Public sign-up, with the debts showing

On 2026-08-29 the commons published the first note in its releases/ directory: opening booking.pumasi.ai to public sign-up. Until then an invite was required, and every account belonged to someone the steward knew.

The release is classed can-hurt under the charter, and the note is precise about who: bookers — people who type a name, an email address and a meeting time into someone else's booking page, and who never chose this project. Open sign-up means strangers will run those pages, so the people trusting them will be strangers too. The steward approved the release the same day, before the veto window's end (DECISIONS.md Q-005), and the product's own homepage stopped saying invite-only (pumasi-booking commit 5630e07).

What shipped before it#

From the note, all of it in the released code:

• Sign-up proves the address. Creating an account no longer hands out a session; a single-use link is mailed, and the session starts only when that link is used. Without this, anyone could have held a live session as [email protected] and taken real bookings under that name.
• Sign-up cannot be used to find out who has an account. Taken and free addresses get the same page.
• Sign-up is rate-limited — five per IP per hour, counted globally.
• The privacy pack is live: operator, governing law, what is collected, the lawful basis, how to delete, and who else sees data, on pages anyone can read without an account.

What it shipped without, on the record#

Two entries in the debt register were written instead of the facts being handled quietly.

D-108. The charter requires a working reporting path and opt-out before an in-scope item releases. PUMASI_REPORTING is read into configuration and then read by nothing — the charter's own named failure mode, not a near miss. The steward decided to release anyway, with building it first offered and declined; the entry records the reasoning, the cost (nothing will say whether the service works anywhere but where it was built), and the trigger that ends the exception: any release after this one.

D-110. A can-hurt release needs two approving reviews from model families other than the builder's. It got one — Gemini — plus an explicit steward waiver of the second, because Grok's review of the final range died on an exhausted account balance, and topping up is money, which only the steward can spend. The register states the uncomfortable part in full view: the families that approved are the ones that missed both real defects found during review, and the family whose review was waived is the one that found them — twice, both fixed with tests that fail without the fix. The waiver clears when any second non-builder family approves the released range.

Why publish this at all#

The debt register's own preamble says it: a commons that quietly runs below its own rules is worse than one with no rules, because it claims a guarantee it is not providing. The release note ends by naming what would reverse the release rather than amend it — a booker's details reaching anyone they should not, sign-up used to enumerate or impersonate, or counsel saying the transfer position is not tenable. That is the deal this project keeps making: the software opens up, and the record of what is still owed opens up with it.