TASK-121 — a check that reads live project state as its expected value is now found mechanically

.perry/events.jsonl

@@ -579,3 +579,15 @@
 {"ts": "2026-08-20T23:35:10", "event": "evidence", "id": "TASK-122", "title": "the repair path the tools advertise leaves the file needing a whitespace fix", "track": "main", "actor": "agent", "from": "—", "to": "evidence/2026-08/TASK-122-spec.md"}
 {"ts": "2026-08-20T23:37:40", "event": "status", "id": "TASK-140", "title": "every mode contract slot is assigned to an axis, and the spine-to-unit map is written down", "track": "main", "actor": "agent", "depends_on": [], "from": "in_progress", "to": "review", "reason": ""}
 {"ts": "2026-08-20T23:37:40", "event": "evidence", "id": "TASK-140", "title": "every mode contract slot is assigned to an axis, and the spine-to-unit map is written down", "track": "main", "actor": "agent", "from": "evidence/2026-08/TASK-140-dispatch-2026-08-21.md", "to": "evidence/2026-08/TASK-140-dispatch-2026-08-21-result.md"}
+{"ts": "2026-08-20T23:38:54", "event": "start", "id": "TASK-141", "title": "a row stays blocked after its blockers close, because the stored status masks the computed one", "track": "intake", "actor": "agent", "from": "not_started", "to": "in_progress"}
+{"ts": "2026-08-20T23:38:54", "event": "evidence", "id": "TASK-141", "title": "a row stays blocked after its blockers close, because the stored status masks the computed one", "track": "intake", "actor": "agent", "from": "—", "to": "evidence/2026-08/TASK-141-spec.md"}
+{"ts": "2026-08-20T23:52:06", "event": "status", "id": "TASK-120", "title": "the linkage edges are read but never folded into KR progress", "track": "main", "actor": "agent", "depends_on": [], "from": "in_progress", "to": "review", "reason": ""}
+{"ts": "2026-08-20T23:52:06", "event": "evidence", "id": "TASK-120", "title": "the linkage edges are read but never folded into KR progress", "track": "main", "actor": "agent", "from": "evidence/2026-08/TASK-120-spec.md", "to": "evidence/2026-08/TASK-120-dispatch-2026-08-21-result.md"}
+{"ts": "2026-08-20T23:52:24", "event": "add", "id": "TASK-144", "title": "the event log timestamp has no zone and the register has one, so ordering them is a guess", "track": "intake", "mode": "queue", "priority": "P1", "actor": "agent", "summary": "", "depends_on": [], "from": null, "to": "not_started"}
+{"ts": "2026-08-20T23:52:24", "event": "add", "id": "TASK-145", "title": "the contract shape baseline is stale against its own recorder", "track": "intake", "mode": "queue", "priority": "P2", "actor": "agent", "summary": "", "depends_on": [], "from": null, "to": "not_started"}
+{"ts": "2026-08-20T23:52:24", "event": "add", "id": "TASK-146", "title": "the viewer renders a KR current with no provenance because it does not go through the shared derivation", "track": "intake", "mode": "queue", "priority": "P2", "actor": "agent", "summary": "", "depends_on": [], "from": null, "to": "not_started"}
+{"ts": "2026-08-20T23:53:36", "event": "start", "id": "TASK-143", "title": "two PRs each green on their own base merged into a red tree, and nothing checked the pair", "track": "intake", "actor": "agent", "from": "not_started", "to": "in_progress"}
+{"ts": "2026-08-20T23:53:36", "event": "evidence", "id": "TASK-143", "title": "two PRs each green on their own base merged into a red tree, and nothing checked the pair", "track": "intake", "actor": "agent", "from": "—", "to": "evidence/2026-08/TASK-143-spec.md"}
+{"ts": "2026-08-21T00:03:14", "event": "status", "id": "TASK-122", "title": "the repair path the tools advertise leaves the file needing a whitespace fix", "track": "main", "actor": "agent", "depends_on": [], "from": "in_progress", "to": "review", "reason": ""}
+{"ts": "2026-08-21T00:03:14", "event": "evidence", "id": "TASK-122", "title": "the repair path the tools advertise leaves the file needing a whitespace fix", "track": "main", "actor": "agent", "from": "evidence/2026-08/TASK-122-spec.md", "to": "evidence/2026-08/TASK-122-dispatch-2026-08-21-result.md"}
+{"ts": "2026-08-21T00:03:52", "event": "add", "id": "TASK-147", "title": "nothing outside describe_cell proves the table and bullet paths stay separated", "track": "intake", "mode": "queue", "priority": "P2", "actor": "agent", "summary": "", "depends_on": [], "from": null, "to": "not_started"}

