click it, instead of giving no warning.Claude account 1

Thursday, September 17 — 3 changes

11:50 PMThe 'total hours worked' figure went missing from the board in last week's redesign — it was still being calculated every time, just never shown anywhere. It's back: about 205 hours on this project across 20 sessions (your own count runs higher, since nothing here can see time spent in chat).Claude account 1
11:46 PMMK11's client card was pointing at a temporary test address that stopped answering days ago. It now shows the real, permanent address. The Modern Sales pilot never had a real address to begin with — its card now says so honestly instead of showing a dead link.Claude account 1
11:35 PMAlyssa added about $20 of credit to the account during the session. Nothing was spent on top of that tonight's ordinary work (well under a dollar, ledger-side) — so if the account balance looks higher next time it's read, that's the $20, not a mystery.Claude account 1

Wednesday, September 16 — 7 changes

11:26 PMSession closed for a laptop restart to free disk space, around 11:30 PM. Everything from tonight is saved and committed, nothing was deployed and nothing was spent. A handoff note is written so the next session starts exactly where this one stopped, with the open decisions listed first.Claude account 1
9:37 AMRead every DNS record on s3eliteai.app from the Squarespace panel — fourteen rows, no custom records. Three of them could not be seen from outside: a lookup only finds names you already know to ask for, and the apex HTTPS record is a type not every provider imports. It is now on the move checklist by name instead of being discovered missing later. The email risk is gone for good: there are no MX records at all, SPF says nothing may send as this domain, and DMARC rejects. The loudest warning in the domain plan — mail stopping quietly and nobody noticing for a day — cannot happen here. One note for the day a portal needs to send mail: DMARC's sp=reject covers subdomains and will refuse it.Claude account 1
9:05 AMChose Cloudflare over Squarespace for client subdomains, and the deciding reason is privacy, not convenience. Staying on Squarespace has no way to prove ownership automatically, so every client would get their own certificate — and every certificate is published to public logs permanently, which means every client's name would be. Cloudflare allows one wildcard, and no client name ever appears. Checked today: only s3eliteai.app and www are in those logs, so nothing has leaked yet. Also built the part that needs no DNS — what makes a valid subdomain, reserved names like www and admin, and a planner that refuses loudly rather than quietly taking a hostname already in use. It will not bind a hostname for a client without a licence.Claude account 1
8:30 AMDeployed. The payment webhook is reachable on the live app for the first time — before today it answered "not found", which Stripe retries for three days and then disables the endpoint, emailing the account holder. It now refuses honestly instead: it has no keys yet, so it says so, and writes a trace. Also live for real learners: 577,129 bytes down to 528,181 and first text 932ms to 828ms on a phone, thirty-five failing text pairs down to zero, and the audio cache on a disk that survives a deploy. Billing is NOT live — the endpoint exists and refuses everything until Alyssa creates the endpoint in Stripe and adds the two keys.Claude account 1
8:10 AMThe billing tables reached the live database — and the check that caught the problem is worth more than the fix. A report had said they were already applied. They were not: the live database was at 58 and the repo at 59, and the report had been reading the laptop and calling it production. Deploying does not apply database changes, so shipping then would have put a payment endpoint on a database with no tables to write to — and every real payment would have failed. Applied through a guarded path that refuses any target that is not the live database and refuses unless the count beforehand is exactly right. Verified after: 59, both tables, and the key that stops a repeated payment message granting a second licence.Claude account 1
7:40 AMHired a subscription tracker, and the reason it can work is that the last one could not. The Console balance reader needs a signed-in browser and an awake laptop, and has produced zero readings in every slot since it was built. This one reads the voice allowance straight from the supplier, unattended. For the Console balance it deliberately does not try — it quotes the other role's reading with its date and says when it is stale. The part that matters: the field it writes when a reading fails has something that reads it, from day one. The last role was told to say when it could not look, and the place it was told to write that existed nowhere else in the project — so an honest failure and a silent death looked identical.Claude account 1
7:20 AM/money/ now answers how much talking is left: 34,925 letters of 40,000, about 14 lessons read aloud or 17 practice sessions, and it says plainly that there is no way to buy more — when it runs out the voice stops mid-lesson, on a customer, and you are not charged extra. The reading is 34 hours old, so the card shows its age and says which way the error runs: the real figure is this or lower, never higher. When there is no reading at all it shows no bar and no number, only words — a comforting figure with nothing behind it is the failure this page exists to avoid.Claude account 1

Tuesday, September 15 — 3 changes

5:04 AMBoard-keeper pass: the 'waiting on you' card was stale (the Console-keys reminder was already done, deleted 00:55 Sept 14) — cleared, and replaced with the four things actually open tonight: five nav pills vs. four, /simple/'s fate, the 30-minute balance reader that has never once fired, and the unexplained $9.23 balance rise. Step 5's card said 'the writing is running now'; that run ended the night of Sept 12 — fixed to say so and marked HELD pending Alyssa's own words, since 'hold it entirely' is on record from the lead's account only, not BUILD_LOG or an earlier board. The 'Tests passing' tile's parser expected passed-before-failed and silently vanished on a red run instead of turning red (tonight's real number, 1,986 passed / 15 failed / 2 skipped, disappeared rather than showing); fixed to read either order and to say 'not read' rather than nothing when it can't.Claude account 1
1:43 AMBalance re-read in the Console: $43.24, not the $34.01 this board had been showing since 1:12 AM on the 14th. I used the stale figure to tell her a $31.70 run would leave her with $2.31; the real worst case is $11.54. She caught it. The staleness had been flagged by the silent-fail audit hours earlier and I quoted the number anyway.Claude account 1
12:50 AMOwner cleared the decision queue: leak-test learner fix APPLIED (leak suite 95/95 green, and it is the CI-only collision that has been failing since Aug 22), arcade 1g and boards 1h both released, tutor prompt approved with the embedding spend held until it is quoted, Console keys moved to the morning brief. Six waiting became one, and that one is a two-click errand.Claude account 1

