Feature-230 time study: 13 ranked pipeline-efficiency changes

jbr870 commented

2026-08-14 13:01:43 +00:00

Owner

Tracking issue for the thirteen ranked changes that came out of the feature-230 time study — a
minute-by-minute reconstruction of Verity issue #230 (release-notes page + unread indicator) from
first idea to v0.2.1 live in production, read off four Claude Code session transcripts and their
42 subagent logs.

The measurement


Calendar, idea → production	25h 37m (2026-08-13 10:10 → 2026-08-14 11:47 local)
Active session time on #230	13h 15m
Watching test suites	4h 04m — 31% of the feature
…of which 7 full Playwright runs	2h 29m, six returning an identical 469/3/0
Four fix rounds on one defect class	2h 21m, still deferred to verity#235
Blocked on a human	~2h 43m (1h 17m of it the UAT gate, working as designed)
Delivered	30 acceptance criteria, 86 files, 13,604 lines

Phase split: requirements + technical-plan 1h 31m · /develop 4h 17m · QA 4h 02m ·
/integrate 56m · UAT 1h 17m · /promote 42m.

The headline: QA cost as much as /develop did, and roughly a third of the whole feature was
spent waiting for suites that had already passed.

Scope of this programme

Items 1–13 of the study's ranked list. Deliberately excluded by the operator:

14 — small-feature lane. Rejected on principle: for production-grade applications everything
gets checked. Not filed.
15/16/17 — accounting guidance and two "do not change" findings (keep QA; keep the lead's
verify-don't-trust discipline). Recorded here as findings, not as work.

The ranked list

Ranked by expected saving. Items are independent — the ordering is the intended work order, not
a dependency chain, so nothing here is blocks-linked. Take them top down.

#	Issue	Change	Est. saving
1	#38	/develop + QA: run the full browser suite at phase boundaries, not after every unit and every fix round	~2h
2	#39	/develop: dispatch work units by declared dependency, not in list order	~1h
3	#40	QA fix loop: stop patching after a second same-class re-validation finding	~45m
4	#41	/develop: scope the per-unit gate to the packages the unit touched	~40m
5	#42	requirements/technical-plan: one question round per phase, not four inside eight minutes	~30m
6	#43	QA: intra-run lane-ownership primitive, plus a void-vs-flaky triage rule	~20m + the false-alarm class
7	verity#237 (other forge)	migration-concurrency teardown hook times out under load	~19m across 3 encounters
8	#44	wait-discipline: an idle signal is not completion — name the tree-state check	6 occurrences/run; harness defect
9	#45	Phase Outcomes: no supersede path when a posted outcome's premise turns out false	~25m rework + a false record
10	#46	wait-discipline: no retry/resume protocol for transient API errors or quota stalls	~9m attended; unbounded unattended
11	#47	release-pipeline-status.sh --ref returns a stale pipeline — a false green immediately before a production tag	correctness — near-miss before a production tag
12	#48	/develop: validate against the real committed content, not only fixtures	caught the one blocking defect all gates passed
13	#49	Baseline rule: a gate's output goes to a file and its exit code comes from the command, never a pipeline	~6m + the false-signal class

Recurrence note

Items 6 and 10 promote existing watchlist entries (RC-W17/RC-W18 and RC-W2) on a second independent
sighting. Item 8 corroborates the report-delivery friction already noted on #11. The rest are
first sightings from this run; several are structural defects in skill behaviour rather than
one-off adaptations, which is why they are filed rather than watchlisted.

Not filed but worth noting: this run also gave RC-W16 (no compaction protocol) a second
sighting — session e9274dc9 ran out of context and needed /compact mid-feature, between QA and
/integrate.

Source

Transcripts 915e1a6d, 6b82bebc, c045bb58, e9274dc9 in
~/.claude/projects/-home-jochem-Coding-verity-worktrees-slot4/ on jochems-forge.