perry/BOARD.md

@@ -36,18 +36,19 @@
 | TASK-102 | Evidence becomes a typed relation: {path, kind, round}, not one prose cell | Coding Agent | not_started | — | — | V4 | TASK-090, TASK-092 | main |  |  |  |  |  |  |
 | TASK-114 | aiMark reads Perry through the current contracts instead of a pin nine versions old | Coding Agent | in_progress | delegated to an aiMark coding agent; awaiting paste-back | evidence/2026-08/TASK-114-delegation-prompt.md | V4 | — | main |  |  |  |  |  |  |
 | TASK-119 | the linkage graph is documented as machine-written and no tool writes it | Coding Agent | not_started | — | — | V3 |  | main |  |  |  |  |  |  |
-| TASK-120 | the linkage edges are read but never folded into KR progress | Coding Agent | in_progress | dispatched to claude-subagent; worktree pinned to 7c0bb99; state-schema.json scoped out so the gate passes without a release | evidence/2026-08/TASK-120-spec.md | V3 | — | main |  |  |  |  |  |  |
+| TASK-120 | the linkage edges are read but never folded into KR progress | Coding Agent | review | PR #24 — contract 2.1; four findings handed back, two worth their own rows | evidence/2026-08/TASK-120-dispatch-2026-08-21-result.md | V3 | — | main |  |  |  |  |  |  |
 | TASK-121 | the sweep that found four more live-state assertions runs once and then is thrown away | Coding Agent | not_started | — | — | V3 |  | main |  |  |  |  |  |  |
-| TASK-122 | the repair path the tools advertise leaves the file needing a whitespace fix | Coding Agent | in_progress | dispatched to claude-subagent; worktree pinned to 6c01b93; spec carries a live reproduction | evidence/2026-08/TASK-122-spec.md | V3 | — | main |  |  |  |  |  |  |
+| TASK-122 | the repair path the tools advertise leaves the file needing a whitespace fix | Coding Agent | review | PR #25 — item 2 proved both ways on one run; two questions handed back | evidence/2026-08/TASK-122-dispatch-2026-08-21-result.md | V3 | — | main |  |  |  |  |  |  |
 | TASK-123 | the goals writer takes the file as truth and derives the store, which is the opposite direction from the KR | Coding Agent | not_started | — | — | V4 |  | main |  |  |  |  |  |  |
 | TASK-126 | closing the dangling-id row requires writing the record that re-dangles it | Coding Agent | review | PR #22 — the suite is fully green; verify the strong anti-vacuity case survives review, then close at V3 | evidence/2026-08/TASK-126-dispatch-2026-08-21-result.md | V3 | TASK-112 | main |  |  |  |  |  |  |
 | TASK-129 | Agent is five strings that do not join, and role has never once been written | Coding Agent | not_started | unblocked: work owns .perry/agents.jsonl → .perry/roles/ as of the 2026-08-20 signature; needs a spec, then dispatch | — | V3 | TASK-128 | main |  |  |  |  |  |  |
 | TASK-135 | a track can be declared but no existing row can be moved onto it | Coding Agent | not_started | — | — | V3 |  | main |  |  |  |  |  |  |
 | TASK-136 | a queue track SLA is parsed, stored and never measured against anything | Coding Agent | not_started | — | — | V3 |  | main |  |  |  |  |  |  |
 | TASK-140 | every mode contract slot is assigned to an axis, and the spine-to-unit map is written down | Coding Agent | review | PR #23 — three open questions for the user, incl. whether an empty illegal-pair list discharges § 7 risk 2 | evidence/2026-08/TASK-140-dispatch-2026-08-21-result.md | V3 | — | main |  |  |  |  |  |  |
-| TASK-141 | a row stays blocked after its blockers close, because the stored status masks the computed one | Coding Agent | not_started | — | — | V3 |  | intake | triaged |  | 2026-08-20 |  |  |  |
+| TASK-141 | a row stays blocked after its blockers close, because the stored status masks the computed one | Coding Agent | in_progress | dispatched to claude-subagent; worktree pinned to f42e84b | evidence/2026-08/TASK-141-spec.md | V3 | — | intake | triaged |  | 2026-08-20 |  |  |  |
 | TASK-142 | triage has no check for a row stranded by a process bug, and the one signal that fired was read as prose hygiene | Coding Agent | not_started | design question answered 2026-08-20: it belongs in conformance, which triage already reads at step 0.5 — not as a new triage feature | — | V3 | — | intake | triaged |  | 2026-08-20 |  |  |  |
-| TASK-143 | two PRs each green on their own base merged into a red tree, and nothing checked the pair | Coding Agent | not_started | — | — | V3 |  | intake | triaged |  | 2026-08-20 |  |  |  |
+| TASK-143 | two PRs each green on their own base merged into a red tree, and nothing checked the pair | Coding Agent | in_progress | dispatched to claude-subagent; worktree pinned to a10f897 | evidence/2026-08/TASK-143-spec.md | V3 | — | intake | triaged |  | 2026-08-20 |  |  |  |
+| TASK-144 | the event log timestamp has no zone and the register has one, so ordering them is a guess | Coding Agent | not_started | — | — | V3 |  | intake | triaged |  | 2026-08-20 |  |  |  |
 
