Task: Health review 2 — sprint 22 mid-sprint analysis

Mission: "Continue commissioning the ores.refdata entity library and correcting codegen C++ generation drift, carrying postponed sprint 21 stories." Two goals in practice: (1) entity commissioning (Qt/shell/CLI/manual per entity), (2) codegen drift correction.

Most active work maps cleanly to goal (1): currency, counterparty, party_status, book, purpose_type are all commission-epic stories. Goal (2) has one active story (composite hierarchy Qt widgets) but its large backlog (refactor_codegen_cpp, codegen-unification-blockers, codegen_ci_zero_diff, codegen_safety_guardrails — none STARTED) is untouched this sprint despite being called out explicitly in the mission. The market-data epic (synthetic data collections) is large and active but only tangentially related to either stated goal — it is carried sprint-20/21 momentum, not new-to-sprint-22 scope.

The most serious finding here isn't miscoverage of a goal, it's invisible coverage: Commission: business_day_convention_type (bright_faraday worktree, task_state=DONE per fleet) has no story or task document reachable from main — compass show fails to resolve both its story and task UUIDs even after a fresh index rebuild. This work legitimately serves goal (1) but is entirely unaccounted for in sprint tracking; if that worktree is lost or its branch abandoned, the work and the decision trail behind it disappear with it.

Day 3 (114 commits) was a clear outlier — more than 3x day 1's rate. Whichever projection is used, the sprint is trending over its 300-commit target; the only question is by how much. This is not a one-off spike: it lines up with the same period the drifted stories below started accumulating undocumented WIP, consistent with a sprint that took on more concurrent work than its own paperwork could keep up with.

45 PRs merged since sprint start (2026-07-02) — 12 on day 1, 15 on day 2, 14 on day 3, 4 so far on day 4. That is a strong raw throughput, comfortably above the ≥1/day floor.

But raw merge count hides a WIP problem: of the 10 fleet worktrees, 7 have no open PR at all, and several are carrying large unmerged/ unsynced deltas — clever_dijkstra (2 ahead, 29 behind), jolly_knuth (0 ahead, 27 behind), solid_dirac (17 ahead, 16 behind), eager_maxwell (10 ahead, 16 behind). A branch 27-29 commits behind main with no PR in flight is not "about to merge" — it is stale WIP that will cost a non-trivial rebase whenever it does surface.

Story-level DONE ratio across all 51 stories in sprint 22: 14 DONE / 51 total = 0.27 — below the 0.3 RED threshold.

Five stories with real, active branch work show a STARTED task but a story-level state of BACKLOG and "Now: Not yet started" — the exact symptom that triggered this review:

Separately, one story drifted the other direction: codegen-sql- variability-marketdata's own story.org says State=DONE, "All four tasks complete; PR pending", but sprint.org still lists it BACKLOG — completed work not yet reflected upward.

Follow-up (PR #1444): only Commission: party_status from the five above was fixed in this checkout, plus five unrelated stories this review didn't flag (book_status, refactor_codegen_cpp, codegen-unification-blockers, fix_orphan_docs_and_tag_taxonomy, qt-upgrade-6.9) that turned out to share the same BACKLOG-with-DONE- tasks symptom on a broader sweep. Commission: book, Commission: purpose_type, and Composite child-entity/hierarchy Qt widgets remain exactly as drifted as described above — their task docs are either absent or themselves still BACKLOG in this checkout, so they need fixing from the worktree that actually holds the work (clever_ dijkstra, eager_maxwell, solid_dirac respectively), not from here. Commission: business_day_convention_type remains entirely untracked.

Nominal STARTED count in sprint.org is 5 (Open sprint 22, Commission: currency, Commission: counterparty, Compass GitHub Actions support, Synthetic data collections). But fleet shows 9 distinct stories with live branch activity right now once the drifted ones and the untracked business_day_convention_type are counted — nearly double the visible number, spanning agile tooling, currency, counterparty, party_status, book, purpose_type, composite Qt widgets, synthetic data, and an entirely separate commissioning story invisible from sprint.org.

This task itself is one more data point: the counterparty Qt verification task is now BLOCKED because the composite-hierarchy- widgets work (solid_dirac) and the Commission: party work are touching the same Qt entity-detail code the counterparty task needs to verify — a direct collision between two of those nine concurrent threads.

Stories closed this sprint so far: 14 (party_type, compass QoL, client start/stop, market_series enum fix, NATS port fix, market.txt duplicate-key fix, FX status badge, GMM improvements, sub-registrars, SQL variability marketdata (bookkeeping pending), synthetic librarian FX, shell market-data import, canary CTest cancellation, macOS relevance-guard hotfix). That is a healthy closure rate for 3.5 elapsed days, but it is concentrated in small, well-scoped stories; none of the larger commission-epic stories (currency, counterparty) have closed, and the two codegen-epic goal stories named in the mission haven't been touched at all.

Overall: RED. Two RED dimensions alone would trigger this, and here there are three, plus two AMBER. The single most important concern is that sprint bookkeeping has stopped reflecting reality: five stories are being actively worked without their state reflecting it, two of those have no task documents at all, and one entire story (business_day_convention_type) is invisible from main. This isn't just a paperwork nuisance — it is the direct cause of the collision blocking the counterparty task right now, since nothing in sprint.org would have surfaced that composite-hierarchy-widgets and Commission: party were both actively touching the same Qt code before the collision was hit. The one fix that would most improve sprint health: stop starting new stories and spend a cycle reconciling every worktree's actual state back into its story/task docs — starting with the five drifted stories and the untracked business_day_convention_type story — before picking up anything else.