Task: Health review 1 — sprint 29 overrun analysis

The mission has one goal: finish the work carried out of sprint 28 before drawing anything new from the product backlog.

Goal	Coverage	Stories (DONE/STARTED/BACKLOG/ABANDONED)	Verdict
Finish the carried work	58 stories, 127 of 332 tasks DONE	3 / 13 / 41 / 1	RED
Draw nothing new	68 new stories, 380 of 491 tasks DONE	32 / 25 / 10 / 1	RED

The sprint did the opposite of its mission. It finished 3 of about 58 carried stories (5%) and 32 of 68 new stories (47%). The carried count is an estimate. It counts a story as carried when its #+created: date is before the sprint or its title ends "the remaining work". A cross-check supports the size: the sprint 28 close sent 62 stories to the inbox, 37 whole and 25 split.

Some new work has a clear cause and does not break the mission. Hotfixes and CI repairs are examples. The * Hotfixes table does not show how many: it has 34 rows, and many of them are features, such as "Support volatility surfaces" and "Hold the session at the door". The table name does not separate unplanned repairs from new scope.

The HTTP server arc, the volatility surfaces and the IR curves are new capability that no carried story asked for. The mission was not revised to admit them. Either the mission was wrong or the sprint ignored it.

Verdict RED.

Metric	Value	Target	Status
Commits since 2026-10-24	4961	≤ 1000	RED
Average	310 per day over 16 days	≤ 143 per day	RED
Commits to the planned end	3434	≤ 1000	RED
Window	day 16 of a planned 7	7 days	RED

Commits per day from 2026-10-24: 234, 662, 249, 488, 536, 245, 438, 582, 411, 345, 381, 151, 19, 85, 129, 6. The last figure is a part day.

The sprint is five times over the commit ceiling, and it was three times over at its planned end. A projection has no use because the planned end has passed. The rate fell sharply after 2026-11-03: the four days from 2026-11-04 to 2026-11-07 average 96 commits. The data does not show whether work stopped or moved to unmerged branches.

Part of the count is bookkeeping: 1629 of the 4155 commits that are not merges (39%) have the [agile] prefix. Without them the sprint is still about 2500 commits. The 2026-10-24 count includes the close of sprint 28 before sprint 29 opened at 09:16.

Verdict RED.

Metric	Value	Notes
Merged PRs	786	Over six times the 100–120 band. The first day overlaps the sprint 28 close.
Rate	49 per day	146 at the peak on 2026-10-31, 3 at the low on 2026-11-05.
Days with no merge	0
Documentation PRs	199 of 786	The [agile], [doc] and [docs] titles.
Merge cycle	median 0.06 h, p90 3.28 h, max 79.6 h	The median is about 3 minutes.
Open PRs	2	#4492 is new. #4471 opened on 2026-11-06.
Long-open WIP	1	#4471 is over a day old. Its only review came before a full rebase on 2026-11-07.

A three-minute median is less than half of the 0.14 h of sprint 28. Most PRs merge with no time for a review before the merge. So the PR count measures how commits are grouped, and not how much change a reviewer read.

The fleet shows wider WIP than the two open PRs. Seven worktrees on a branch are 369 or more commits behind main. These figures use each worktree's last fetch. A branch with no PR is invisible to this metric.

Verdict RED on the count, AMBER on flow. The skill's rule puts a total over about 150 at RED.

Metric	Value	Notes
Stories	126: 35 DONE, 38 STARTED, 51 BACKLOG, 2 ABANDONED
DONE : open	35 : 89 ≈ 0.39	The skill's AMBER band is 0.3–0.5.
Tasks	823: 507 DONE, 250 BACKLOG, 45 STARTED, 21 ABANDONED	62% DONE.
Undecomposed stories	3	All BACKLOG.
Blank Acceptance	11	3 of them are DONE, so nothing states what they promised.
STARTED with no task DONE	5	Listed below.