Monday, September 14 — 31 changes

6:05 PMThe standalone role-play demo is live on a permanent address that does not involve Alyssa's laptop — it survives sleep, restarts and wifi, and its name never changes. A neutral S3 Elite AI tenant, deliberately not MK11-branded, so a prospect is never shown another client's training. The link and password are with her.Claude account 1
6:03 PMSomething was found while checking that demo link that had been true for days: the staff console was answering on the same address. A real login page, password-protected, but a door that should not have existed on a link about to be sent to a business partner. Our own deployment runbook says all six staff addresses must return nothing and says to stop before handing out a link if they do not. Fixed with Alyssa's approval and confirmed from outside: all six now return nothing, and the learner side still works.Claude account 1
6:00 PMSigning in was genuinely broken this afternoon, and it was the lead's doing. Rotating the key that signs logins left the matching key inside the database unreadable, because it had been locked with the old one. Forty-eight tests were failing on it. The unusable key was removed by name and the system made a fresh one; the full suite is green again at 1,800 tests. When Alyssa said 'nothing works', she was more right than the first answer she got.Claude account 1
5:58 PMMK11's nine empty modules have a decision: seven can be written from the client's own documents, two are deleted because their content genuinely lives in other lessons — verified by reading those lessons, not by trusting the note. That takes the course from twenty-four modules to twenty-two, every one of which opens onto something real. The seven pages are being checked claim by claim against the source documents before anyone sees them.Claude account 1
1:34 PMTwo of the six problems found in last night's portal walk do not survive measurement. Both were seen during the window when the lead had mistakenly put two reviewers on the same browser window, so each was describing a page the other had moved. That mistake is on the record; this is what it cost. The two genuine blockers from the same walk were real and are both fixed.Claude account 1
1:32 PMThe lesson reader no longer pretends. Pressing Listen used to show 'Pause' as though it were reading aloud while a single word sat highlighted for nearly twenty seconds in silence. It now says 'Preparing' for exactly as long as it is fetching the voice, and only claims to be playing once sound is actually coming out — measured at five thousandths of a second after the audio truly starts, where it used to be half a second ahead of it.Claude account 1
1:30 PMThe biggest correction of the day: MK11 did give us their sales training. It is in one of their own documents — the four factors, the five steps to a conversation, all eight work habits — and three lessons that were meant to teach it failed to write, most likely because our own scanning scrambled the words on those pages. Every document since three this morning said the opposite, including the note that went to Dave. The letter asking MK11 for that material has NOT been sent, because Alyssa chose to read it first.Claude account 1
10:55 AMWith Alyssa's approval, publishing now refuses a course that contains a module a learner cannot open — the defect that let MK11 ship with nine named modules and nothing inside them. The rule was rewritten once before it shipped: the first version would have flagged 204 modules across the database when only 21 were genuinely empty, telling people to write lessons that already existed. Checked against every course we hold: 18 flagged, none of them wrongly, and MK11 comes out at exactly the nine we knew about.Claude account 1
10:50 AMTwo abandoned tunnels into the laptop were closed, leaving one door where there were three. The demo link was confirmed working before and after. Five further tunnels were found pointing at a port with nothing behind it; those were outside what Alyssa approved, so they were left alone and raised with her rather than assumed.Claude account 1
10:45 AMThe practice page is fixed. The four customer-type cards were being squeezed into a strip the width of a business card, clipping the name and breaking the description into a column two words wide, on the screen where a nervous new hire chooses who they are about to practise against. They now sit two across with room to spare, and the Start button is back on the first screen. The same fix closed a second fault at phone width that last night's walk had no way to test.Claude account 1
10:40 AMThe COO's plan of action is written. Its headline is that the original plan assumed a client's folder contains their training, and MK11's does not — so the step that reads 'regenerate the course for about ten dollars' is a step money cannot buy. It commits on MK11 rather than deferring: send the gap list, and publish an honest first part this week without waiting for the client's answer. The week's single target is an invoice or signed quote in MK11's hands by Friday.Claude account 1
6:58 AMThe first organisation-wide all-hands finished and its write-up has gone to Dave and Alyssa as both a Google document and a slide deck, so the two of you can judge which format works better. The meeting produced one uncomfortable and useful conclusion: fifteen separate-looking mistakes this week were one mistake wearing different clothes — we keep treating a system's own report about itself as proof that it is working. The new rule is that a claim about the outside world needs a witness outside the thing making the claim. The write-up also names what nobody said all night: not one of six documents mentions a customer, a date, or a dollar coming in.Claude account 1
3:52 AMThe third and last of tonight's learner fixes is done: our internal database codes no longer appear in the words a new hire reads. A lesson that said 'Riley Gibbs is President of MK11' followed by a long code now just says 'Riley Gibbs is President of MK11.' Nothing was hidden or deleted — the sources list at the bottom of every lesson and the record of where each fact came from are untouched; we simply stopped printing the part that was only ever meant for us. Checked against every lesson in the database, all 321 of them: 42 were affected, and afterwards none show a code. The underlying cause is one missing line in the instructions given to the writing model, and that file is protected, so it is waiting for Alyssa rather than being changed quietly.Claude account 1
3:42 AMTwo of tonight's three learner fixes are finished. The sentence in the pay lesson that stopped dead at 'a minimum of 5' now reads as one whole sentence and says what five means; the same repair restored the missing verb in the commission lesson, with no word of any definition changed. And the progress page now lists the same fifteen modules the course actually offers, instead of twenty-four — nine of which led nowhere. A new check that catches this kind of disagreement was run against all forty-three courses in the database and was correct on every one.Claude account 1
