Repository navigation
0.2.0 “Evidence Observatory”: the owner's request of 2026-09-30, the candidate specification, and how it maps onto this repository #47
Description
The tracking issue for the next major release, filed by the review of 2026-09-30 so that the plan has a public record here; until now it exists only on Drive.
Where it comes from
- Owner request, 2026-09-30, to the research process: a large next major patch ("сделай огромный мажорный следующий патч"), preserved in Drive
01 - InputsasOWNER_REQUEST_2026-09-30. - Candidate specification
PV-RELEASE-020-001(research run 410), "Devostasis 0.2.0 — Evidence Observatory", candidate 1: 25 sections, 16 implementation packagesR020-W01..16, 73 acceptance scenariosR020-S001..073. Status PRODUCED, independent review required: the composition, the new CLI, the source-format profiles and budgets, the conformance carrier, date-only precision, the Explorer boundary and the exit codes are six open review questions (RD-01..06) for the gateG1. The accepted contracts it builds on are not reopened.
What 0.2.0 would be
Four Instruments beside the seven Vitals (Coverage, TestState, Deployment, Work; never an eighth Vital, never an input to Integrity), GitLab as a first-class provider, one bound provider context for endpoint, credential, cache and representation, strict admission of every input before defaults, exact-lineage verification and replay, crash-, concurrency- and rename-safe history with durable CI contribution, explain / doctor / timeline commands, and an offline HTML Explorer derived from verified bundles. No health score, no ranking, no model in the runtime, no aggregate: the non-goals of 0.1 stay.
How it maps onto this repository
The dossier's first instruction is to reconcile #29 and #31 before anything else; that is the precondition, not a package.
| Wave | Package | What | Already tracked here |
|---|---|---|---|
| A | W01 strict admission and configuration | complete schema before defaults, duplicate-safe decoding, saved-observation lineage | #35 (config/register shape, observation lineage), #38 (saved observation sets), #45 (configuration merge and schema) |
| A | W02 exact time and lifecycle | one instant language, chronology before derivation | #35, timestamp and chronology family (runs 382..395) |
| A | W03 provider context and bounded HTTP | base/TLS, endpoint binding, cache context, response size, pagination Link | #35, runs 397..403 and PV-AUDIT-GITHUB-PAGINATION-LINK-001 |
| A | W04 historical lineage and whole-bundle verifier | exact lineage dispatch, identity copies present and equal, receipt/observation coherence | #35 verifier family, #19, #46 (historical renderers); the renderer_version bypass is repaired in #31 |
| A | W05 history transactions and durable CI | same-project serialization, rename recovery, PV-HIST-002 |
#34 (B3), #35 (concurrency, rename recovery), #12 finding 6, #41 (backfill and id-less binding) |
| A | W06 GitHub payload and planning completeness | per-field validation, marker grammar, subset trust | #35 payload families (mostly repaired in #31), marker syntax; PV-REV-PR-031-003 repaired in #31; #33 (Direction closed-target); #39 (PARTIAL trust in Pulse and Debt); #40 (which runs are verification) |
| B | W07 instrument envelope and successor carrier | config.v2, effective-config.v3, instruments.json, receipt identity |
#34 (C3), #19 |
| B | W08..W12 artifact profiles and the four Instruments | Cobertura/JaCoCo/JUnit profiles, COV, TSTATE, DEPLOY, WORK cases |
#34 (C3) |
| B | W13 GitLab | GLA-01..29 |
#34 (C1) |
| C | W14 explain, doctor, timeline | new read-only CLI | new |
| C | W15 literal reports and Explorer | escaping, observability truthfulness, render.v5 |
#35 (PV-AUDIT-RENDER-OBSERVABILITY-TRUTHFULNESS-001), #44 (live Markdown titles, dropped qualifiers) |
| D | W16 conformance, migration, release | lossless conformance carrier, compatibility policy, migration rehearsal | #20, #23, #34 (conformance surface, PV-COMPAT-002) |
What this issue asks for
- Ship 0.1.9 first. Nothing has been released since v0.1.8 on 2026-09-07; the release branch carries three weeks of adoption and repair. 0.2.0 starts from a master that has them.
- Do 0.2.0 as a sequence of small pull requests under this issue, one package or less each, as the dossier's section 18 says. The lesson of Adopt every delivered research judgement and repair the review findings (0.1.9) #31 is concrete: one pull request of 53 files and some 7,000 lines stayed open for eight days, collected four research reviews against moving heads, and two of its blockers were recorded only on Drive.
- Gate
G1before the new surfaces freeze: the research review of the composition andRD-01..06. Wave A implements accepted contracts and does not wait for it; the CLI names, exit codes, Explorer and budgets do. - The fleet is down (The fleet has produced no bundle since 2026-09-10: the daily Observe job is refused before it starts #32), so no package can be measured against the store until billing is fixed; the 0.2.0 shadow-mode rollout the dossier describes needs it.
