Sprint tracking: triaged backlog — valid + rescope (board mirror)

[Yaraslaut]

The board is authoritative. If this issue and the board disagree, the board is
right. This is a mirror, not a queue — it exists because reading a project
board needs a token with project scope, and a CI job or a repo-scope session
does not have one.

• Board: https://github.com/orgs/LASTRADA-Software/projects/2
• Scope to read it: gh auth refresh -s project

---

Snapshot — 2026-09-02

First snapshot.

Phase	Items
1 — A worktree builds and the finished fixes are on master	3
2 — No shipped surface lies about what it accepts	4
3 — Ledger enforces the boundaries it documents	1
4 — Every comment names something that exists	2
5 — Coverage measures the whole tree and a regression fails the build	4
6 — The library and the ladder meet the coverage bar they claim	5
Deferred	3
Backlog	5
Total	27

Progress: 0 / 19 sequenced done. #386 and #383 are In Review.

The sequenced queue, in Order

#	Issue	Lane	Phase
1	#372 build: a worktree under .claude/worktrees/ breaks every AUTOMOC target	T tests	1
2	#386 forms: enum class renders as free text — PR #395	L forms	1
3	#383 mutator never flips != back to == — PR #394	T tests	1
4	#403 coverage.sh profiles 3 of 9 test binaries; include/morph/net scored by nothing	T tests	5
5	#402 every codecov status is informational; two components below target are green	T tests	5
6	#404 120 partial lines are branches nothing takes both ways	T tests	5
7	#405 the framework's C++ has never been measured against a mutant	T tests	5
8	#388 DynamicForm mistypes array-valued keys assigned as QVariant	L forms	2
9	#387 spurious onOptionsReceived warning on every controller without a Choice	L forms	2
10	#368 an unattached BoardModel replies stoull; two guards never fire	N ladder	2
11	#362 RunReportJob and UndoTransaction cannot be driven by any client	N ladder	2
12	#382 any authenticated principal can read and write any book	N ladder	3
13	#355 four header comments cite tryFireImpl/SubscriberEntry, neither exists	D framework	4
14	#340 ten comments cite findings by bare id the lint cannot see	D framework	4
15	#406 148 throw sites, 81 catch arms, nothing measures how many are driven	D framework	6
16	#409 include/morph/offline at 85.20%; one header reads 57.04%	D framework	6
17	#410 include/morph/forms at 94.37%; forms.hpp holds 36 of 41 misses	L forms	6
18	#411 examples/common fell to 90.39% from the 95.85% its target came from	N ladder	6
19	#412 crm: two models' journaling paths executed by no test	N ladder	6

Deferred — sequenced, held behind something unscheduled

Issue	Held on
#391 bookmarks fetch-timer segfault	No reproduction. One CI hit on a loaded runner; 40/40 green locally, ASan+UBSan clean on the same run.
#373 ledger cross-book category link	Repo-owner decision: refuse it, or keep ignoring it downstream. The corpus currently documents it as accepted.
#341 r5-003 disposition	Repo-owner decision. examples/FINDINGS.md says the owner decides and the ladder never self-triages.

Backlog — real work, not yet sequenced

#226 (replay-family decision), #204 and #198 (surveys whose gaps are already
split out or parked), #407 (SQLite-shaped testing charter), #408 (mutation gate
in CI).

---

Open questions

Stated honestly, because these are what the sprint does not yet know.

1. Every coverage figure in phase 6 is provisional until coverage: scripts/coverage.sh profiles three test binaries out of nine, so include/morph/net is scored by nothing at all #403 lands.
   include/morph/net — 955 lines across two headers plus detail/, holding 42
   of the library's 148 throw sites — contributes zero files to the report.
   sqlite_offline_queue.hpp's 57.04% and include/morph/qt's 13 reported lines
   are measured with their own test binaries unprofiled. The library's headline
   95.69% is therefore computed over a tree that silently excludes its network
   stack. The targets in phase 6 will move, possibly a lot, and in both
   directions.

2. The 99% goal has not been costed against the real denominator. From
   95.69% it means closing ~190 of 246 unhit lines — but that arithmetic is over
   the wrong tree (see 1). It is also not obviously reachable everywhere:
   codecov.yml documents an llvm-cov artifact class (a "control reached past
   this block" counter on certain closing braces, with no LCOV_EXCL_LINE
   equivalent) that puts several components' honest ceiling below 100%.

3. No mutation tool is known to work here. mull, dextool and friends are
   absent, and their LLVM support has historically lagged; CI pins clang 20 and
   this workstation runs clang 22. mutation: the framework's own C++ has never been measured against a single mutant, so an inverted if in include/morph fails no known test #405's first deliverable is settling whether
   any tool builds against this tree. That is the most likely way the
   mutation workstream dies, and it is named in the ticket rather than
   discovered later.

4. A literal 100% mutation score is not a coherent target. Equivalent
   mutants are undecidable in general, so the honest goal is 100% of
   non-equivalent mutants with an allowlist carrying a reason per entry — the
   shape scripts/scenario/coverage_allowlist.json already uses.