 ## P2
 
@@ -68,6 +69,9 @@
 | TASK-132 | the parity check cannot see 23 keys because Perry own state leaves four collections empty | Coding Agent | not_started | — | — | V3 |  | main |  |  |
 | TASK-137 | a new queue row is born in the second stage, not the first | Coding Agent | not_started | — | — | V2 |  | main |  |  |
 | TASK-139 | a design back-reference lives in a cell the close path clears, so a finished design reports as never handed off | Coding Agent | not_started | — | — | V3 | TASK-102 | intake | triaged | 2026-08-20 |
+| TASK-145 | the contract shape baseline is stale against its own recorder | Coding Agent | not_started | — | — | V2 |  | intake | triaged | 2026-08-20 |
+| TASK-146 | the viewer renders a KR current with no provenance because it does not go through the shared derivation | Coding Agent | not_started | — | — | V3 |  | intake | triaged | 2026-08-20 |
+| TASK-147 | nothing outside describe_cell proves the table and bullet paths stay separated | Coding Agent | not_started | — | — | V3 |  | intake | triaged | 2026-08-21 |
 
 ## Cadence (recurring; doesn't consume P0 slots)
 

perry/evidence/2026-08/TASK-120-dispatch-2026-08-21-result.md

@@ -0,0 +1,106 @@
+# TASK-120 — result
+
+> Date: 2026-08-21 · Executor: claude-subagent · PR: https://github.com/ranjiao/Perry/pull/24
+> Branch: `coding/task-120-kr-progress-provenance` · Cycle time: ~35 min
+> `perry-goals/list` **2.0 → 2.1**, additive: four keys added, none removed or
+> retyped. `schema/state-schema.json` **untouched** — verified by diff, and the
+> hard stop was live, so this is a real avoidance rather than luck.
+
+## The shape: Perry exposes the contradiction, it does not resolve it
+
+One derivation in `bin/lib/__init__.py`, three sibling keys per KR, emitted by
+both payloads. Verified on this repository:
+
+```
+P-O1.1  current 0.0  target 1.0
+        provenance  state=asserted  measured=false  source=linkage-register
+        completion  total 4  done 4  open 0
+P-O2.2  current 0.0  target 0.0
+        provenance  state=asserted  measured=false
+        completion  total 2  done 0  open 2
+```
+
+`P-O1.1` no longer reads as 0-of-1 progress: it reads as **an author's assertion
+of 0 against four closed tasks**, and Perry resolves neither. `P-O2.2` can no
+longer be read as met, because nothing claims the zero was measured.
+
+**No `met` / `achieved` / `progress` / `ratio` key exists** — asserted as an
+absence in the tests. No percentage is emitted anywhere: counts only, in their
+own unit, so the tally cannot be misread as a metric value. That is the line the
+spec drew and it held.
+
+## What it rejected, and one of the rejections is the interesting one
+
+- any ratio or percentage;
+- a conformance entry for *"`current` disagrees with its tasks"* — that would be
+  Perry inferring the metric from the edges, which is the forbidden move one
+  step removed;
+- **any new authored field in the register.** `asserted_at` reuses the
+  register's existing top-level `updated`, and `asserted_scope: "register"` is
+  emitted so a reader cannot mistake it for a per-KR date. **That choice is what
+  kept `schema/state-schema.json` out of the change** — the agent found the
+  cheap path around a gate rather than asking for a release.
+
+## The tally is what flips P-O1.1, not the staleness check
