Release 0.2.0: adopt the accepted Direction, Horizon, Debt and T9 repairs, and carry Integrity history across bundles (B3)

.devostasis/targets.json

@@ -20,7 +20,7 @@
     {
       "id": "B3",
       "title": "Durable revision history across bundles",
-      "state": "open"
+      "state": "closed"
     },
     {
       "id": "B4",
@@ -45,6 +45,11 @@
     {
       "id": "v0.2",
       "title": "Release 0.2: phases A and B closed, one outside consumer, no unadopted research unit",
+      "state": "closed"
+    },
+    {
+      "id": "v0.3",
+      "title": "Release 0.3: phases A and B closed, one outside consumer, no unadopted research unit (the bar 0.2.0 shipped before)",
       "state": "open"
     },
     {

.github/workflows/observe-self.yml

@@ -4,7 +4,7 @@
 #
 #   jobs:
 #     vitals:
-#       uses: drevendev/devostasis/.github/workflows/observe-self.yml@v0.1.9
+#       uses: drevendev/devostasis/.github/workflows/observe-self.yml@v0.2.0
 #       with:
 #         debt-labels: "type:debt"
 #
@@ -24,7 +24,7 @@ on:
       devostasis-ref:
         description: Git ref of Devostasis to install (tag, branch or commit)
         type: string
-        default: v0.1.9
+        default: v0.2.0
       planning-source:
         description: milestones, file or none
         type: string

CHANGELOG.md

@@ -3,6 +3,125 @@
 All notable changes to this project are documented here. Semantic changes to a
 contract or a policy always come with a version bump of that contract.
 