5. Whether the ladder's coverage gates should start blocking is a decision, not
   an oversight. codecov.yml records the policy — finish the ladder, then
   raise the gates together — and examples/rungs.txt now lists seven rungs.
   coverage: every codecov status is informational, so a component sitting below its own target still produces a green build #402 asks for that to be resolved one way or the other, because
   examples/IMPLEMENTATION.md's rung-exit criterion 5 already calls the gate
   "blocking" and no component is.

6. Three Deferred items need a decision from the repo owner, not evidence.
   Each is named above with its exact question. They are not blocked on work.

[Yaraslaut]

Board update — 2026-09-03

Added lane X triage-valid-sweep, a one-off addition at the user's explicit
request: work all triage: valid issues that are not already blocked, as one
branch and one PR, outside the normal one-lane-per-component /
one-ticket-at-a-time shape.

In the new lane (Order 1-6, Phase Backlog, lane X triage-valid-sweep):

Order	Issue	Was in
1	#430 coverage.sh merges every stale .profraw recursively	D framework / Backlog
2	#409 include/morph/offline at 85.20%, one header at 57.04%	D framework / Phase 6
3	#410 include/morph/forms at 94.37%, forms.hpp holds most misses	L forms (retired) / Phase 6
4	#422 kanban BoardModel::_projectIdStr conflates two roles	N ladder / Backlog
5	#406 148 throw sites, 81 catch arms, nothing measures reach	D framework / Phase 6
6	#407 no written testing charter	D framework / Backlog

Excluded from the new lane, left as-is: #431 and #432 are also
triage: valid but stay Blocked in D framework — #434 (still being
triaged) found that the mutation-survivor list they're written against
produces false positives when sampled by hand, so their premise is under
re-verification. #410 is kept live rather than closed as "superseded by
#432" (as #410's own text suggests), because #432's premise is exactly
what's in question.

A developer session is running this lane now, targeting one PR that closes
all six of #430, #409, #410, #422, #406, #407.

Separately, a parallel triage pass (not part of this board) is running
/triage-issue over #435, #434, #396, #393, #392 — the open issues that
carry no triage: label yet.

The board is still authoritative; this is a note of what changed, not a
restated snapshot.

[Yaraslaut]

Lane update — 2026-09-03: X triage-valid-sweep

PR #439 opened, closing #430, #422, #406, #407 on merge (verified via
closingIssuesReferences). CI running, no failures so far (12 SUCCESS, 6
QUEUED at last check). Not merged — merging needs CI green, per the board's
own "what done means."

#409 and #410 stay open, now Blocked. Both need include/morph
coverage re-measured against a full CI-shaped build (Qt + ladder +
offline-sqlite), which the lane's environment could not do. While
attempting it, the lane found a real, previously-unlabeled defect:

New issue #440 — CMakeLists.txt:404 sets the synthesised
SQLite3::SQLite3 target's IMPORTED_LOCATION from SQLite3_LIBRARIES
(a list) instead of SQLite3_LIBRARY (the single path IMPORTED_LOCATION
requires), breaking -DMORPH_BUILD_OFFLINE_SQLITE=ON for any SQLite3
discovery whose _LIBRARIES isn't trivially equal to _LIBRARY — reproduced
twice (vcpkg's port, a manual system install), latent rather than an
environment fluke; CI's specific apt-based discovery happens not to trigger
it. Filed, added to D framework / phase 6, Order 18, Todo. #409 and #410
are now Blocked by #440.

Also found and fixed inline, before the PR was opened (code-review
findings, all in the six tickets' own files — see PR #439's description for
the full list): an unquoted $PROFILES word-splitting bug in
scripts/coverage.sh that would have silently defeated #430's own fix; a
doc comment on BoardModel overstating parity with LedgerModel; a dead
return value in check_branch_coverage.py; a stale comment after a find
moved files. One review-suggested guard on BoardModel::attachActionLog
was proposed, then proven structurally unreachable by a dedicated
verification pass, and reverted rather than kept out of caution.

Net for this lane: 4 closing (pending merge), 2 reopened as Blocked, 1 new
issue filed and scheduled. Nothing silently deferred, nothing silently
widened, per protocol.

[Yaraslaut]

PR #439 merged — 2026-09-03 21:09 UTC

Merge commit 3c456c03, CI 44/44 green before merge. #430, #422, #406, #407
closed automatically and marked Done on the board.

#409 and #410 remain Blocked in the same lane, waiting on #440 (the CMake
SQLite3_LIBRARIES/SQLite3_LIBRARY fix) as reported earlier.

[Yaraslaut]

PR #442 merged — 2026-09-03 22:49 UTC

Merge commit bc09f725, CI 44/45 green + 1 correctly skipped (Header ↔ spec sync — exempted via no docs update, verified this was warranted:
the PR only adds new private helpers absent from every spec doc, and
preserves BRIDGE_REGISTER_ACTION's public contract byte-for-byte). #441
closed automatically and marked Done on the board.

Lane Y core-util-simplify is now complete: 7 duplicated-code extractions
and 8 comment-convention fixes landed in include/morph/core and
include/morph/util/rational.hpp, all behavior-preserving, full
morph_tests suite green throughout (1302 test cases, 0 failures).