[rusholme]

The day-one page's "What needs Project-Admin today" brought to the tips of rapid_systems main (a17cd61) and rapid rebuild (ca322be7), read 2026-10-27.

Changes, each with its source

• The Batch row now says a night runs under your own workstation's instance role with no administrator profile, naming rapid_systems docs/workstation-submission.md rather than repeating it. Read from that page's What you need (lines 11–20: all five workstation roles on the roster since 2026-10-26, "no AWS_PROFILE, no administrator session") and rapid docs/foundation-smdc-step.md Current procedure, The environment ("A night runs on a workstation under its own instance role … An administrator profile is needed only for the image … and for a statement as postgres"). The clause about adding a role to the roster stays, reworded for a new member's workstation: cloudformation/rapid-scratch-identities.yaml line 115–117, WorkstationSubmitterRoleNames, is a stack parameter.
• The roster bullet under Open is removed: the same What you need, lines 13–16, and rapid-scratch-identities.yaml line 117 (the parameter default lists all five roles).
• The lead-in date becomes 2026-10-27, the date of this read.

Kept, with the reason

• The second-admin bullet: rapid_systems docs/team-access-policy.md lines 30–32 ("A second Project-Admin is required as a backup … no one is designated yet").
• The access-question bullet: rapid_docs system/migration.md lines 171–174 still list Access and "who builds the next one".
• The image and schema rows: the release hook still passes CAPABILITY_NAMED_IAM (cloudformation/release-hooks/deploy line 69), and Current procedure still names the administrator profile for the image and for postgres.
• "about five minutes and needs 16 GB of memory": task 1 of this brief ended BLOCKED and merged no README PR, so the README's numbers, and this page's, are unchanged.

Files

• project/day-one.md

Measurements

None; a docs read.

Improvements seen

• Task 1 of this brief found that the README's one-epoch walkthrough no longer passes on the newest image: the SFFT difference job exits 70 locally (botocore NoRegionError fetching the RuBR-AT model through SSM since rapid PR 493), and the README's newest-tag query prints one tag per 100-image page. This page's "One epoch locally … anyone with the two checkouts and a RAPID host" row and its One epoch section rest on that walkthrough. They are left as they are: this PR is scoped to Project-Admin, and the README tip has not changed. They follow whatever fix comes of task 1's finding.

Session notes

Task 2 ran even though task 1 was BLOCKED. Its only dependence on task 1, the README's numbers, is conditional on a merged README PR, and there is none.

Anything left

Task 1's walkthrough failure, above.

🤖 Generated with Claude Code