+
+Worth recording because it is counter-intuitive: P-O1.1 is **not stale by any
+timestamp test** — all four of its tasks closed *before* the register's
+`updated`. The contradiction is visible only because the completion tally sits
+beside the number. A design that shipped staleness alone would have left that
+KR reading exactly as wrongly as before.
+
+## Staleness, both directions on one fixture
+
+```
+not stale  "no linked task has changed state since 2026-08-15T12:00:00"
+stale      "1 linked task changed state after 2026-08-15T12:00:00:
+            TASK-003 (in_progress → done)"
+           moved_tasks: [{id, from, to, at}]   ·  only that KR goes stale
+```
+
+The fixture also carries a `next` event dated after the assertion, **so a check
+keyed on the event name rather than on whether `to` is a status would redden.**
+
+## Parity, stated separately as instructed
+
+`perry-goals/list` **before**: 0 documented-not-emitted, **5** emitted-not-documented.
+**After**: 0 and **5** — the same five, byte-identical, TASK-131's, not hidden
+inside. Documented 54 → 77, emitted 59 → 78. Repo-wide total unchanged at 17.
+
+## Four findings handed back
+
+1. **The `Z` problem, unresolved and documented in the contract.** The register
+   writes `updated` as ISO with a `Z`; `.perry/events.jsonl` writes `ts` as naive
+   local time. There is no honest conversion, so the `Z` is **stripped rather
+   than applied**, and a register written within a few hours of a task move can
+   order wrongly. Fixing it means deciding what the event log's `ts` means — a
+   row of its own, touching every consumer.
+2. **`tests/fixtures/contract-shapes.json` is stale w.r.t. its own recorder.**
+   `--record` wants to add an `empty_lists` block to two contracts and drop a
+   trailing newline. The agent refused to re-record rather than hide unrelated
+   drift in this row. **Someone will eventually re-record it into an unrelated
+   diff.**
+3. **`viewer/serve.py` renders the chain view from `viewer/parsers.py`**, not
+   through `bin/lib`, so the viewer still shows `current` with no provenance.
+4. **`measured` is `false` everywhere, by construction, until something re-runs
+   a metric.** Honest — and it means `stale` is Perry's only mechanical opinion
+   about a KR number.
+
+## Two notes on this project's own rules
+
+- The `current: 0` default the spec targeted lives in
+  `goals/state/linkage_TEMPLATE.md:13`, which writes it into every new register.
+  That is **authoring**, belongs to TASK-119's writer, and was correctly not
+  touched. `_num()` already returned `None` for an unwritten value, so V3 item 2
+  was true but untested; it is now locked by four tests.
+- **TASK-091's Definition of Done makes `bin/` a place where the history of that
+  defect cannot be written down.** The agent's first draft of a comment named the
+  deleted symbol and reddened `test_goals_writer`; it rewrote the comment to
+  describe the symbol without naming it. Same family as TASK-126 and TASK-112 —
+  a guard that forbids describing the thing it guards.
+
+## Process error, mine
+
+The worktree was cut from `feat/work-modes` **before** I committed the spec, so
+the agent had to fetch it with `git checkout e71d7c0 -- <path>`. Corrected for
+TASK-122 and TASK-141: commit the spec, *then* cut the worktree.