3:38 AMThe local web server was found dead across the whole machine, answering an error to every page, knocked over by this morning's disk exhaustion corrupting its build cache. No portal walk could have succeeded while it was down. It was restarted and confirmed working by a real request rather than by checking that the process exists; the public demo link answers again. The tunnels were left running throughout, and reconnected on their own.Claude account 1
3:30 AMThis board's own record was found emptied to zero bytes at 3:04 AM, wiped mid-write when the laptop ran out of disk, and restored intact from the last saved copy with nothing lost. A first attempt to add tonight's entries was then silently overwritten, because three separate scripts each read and rewrite this file and an edit made between one script's read and another's write disappears. Both holes are now closed: the file is written completely and swapped into place in one step, so it can never be left half-written, and a write that has lost the history is refused outright instead of going through quietly.Claude account 1
3:15 AMThree learner-facing defects are being fixed now, each with a test written first that was watched failing. A sentence in the pay lessons is cut in half by a display bug, so the course tells a new hire to close 'a minimum of 5' and never says five what. The progress page lists a course that does not exist — twenty-four module names, only one of which is in the course the learner is actually taking. And 132 of our internal database codes are printed in the visible text of nine lessons.Claude account 1
3:12 AMThe laptop ran down to 328 megabytes of free space out of 466 gigabytes. Three separate reviews could not write their files and one stopped mid-way. About nine gigabytes of caches that rebuild themselves were cleared to get work moving again; nothing of yours was touched and nothing was emptied from the Trash. The drive is still 98 percent full and the rest is your own files.Claude account 1
3:09 AMTonight's security review found that four credentials exposed in last night's leak were never actually replaced and are still the live values in production — including the one that signs users' login sessions. Two outside vendor keys were rotated last night; these four were ranked lower and that ranking was wrong. It needs twenty minutes with Alyssa awake. The review also confirmed the leak itself is genuinely sealed, by rebuilding the package from scratch and deliberately breaking the new guard three ways to prove it refuses. Update, October 14: this is now fixed in full. Every credential from that leak has been replaced and the old ones no longer work; see the October 13 and 14 entries.Claude account 1
3:07 AMThe demo link currently being shared is a development server open to the whole internet with no password on it, through a temporary tunnel, with two further stale tunnels to the same server also still live. This is the same root cause as the portals timing out. Nothing was shut down, per the standing rule, but the link should not be shared further until it sits behind a real address with a sign-in.Claude account 1
3:05 AMThe first full learner walk of MK11's portal, alongside the first curriculum review, found the course has eight modules with no lessons in them at all, and that it asks a new hire to explain seven things to earn $400 of training pay while teaching only two of them. The two computer systems the job runs on are named twenty-six times and explained nowhere. This is not the AI writing badly — the client's documents contain their rules and their product knowledge but not their sales training, so no amount of regenerating can produce it. A decision for Alyssa in the morning.Claude account 1
3:03 AMAlso found working and worth keeping: the portal's front page answers 'what do I do next' within three seconds, lesson one has a real teaching voice that corrects a wrong assumption a new hire would make, and the whole energy-knowledge section teaches each idea before the next one needs it — start to finish, with no gaps.Claude account 1
3:00 AMBefore bed, three things were checked with you rather than assumed: the 7 AM all-hands email is approved to send as both a document and a slide deck, an overnight spending ceiling of $5 was set, and a leftover backup file from the earlier key rotation was safely deleted.Claude account 1
2:57 AMThe portal-walker role (which reviews every finished build) now also judges the actual writing quality — whether ideas build logically and a real teacher voice comes through — and can trade notes with the curriculum team while it walks. It now launches itself automatically after every build.Claude account 1
2:38 AMThe new team-oversight role (created earlier tonight) was renamed AI Resources, and can now pause an underused team role instead of deleting it, so nothing it has learned is lost — it can come back exactly as capable as when it left.Claude account 1
2:23 AMAdded a dedicated graphic-designer role for documents and reports, separate from the app's screen designers, working in both Google Docs and a presentation tool in the company's own brand. Tomorrow's 7 AM all-hands will use both formats side by side as a real test of which the team prefers.Claude account 1
2:12 AMFixed a bug where the deployed portal falsely reported itself unhealthy and never came online, because its health check demanded document-scanning tools this deployment doesn't include yet. It now correctly reports itself healthy for serving lessons, while being upfront that document upload isn't available on this deployment yet.Claude account 1
2:00 AMTracked down the real reason the overnight production build kept failing: parts of the app were trying to check settings and reach the database while merely being built, before either is ever available. Fixed so building the app never needs the database or any settings — only running it does — and the production build finally completes cleanly.Claude account 1
1:31 AMA review of four days of how the boards actually get used found they only ever do two things for you: catch a board that's lying, or hand you a decision. Everything else — hourly tallies, weekly breakdowns, vanity numbers — is being trimmed so the boards stay focused on just those two jobs.Claude account 1
1:15 AMBoth outside services affected by tonight's credential leak had their access keys fully replaced — tested working before and after the swap, with no downtime — and every old, exposed key was permanently revoked. The leaked image itself can't be erased from the host's history, but nothing inside it still works.Claude account 1
12:00 AMAdded a standing security-review role plus an automatic check that inspects the deployment package, database access and credentials on a schedule — daily, before every deploy, and right after any incident — instead of relying on someone remembering to ask for a review.Claude account 1

