Correct #469 move-diagnostics documentation

.claude/agent-memory/atomic-executor/MEMORY.md

@@ -39,6 +39,7 @@
 - ["Skip the pointless drain" note makes a negative test vacuous](project_preflight_drain_scope_optimization_note_makes_test_vacuous.md) — a cosmetic "path
 - [Sanitisation task cannot sweep its own record](project_sanitisation_task_cannot_sweep_its_own_record.md) — exactly one resi
 - [Conjunctive criteria break the one-artifact citation rule](project_preflight_conjunctive_criterion_citation_gap.md) — "cites exactly o
+- [Pre-edit gate cites the post-edit replacement table](project_preedit_gate_cites_postedit_replacement_table.md) — baseline asserts s
 - [Caller-stated preflight count drifts before execution](project_caller_stated_preflight_count_drifts_before_execution.md) — 42 quoted vs 48 measured; read the plan's own clause
 - [Orchestrator override does not satisfy an AC](project_orchestrator_override_does_not_satisfy_an_ac.md) — proceed past the gate,
 ## Build / toolchain environment

.claude/agent-memory/atomic-executor/project_preedit_gate_cites_postedit_replacement_table.md

@@ -0,0 +1,25 @@
+---
+name: preedit-gate-cites-postedit-replacement-table
+description: A Phase-0 pre-edit citation gate that sources its literals from the plan's own "replacement text" table asserts post-edit strings before the edit, so it can only fail
+metadata:
+  type: project
+---
+
+A baseline citation-re-verification task that says "the N pre-edit tokens named in the R<k> table
+each have count 1" is unsatisfiable when the R<k> table is the plan's *replacement* text table,
+whose value column holds post-edit text. Before any edit those literals have count 0, so the gate
+fails and the executor halts in Phase 0. The pre-edit literals normally live in a separate
+"Currently reads" / verified-facts table earlier in the plan.
+
+The companion signal is that the task's command block contains no command producing those N counts,
+while the acceptance bullet demands them — an acceptance condition naming an observation the task
+never makes.
+
+**Why:** Plans that renumber or restate literals in two tables (current text and replacement text)
+invite a cross-reference to the wrong one. Observed on the issue #469 comment-accuracy plan at
+`[P0-T15]`, round 3, after the same text had survived two earlier preflight rounds.
+
+**How to apply:** For every pre-edit / baseline assertion, check which internal table the plan cites
+and confirm that table's value column holds the *pre*-edit spelling. Then check the task's command
+block actually emits each asserted count. Related: [[project_preflight_gate_literal_extract_from_plan_not_retype]],
+[[project_exact_count_gate_vs_remediation_loop]].

.claude/agent-memory/atomic-planner/MEMORY.md

@@ -19,6 +19,7 @@
 - [#677 R1–R8](project_677_keyboard_focus_leak_plan_seams.md) — ctor param REJECTED (5 reflection-arity tests); typed harness for compile-red; internal 9-arg ctor, never ambient SetSynchronizationContext; per-file non-vacuity floors
 - [#635](project_635_reflective_caller_audit_plan_seams.md) — evidence-only audit: tracked plan inflates its own sweep; scan hits its own pattern list; spec said six sites, tree has eight; pathspec breadth inflates a count
 - [#440 R1–R4](project_440_breadcrumb_left_arrow_plan_seams.md) — deletion-only change voids a diff-derived changed-line gate; `(Rebuild target(s))`, NOT `(Rebuild target)`; `.csharpierignore` matches the `.cobertura.xml` suffix; `.dotnet-sdk` IS gitignored (`.gitignore:350`); cite an AC by sentence only after counting its sentences; `Include` resolves against the declaring project's dir
+- [#469 R1–R3](project_469_comment_accuracy_plan_seams.md) — defect-number SWAP voids whole-file token gates; `- [x] AC1` is a prefix of `AC10`; unconditional base-merge task; Phase 0 insert renumbers artifact FILENAMES; pre-edit vs post-edit table cited wrong; evidence dir rule ≠ recorded path form
 - [#644](project_644_ac16_referral_revision_seams.md) — AC's named instrument prints no figure; substitute's noise (~0.028pp) exceeds the 0.01pp delta; check-off task → referral task pinning the UNCHECKED state
 - [#644 cycle 2](project_644_cycle2_sweep_gate_evasion_seams.md) — rewording out of a detector's match set is gate evasion; SHA-256 pair for untracked prior artifacts; bound the porcelain `??` scope
 - [#644 PA-7](project_644_pa7_redaction_plan_seams.md) — untracked audit artifact still enters main; name-status diff blind to it; mandated literal contradicted its own stated intent
@@ -66,6 +67,7 @@
 - [CSharpier "Formatted N files" is processed count](csharpier-formatted-n-is-processed-count.md) — a restart-on-rewrite loop keyed on it never terminates; define rewritten-count via before/after SHA-256
 - [Repo-wide csharpier format breaks zero-diff ACs](csharpier-repowide-format-breaks-zero-diff-acs.md) — scope the mutating pass to the plan's own path list
 - [.csharpierignore scope](csharpierignore-scope-packages-config.md) — only `*.csproj`/`*.props`/`*.targets` are excluded; `packages.config` is NOT
+- [.gitignore bracket classes defeat a literal grep](gitignore-bracket-classes-defeat-literal-grep.md) — `[Tt]est[Rr]esult*/` does ignore `TestResults/`; never claim a path is tracked from a literal search
 - [`/Logger:trx` needs `/ResultsDirectory`](trx-needs-resultsdirectory.md) — TRX lands in `TestResults\` relative to cwd; give each run task its own `p#-t#` subdirectory
 - [`[expect-fail]` needs a synchronous seam](expect-fail-needs-a-synchronous-seam.md) — async-void boundaries false-GREEN; re-run RED analysis after scoping `Times.Never()`
 - [Invoke-MSTestWithCoverage.ps1](reference_invoke_mstest_with_coverage_script.md) — canonical full-suite Cobertura runner