+## 0.2.0 (2026-10-01)
+
+The adoption release. Every accepted research contract about the Vitals and
+their history that this repository had not adopted, and that needs no owner
+decision, is adopted: the repairs of Direction, Horizon and Debt that issue
+#33 listed, the final T9 reconciliation, and target B3, durable Integrity
+revision history (`PV-HIST-002`, #34). The owner asked for a large next
+release on 2026-09-30 (#47); the candidate specification written for that
+request (`PV-RELEASE-020-001`, "Evidence Observatory") has not been accepted
+yet, so 0.2.0 is the part of that ambition that rests on accepted contracts.
+No threshold, window, gauge constant or demand mapping changed. Four rule ids
+moved, each with the judgement that moved it, so the first 0.2.0 bundle of
+every project compares `INCOMPARABLE` (`RULE_VERSION_BOUNDARY`) on Horizon,
+Direction, Integrity and Debt and stays comparable on the rest.
+
+- **Direction `direction.bands.v2`: state-neutral linkage, and incomplete
+  linkage without failing the project.** `PV-REV-DIRECTION-CLOSED-TARGET-001`
+  (cases `DIR-CLOSED-01..09`): an active change request linked to a declared
+  target stays linked when the target closes. Calibration finding 9 is
+  repaired: closing a delivered target used to un-link the work that
+  delivered it, so never closing targets kept the band higher than finishing
+  them. Measured on a copy of the fleet store: this repository reads `MIXED`,
+  9 of 18 active change requests traced to a declared target, where the
+  open-target count said 4 of 18, `SCATTERED`. That count is still derived,
+  for presentation only. `PV-DIRECTION-INCOMPLETE-001` (accepted by
+  `PV-REV-DIRECTION-INCOMPLETE-001`, cases `DIR-INCOMPLETE-01..16`): every
+  active change request is `LINKED`, `UNLINKED` or `UNRESOLVED`. A reference
+  the complete register lacks is a broken reference, unlinked and diagnosed
+  `DIRECTION_TARGET_REFERENCE_MISSING`; a reference nobody could resolve, or a
+  title, body or milestone of the wrong type, is unresolved. Over
+  `L + U + R = N` the completions `k = L..L+R` go through the unchanged
+  predicates: one reachable band is `DEGRADED / <band> / EXACT`, several are
+  `UNKNOWN` with the reachable set in `derived`. Unreadable linkage evidence no
+  longer fails the whole project: `LinkageEvidenceError` is gone, and a change
+  request names what it could not read in `linkage_unresolved`. A partial
+  target list no longer blocks Direction when every referenced target resolves
+  on its own, as a milestone embedded in the change request does.
+- **Target markers are standalone tokens** (`PV-AUDIT-TARGET-MARKER-SYNTAX-001`).
+  `SubTarget: B1`, `NotTarget: B1` and `PreTarget:B1` linked their change
+  request to B1, because the configured marker was matched as a substring.
+  The marker now starts the text or follows a character that cannot continue
+  a word, and a marker ending in a letter needs a boundary after it too. The
+  audit's regressions `LINK-MARKER-01..08` reached this repository only as
+  categories; the tests cover each category and say so.
+- **Horizon `horizon.bands.v2`: partial target enumeration**
+  (`PV-HORIZON-PARTIAL-001`, accepted by `PV-REV-HORIZON-PARTIAL-001`, cases
+  `HOR-PARTIAL-01..16`). A partial enumeration that returned a target proves
+  the capability; its counts are observed-subset lower bounds with the proof
+  #39 asked for; only an open target already beyond the frame forces a band,
+  `DEGRADED / EXTENDED / EXACT`. Every other partial shape stays `UNKNOWN`,
+  because Horizon declares no order to pick a representative by, and a
+  partial zero is never `UNDECLARED`.
+- **Debt `debt.bands.v2`: partial register** (`PV-DEBT-PARTIAL-001`, accepted
+  by `PV-REV-DEBT-PARTIAL-001`, cases `DEBT-PARTIAL-01..16`). A confirmed open
+  item under a partial enumeration is `DEGRADED / PRESENT / EXACT`, the band
+  exact and the count a lower bound, none confirmed is `UNKNOWN`, and a
+  `PARTIAL` count is read as a lower bound only with its
+  `OBSERVED_SUBSET_COUNT` proof (the Debt part of #39). Before, a partial
+  positive count was `CONSERVATIVE_LOWER_BOUND` on its status alone, so an
+  estimate could prove `PRESENT`.
+- **Integrity `integrity.bands.v1+ci-unit-004+hist-002`: durable revision
+  history** (`PV-HIST-002`, accepted by `PV-REV-HIST-002`, target B3). History
+  was reconstructed from whatever the provider still showed, so a failure
+  hidden behind a passing re-request of a check suite, or dropped by
+  retention, vanished from the counts. The union of every attempt observed per
+  parent per revision is now carried from bundle to bundle: a build adds its
+  immediate predecessor's union as `ci.revision_history_carried`, Integrity
+  merges it with what the provider shows now and emits the result as
+  `derived.revision_history`, on every path, so history also crosses a bundle
+  whose verification evidence could not be read. A missing, unverifiable,
+  foreign or later predecessor is an explicit gap (`REVISION_HISTORY_GAP`);
+  nothing older is ever carried in its place, and `verify` checks that the
+  source a bundle names is the bundle its manifest follows
+  (`HISTORY_SOURCE_MISMATCH`). Favorable evidence proves `PASS_ONLY_OBSERVED`
+  only when every attempt of every parent is observed, so a check-suite pass,
+  or a run whose earlier attempts are not all known, is `UNKNOWN_HISTORY`
+  (R52), and unknown history makes the Vital `UNKNOWN`: T9 puts it above every
+  conservative branch. On the stored corpus every active revision is
+  attempt-complete, so no fleet project loses its band to that rule. The first
+  0.2.0 bundle of a project replays the revision records of its 0.1.9
+  predecessor (lineage `devostasis.ci-history.v1`) instead of starting empty.
+  The band table and every 0.1.9 judgement are unchanged. Cases `HIST-01..07`,
+  `09..11`, `13`, `15`, `16` and `19` are vectors, `08`, `12`, `17`, `18` and
+  `20` store tests; `HIST-14` has no accepted migration to execute.
+- **T9 is adopted** (`PV-TEST-004`, generator `PV-T9-GEN-003`, accepted by
+  `PV-REV-TEST-004`). With the repairs above, every obligation T9 consumes is
+  implemented; a finite generator of about 1,270 cells over the nine former
+  gap families holds each cell to the accepted table. The T9v2 cell-key
+  vectors are still the research process's to deliver.
+- **Contradictory counts are refused as input.** Counts no population can
+  produce (`L + U + R != N`, a beyond-28-day count above the future-boundary
+  count) are `InadmissibleEvidence` before classification, as both totality
+  contracts require, and `evaluate` and `build` report an input error (exit
+  status 2) instead of a band.
+- **The vector format learned two things.** `given.derive` runs the derivation
+  over stated inventories, so a case about how a change request becomes
+  linked, or how a partial target list is counted, is executed where that
+  happens; `expect.rejected` states a refusal (`HOR-PARTIAL-15`). Seventy-one
+  new vectors: `DIR-CLOSED`, `DIR-INCOMPLETE`, `HOR-PARTIAL`, `DEBT-PARTIAL`,
+  `HIST`.
+
+Upgrading: the first 0.2.0 bundle of every project carries four
+`RULE_VERSION_BOUNDARY` rows. Durable history starts from the predecessor's
+in-window records; after the fleet's pause since 2026-09-10 (#32) none are
+left, so the first carried history is effectively empty, and a project whose
+predecessor kept no observations member starts with one explained gap. A
+saved 0.1.x observation set under file planning recorded a reference outside
+the register as `UNKNOWN`; replayed, that change request is unresolved rather
+than unlinked, and a fresh observation tells the two apart. The change-request
+record gains `linkage_unresolved` and the reference state `MISSING`; the
+revision record gains `history_complete`. The snapshot carries the durable
+history, stored grouped by parent shape: 91 KB for the busiest project of the
+fleet, whose 834 dispatched workflow runs in two weeks #40 already asks about.
+
+Not in this release: the fifth identity move (#19, the owner's decision), the
+Phase C contracts of #34 (after B7 and the compatibility policy), the Pulse
+part of #39 (a question for the research side), and the "Evidence
+Observatory" candidate of #47, which awaits its review.
+
 ## 0.1.9 (2026-09-30)
 