Sunday, September 13 — 22 changes

10:39 PMSecurity incident: a copy of the app's private settings file — holding API keys, the database password, and other credentials — was accidentally built into the deployment package sent to the hosting service. It was caught almost immediately by an unrelated safety check that refused to let the app start with it, before any known harm. A permanent block is now in place so that file can never be included in a deployment again.Claude account 1
9:30 PMA review of the whole AI team found eight roles had never produced anything useful despite being on the roster, and a scheduled safety review was skipped before tonight's first live deploy attempt. Separately, the hosting deploy hit three build failures in a row overnight and work was stopped there rather than guessing further; nothing is running or being billed on the host yet.Claude account 1
9:00 PMTonight's priority was getting a permanent web address for the demo that survives the laptop sleeping, since the temporary one won't even load on your own home network — that needs one browser sign-in step from you. With the hosting side confirmed unblocked, the approved deploy (about $6/month) was started, and a serious near-miss was caught first: the recommended database connection would have let the app bypass the safeguard that keeps one client's data separate from another's. An existing guard refused to start rather than allow it, so it stayed a near-miss, and a properly restricted database login was set up before continuing.Claude account 1
4:35 PMThe spending safety guard now covers every script that can spend money, not just one of them, and always uses whichever cost estimate is more cautious — measured or manually entered — so a projection can never be too optimistic.Claude account 1
3:14 PMAdded four new customer role-play characters and voices for practice conversations, at your request for more diversity, bringing the roster to seventeen with a more even mix of personality types.Claude account 1
2:08 PMCaught and fixed a board display bug: the money figure looked live because its timestamp kept moving, but it was actually showing a reading from sixteen hours earlier. It now shows the latest reading.Claude account 1
12:30 PMFound, but not yet fixed (another session is mid-edit on the board files), a display bug: on a phone, the accessibility settings panel on the team-chart board opens running off the left edge of the screen. The cause is identified and a one-line fix is ready to apply.Claude account 1
12:24 PMConfirmed the navigation-overflow fix from earlier this week is already safely committed, not lost. A bigger phone-navigation redesign stays on hold pending your word; a smaller fix for larger tap targets is queued as the next code change.Claude account 1
12:08 PMFixed a real inconsistency where two different voice pickers could silently override each other, so a learner could hear two different voice identities in one session — there is now one shared, consistent control everywhere. Also fixed a silent failure where the microphone could stop working with no on-screen message, and made the currently selected voice actually visible and readable in both light and dark mode, backed by a new automated check.Claude account 1
12:07 PMBLOCKED: the three automated browser checks all pass individually but fail intermittently when run together in one process. Work was stopped there per the two-strikes rule rather than guessing further; every measurement taken shows no evidence the actual product changed.Claude account 1
10:55 AMA cross-team review found and fixed six issues; the most serious was that the public board had shown a work session as "live" for three and a half hours after it had actually ended, because the shutdown process didn't fully stop a background watcher. Fixed so closing a session is verified for real, not assumed.Claude account 1
6:49 AMCorrection, and it is mine: this board publicly said 'I am in a working conversation with Alyssa right now' for three and a half hours after the session ended at 3:18 AM. The close-out did turn the light off, and seventeen seconds later the live watcher turned it back on because a background agent was still writing. The watcher was never stopped by the close-out. It is stopped now, the board is corrected, and the close-out has been fixed so this cannot repeat.Claude account 1
3:18 AMSession closed cleanly at 3:18 AM after 26 hours. Everything is committed, including a labelled safe point for the work still under review. Nothing was lost. The boards will say 'not working' until Alyssa is back, which is the honest state.Claude account 1
3:10 AMNew role at Alyssa's word: a silent-fail auditor whose only job is what broke without saying so — checks that cannot fail, errors swallowed in silence, jobs that died with nobody noticing. It fixes what is safe to fix, escalates the rest with the mechanism named, and writes every other agent a short note on what they missed so the team improves instead of just being corrected. Its scanner is proven able to fire before any clean result is believed. First sweep running now.Claude account 1
2:56 AMNew standing rule at Alyssa's word: the board keeper runs whenever Claude is working. It now starts automatically with every working session, checks every five minutes whether the boards actually match what is running and what has been spent, redeploys, and writes anything only a person can judge into a queue the lead must read. A test fails if anyone removes it. Building it caught two of my own bugs first, including a check that read the wrong day's figure.Claude account 1
2:50 AMAll thirteen role-play voices are on the ElevenLabs account and reachable from the portal, verified one by one by name. Clicking the voice pages was failing about half the time at 2 AM, so they were added through the API instead; the one permission that needed widening was flipped by Alyssa in the Console, and the script that does it is in the repo for next time.Claude account 1
2:12 AMRole-play customers now speak like people: every character's verbal habit is occasional, not a sentence opener, and the buyer's persona text carries a standing line to vary openings and shift mood (Alyssa caught Earl saying 'well now' every line).Claude account 1
2:12 AMPractice coach gains a live conversation: the mic stays on, a finished sentence is sent by itself, the mic holds while the customer answers in their own voice, and a 'Pause for coaching' button stops it to ask the coach. Committed; walked by hand next.Claude account 1
2:12 AMA shareable demo link for the MK11 portal now serves the dev build with everything from the weekend (reader, Aa panel, DISC practice, brand pass). It runs on Alyssa's laptop, which is kept awake tonight at her word. Address and demo login are in the demo-accounts file, never on this board.Claude account 1
2:12 AMThe compact reader dock is built and measured by the UI designer (its own column beside the text on desktop, a top bar on phone, Adam and Clara only). Round two is fixing its own test walk; the design critic judges after.Claude account 1
1:18 AMRole-play customers are now thirteen voiced characters across the four DISC behavioural types, each with a briefing (who they are, their business, how they talk, what is going on today). A learner picks a type, meets a random character of that type, reads the briefing, then starts; the customer answers in that character's own studio voice. Proven signed in on the test tenant. Alyssa's thirteen voices are being added to her account page by page.Claude account 1
12:20 AMLesson reader, round two: Alyssa's two chosen studio voices (Adam and Clara) are on the account and listed first; the word being spoken is highlighted on the page as the audio plays, driven by the voice service's own timings for studio voices and by the browser's word events for device voices; five speeds for both. Proven signed in: highlight lit on 'Two', moved to 'energy', cleared on Stop.Claude account 1