.claude/agent-memory/atomic-planner/gitignore-bracket-classes-defeat-literal-grep.md

@@ -0,0 +1,26 @@
+---
+name: gitignore-bracket-classes-defeat-literal-grep
+description: Never conclude a path is untracked from a literal grep of .gitignore — this repo writes bracketed character classes ([Tt]est[Rr]esult*/, [Bb]in/, [Oo]bj/) that a literal search misses
+metadata:
+  type: feedback
+---
+
+Before asserting in a plan that some path is or is not gitignored, search `.gitignore` with a
+character-class-tolerant pattern (or read the file), never with the literal directory name.
+
+**Why:** This repository's `.gitignore` uses the Visual Studio template's bracketed-case form. A grep
+for the literal `TestResults` returns nothing even though `.gitignore:39` is `[Tt]est[Rr]esult*/`,
+which does match `TestResults/`. On the #469 plan this produced a false factual claim in the plan
+prose ("`TestResults` and `*.trx` are NOT ignored") that survived into a self-review enumeration and
+was only caught by [[trx-needs-resultsdirectory]], which had recorded the real line. A plan that
+states a wrong tree fact is a defect even when the command it justifies happens to be harmless.
+
+Known bracketed entries at the time of writing: `:26` `[Bb]in/`, `:27` `[Oo]bj/`, `:39`
+`[Tt]est[Rr]esult*/`, `:40` `[Bb]uild[Ll]og.*`. Plain entries include `:144` `coverage/*`.
+
+**How to apply:** Any plan task whose acceptance depends on a clean tree, or whose prose explains
+where a tool's output lands, must cite the `.gitignore` line number and quote the entry verbatim in
+its bracketed form. Grep with a pattern like `[Tt]est|[Bb]in|[Oo]bj` or just read the first ~60 lines
+of `.gitignore`. Relatedly, do not route tool output somewhere merely because a literal grep
+suggested the default location was tracked. See [[agent-memory-is-tracked-scope-git-gates]] for the
+converse trap, where a path that looks like tooling scratch actually IS tracked.