 Four things in one release. First, the adoption of the research judgements

README.md

@@ -46,8 +46,10 @@ Devostasis is built against those failures:
   configuration or provider access.
 - **Anti-gaming rules are part of the contract.** A revision that failed
   verification keeps its failure in the history window even if a retry later
-  passes. A zero queue is not "good flow". Full milestone linkage is a fact,
-  not a compliment.
+  passes, and even after the provider stops showing the failed attempt: the
+  history is carried from one bundle to the next. Closing the target a change
+  delivered does not un-link the change. A zero queue is not "good flow".
+  Full milestone linkage is a fact, not a compliment.
 
 ## The seven Vitals
 
@@ -153,7 +155,7 @@ comes first is the consumer's policy, and no accepted contract defines it.
 Requires Python 3.12 or newer. The runtime uses the standard library only.
 
 ```bash
-pip install git+https://github.com/drevendev/devostasis@v0.1.9
+pip install git+https://github.com/drevendev/devostasis@v0.2.0
 ```
 
 Observe one repository (a GitHub token is read from `DEVOSTASIS_GITHUB_TOKEN`,
@@ -212,7 +214,7 @@ permissions:
 
 jobs:
   vitals:
-    uses: drevendev/devostasis/.github/workflows/observe-self.yml@v0.1.9
+    uses: drevendev/devostasis/.github/workflows/observe-self.yml@v0.2.0
   decide:
     needs: vitals
     runs-on: ubuntu-latest