Saturday, September 12 — 47 changes

11:48 PMStudio voices are live on the dev server: the lesson reader and the practice coach can use the voices on Alyssa's ElevenLabs plan, with a spending guard (the last 5% of the plan is never spent) and a cache so a replayed lesson costs nothing. Proven signed in: 21 voices listed, one phrase synthesized, the repeat served free.Claude account 1
11:40 PMMK11 brand pass shipped after two critic rounds: the client's fonts on every page, the client's name in the browser tab, the sign-in card and inputs in the brand's navy, the footer note readable, no stray serif fallback. The critic then caught the light-mode palette still carrying the brown; fixed and re-applied. Studio voices (ElevenLabs) are wired for the lesson reader and the practice coach and wait only on the key Alyssa is creating.Claude account 1
11:03 PMAlyssa caught the MK11 portal painted brown where the client's site is navy. Audit: the brand tool built dark backgrounds from the primary colour's hue, so a gold brand always came out brown; the site's real background was never an input. Fixed in the tool (a measured background now drives the dark palette), re-applied to MK11 (navy), and the brand steward and design critic were taught a one-word ground check before anything is called 'applied'. The design critic separately sent the director's brand pass back with four measured defects; round two is in progress.Claude account 1
10:24 PMLesson reader shipped: every lesson can be read aloud in the browser's own voices, with a voice picker and five speeds, remembered per device — $0, nothing leaves the laptop. Reader accessibility controls to match the S3 site: text size in four steps up to 150%, bold text, light/dark, skip-to-content, a visible focus ring; saved in the browser. A signed-in walk proved them and caught one bug on the way (the side menu could not scroll at 150%), fixed.Claude account 1
9:46 PMAlyssa opened the MK11 demo and saw the August course in front and a test course beside it — the new course had been demoted by the home page's 'largest course first' rule. On her word both old courses were taken off the demo portal (audited, reversible). A new Portal walker role now signs in as the learner and checks every portal before any link is handed over.Claude account 1
9:00 PMAlyssa restarted the MK11 tunnel from her terminal; the MK11 walkthrough portal (with the new course's preview copy) is reachable on a new public address. Pilot link still down.Claude account 1
8:55 PMFound while setting up the demo-accounts keeper: both public demo links died at 3 AM Sept 8 and have been retrying since (10,000+ attempts). The MK11 tenant now answers at a local address on the laptop so the preview can be read tonight; restarting the public tunnels is Alyssa's action.Claude account 1
8:28 PMAt Alyssa's word, a 'Preview, unreviewed' copy of the new MK11 course is published on the MK11 walkthrough portal: 24 lessons, 20 quizzes, 134 questions, badged on every page. The original stays in the review queue untouched; nothing was approved. $0.Claude account 1
7:59 PMMK11 rebuild complete: 38 lessons through outline, writing, quizzes (134 questions, 20 quizzes, one final exam) and fact checks (28 checked, none below the line). 24 written, 3 flagged for the client's expert, 11 refused by the free pre-flight — 4 of those were the model mangling its own source ids, caught by the tenant guard; no other client's material involved. Billed $16.32 today (Console), approved $18 held.Claude account 1
7:57 PMMoney roles built: the CFO now audits the Console against the ledger every tick (1.16–1.22× measured gap, mechanisms named); a cost estimator prices runs at the bill; a budget controller stops a run between stages when it would cross the approved figure. Every new run needs a written budget before it starts.Claude account 1
6:52 PMMK11: four signature lessons folded in, 38 lessons, shape check clean. Authoring started on the Batch API.Claude account 1
6:39 PMThe MK11 outline was accepted and saved (take two, with the tightened direction). $8.00 of the $15 cap spent so far; the free shape check is next, then lessons.Claude account 1
6:10 PMThe Admin key was created correctly, but Anthropic's billing API is not enabled for this organization (every call answers 'missing permissions'). The spend check stays a hand-read of the Console at every close. No key was ever seen by Claude.Claude account 1
6:03 PMAlyssa: cap to $15, run it all, tighten the outline (30–34 lessons, orientation second, signature pages folded). The rebuild resumed.Claude account 1
5:21 PMChecked, not assumed: no free recovery exists. The outline and repair calls were interactive, not batch, and a refused outline's text is never kept — only the printed copy survives. The run is stopped at $4.23 plus about six cents landing; nothing more is spent without Alyssa's word.Claude account 1
5:20 PMThe outline resume re-bought the summaries instead of re-reading the batch and was stopped before a third outline call. The refused tree is not recoverable by resume; a free recovery is being checked; the paid choice waits for Alyssa.Claude account 1
4:48 PMMoney, from the ledger: the first outline call finished at $3.20 and a second outline call started at 4:46 PM. Session total $3.25 of the $10 cap. The Console has not been re-read since 11:52 AM; it is the bill and is read at the close.Claude account 1
4:46 PMBoard-keeper's first pass after the outage: the run's start was missing from the path, three answered decisions were still listed as needing Alyssa, the critic was still shown as 'not in the loop', the session's cap read $0. All corrected against the log, the report, the ledger and the commit history.Claude account 1
4:36 PMIncident: the laptop's disk filled (under 1 GB free); the database inside Docker stopped at 3:55 PM; the boards were last updated at 3:34 PM and not again until 4:39 PM. Caches cleared (11 GB freed), Docker and the database restarted, the outline batch untouched. The boards now survive a database outage and a board-keeper role owns their accuracy.Claude account 1
3:43 PMBrand-voice wrote MK11's voice guide from their website and the ingested source (education first; the voice is applied only where it costs the lesson nothing). The tutor prompt and an embedding-pass quote are written as proposals for Alyssa. Ledger: 20 summaries for six cents; the outline call went to the Batch API.Claude account 1
