Completed items in this epic. All four shipped together in v1.27.0 (2026-08-14).
Outcome: methodology/09_git_workflow.md trimmed 1,026 → 798 lines (−22%), 24 → 20 sections,
under the charter’s 800-line exit criterion. Full analysis, execution log, and clarity assessment:
2026-05-09-git-workflow-decision.md.
Verification (all four items): repo-wide anchor check across 86 tracked markdown files — zero
broken anchors; 85 inbound references to 09_git_workflow.md verified. Content-loss check confirmed
every rule, table, and reasoning passage survives; only second tellings were removed.
Authority note: BL-0012 touches methodology/ and is T2/T3 under the tier matrix. It was
executed at the maintainer’s explicit direction (“finish and proceed do not stop until all done”),
which is what satisfies the maintainer-authored requirement — the change was human-directed, not
loop-initiated.
| Field | Value |
|---|---|
| Epic | E03-git-workflow-trim |
| Pillar | P2 |
| Priority | P2 |
| Effort | S |
| Status | done |
| Test | pass — decision doc exists, costs both options, commits to one with reasoning |
| Deps | — |
| Lock | — |
Why / Description: Score each top-level section of 09_git_workflow.md for trim potential vs.
split coherence, and decide which path BL-0012 executes.
Done means:
self-development/evaluations/2026-05-09-git-workflow-decision.md.Resolution: Trim chosen. The measurement drove it: the doc had grown to 1,026 lines (98% of the soft cap, 40 more than the charter recorded), but its mean section was only ~43 lines — no section was bloated. Length came from 24 genuine facets of one cohesive topic, which is the profile that argues against splitting. Meanwhile ~120 lines were outright duplication (three appendix sections restating content given better inline earlier). Splitting would also have spent a T3 restructure, broken the README’s “fourteen short docs” identity claim, and scattered anchors across three files. The split layout is costed in the decision doc for whoever faces this again.
Also corrected a measurement trap for the record: a naive ^## scan reports 29 sections; five of
those are template headings inside code fences. The real count is 24.
| Field | Value |
|---|---|
| Epic | E03-git-workflow-trim |
| Pillar | P2 |
| Priority | P2 |
| Effort | L |
| Status | done |
| Test | pass — 798 lines (<800 criterion); content-loss check clean; every removal logged |
| Deps | BL-0011 |
| Lock | — |
Why / Description: Execute the trim decision: bring 09_git_workflow.md under 800 lines without
losing any rule, table, or reasoning.
Done means:
methodology/09_git_workflow.md is under 800 lines (798) AND every removed section or
paragraph is logged in 2026-05-09-git-workflow-decision.md.worktree prune, filter-branch, reset --hard, --force-with-lease,
clean -fd, and the revert type all survive.Resolution: −228 lines (50 insertions, 277 deletions) across six cuts: three whole duplicate appendices (commit examples, PR body skeleton, worktree command reference), the merge of the destructive-command list into the affirmative list, a condensed operational-work section, and a prose-density pass that collapsed second tellings.
The structural win beyond length: destructive operations now have one home. Previously an agent consulted a negative list and an affirmative list that disagreed by omission on six operations; now a single table answers “may I run this?” with ✓/⚠/✗ for every operation the doc covers.
Deliberate non-removal: the “Common mistakes” table, several rows of which restate rules stated above. That repetition is the point — it is a corpus-wide convention that restates each rule in failure form, and it appears in every methodology doc.
| Field | Value |
|---|---|
| Epic | E03-git-workflow-trim |
| Pillar | P2 |
| Priority | P2 |
| Effort | S |
| Status | done |
| Test | pass — repo-wide anchor scan across 86 files, zero broken |
| Deps | BL-0012 |
| Lock | — |
Why / Description: Catalog inbound references to 09_git_workflow.md, update those the trim
invalidated, and confirm no new broken links.
Done means:
methodology/09_git_workflow.md resolve (85 references checked).Resolution: Merging two sections renamed one anchor, breaking three inbound links — in
11_human_roles.md (the decision-ownership matrix’s “pairs with” note) and twice in
self-development/AUTONOMOUS_LOOP.md (Constraint 3 and the history-rewrite negative-list row). All
three now point at #what-ai-agents-can-and-cant-do-in-git--the-operation-table.
One reference was deliberately not updated: evaluations/2026-05-first-pass.md cites the
affirmative list by name and line number. It is a dated historical eval; rewriting it would falsify
what that eval actually examined.
| Field | Value |
|---|---|
| Epic | E03-git-workflow-trim |
| Pillar | P2 |
| Priority | P2 |
| Effort | S |
| Status | done |
| Test | pass — closure note compares post-change to the v1.6.0 baseline on five measures |
| Deps | BL-0012, BL-0013 |
| Lock | — |
Why / Description: Author the clarity assessment the charter requires, comparing the trimmed doc to the v1.6.0 baseline.
Done means:
CHANGELOG.md has the E03 closure entry (v1.27.0).EPICS.md rollup reflects E03 done.Resolution: The assessment’s key finding is that mean section length stayed flat (~41 → ~40) while total length fell 22%. That is the signal that the reduction came from deleting duplicate sections and collapsing second tellings rather than thinning uniformly — a uniform thinning would have shown a falling mean and a shrinking longest section. The reference tables that carry the doc’s load are all intact at full size.
Headroom went from 2% to 24% of the soft cap — roughly six releases of runway at this doc’s
historical growth rate. Honest caveat recorded: 09 is still the corpus’s longest doc, and the
duplication that made this trim easy will not be available next time. The split analysis stands
ready if it recurs.