The undecomposed stories:

• Bring the FX spot conventions onto the market standard
• Draw every coded value as its badge — the remaining work
• Read a batch of rows in one query in the generated services

The STARTED stories with no task DONE, all created before the sprint:

Story	State	Tasks	Age	Notes
Author the report configurations on guided screens	STARTED	7	20 days	4 STARTED, 3 BACKLOG.
Complete the ACME report inputs and prove they run	STARTED	9	26 days	1 STARTED, 8 BACKLOG.
Draw the node heatmap and carry the engine run statistics	STARTED	6	26 days	2 STARTED, 4 BACKLOG.
Follow the child run a workflow hands off	STARTED	2	22 days	2 STARTED.
Scope a report by regulatory book type	STARTED	2	18 days	1 STARTED, 1 BACKLOG.

Oldest STARTED story: "Account for the grid's usage", 32 days, 1 of 3 tasks DONE.

Sprawling stories. Each row applies the skill's split test: the open tasks are read against the story's own Acceptance. The test used the task titles and the Acceptance text. It did not read each task's * Result or the code, so each row is a recommendation for the owner.

Story	Tasks	DONE	Open	Reading	Recommendation
Rework the market data feeds	65	43	21	Several open tasks are the unmet Acceptance: the rename, the marks of the simulator, the screens, the shell and the side-by-side check. The sensitivity and greek ladder tasks appear in no criterion.	Partial split: move the ladders, keep the story STARTED.
Replace sqlgen and reflect-cpp	62	58	3	Removing reflect-cpp and measuring the build are Acceptance. The pgbouncer task is not.	Partial split: move pgbouncer, carry the rest.
Support volatility surfaces	34	23	11	The Acceptance asks only for the architecture and the named tasks. The open tasks are the build.	Close and split, after a check that the architecture page exists.
Make every configuration screen follow one standard	27	21	6	Sign-off of a change that alters the numbers is Acceptance. The other five open tasks are not.	Partial split.
Populate the ACME books	27	24	1	The open wizard task is in no criterion. The sprint table already says DONE and the file says STARTED.	Close and split, after a check of each criterion.
Prove the report grid run end to end	26	25	1	A run on generated trades is in no criterion.	Close and split, after a check of each criterion.
Carry a market-wide arrival regime	23	21	1	The volatility loading is in no criterion. "Let a busy market move prices" already covers that subject.	Close, and move the task.
Split the types that carry too many columns	12	9	2	Both open tasks are the "every type is split" criterion.	Carry.
Build the IR curves of every supported currency	11	4	6	The open tasks are the core deliverable.	Carry.

Verdict AMBER by the skill's rule. To close the eight STARTED stories that have every task closed would move the ratio to about 0.53.

Metric	Value	Verdict
STARTED stories	38	RED: the skill's red line is 6.
Stories with a STARTED task	24	9 of these are BACKLOG stories.
STARTED stories over 7 days old	28 of 38	An estimate from #+created:.
STARTED with all tasks closed	8	They need a close, not more work.
STARTED tasks last touched before the sprint	32 of 45	Most STARTED states are stale.
Worktrees	15
Themes	market data, web UX, reporting, trading, conventions, codegen, HTTP server, workflow	Divergent.

The count rose from 27 at the sprint 28 review to 38. That review named a limit on stories in flight as the one change to make. Nothing enforced it, and the sprint started more stories.

The stale states overstate the real concurrency. Tasks touched or created since 2026-11-03 fall in 18 stories, and that is still three times the red line. This reading is weak, because many documents do not update Last touched.

Verdict RED.

Metric	Value	Notes
Stories closed DONE	35 of 126	28%. Sprint 28 had 48% at its review.
Tasks closed DONE	507 of 823	62%. Sprint 28 had 58%.
Tasks in the sprint	823	534 in sprint 28.
Tasks created during the sprint	617 of 823	By #+created:.
Window	day 16 of 7	Eight days over.