3:37 PMAlyssa: 'Start it now.' The MK11 regeneration began — outline on the Batch API, then the free shape check, then the lessons overnight; cap $10. Her direction (Welcome first, a short who-we-are introduction from the documents and the website source, the rest folded to 31–34) rides in the course notes; no prompt file changed.Claude account 1
3:34 PMAlyssa decided by multiple choice: address accepted, MK11 link + brand proposal, author overnight without waiting, exam as recorded, cap $10, domain work Sunday with her, Admin key now, tutor prompt to draft, stray script deleted. The leak-test one-liner failed locally and was reverted with the real fix proposed.Claude account 1
3:31 PMAlyssa's decisions, batch 2: her words on the outline ARE the review (no mark-up sheet); cap $10 on the Batch API; domain work Sunday afternoon with her; the Admin key now, typed by her; a new brand-voice role — education first, brand voice only where it makes sense.Claude account 1
3:30 PMBoards 1d/1e/1f: one nav, one heading system and the same reader controls on all three boards; the team chart fits a laptop screen; the simple board's tiles hold at every text size; history folded at ten. Two strikes; the better-measured version kept; the rest is 1h.Claude account 1
2:54 PMAlyssa's decisions, batch 1: office address confirmed and the finding accepted; MK11's website recorded as its one allowed link and the brand tool run in propose-only mode; the leak-test one-liner applied, failed locally, reverted with the real fix proposed; the eval key on GitHub was already gone. Arcade 1c under the fixed sky: one stage box, a coin glyph everywhere, focus follows the arrows; two strikes; kept; the rest is 1g.Claude account 1
2:25 PMAlyssa released the boards' remaining design lists (1d, 1e, 1f) and ruled 'Fixed sky' for the games. Both design loops are running.Claude account 1
2:04 PMBoth boards through the design loop on a phone: a phone tier, an honest spend bar (it had drawn 74% for an 83% ratio), AA contrast in light, 'needs a person' first. Each loop stopped at two strikes with the work kept; the remaining lists became 1e and 1f.Claude account 1
1:55 PMTonight's MK11 run plan written by factory-ops (outline → free shape check → author; exam settings from MK11's own intake). The security check passed with fixes first: the live watcher had been publishing agents' last sentences to the public board — it now publishes a fixed vocabulary only, and status updates refuse anything that looks like a path, host, id or key.Claude account 1
1:48 PMQueue 6 opened. The eval workflow's push trigger removed (it runs on demand only; Alyssa's veto word is 'keep the trigger'). Every agent must have a box on the team chart — enforced by a test and by the renderer refusing an incomplete chart.Claude account 1
1:45 PMNew machinery Alyssa asked for: the critic gate (a send-back records a strike and blocks the lead's turn until it is answered; two strikes stop the item), the design loop (critic and designer iterate without the lead), the live watcher (working/idle read every 30 seconds), and the COO + chief-of-staff roles with tonight's agenda and the day plan.Claude account 1
12:29 PMThe team chart: every agent as an org chart in the starry sky, glowing when working and grey when idle, each with a plain-words update. The daytime brief became a Google Doc at Alyssa's request.Claude account 1
12:21 PMDaytime session closed: five of seven items done, one stopped for Alyssa (arcade 1c), $0 spent; the daytime brief for Sept 12 was written; the simple board is live at /simple.Claude account 1
12:00 PMArcade fix list: five of the critic's points fixed and measured; the fourth review sent it back again, so it stops (two strikes) and the rest is 1c, yours to release.Claude account 1
11:52 AMConsole re-read after Alyssa's top-up: credit is $50.78 (was $30.79), 30-day total $262.99, today $0.01. A second, plain-language board now lives at /simple — the full board stays.Claude account 1
8:32 AMAlyssa approved the outline rule at 30–40 lessons; the outline prompt is at version 12. Next paid step is hers: mark up the side-by-side, then regenerate.Claude account 1
3:10 AMOvernight run closed: 15 of 17 items, $0.002 spent, brief written through the CFO and plain-English roles.Claude account 1
3:03 AMFailed model calls now keep their price; dev probes refuse to run unledgered; three stale runs closed by id; Alyssa's 33 lessons laid beside the factory's 60 for mark-up.Claude account 1
2:58 AMLearners can search their own company documents from the portal and hear a lesson's key points read aloud. The AI-written answer waits on a prompt approval and an embedding pass — both Alyssa's call.Claude account 1
2:54 AMTen measured palette presets on the branding console, researched by two design sub-agents — every one failed the dark-mode button test on the first pass and was corrected by measurement.Claude account 1
2:43 AMManagers can post announcements that appear on every learner's dashboard; pinned ones stay on top.Claude account 1
2:38 AMLearners and managers can message each other inside the portal — proven both ways through the real pages, with audit rows.Claude account 1
2:32 AMManagers see time this week and weak spots on the roster, and each learner's standing, time per lesson and arcade bests on their page — the same numbers the learner sees.Claude account 1
2:29 AMLeaderboard: arcade bests, best average check score, most study time this week — per-client switch, names as first name and last initial. The nav now marks the current page on every portal screen (it never had).Claude account 1
2:20 AMLearners can see where they stand: accuracy per module, graded exactly as their checks were, and time per lesson. The design critic sent the arcade back a second time; it waits for Alyssa.Claude account 1
2:14 AMTime in the portal is tracked per lesson and per day, proven by a learner reading a lesson for half a minute.Claude account 1
2:07 AMThe dashboard draws the journey as a rail through numbered phase nodes, like Alyssa's own MK11 portal. The arcade was sent back by the design critic (its buttons rendered as bare text) and reworked.Claude account 1
1:48 AMChill Study Mode: Mike's six games ported and fed by the client's own lessons and question bank; scores table under RLS; seen light and dark.Claude account 1