perry/evidence/2026-08/TASK-121-spec.md

@@ -0,0 +1,85 @@
+# TASK-121 — the sweep that finds checks reading live state runs once and is thrown away
+
+> Source: `perry/evidence/2026-08/TASK-113-dispatch-2026-08-20-1813.md`
+> Dispatch mode: auto
+> Executor: claude-subagent
+> Estimated cycle: medium
+> Subjective verification: no
+> Touches architecture: no — it adds a guard over the test suite
+> Deployed: no
+
+## Schema
+
+- **Owner**: Coding Agent
+- **Priority**: P1
+- **Attribution**: unlinked
+
+## The class, and its eight known instances
+
+A check that reads **the project living around it** as its expected value goes
+red on ordinary progress, and green again for reasons that have nothing to do
+with what it measures. Every instance below is real and dated within four days:
+
+| # | check | what moved under it |
+|---|---|---|
+| 1–3 | `test_diagnose`, `test_one_line_break_rule`, `test_v5_signoff` | TASK-113 found and fixed three |
+| 4 | a fourth, handed over mid-run | same row |
+| 5 | a fifth the agent found itself — `DESIGN-900` | same row |
+| 6 | `test_md_store § test_config_including_its_prose_section` | asserted every config record is a `setting`; **declaring one track reddened it** |
+| 7 | `test_track_attribution § TestPerrysOwnProjectIsUnmoved` | asserted Perry itself has no track register; same declaration reddened it |
+| 8 | `test_state_cost` ×2 | asserted `perry/tasks.jsonl` is unclaimed and `.perry/events.jsonl` rolls up under `.perry/` — **both true until PR #14 declared the two store files owned** |
+
+TASK-113 fixed instances 1–5 **by hand, in one pass, and the pass was thrown
+away.** Instances 6–8 arrived afterwards. There is no mechanism; there is a
+memory of having looked.
+
+## Deliverable
+
+A guard that finds this class **mechanically**, so the next instance is reported
+rather than discovered by a human running the suite after a merge.
+
+**What "this class" is, precisely, is the hard part of this row** — and getting
+it wrong in either direction makes the guard worthless:
+
+- too broad, and it flags every test that reads a fixture, which is all of them;
+- too narrow, and it is a list of the eight above wearing a regex.
+
+The instances give you the shape to generalise from: each one asserted a
+**literal about the project's current state** — a count, an id, a set membership,
+a filename — where the *property* being tested was true independently of that
+literal. Note that instance 8's literals were about **which paths the schema declares
+Perry owns**, not about a board row — so a guard keyed only on `BOARD.md` or the
+task store would have missed it.
+
+Report what you decided the class is, in the guard's own docstring, in the voice
+of the surrounding modules — and **name what it deliberately does not catch.**
+
+## Verification — V3
+
+1. **It finds instances it was not shown.** Reconstruct at least three of the
+   eight from git history — `test_md_store` and `test_track_attribution` before
+   their 2026-08-21 fixes, and one of TASK-113's — and show the guard flags them.
+   Reconstruct, do not hand-write an approximation.
+2. **It does not flag the fixes.** The same three, after their repairs, are
+   clean. A guard that still flags the repaired form is measuring the wrong
+   thing.
+3. **False-positive floor, stated as a number.** Run it over the whole suite as
+   it stands and report **every** hit. If the count is not zero, each survivor
+   is either a real instance — open a row for it — or a false positive you must
+   name and explain. **Do not silence one to reach zero.**
+4. **Reverting the guard reddens its own test.**
+5. `python3 tests/parallel -j 4`, `bash tests/run`, `python3 bin/perry-lint`,
+   `git diff --check`.
+
+## Files in scope
+
+- the guard, as a new test module or a check under `tests/`
+- its own tests and fixtures
+
+## Out of scope
+
+- **Fixing any instance you find.** Report them; each is its own row. This row
+  ships the mechanism, not the repairs.
+- `bin/perry-diagnose` and `tests/test_diagnose.py` — an unmerged branch (PR #22)
+  is editing both. Cutting across it would conflict.
+- `perry/` — no project state changes; `git diff -- perry/` must end empty.