Task: Health review 1 — sprint 25 mid-sprint analysis

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

Challenge: "Progress IR curves" is covered but the IR curve bootstrapping story has been STARTED since sprint 24 (carried forward) — the sprint table says BACKLOG while the story file says STARTED. Are we actually making forward progress on IR curves, or has momentum stalled? The IR rates story has 19 tasks (some DONE from prior sprints) but is STARTED with no end date — progress is opaque.

The unmapped stories are not noise — they're substantive: holding company treasury (6 tasks, 4 DONE), NATS config passthrough (9 tasks), synthetic data scope (4 tasks). They advance the product but don't serve the sprint mission. This is characteristic of a sprint that pulled in backlog items opportunistically rather than staying focused. Not harmful at day 5 given the drift goal is well-served, but worth watching at close.

The commit rate is sustainable — no signs of bursts followed by stalls. Day-over-day rate has been consistent. The load target is easily met.

Verdict: GREEN. Merge rate is healthy and there's no WIP accumulation. The projected total of 70 is under the 100–120 soft target, which may reflect the sprint being front-loaded with smaller, focussed PRs rather than a genuine velocity problem.

DONE/BACKLOG+STARTED ratio: 13 DONE / (7 STARTED + 7 BACKLOG) = 0.93 → GREEN on ratio.

However, three concerns:

1. Sprawling stories: Refactor codegen C++ (21 tasks, STARTED for 69 days) and IR Rates (19 tasks, STARTED for 69 days) are classic sprawling stories — many tasks DONE but the story itself not closed. Both were carried from sprint 24. The DONE work is real and delivered, but keeping them STARTED makes the sprint look like it has more open work than it actually does. Both should be evaluated against their Acceptance criteria — if the DONE tasks satisfy the criteria, close them and split remaining tasks into new stories.
2. Ready-to-close stories closed: Investigate grants (1/1 DONE) and Normalise NATS protocol (1/1 DONE) were closed in the bookkeeping pass accompanying this review — both now DONE in story files and sprint table.
3. New undecomposed stories: Fix codegen template drift, Close systemic codegen gaps, IR curve follow-ups, Entity classification, and two hotfix stories were added to the sprint table but not decomposed into tasks — they show BACKLOG with no tasks at all. This is acceptable for stories promoted late in the sprint but they count as undecomposed work.

7 simultaneously STARTED stories (was 9; 2 closed in bookkeeping pass) across 6 distinct themes far exceeds the GREEN threshold of 1–3. Several are owned by other environments (bright faraday has 3, solid dirac has 1, swift curie has 1, festive dijkstra has 2, prime origin has 1). The fleet is spread thin — brave hopper alone has 1 STARTED story (Docs improvements) after the bookkeeping pass.

This is the sprint's most significant risk. Focus is sacrificed incrementally — each "just one more" story compounds context-switching cost. The diversity of themes (finance, quant, infra, docs, NATS) means no one person can hold the full sprint in their head.

The 5 table disagreements are a pattern: stories that were promoted or started mid-sprint had their story files updated but the sprint table wasn't synced. Three of the 5 are carried-forward stories (IR curve, IR rates, synthetic data scope) whose sprint table entries were never updated when the stories were carried. The 12 stale Now fields on closed docs are a known pattern — DONE stories often have a summary in Now that isn't cleared to "Nothing."

The sprint is productive — drift resolution is happening at pace (50 PRs, 383 commits). The focus signal improved from RED to AMBER during the accompanying bookkeeping pass (2 stories closed, STARTED count 9→7). The most important remaining fix: evaluate the two sprawling stories (Refactor codegen C++, IR Rates) for split-and-close. The 5 sprint table disagreements and stale Now fields have been resolved.