Friday, September 11 — 11 changes

9:50 PMAlyssa's notes on the morning brief answered on the board; tests can no longer reach the paid API; the nightly eval disabled at the repository level (it had run again this morning); onboarding band set to 30–40 lessons; CFO, plain-English and factory-ops roles added; spend:audit.Claude account 1
2:47 PMManager screens wear the dashboard's header; learner page opens with four tiles. Queue 3 closed.Claude account 1
2:37 PMCI unit tests green on GitHub for the first time since Aug 22. The leak job's own, older failure is diagnosed and waiting on Alyssa (protected test).Claude account 1
2:33 PMCI had been red for three weeks. The cause: a test was quietly paying for a model call on Alyssa's API key to pass locally — and had no key on the runner. The judge now skips lessons with nothing to judge; the test can never reach a model.Claude account 1
2:20 PMThe Content Verification Audit is now a document a client can sign: headline, do-these-first, five parts, sign-off — generated in one command, never carrying an internal id.Claude account 1
2:20 PMContent Verification Audit renders as a five-part client document; Resources page shows the client's contacts as cards.Claude account 1
2:13 PMCoverage regression pin proven: broke the metric on purpose, the test failed; restored, it passed.Claude account 1
2:12 PMResources page: the client's own lines become cards — who to call, help lines, the chain of command — and what the scan could not read is counted, not shown. On MK11: two real contacts, one help line, 23 unreadable lines named as such.Claude account 1
2:03 PMResumed on schedule at 2:03 PM EDT. Eight-hour window; Queue 3; nothing will be spent.Claude account 1
12:20 AMOwner's review: the monthly spend and the Numbers tiles were stale, hand-typed values. Now every tile is measured at rebuild and names its source; months follow the Console with the ledger beside; the board stamps its own clock.Claude account 1
12:01 AMPaused at the owner's request; scheduled to resume at 2:03 PM EDT with Queue 3.Claude account 1

Thursday, September 10 — 29 changes

11:59 PMLoop closed: queue empty. No session is running; nothing is being spent.Claude account 1
11:58 PMMorning brief written. Night total: 22 queue items, ~45 commits, $0.003 spent on the factory key; three defects of my own found and fixed on the record.Claude account 1
11:55 PMOnly 28% of batch-routed calls ever used the Batch API. The stage scripts now default to it; immediate delivery is an explicit opt-in. Regeneration will cost roughly half.Claude account 1
11:53 PMWhere the money went: read from the console by API key — $249 of $259 on the factory key in 30 days; $69 before the ledger existed; a nightly cloud eval paid daily and failed daily (now off); the rest under-recorded. Credits left $35.Claude account 1
11:46 PMContent audit: enumerations (the source counts 8 Great Work Habits; extraction recovered 6) and P3-17's 'referenced but never explained' folded in — 25 such gaps on MK11.Claude account 1
11:41 PMLedger defect found and fixed: ~2,000 calls delivered interactively were priced at the 50% batch rate. Board now shows the ledger as written plus the correction, and a section on what it cannot see.Claude account 1
11:36 PMContent audit: one command lists what a client's documents contained that the course never used — 65 real gaps on MK11 after triage (harassment policy, the contact directory, the four factors of impulse).Claude account 1
11:31 PMSpend: Alyssa's API top-ups entered ($298.20 across nine purchases, Aug 20 – Sep 7). The factory's ledger recorded $80.37 of calls; the gap is unlogged calls plus unspent balance.Claude account 1
11:29 PMOutline gate: headings that are really OCR word-salad or a scrambled running title no longer create phantom sections — 81 of 207 MK11 headings reclassified, measured at $0.Claude account 1
11:20 PMGrounding: tables and half-citations no longer count as unsupported claims — 159 → 3 artifact assertions on the pilot, measured without a model call.Claude account 1
11:17 PMThis board: total project spend — model calls measured from the ledger by month and by kind of work; a slot for costs outside the ledger.Claude account 1
11:14 PMThis board: the site's galaxy backdrop, and hours on the project measured from git.Claude account 1
11:14 PMReference snapshot: MK11's course exported outside the repo (2,562 rows) with a content-free manifest committed; restore drilled and verified on a fixture. A database reset now costs a command, not $5.Claude account 1
11:05 PMClients: a run stuck at 'running' for over a day is now called a stale run, with the fix named. Nine such rows found from earlier sessions.Claude account 1
11:02 PMClients: one bucket per client on the factory — stage, headline, next step, deep links, each course's minutes per lesson.Claude account 1
10:51 PMManager walk: a learner's completion shows on the manager's roster and learner page — walked through the real pages, light and dark.Claude account 1
10:50 PMThe Build Board (this page) — rebuilt by every session, secret-guarded.Claude account 1
10:47 PMshape:check — the course ruler runs on any built course in one command. All three MK11 courses ever generated refuse it (60×2.1 min, 17×6.8, 27×4.0).Claude account 1
10:39 PMPixel proof: chips, gate subtitles and shop CTA measured on brand in light and dark. Shop-CTA premise corrected.Claude account 1
10:28 PMM63 closed — roster suite owns its tenant; full suite 1,270 passed, 0 failed.Claude account 1
10:17 PMdemo:share — one command hands a portal to a reviewer (prints the link, never a password).Claude account 1
10:16 PMClient notes reach the outline for every archetype; shape-gate corrected under the security suite.Claude account 1
10:06 PMFactory screens hide test and seed tenants by default.Claude account 1
10:00 PMThe organisation as 8 durable agent roles (VP Review runs on Fable).Claude account 1
9:57 PMPractice calls: an 18-turn target with 4 turns of grace, judged before any model call.Claude account 1
9:48 PMA tenancy violation fails one lesson, not the whole author batch.Claude account 1
9:39 PMPhase timeline renders numbered chips with lock reasons, not a count.Claude account 1
9:32 PMPhase gates persist; MK11's six backfilled.Claude account 1
9:20 PMTHE PLAN v2 is the single source of truth; the shape gate; M60; verified cloud backup in one command.Claude account 1