The sprint finishes tasks at a good rate and finishes stories at a poor one. It creates tasks faster than it can close its stories, so a story stays open while its scope grows. Eight stories only need a close, and that would raise the figure to 43 (34%).

The skill gives no threshold for this dimension. The verdict RED is a judgement: story closure is well below the sprint 28 figure, which that review rated AMBER, and the sprint is eight days late.

This review reports these findings and repairs none of them.

Finding	Count	Evidence
Audit findings	105	compass sprint audit.
STARTED stories with every task DONE	8	Audit, first group. It also names one BACKLOG story.
Open tasks whose PRs have all merged	25	17 STARTED, 8 BACKLOG. A merged PR does not prove a task is finished.
Sprint table disagrees with the story file	15	Ten say BACKLOG and the file says STARTED. Four say DONE and the file says STARTED. One says BACKLOG and the file says DONE.
Stories absent from the sprint table	6	The audit does not report this. Listed below.
Sprint table rows with no Start	90 of 120	Story age cannot be read from the table.
Closed documents whose Now is not "Nothing."	56	Audit, last group.

The stories absent from the sprint table:

• Book a trade as a draft
• Carry a large body without holding it
• Complete a structure from a screen
• Configure every seeded report from the ORE examples
• Model suppliers, sessions and feeds
• Eight columns in the trading models hold JSON blobs

"Seed a relative-time trade population with the market data it needs" is named ready to close again. Reviews of sprint 27 and sprint 28 rejected that close, because its one task only describes the story. The same reason applies.

The sprint * Status table is also stale. It says "Pick the first story" and was last touched on 2026-10-24.

The owner asked for a cleanup to prepare the close, after this review. The numbers in the sections above are the state before the cleanup. The cleanup uses the rules of the sprint 28 close. The audit reported 105 findings before it and reports 1 after it. The 1 finding is the live story "Serve what the BFF serves, then retire it".

Outcome	Stories	Notes
DONE	61	36 before the cleanup, plus 20 split and 5 closed.
ABANDONED in the table	60	58 carried to the inbox, plus the 2 already abandoned.
STARTED, left in place	6	A live branch owns each one.

Tasks left in the sprint: 547, of which 503 DONE, 21 ABANDONED, 22 BACKLOG and 1 STARTED. The open tasks all belong to the six live stories.

Two tasks closed DONE, because their PRs merged and each Result records the work: "Put the numbers-changing engine run settings through sign-off" (#4011) and "Stand up the synthetic adapter and serve its first collection" (#4493). The other open tasks with merged PRs stayed open. Each has no Result, or its own text says the work is partial.

Twenty stories split. Each closed story keeps its done tasks and narrows its Acceptance to them. A new story in the inbox takes its open tasks, 82 in total, and its original Acceptance. A story that was already "the remaining work" keeps that title, and its new folder ends in -remaining-2. Each closed story names the new story in its Decisions. A split records what shipped. It does not claim that the original Acceptance is met.

Five stories closed DONE on their finished tasks: "Complete a structure from a screen", "Hotfix: Keep the build warning-free on every toolchain", "Make main green after the sqlgen removal", "Model suppliers, sessions and feeds" and "Fix the nightly platform failures".

Fifty-eight stories moved to the inbox with their :ID: intact, because they have no done task other than a scaffold, or because of the four cases below.

The audit named four stories ready to close, and this cleanup rejects them. Every task is done, but those tasks do not deliver what the story promises:

Story	Tasks DONE	Why it is not closed
Let a busy market move prices, not only the clock	2	A document and the sampling change. The volatility channel, the tails and the control plane are not built.
Validate that every trade type is modelled as ORE reads it	1	The audit of the tests only. The contract for each type and the level check are not built.
Eight columns in the trading models hold JSON blobs	2	The basket constituents only. The Goal names eight columns.
Seed a relative-time trade population with the market data it needs	1	The one task describes the story. Reviews of sprints 27 and 28 rejected this close too.

