Task: Health review 1 — sprint 27 close analysis

The sprint carries a mission but no separate goal list: "Finish the UX pass: land the remaining journey groups in ores.web, and bring the record screens to the standard." Read as two goals, both advanced and neither closed. The sprint's own * Stories table mixes Product and Hotfix work in one table, so the mapping below is by the mission's two clauses.

Verdict AMBER: each goal advanced; each holds an open remainder. The System 2 challenge is that the mission covers only part of the sprint. Around a third of the table — the money-exactness story, the curve-support story, the rates and credit instrument reworks, the cleandown of ores.scheduler and ores.synthetic, the drift-gate and canonical-protocol stories, the isolation stories and roughly forty hotfixes — serves no clause of the mission. That is not automatically wrong for a sprint that also acts as the release vehicle, but it means the mission does not describe what the sprint spent its capacity on.

The tree carries about three and a half times the commit ceiling. The load is not uneven in the way sprint 26's was: there is no idle day and no single burst. The rate rises to a peak of 560 on 2026-10-08 and stays between 277 and 459 on every other day. This is a sustained overrun, not a stall and not a spike.

Verdict RED: velocity is the load's twin. The sprint merged about five times the planned PR count, and no WIP piled up — but the cycle sample is a mediocre signal: if the true median really is around half a day, it is materially slower than sprint 26's median of 0.3 h, and the sample's own cap is a defect in the chart collector worth fixing.

The six undecomposed story folders:

• Paint every web badge from the catalogue
• Session activity: record the size of every NATS message per server
• Keep counting failed attempts on a locked account
• Link every reference to its record
• Move the installation's remaining reads onto its own class
• Draw the menu by reads alone (DONE, closed with no tasks)

Verdict AMBER: the ratio is comfortably green, but six stories carry no tasks, five of them BACKLOG, so no one can see their shape.

Verdict RED: thirty-one stories remain in flight across at least seven themes, and eleven other worktrees hold their own branches at the same time. The close-of-five reduced the count by only five. The count is not the whole story: several of the STARTED stories are the decade-long infrastructure programme (clean standards, drift gate, canonical protocol) that the sprint carries every time, so the sprint's focus is diluted by construction rather than by accident.

Verdict AMBER: throughput stayed high to the last day — 314 commits and 55 merged PRs on 10-13 alone — but the sprint closes with about a quarter of its stories and tasks open, one day past its planned window.

The audit named six stories whose every task is DONE. Each was read against its own * Acceptance before closing. Five satisfy it and are now DONE. One is a false positive.

Three Now cells on the five closed stories held in-progress prose ("The toggle is built…", "The branch … is ready for review", "Design in review (PR #3115)"). The prose is superseded by the merges above, so it is dropped and Now is set to exactly Nothing. The split-accounts note named PR #3115; the work merged as #3124.

• 18 sprint-table rows carried a State that disagreed with the story file. Each was reconciled to the story file's own State, except where the story was closed above. Thirteen rows were BACKLOG in the table against STARTED in the file (the DQ artefact tables, the existing-screens sweep, the four-eyes framework, the standard party page, the record-screen updates, the change-event key, the feed organisation, the ACME reports, the trading guards, the grid samples, list ordering, the ER-diagram split, market-data bindings); one row was BACKLOG against DONE (the IAM isolation gaps); the two DONE rows against STARTED are two of the five stories closed above.
• Three duplicate story rows were removed: 3541A81F (super administrator journeys) appeared twice with opposite states, and D7A13BC8 (polish the audit sign-ins screen) appeared three times. The richer, DONE row survives.
• The blank lines that split the * Stories table into four fragments were removed, so the table parses as one.
• The four chart PNGs were generated. The sprint page linked prs_commits.png, line_churn.png, pr_cycle.png and stories_done.png, and none existed, so every chart link on the page was broken before this review.
• The review story row was added to the sprint's * Stories table.

