Portal Factory · the board
How much is left, and how long does it last?
$46.67 left — about 3.7 days
How much is left
At $12.71 a day — and the bill's daily figures stop 57 days ago, so anything spent since is not in that rate.
Credit left on the account. Read by hand on Oct 14, 1:15 AM ET— 26 days ago
How this number is made — 2 things
$46.67 ÷ $12.71 a day = 3.7 days. The 7 days run September 6–September 12, and they stop 57 days ago.
The unspent balance on the API account — only the Anthropic console shows it, so 'not in the ledger' is a ceiling, not a bill.
Talking voice: 35,571 letters left of 137,673
About 12 lessons read out loud, or 14 practice sessions. 102,102 used since the month began; it all starts over on November 1.
read 33 days ago, by hand — anything spoken since is not counted in it, so the real figure is this or lower — never higher
There is no way to buy more. When the letters run out the voice stops in the middle of a lesson, on a customer — you are not charged extra. 6,884 are held back so a demo never lands on empty, leaving 28,687 to spend.
One client of ordinary size would need about 6.2 times this whole month. A lesson read aloud costs about $0.34 of it; a practice session $0.50 all in, 59% of that the voice.
What goes on the card every month
Not one line below has ever been seen on a bill. Each says which it is instead — your own words, written down or worked out, never billed — and none of them are counted in the spending further down, which is model calls only.
- The talking voice $6.00 reads lessons aloud, speaks the practice partner · your own words, written down · confirmed Sep 15
- The machine the portals run on $5.92 one small always-on machine · worked out, never billed · confirmed Sep 15
- Database, document storage and background jobs $0.00 free plans, measured to be well inside them · worked out, never billed · confirmed Sep 15
- Taking card payments, and sending email not set up nobody can be charged, and nothing is being sent · nothing to bill
Keeping the portals running costs $11.92 a month, on top of the model calls — and 2 of those 3 figures are worked out rather than billed.
Not part of running the portals
- The Command Center site $9.59 a different product, on a different host · your own words, written down · no date on file — only you can say when
Kept out of the figure above on purpose: business overhead, never a cost of serving a portal client.
This week, and today
The bill, this week so far
not in yet
The Console's day-by-day figures stop at September 12, so there is nothing for this week yet. This is not a zero — it is no reading.
Since the last brief
$27.87
Video pictures (Google image generation, by our ledger, an estimate): $14.74 through Oct 7, plus about $3.43 of Module 3 pictures on Oct 10, about $18.17 in all of the $35 image budget Alyssa set on Oct 7. Narration (ElevenLabs) is now on the Pro tier: 624,220 characters available, resetting Nov 10; Module 3 lessons 1 and 3 used about 13,100. Not in that figure, by the work log (estimates, not the Console): about $1.41 of arcade cover art and about $0.24 of avatar tests on Oct 12 and 13, and about $1.12 of Ask-tutor checks on Oct 12. The Anthropic Console balance on this board was last read on Oct 14 (1:15 AM ET) and is carried forward, not re-read; the automatic reader could not get into the Console page on Oct 16. The October 13 and 14 security work ran no paid model calls; its only cost was hosting releases. Tonight (Oct 18), by our ledger and an estimate rather than the Console: about $2.02 for the nine new MK11 lessons and two quizzes written on the laptop, inside the $7.60 high estimate and the $30 cap Alyssa set. That run was still going at last report. Oct 19: the glossary summary trials cost about $0.17 by our ledger (not the Console), capped at $4; a later trial was interrupted by the Docker outage. The Console balance has still not been re-read since Oct 14. Oct 21: nothing paid today (no model calls, no image or narration runs); release 105 cost only hosting. The Console balance has still not been re-read since Oct 14. Oct 22 night run (12:00 AM to about 3:57 AM): nothing paid (no model calls, no image or narration runs; the reviews and the document conversions ran on this machine). The Console balance has still not been re-read since Oct 14. Oct 22 morning (to about 7:43 AM): nothing paid (no model calls, no image or narration runs); release 106 was built on this Mac. The Console balance has still not been re-read since Oct 14. Oct 23 evening: release 110 (interview role plays) cost only hosting; no model calls, image or narration runs were bought for the release. The Console balance has still not been re-read since Oct 14. Oct 24: release 111, the manual upload and the What's New post cost only hosting; no model calls, image or narration runs. The Console balance has still not been re-read since Oct 14.
What the whole thing has cost
Paid in
$363.19
13 purchases, August 20 – October 2.
Spent
$316.52
Paid in minus what is still on the account. Built on the Console balance read 26 days ago — not today's balance.
Costs outside any ledger — subscriptions, the domain, hosting — until Alyssa enters them.
Where it went
By month — the bill, with our own notebook beside it
By kind of work
Model calls that failed before returning a token count (147 rows) — billed or not, their cost is unknown to the ledger.
Calls made by test and eval runs whose rows were deleted at teardown, and one paid batch orphaned on Aug 20.
How this is counted — 4 things worth knowing
The Console is Anthropic's bill and it wins every disagreement. The ledger is the notebook the factory keeps of what it asked for and what that should have cost; it is checked against the Console after every run. Where the two differ, the difference is shown rather than hidden.
Money you put in
- August 20
- August 20
- August 21
- August 22
- August 25
- September 4
- September 4
- September 5
- September 7
- September 12
- September 18
- October 1
- October 2
Where the difference went
- Aug 19–21: $68.81 on the factory key before the ledger existed — and the two database resets erased the rows that did exist.
- A nightly cloud eval ran every day at 09:00 UTC with the real key, on a database that dies with the run, and FAILED every day since Sep 5 — about $1–4 a day, plus five push-triggered runs on Sep 6–7. Switched off Sept 11; re-enable only after a green on-demand run.
- Sep 5–7 overnight work under-recorded: interactive calls priced at the batch rate (fixed), rows deleted at test teardown, 147 failed calls with no token count.
- Almost no batch discount was ever earned: $2 off $260 — the pipeline ran interactively nearly every time. The stage scripts now default to the Batch API.
Money Alyssa put into the API account (credit purchases, tax included), entered by her on Sept 11.
Our notebook was last measured Nov 8, 10:54 PM ET. Subscriptions, the domain and hosting are in no ledger the factory can read; the monthly ones are listed higher up this page, from what you have told us and from arithmetic, and they are deliberately not added into the figures in this section, which count model calls only.
Other things this board cannot see
- The AI illustration tool (Gamma) runs on its own prepaid image credits, separate from everything above — tonight's own working notes say about 428 of 2,000 are left, roughly six more illustrations, but that figure is reported by the session, not re-read from the vendor the way the talking-voice number above is, and there is no line for it here yet.
- The local database is offline tonight. Nothing customer-facing breaks, but the new fact-checker teammate cannot do its job — tracing a slide's claims back to the client's documents — until it is back.
That's the money. Back to Now.