Six stories stay STARTED in the sprint. Each has a worktree with commits ahead of main or an open PR from the last few days, and a move would pull the story from under that branch:

Story	Worktree	PR
Replace sqlgen and reflect-cpp with generated code	brave_hopper	#4492
Describe every product type on one page	merry_newton	#4471
Split the types that carry too many columns	bright_faraday	none
Serve what the BFF serves, then retire it	clever_dijkstra	none
Prove generated trades value against the ORE samples	swift_curie	none
Build the IR curves of every supported currency from public data	prime_origin	none

The sprint table now agrees with every story file. It gained the six missing rows. Closed documents in the sprint now state "Nothing." in their Now, Waiting on and Next cells. The inbox index is regenerated.

The cleanup did not read the code behind any closed task. The closes rest on the task's state, Result and PR. Two older closed stories have fewer table rows than task files: "Let the plan state domain rows, not a bag" and "Trade Blotter — the remaining work". The closed "Rework the market data feeds" also has one row fewer than its files. This cleanup left these as they were.

The owner then asked for five of the six live stories to close this sprint on their merged work, with the rest in a follow-up story for the next sprint. The sqlgen story stays STARTED, because PR #4492 is still in progress.

Two PRs merged on main after the split, and the rebases record them. PR #4471 finished "Classify each product type as flow or non-flow", so that task returned to its closed story, and the table above counts it there. PR #4494 shipped part of "Seed ORE's built-in index conventions". That task stays in the follow-up story, because the database-backed test and the tenant override remain.

Story	DONE tasks kept	Tasks moved	Follow-up holds
Describe every product type on one page	7	7	The page, the serve, the joins and the retirement of the duplicates.
Split the types that carry too many columns	9	2	The market series identity and the instrument parts.
Serve what the BFF serves, then retire it	7	0	The original Acceptance only. It needs tasks.
Prove generated trades value against the ORE samples	3	4	The trade builders, the seed and the references.
Build the IR curves of every supported currency from public data	4	6	The adapter, the calibration and the coverage.

Each closed story narrows its Acceptance to its done tasks, so a close records what merged and does not claim the original Acceptance. Three moved tasks have unmerged work on a live branch, and their ids are unchanged: "Classify each product type as flow or non-flow" (#4471, merry_newton), "Split the market series identity by series kind" (bright_faraday) and "Seed ORE's built-in index conventions" (swift_curie). Those branches must rebase onto the moved paths.

After this pass the sprint table holds 66 DONE, 60 ABANDONED and 1 STARTED. The audit reports no finding.

Still to do for the close: resolve the sqlgen story, write the Achievements and the Retrospective, regenerate the charts, write the release notes, and mark the sprint DONE.

• It did not read the * Result of each task, so the split table is a recommendation.
• It did not read the 823 task files by hand. A script read their state fields.
• It did not count how many PRs had a review before merge.
• It did not check the health of CI on main.
• It did not regenerate the sprint charts. A run needs compass sprint charts --sprint 29 --end-date 2026-11-08, because the default window stops one day after the planned end.
• It did not triage the one capture in the sprint * Captures table.
• PR counts use UTC days. The sprint 28 review used local days.

Dimension	Verdict
Goal alignment	RED
Sprint load	RED
PR velocity	RED (flow AMBER)
Story and task balance	AMBER
Focus signal	RED
Velocity	RED
Overall	RED

The sprint is RED, and worse than sprint 28 on most measures. It is eight days past its planned end, with five times the commit ceiling, six times the PR band and 38 stories in flight. Its mission was to finish carried work, and it closed about 5% of that work and 47% of the new work it added. Output is high and flow is fast, but the sprint boundary, the mission and the limit on work in flight do not constrain anything. The most important concern is that the sprint 28 review gave the same finding and the count still rose, so a written recommendation is not enough. The one change that would most improve health is to close this sprint now and to enforce a limit on STARTED stories in compass, at the point a story starts.