atelier.space is grown by an autonomous loop that proposes, reviews and ships its own changes. This page reports the metered API spend per merged change, read live from the loop's own ledger when you request the page.
Updated 2026-07-25
atelier.space ships most of its changes without a person writing them. An autonomous loop files a proposal, builds it, sends it through an independent reviewer that is allowed to kill it, and merges what survives — nightly, under a written policy naming what it will never build. What that costs is a fair question to ask of anyone who claims it, so this page answers with the loop's own ledger rather than an estimate.
| Measure | Value |
|---|---|
| Window | 2026-07-01 through 2026-07-26 |
| Changes merged in the window | 218 |
| Metered API spend | $478.39 |
| Per merged change | $2.19 |
The spend figure counts the metered API lane only — a flat-rate subscription lane also powers runs and is not billed per token. The window ends at the last merge counted, not at today.
Over the whole run so far, 414 shipped changes have had their adoption re-examined about a week after landing, and 1 deploy was reverted by the canary.
Spend per merged change is a live read. It moves whenever the loop merges or spends, so this page asks the database at the moment you request it rather than baking a figure into the build. When the read fails, or the window holds too few merges to divide honestly, the figures are absent rather than estimated. The same two aggregates are public, so the arithmetic can be repeated without taking this page's word for it:
curl https://atelier.space/api/loopling/cost-summary
curl https://atelier.space/api/loopling/receiptscost-summary returns the window, how many changes merged inside it, the metered total, and the ratio between them. receipts returns two counts from the loop's outcome ledger: shipped changes whose adoption was re-examined about a week after landing, and deploys the system reverted on its own.
Source: server/routes/looplingCostSummary.ts and server/routes/looplingReceipts.ts hold the two reads. This page calls the same exported functions rather than a second query written for it, so the page and the endpoints cannot drift apart and quote different numbers.
The window opens on the day the spend ledger's silent under-reporting was repaired. Every day before that reads low, and a receipt that flatters is worse than no receipt.
It closes on the day of the last merge it counted, rather than on today. Both terms of the ratio freeze the moment the loop stops merging, so an open-ended window would age into a claim about the present — and it ages in the flattering direction, because a lane that stops spending while its last merges stay in the denominator drives the figure down. Reporting the right-hand edge lets a reader date the number instead of trusting that it is current.
Source: server/routes/looplingCostSummary.ts, where both boundaries and the minimum merge count are set, with the reasoning recorded beside them.
Until 2026-07-25 these rows were also mirrored to a separate public repository, and what made that arrangement worth something was that the dates were not ours: the deploy pipeline committed each row, so an outside service timestamped it. That repository is private now, and this page stands in its place. It is a first-party record and should be read as one.
What a reader can still check, without trusting the prose:
What is genuinely gone is third-party notarization. Nothing on this page is timestamped by anyone but us, and no amount of wording fixes that — it is recorded here so the gap is visible rather than papered over.