🤖 Generated with Claude Code
Activity
drevendev commentedon Sep 30, 2026
#53 is the 0.2.0 the owner asked for, scoped to what rests on accepted contracts: Wave A's W05 durable CI contribution (
PV-HIST-002, target B3) and the W06 parts that were accepted (the marker grammar, subset trust for Debt and Horizon, and the Direction repair of #33), plus the Horizon, Debt and T9 adoptions. It is one pull request because the request was for one major release, cut into nine commits that can be reviewed one by one; about 1,300 lines are source and most of the rest is the generated vector corpus.Everything that needs gate G1 (the instruments, GitLab, the new CLI, the Explorer, the conformance carrier) is not in it, so the roadmap now names the condition it had set for 0.2 (Phase B closed, an outside consumer, no unadopted research) as the bar for 0.3. The fleet was still down (#32), so the release was measured with a local read-only run into a copy of the store rather than the scheduled job.
🤖 Generated with Claude Code
drevendev commentedon Oct 14, 2026
Manual release recovery and implementation handoff — 2026-10-14
RUN-433 /
PV-MANUAL-REPOSITORY-HANDOFF-20261014-001.This updates the existing release tracker; it is not a new release proposal or approval to merge.
What is now delivered
The current default branch is
master@3bd60db707c5022f5df6034458cb23b949911dad, the September 30 merge of #29 / 0.1.9. The old prerequisite “ship 0.1.9 first” is therefore satisfied. The later 0.2–0.6 pull requests remain release candidates, not shipped versions.Five source-backed handoffs were published and individually read back today:
filter=all. The release chain already usesdevostasis.ci-history.v2; do not silently reuse that identifier for new semantics.The latter semantic proposals remain candidates for independent acceptance. Publication, semantic acceptance, implementation and successful verification are four different facts.
Integration sequence, not another expanding release
Live verification gate
An exact-head query for
09d497f16520b6dc20f5dd42077ff76774e1f5e1returned one upstream workflow run: CI #37777635756, created October 8, with conclusionaction_required, not success. Active master rules requireTests (Python 3.12),Tests (Python 3.13)andTests (Python 3.14). An operator must resolve the workflow authorization/execution gate and obtain the required results before release integration; this research handoff neither approves nor reruns workflows.Manual repository communication writes succeeded today. That does not prove scheduled-agent writes are repaired or explain their earlier failures. No code, branch, release tag, repository setting or workflow was changed in this manual research run.
Publication receipt:
DV-MANUAL-20261014-I47-f0c267d9.drevendev commentedon Oct 16, 2026
Manual release handoff delta — 2026-10-16
The new material since the October 14 handoff is three additional independently reviewed admission findings delivered on PR #75: CI attempt-state uncertainty, malformed timestamps crossing durable CI history, and unjustified offline source exclusions (including the limit of self-declared blob-size evidence). They now have a public implementation handoff and belong in the finite repair set, not in another expanding release concept.
Recommended implementation sequence:
masterremains3bd60db707c5022f5df6034458cb23b949911dad(0.1.9). PR #75 is open and GitHub reports it mechanically mergeable; that is not reviewed, tested release readiness. Fresh GET of upstream CI run 37777635756 still reportscompleted / action_requiredat exact head09d497f16520b6dc20f5dd42077ff76774e1f5e1, not a successful suite. The operator's workflow-authorization gate and subsequent required checks remain distinct from this research delivery. No approval, rerun, merge or tag was performed here.Keep candidate semantic/version decisions explicit. A complete package run and the four-bundle timestamp regression were not executed by this manual research pass. Manual discussion and Drive writes working now do not establish that scheduled executions are repaired.
Publication receipt:
DV-MANUAL-20261016-I47-419e7b6f.andy-zen-dev commentedon Oct 22, 2026
Owner-requested repository delivery / retained debt reconciliation — 2026-10-22
This updates the delivery plan after the owner's request to review open PRs, integrate fixes and move useful accumulated work into GitHub. It does not replace accepted research, reopen the historical Drive control plane, or mark candidate work released.
Work actually persisted
Retained artifact disposition
source_selection; keep final acceptance under #72/#73/#81, not a duplicate new defectThe retained .crab ledger's 18 entries are proposals/receipts, not eighteen accepted product tasks. Concurrency and Dependabot map to #37/#63; release automation to #24; OS runners and job timeouts to the bounded #84 scope; docs site to #85. Cache tuning, generic badges/CODEOWNERS/PR-template suggestions and upstream dependency lists are not release blockers. Do not install the upstream project's 38 dependencies into this stdlib-only runtime merely because a tool suggested them.
Next delivery order
A matching ROADMAP.md update and consolidated repair package are being prepared locally for a maintainer-capable checkout. This comment is the durable plan already saved in GitHub; it is not a claim that ROADMAP.md was committed.