.claude/agent-memory/atomic-planner/project_469_comment_accuracy_plan_seams.md

@@ -0,0 +1,125 @@
+---
+name: project-469-comment-accuracy-plan-seams
+description: Issue #469 plan seams — defect-number SWAP voids whole-file token gates; digit-only edits make numstat exact; terminal clean-tree acceptance unreachable; scoped format vs repo-wide check contradiction; base-merge task + Phase 0 renumber; pre-edit vs post-edit table mix-up; evidence dir rule does not constrain recorded path form
+metadata:
+  type: project
+---
+
+Issue #469 turned out to be documentation-accuracy only: three of four defects were already merged,
+and the fourth's residual action is open issue #629. The plan is comment/XML-doc/`because:`-string
+edits with zero executable-line change.
+
+**A renumbering SWAP makes every whole-file token gate vacuous.** Both `Issue #469 defect 1` and
+`Issue #469 defect 2` already existed in BOTH edited files at branch head, so "the file contains
+`Issue #469 defect 2`" passes before any work. Every gate had to become a combined single-line token
+pairing the defect number with its distinguishing text (`Issue #469 defect 2: exactly one diagnostics
+line`), plus the complementary must-become-zero token. This generalises to any A-to-B relabelling
+where both labels are already present.
+
+**Why:** a swap conserves the multiset of tokens; only their pairing with surrounding text changes.
+**How to apply:** for any swap/rename plan, gate on the PAIRING, and always author the zero-match
+companion alongside the one-match assertion.
+
+**A single-character substitution makes exact `--numstat` derivable.** All eight renumbering sites
+were one-digit changes on one physical line each, so line length and line count are invariant and the
+plan could assert exactly `2 2` and `6 6` per file. Verify the digit-only property by reading each
+line before promising an exact numstat; a rewrap would void it.
+
+**Spec line counts were off by one and research line counts were wrong.**
+`QfcCollectionControllerDefects468MoveTests.cs` is 497, not the spec's 498;
+`QfcHomeController.Metrics.cs` is 215, not the research doc's "232, approximate" and not the 216 the
+plan's own first pass asserted in four places; and the research cited `:351` for a site that is
+actually `:352`. Re-derive every count and citation even when two upstream documents agree, and count
+with a whole-file line count rather than by reading a tail window — a `cat -n` tail read of a file
+ending in a final newline is easy to misread as one line longer than it is. See
+[[verify-test-provenance-before-planning-deletion]].
+
+**Preflight round 1 (version 0.3) seams — the ones a first pass will miss:**
+
+- **A terminal "clean tree" acceptance is unreachable when the task must tick its own checkbox.**
+  P7-T17 asserted empty `git status` over a pathspec containing the plan file it must mark `[x]` and
+  the artifact it must write. Committing either re-dirties the other. Author the acceptance as
+  "names no path other than the plan file and this task's own artifact" and state that committing
+  those two is the orchestrator's step after plan completion.
+- **A scoped format pass and a repo-wide `check .` contradict each other.** If the format task
+  declines to sweep pre-existing drift (to protect a zero-executable-line-change AC), the check task
+  cannot demand repo-wide exit 0. Make the check baseline-relative: reported set must be a SUBSET of
+  the Phase 0 enumeration. Then re-check the clean-pass declaration task, which will still say "no
+  failure" and now needs "acceptance held, not exit code 0".
+- **Branch on the exit code, never on an unobserved output literal.** The first pass keyed the format
+  branch on counting lines containing `Was not formatted` at CSharpier 1.2.6. No run was cited. If
+  the spelling differs the count is 0, the repo-wide mutating branch is taken silently, and the AC it
+  protected disappears.
+- **Evidence captured before a mutating formatter is stale.** The changed-line-classification and
+  numstat evidence all ran in Phases 2-5; Phase 6's `format` then rewrites the same four files.
+  Re-assert the per-file numstat inside the format task itself.
+- **`- [x] AC1` is a prefix of `- [x] AC10`..`AC13`.** Assert the spec's em-dash form
+  `- [x] AC1 —`. True-only-because-it-runs-first is not an assertion.
+- **A `git diff origin/main` footprint enumeration must include what earlier branch commits added.**
+  This branch's own `issue.md` and `research/` document appear in every `origin/main`-anchored diff,
+  so an exact enumeration omitting them is unsatisfiable. Also scope the pathspec: tracked
+  `.claude/agent-memory/` modifications by other agents in the same worktree otherwise show up. See
+  [[agent-memory-is-tracked-scope-git-gates]].
+- **AC10 named `vstest.console.exe /EnableCodeCoverage`, which no task runs.** The runsettings file
+  declares no coverage `DataCollector` and the pipeline is `dotnet-coverage`. Record the AC-to-task
+  realisation mapping and state the wording divergence explicitly rather than silently substituting.
+- **Two contradictory totals in one sentence.** "must be 20" and "14 added and 14 deleted, 28 diff
+  lines" coexisted; a downstream task already depended on 28. Recompute every arithmetic figure from
+  its per-file components.
+
+**Preflight round 2 (version 0.4) seams:**
+
+- **A plan with `git diff origin/main` gates needs a Phase 0 task that MERGES `origin/main`.** Every
+  numstat/name-only/footprint gate is anchored to `origin/main`, so an upstream advance between
+  planning and execution silently re-scopes them. The task is unconditional — `git merge` exits 0 and
+  prints `Already up to date.` when there is nothing to do — and it must precede every baseline
+  capture, because a merge landing after a baseline invalidates that baseline (most directly the
+  `BASELINE_PASSED:` count). Use `merge`, never `rebase`: the force-push guard rejects the rewritten
+  history. Exit code alone is not evidence; re-assert rev-equality plus the gated line counts.
+- **Inserting one Phase 0 task forces a whole-phase renumber, and the artifact FILENAMES carry the
+  task ID too.** The uppercase cross-references were 32 across 27 lines; the lowercase
+  `p0-t<N>-<slug>.<timestamp>.md` evidence names and a `/ResultsDirectory:TestResults\p0-t<N>` path
+  were a second, disjoint set the reference count did not cover. Remap both, DESCENDING (T14→T15
+  first), so a `replace_all` never collides with a number it has not yet processed.
+- **A conditional branch stated only in prose is not executable.** P6-T1's prose described a scoped
+  `csharpier format` on the non-zero-baseline branch, but its command block held only the repo-wide
+  `format .`. An executor following the block verbatim takes the branch the prose exists to prevent.
+  Every branch a task's prose describes needs its own line in the task's command block.
+- **`acceptance-criteria-tracking` forbids modifying criterion text, so "check it off, appending the
+  evidence path" is a policy conflict.** Reword to record the path in the task's progress output; the
+  criterion line changes only in its checkbox marker.
+- **A base merge silently shifts line citations in files the plan does not edit.** The merge added one
+  `<Compile Include>` to `QuickFiler.Test.csproj`, moving two cited entries from `:135`/`:155` to
+  `:136`/`:156`. After ANY base reconciliation, re-derive every citation into a merged file, not only
+  into the files the plan edits.
+- **A class-level summary that enumerates all four defect numbers is not a defect-N citation.**
+  `QfcCollectionControllerDefects468MoveTests.cs:17-18` reads "issue #469 defects 1," / "2, 3 and 4",
+  so it is invariant under a 1-for-2 swap. Grouping it with the defect-3 sites was a misclassification
+  even though the operative "needs no edit" conclusion was right.
+
+**Preflight round 3 (version 0.5) seams:**
+
+- **A plan that quotes both PRE-edit and POST-edit text in two tables will get the wrong one cited.**
+  P0-T15's baseline bullet said "the eight pre-edit tokens named in the R3 table". The R3 table's
+  third column is headed `Becomes` and holds POST-edit text, every token of which has count 0 at
+  branch head — so a Phase 0 acceptance demanding count 1 for them halts the executor before any
+  work. Name the SOURCE TABLE and its COLUMN HEADER in the bullet, and quote the eight literals
+  inline rather than referring to a table by name.
+- **A re-verification task must produce every count its acceptance asserts.** P0-T15 listed eight
+  token counts in its acceptance but its PowerShell block held no command producing them. Walk each
+  acceptance clause of every command-bearing task back to a line in that task's own block.
+- **Where-artifacts-are-written does not constrain what-paths-they-record.** The evidence-location
+  rule and the `enforce-evidence-locations.ps1` hook both check the DIRECTORY. Any acceptance
+  demanding verbatim tool output (MSBuild diagnostics, CSharpier unformatted-file enumerations)
+  publishes absolute paths carrying the account name, machine name and drive letter once a staging
+  task commits `evidence/`. Add an explicit "recorded paths are repository-relative" clause naming
+  the tasks, and carve out counts/exit codes/summary lines so the assertions keep their observables.
+  See [[../_shared_no_absolute_host_paths]].
+
+**Local facts confirmed this pass:** the CSharpier manifest is `dotnet-tools.json` at the repository
+ROOT (there is no `.config/` directory); `packages/` and `QuickFiler.Test/bin/Debug/` are absent from
+a fresh agent worktree so restore-then-build must precede any test-count baseline;
+`Invoke-MSTestWithCoverage.ps1` calls `Assert-CoberturaLineCoverageThreshold`, which throws below 80%
+BEFORE the Koverage post-processing writes the XML, so a baseline task must record the thrown
+percentage and continue rather than treating it as this change's failure. Related:
+[[project_494_threshold_reconciliation_plan_seams]], [[reference_invoke_mstest_with_coverage_script]].

