Task: Health review 2 — sprint 25 close analysis

Sprint mission: "Resolve sprint drift on as many components as possible and progress IR curves work."

Drift. The program story (Entity classification and drift baseline across all components) is DONE. Four components are drift-free and closed: dq, iam, synthetic and compute. This review listed trading as a fifth; that is wrong. A measurement on 2026-09-26 finds 1751 files of drift in ores.trading, and the sprint 25 story said the component could not join the drift gate. The release notes count five drift-free components because they also count analytics, which is drift-free while its story stays STARTED, as the next sentence says. The redundant-properties pass is also DONE. analytics and reporting have all tasks DONE, but their stories are still STARTED. marketdata is in flight. database, scheduler, workflow and workspace are in BACKLOG. The codegen template fixes, the systemic gap fixes and the has_parent_id fix also serve this goal. This goal is well served.

IR curves. The FOMC short end (DONE) and the oresmd quote-type coverage (DONE) shipped by 2026-08-11. After that date, the IR stories moved slowly:

• IR curve bootstrapping + official curve republish: 5/8 tasks, carried from sprint 24. The engine, the official curve republish (#1879) and the lineage shipped. The open tasks are Qt UI tasks.
• IR Rates synthetic data: 11/20 tasks, carried from sprint 24.
• Add new stochastic processes: 3/11 tasks. Five of its BACKLOG tasks have a merged PR (#1932). The real progress is higher than the file shows.
• IR curve follow-ups: 6/6 tasks DONE, story still STARTED.

The headline deliverable (a bootstrapped curve that is republished as an official stream) shipped before sprint 25 started. In sprint 25 the IR goal got the FOMC short end, six new process engines with no wiring, and bookkeeping that lagged behind the work.

No mapped goal. From 2026-09-19, the work moved to a new direction: Retire Qt, Integrate ores.web, Prototype bootstrap in ores.web, IAM entities in the web client, IAM journeys, canonical NATS entity protocol and the two TypeScript protocol stories. This is a large, strategic change: the product lost its desktop client. It is good work, but sprint 25's mission does not mention it. The skills catalogue (21/51 tasks), the trade entry prototype and the ORE v17 update also have no mapped goal.

Challenge: some of the web work uses codegen, so it is possible to say that it "serves drift". That mapping is wishful. The web pivot is a different goal. It needs its own sprint mission.

The average hides the shape of the sprint. There are three phases:

1. Days 1–10: a normal, heavy sprint. 1225 commits. This phase alone is slightly over the target.
2. Days 11–32: a lull of 22 days with 169 commits. On 8 of these days there were 0 commits, and on 7 days there were 1 to 8.
3. Days 33–52: a second heavy phase with 1461 commits. This phase alone is over the target.

Phases 1 and 3 are each a full sprint. The sprint did not close at the lull, so the second phase went into the same document. This is the root cause of most other findings in this review.

The release notes count 316 PRs; this review counts 319. Both are right for their window. The 319 is every PR merged on or after 2026-08-03, read before #2132 and #2133 merged. The release notes count PRs merged into main after the v0.0.24 tag (2026-08-03 16:05): that excludes #1810 to #1813, merged earlier that day for the sprint 24 close, and #2084, merged into a feature branch, and it includes #2132 and #2133. 319 - 5 + 2 = 316.

PRs merge fast. The median cycle time is under 2 hours, and there is no queue of open PRs. The total is 319 against a target of 100–120, so the verdict is RED. The cause is the sprint length, not the merge process.

WIP signal. merry_newton has 6 unmerged commits on feature/identity-core-series-and-bindings. They are 29 days old and 1608 commits behind main. They belong to Adopt oresmd as the native market data identity (2/9 tasks). A rebase of this branch will be expensive. Decide at close: rebase it, or abandon it and record why.

The ratio is GREEN. The story files are not. compass sprint audit reports 99 findings:

• 10 stories have every task DONE but are not closed: Absorb PR checks locally, analytics bind, reporting bind, eventing flake hotfix, harden eventing chain, feed lifecycle harmonization, generate every protocol message as TypeScript, IR curve follow-ups, the provisioning bundle hotfix and verify compute-grid flows from the UI.
• 12 open tasks have all their PRs merged.
• 11 rows in the sprint table do not agree with the story file.
• 1 DONE story (Compass QoL follow-ups) has 3 BACKLOG tasks.
• The other findings are closed documents with a Now that is not "Nothing.".

Sprawling stories. Four stories are over the sprawl threshold. Apply the split test to each one at close. Read its * Acceptance and compare each open task to it. Do not close a story only because most tasks are DONE.

• Refactor ores.codegen C++ generation: open for 116 days and over three sprints. The drift program now does most of its work. It is probably a close-and-split candidate.
• Unify the LLM skills catalogue: 30 open tasks. It is probably two stories: the part that shipped (the compass- namespace), and a remainder.
• Redesign ores.trading and IR Rates synthetic data: check against acceptance. They possibly carry forward.

Challenge: the Market Simulator client story describes a Qt window (MarketSimulatorWindow, IrCurveEditor). Qt is gone. This story is not only undecomposed. Its premise is out of date. Move it back to the product backlog, or abandon it.

Themes in flight: codegen drift, TypeScript and ores.web, NATS protocol, IR curves and quant, market data identity, trading data model, CI and eventing hardening, skills and documentation, and the ORE v17 update.

The mid-sprint review found 7 STARTED stories and rated focus AMBER. The count is now 31. Part of the cause is bookkeeping: 8 STARTED stories have every task DONE, and 12 open tasks have merged PRs. The real WIP is smaller than 31, but the documents do not show its size. That is a focus problem too.

Delivery is high. 39 product stories closed, and some are large: the removal of Qt, ores.web, five drift-free components, oresmd coverage, and the FOMC short end. But 24 hotfixes is a warning. Most came from code that compiled on Linux clang and failed on MSVC, gcc or macOS. The local-checks story (Absorb PR checks locally) has every task DONE. Check at close whether its checks include MSVC and gcc. If not, file a capture for a cross-compiler check before merge.

The * Achievements section in sprint.org is empty. The * Captures section is empty. The release notes will have to come from the PR list only.

Sprint 25 delivered a lot of work. But it ran as two sprints in one document: a drift and IR sprint in August, a 22-day lull, and a web and TypeScript sprint in September. The mission stayed the same through all three phases. Because the sprint did not close at the lull, the load, PR and focus numbers are all RED, and the documents show 31 STARTED stories where the real WIP is much smaller. The most important concern is that the documents no longer describe the work. The one fix that helps most is a full bookkeeping pass before the close: close the 10 finished stories, apply the split test to the 4 sprawling stories, decide on the merry_newton branch, and give sprint 26 a mission that names the web-client direction. After that, keep to a 7-day close, so that a lull ends a sprint instead of stretching it.