The runbook's timeline step is not applied. Only sprints 20 and 21 carry a timeline/ directory; sprints 22 through 26 do not, so the convention is retired, and generating buckets from 2026-10-05 to now would add roughly 580 files to a sprint the current convention keeps timeline-free. The decision is recorded here rather than acted on.

./compass.sh sprint audit at 82d59f0116 reported 48 findings before the repairs: 6 stories ready to close, 8 open tasks whose pull requests have all merged, 18 table/file drifts and 16 closed documents whose Now is not Nothing.. After the repairs the same command reports 26.

• The 18 table/file drifts are reconciled, except the one row this review itself introduced, which is now closed.
• The six stories the audit called ready are resolved: five closed DONE on merged evidence, one rejected (above).
• The 8 open tasks whose pull requests have all merged were each read. Every one is genuinely open, and each names its remainder: 37289A3C (paging, people links, history timeline), 8063C1A9 (four clean-standard items: B06, V08, P03, V04), 15D464C7 (the literate-template pilot is paused after one file), 775FD373 (not started), A13916C5 (not started, no PR recorded), 48943E90 (not started), 11AE607C (design agreed, build not started) and 2A5A2501 (not started). The audit's heuristic keys on the #+pr: field alone and cannot see that the task is only partly delivered by that PR, so it will keep reporting them until the tasks close.
• The 16 Now prose findings on already closed documents are historical records. This review leaves them, as sprint 26 did before its own close, and records them for the closure pass.

The runbook's shape pass over every DONE task is measured and deliberately deferred to the closure pass, because the volume exceeds what can be resolved accurately here. Over the sprint's task_*.org files at 82d59f0116:

Filling these from the task's own * PRs table is safe and mechanical, but it covers only 15 of 141. Deriving the rest needs a per-task walk of git history and gh: the Task-ID: commit trailer exists but appears in only about a third of recent commits, so it cannot be trusted as a complete key. Writing none for all 126 would assert something false for an unknown share of them. The honest move is the measurement above plus the closure pass, not 126 guesses.

Sprint 27 has the same condition as sprint 26: not a stall, a scale problem. It ran three and a half times the commit ceiling, about five times the PR band, and one day past its window, while keeping its merge cycle short and its WIP at zero. Both mission goals advanced and neither closed, thirty-one stories remain in flight across at least seven themes, and the mission describes only part of what the sprint did. The widest fix available at the close is the one this review executed: reconcile the bookkeeping, close what is genuinely finished, and carry the rest honestly. The next lever is focus — reduce the in-flight count and stop carrying the standing infrastructure programme in the same sprint as the UX pass. Sprint load, PR velocity and focus signal are RED together, so the overall verdict is RED.

The close split every one of the 42 open stories into the part sprint 27 delivered and the part it did not.

• 21 stories were reworked and closed DONE. Each had its Goal and Acceptance narrowed to the shipped scope, and its undelivered tasks moved into a new successor story in the product backlog inbox. 20 successors were created; one closed story — Hotfix: the seeded books have no purpose — had no remainder and needed none.
• 21 stories were carried whole to the inbox. Sixteen had no delivered task at all. Five more had only a scaffolding or story-writing task DONE, which does not satisfy the story's own Acceptance, so they were carried rather than closed: build the retire or reset tenant journey, raise a notification for a scenario waiting for a tester, restore the trading guards the versioned model dropped, seed a relative-time trade population with the market data it needs, and trace a process run and the trade groups it creates.
• 93 tasks moved with their successors. Each closed story names its successor in its * Decisions; each carried story records that it came from sprint 27 at close.

After the pass, sprint 27 holds 147 story files (146 DONE, 1 ABANDONED) and 575 DONE tasks — no open story and no open task. The inbox grew from 326 to 367 entries. The sprint table reads 146 DONE and 22 ABANDONED: the 21 carries plus the abandoned workspaces story. compass sprint audit fell from 26 findings to 16; the 16 are the historical Now prose this review already recorded. The four charts were regenerated at the close.