The longer story on 1 step

The steps themselves are on Progress, one or two lines each. This is the part that did not fit there.

Step 5: Regenerate MK11 Onboarding — 24 of 38 written; 14 wait on MK11

The plan settled at 6:38 PM on Sept 12, after the direction was tightened and one repair round was run: $8.00 of the $15 set aside. The curriculum architect read the shape of it for free and said go, with a note — 42 lessons merged down to 38, because four pairs that were read-then-sign or near duplicates were folded together at no cost, and the shape check passed clean. The lessons, quizzes and fact checks were then written that night on the Batch API; that run is FINISHED, not in progress. Result: 24 lessons written and fact-checked, 3 flagged below the author's word floor for the client's expert to fill in (the source material is thin, not a factory defect), 11 refused by the free pre-flight (four of those from the model mangling its own source ids, caught by the tenant guard). $16.32 billed on the 12th, inside the $18 cap. Four chapters currently have no quiz. We asked for a billing key that would let us read the bill automatically, but Anthropic does not offer one to this organisation, so the money is still read off the Console by hand. The 60 old drafts remain exported and safe.

Where the time went

865.6 h building this, 73.7 h of it this week.

The total is the working clock on this project across 122 sessions, starting July 12. Your own count is over 300 h, and yours is the one that is right — no clock here can see a chat on claude.ai, in Cowork or in Claude Chrome, so treat the figure above as the floor.

By month

November 2026 so far90.9 h
October 2026391.4 h
September 2026293.9 h
August 202686.3 h
July 2026 from July 123.1 h

These are the same working-clock hours as the total above, split by calendar month, so the months add up to it. They are a floor: no clock on this machine sees a chat on claude.ai, in Cowork or in Claude Chrome.

Older

24 earlier briefs

Library build plan and result October 21 · lead

Morning review: the overnight queue October 14 · lead

Role-play customers: where the wait comes from October 14 · research-analyst

MK11 final exam: from the document arriving to live and verified October 14 · lead

Morning brief — Saturday, October 4, 2026 October 4 · vp-review

Morning brief — Thursday, October 1 October 1 · vp-review

Sadie in the Series 65 portal — build plan, direction B September 30 · Claude (the lead)

Morning brief — Series 65, overnight Sept 28/29 September 29 · vp-review

Owner demo login — build plan, stopped before starting September 29 · Claude (the lead)

Series 65 base portal — content and quiz plan September 29 · Claude (the lead)

v70 → main branch sync report September 28 · Claude (the lead)

Series 65 Module 0 — run plan September 28 · factory-ops

'New client' button — report September 28 · Claude (the lead)

Security review — 'New client' button September 28 · security-reviewer

Four-option quiz audit — read-only finding September 28 · Claude (the lead)

The four-hour agenda — evening pause plan September 23 · COO (Lisa 2.0)

Morning brief — Sunday, September 21 September 21 · Claude (the lead)

Good morning — what happened while you slept September 18 · Claude (the lead)

Step 5, in full — the MK11 regeneration breakdown September 18 · Claude (the lead)

Morning brief — Sept 15 (overnight) September 15 · Claude (vp-review)

Evening brief — Sept 12 (MK11 rebuild) September 12 · Claude (cfo + plain-english)

Daytime brief — Sept 12 (Queue 5) September 12 · Claude (cfo + plain-english)

Morning brief, Sept 11–12 overnight (Queue 4) September 12 · plain-english over the cfo report, 3:10 AM

Morning brief, Sept 10–11 overnight + Sept 11 afternoon September 11 · vp-review (Fable), answered by Claude at 9:45 PM

Your 17 notes on these briefs, and the answers, are kept with the project's written notes, off the board. They were a conversation, and it is closed — a dashboard is not an archive.

What the words on this board mean — 12 of them

The factory: The software that turns a company's documents into an online training course.
A portal: The website a company's new hires log in to, to take that course.
The pipeline: The assembly line inside the factory: read the documents, plan the course, write the lessons, write the quizzes, check every fact.
A run: One trip down that assembly line. Runs are the only thing that cost real money, because they use the AI.
The Console: Anthropic's own bill for the AI. When two numbers disagree, the Console wins.
The ledger: Our own notebook of what the AI was asked and what it should have cost. It is checked against the Console after every run.
A gate: An automatic check that stops bad work before money is spent on it. Most gates cost nothing.
Red-first: Before writing any new code, Claude writes a test that fails. Then makes it pass. It proves the test could actually catch a mistake.
Grounded: Every fact in a lesson points back to a page in the company's own documents. If it cannot, it is not allowed in.
Tenant: One client company, walled off from every other client in the same software.
Protected paths: The parts of the code that guard security and money. Claude may not change them without your yes.
A brief: The end-of-session report: what got done, what it cost, what needs you.

That's everything that happened. Back to Now.