.claude/agent-memory/orchestrator/MEMORY.md

@@ -73,6 +73,8 @@
 - [Epic-child nullable fan-in debt is deferred](project_epic_child_nullable_fanin_debt_deferred.md) — cross-child CS86xx fan-in is the
 - [C# coverage has two denominators](csharp-coverage-denominator-two-figures.md) — filtered first-party ~85.9% clears the gate, unfiltered
 - [lines-covered is nondeterministic; lines-valid is not](coverage-lines-covered-is-nondeterministic.md) — same-tree runs drift up to 4
+- [Verify an issue is open in SUBSTANCE](verify-issue-still-open-in-substance.md) — grep the source for the issue number before preparing; the residual may already be its own issue
+- [Reconcile plan numbers against your own measurements](reconcile-plan-numbers-against-your-own-measurements.md) — a stale line count passes a citation-only review and halts Phase 0
 - [Preflight catches vacuous gates](preflight-catches-vacuous-gates.md) — MCP `ok:true` is not enough; executor preflight found 6 gates
 - [Preflight: sweep task ordering + citation arity](preflight-sweep-task-ordering-and-citation-arity.md) — name both sweeps early; a
 - [Revert plans must check test provenance](revert-plans-must-check-test-provenance.md) — verify each test against the pre-change sha; a