Tracking issue for the thirteen ranked changes that came out of the **feature-230 time study** — a minute-by-minute reconstruction of Verity issue #230 (release-notes page + unread indicator) from first idea to `v0.2.1` live in production, read off four Claude Code session transcripts and their 42 subagent logs. ## The measurement | | | |---|---| | Calendar, idea → production | 25h 37m (2026-08-13 10:10 → 2026-08-14 11:47 local) | | Active session time on #230 | **13h 15m** | | Watching test suites | **4h 04m — 31% of the feature** | | …of which 7 full Playwright runs | 2h 29m, six returning an identical 469/3/0 | | Four fix rounds on one defect class | 2h 21m, still deferred to verity#235 | | Blocked on a human | ~2h 43m (1h 17m of it the UAT gate, working as designed) | | Delivered | 30 acceptance criteria, 86 files, 13,604 lines | Phase split: requirements + technical-plan 1h 31m · `/develop` 4h 17m · **QA 4h 02m** · `/integrate` 56m · UAT 1h 17m · `/promote` 42m. The headline: **QA cost as much as `/develop` did**, and roughly a third of the whole feature was spent waiting for suites that had already passed. ## Scope of this programme Items 1–13 of the study's ranked list. Deliberately **excluded** by the operator: - **14 — small-feature lane.** Rejected on principle: for production-grade applications everything gets checked. Not filed. - **15/16/17** — accounting guidance and two "do not change" findings (keep QA; keep the lead's verify-don't-trust discipline). Recorded here as findings, not as work. ## The ranked list Ranked by expected saving. Items are independent — the ordering is the intended **work order**, not a dependency chain, so nothing here is `blocks`-linked. Take them top down. | # | Issue | Change | Est. saving | |---|---|---|---| | 1 | #38 | /develop + QA: run the full browser suite at phase boundaries, not after every unit and every fix round | ~2h | | 2 | #39 | /develop: dispatch work units by declared dependency, not in list order | ~1h | | 3 | #40 | QA fix loop: stop patching after a second same-class re-validation finding | ~45m | | 4 | #41 | /develop: scope the per-unit gate to the packages the unit touched | ~40m | | 5 | #42 | requirements/technical-plan: one question round per phase, not four inside eight minutes | ~30m | | 6 | #43 | QA: intra-run lane-ownership primitive, plus a void-vs-flaky triage rule | ~20m + the false-alarm class | | 7 | [verity#237](https://gitlab.com/jbr870/verity/-/work_items/237) *(other forge)* | migration-concurrency teardown hook times out under load | ~19m across 3 encounters | | 8 | #44 | wait-discipline: an idle signal is not completion — name the tree-state check | 6 occurrences/run; harness defect | | 9 | #45 | Phase Outcomes: no supersede path when a posted outcome's premise turns out false | ~25m rework + a false record | | 10 | #46 | wait-discipline: no retry/resume protocol for transient API errors or quota stalls | ~9m attended; unbounded unattended | | 11 | #47 | release-pipeline-status.sh --ref returns a stale pipeline — a false green immediately before a production tag | correctness — near-miss before a production tag | | 12 | #48 | /develop: validate against the real committed content, not only fixtures | caught the one blocking defect all gates passed | | 13 | #49 | Baseline rule: a gate's output goes to a file and its exit code comes from the command, never a pipeline | ~6m + the false-signal class | ## Recurrence note Items 6 and 10 promote existing watchlist entries (RC-W17/RC-W18 and RC-W2) on a second independent sighting. Item 8 corroborates the report-delivery friction already noted on #11. The rest are first sightings from this run; several are structural defects in skill behaviour rather than one-off adaptations, which is why they are filed rather than watchlisted. Not filed but worth noting: this run also gave **RC-W16 (no compaction protocol)** a second sighting — session `e9274dc9` ran out of context and needed `/compact` mid-feature, between QA and `/integrate`. ## Source Transcripts `915e1a6d`, `6b82bebc`, `c045bb58`, `e9274dc9` in `~/.claude/projects/-home-jochem-Coding-verity-worktrees-slot4/` on `jochems-forge`.

jbr870 added the

type:chore

priority:high

phase/backlog

labels

2026-08-14 13:01:43 +00:00

jbr870 referenced this issue

2026-08-14 13:02:35 +00:00

/develop + QA: run the full browser suite at phase boundaries, not after every unit and every fix round #38

jbr870 referenced this issue

2026-08-14 13:02:35 +00:00

/develop: dispatch work units by declared dependency, not in list order #39