aidanmoller · Sep 14, 2026 · Sep 10, 2026
@@ -28,10 +28,10 @@ author.
 ## Review
 
 Every pull request is reviewed under `REVIEW.md` and needs one approving
-review from an author who did not write it. Branch protection on `main` is the intended
-enforcement mechanism for that rule and for the required checks; this
-repository does not record the host configuration, so confirm it on the
-repository before relying on it.
+review from an author who did not write it. Branch protection on `main` enforces this: a pull request is
+required, one approving code-owner review, the ten required checks green, and
+no force push or deletion. Administrators may bypass, so the rule binds the
+ordinary workflow rather than every possible push.
 
 ## Reporting
 

@@ -22,7 +22,7 @@ this process applies from 2026-09-02.
 | Design | `spec.md`: requirements, design, flagged concerns | same directory; the package-level specification is `dev/specification.md` | flagged concerns resolved before engineering starts; the other author signs off | the authors |
 | Build | `plan.md`: files that change, order of work, risks, proof | same directory, committed before code; a departure is recorded in the same commit as the change that causes it | a reader unfamiliar with the change could implement it from the plan alone | the author of the change |
 | Test | the diff, its tests, the evidence register | `acir/tests/`, `acir/inst/evidence/register.csv`, `.github/workflows/` | the four gates below, locally before the pull request and again in CI | CI, deterministically |
-| Deploy | review findings ranked by severity | the pull-request thread, under `REVIEW.md` | one approving review from an author who did not write the change; branch protection is the intended enforcement, see the correspondence table below | the reviewing author |
+| Deploy | review findings ranked by severity | the pull-request thread, under `REVIEW.md` | one approving review from an author who did not write the change; branch protection enforces it, see the correspondence table below | the reviewing author |
 | Maintain | incident record | a new `intent.md` | a bench breach or a numerical disagreement re-enters at Plan | the authors |
 
 ## The four gates
@@ -90,7 +90,7 @@ and is a review finding in its own right.
 | Feedback loop before reporting done | the four gates, run locally and in CI | in use |
 | Continuous evaluation of agent configuration | not applicable to the repository | not used |
 | Review loop under a written policy | `REVIEW.md`; approval by a human author is required | in use |
-| Hooks as approval gates | intended branch protection on `main`: pull request required, one approving review, the required checks green, no force push | intended; the host configuration is not recorded in this repository and has not been verified here |
+| Hooks as approval gates | branch protection on `main`: pull request required, one approving code-owner review, ten required checks green, no force push or deletion | in use, enabled 2026-09-10 and read back from the host API; administrators may bypass |
 | An agent inside CI/CD | not used; CI is deterministic | not used |
 | Control bands in maintenance | `bench.yaml` compares every run with `baseline.csv`; a breach is triaged by the authors and re-enters as an intent | in use, with a manual response |
 | Recurring security scans | not used; the package opens no network connection (`SECURITY.md`) | not used |

@@ -10,6 +10,7 @@
 ## Cairns
 
 <!-- BEGIN auto-generated cairn list (regenerated by /ledger reindex) -->
+- [2026-09-10 T3 open] [acir 0.2.0 releases the guards and the scoped evidence claims, with the near-zero limitation documented rather than closed](r-package/2026-09-10-release-020-guards-and-scoped-claims.cairn.md)
 - [2026-09-09 T3 open] [The near-zero observation Gram ships documented for 0.1.0, and the scale guard is queued as its own work package](r-package/2026-09-09-near-zero-gram-documented-not-guarded.cairn.md)
 - [2026-09-02 T4 open] [acir 0.1.0 is tagged at parity with the reference implementation, with every performance budget met, before the remaining plan steps](r-package/2026-09-02-release-0.1.0-parity-milestone.cairn.md)
 - [2026-08-14 T3 open] [An unresolved causal influence range is right-censored, and is reported as a bound](r-package/2026-08-14-censored-range-not-missing.cairn.md)

@@ -0,0 +1,115 @@
+---
+id: 2026-09-10-release-020-guards-and-scoped-claims
+schema_version: 1.4
+date: 2026-09-10
+tier: T3
+classification: open
+export_status: local_only
+consensus_mode: none
+domain: r-package
+project: aciR
+status: accepted
+title: "acir 0.2.0 releases the guards and the scoped evidence claims, with the near-zero limitation documented rather than closed"
+tags: [release, versioning, guards, evidence, governance]
+triggers:
+  - adr_class_commitment
+  - release_reconciliation
+  - breaking_change_with_deprecation
+reversal_cost: medium
+decision_pressure: publication
+review_due: null
+review_trigger: "at the next release, or when the observation-scale guard work package lands and changes what is refused"
+supersedes: []
+superseded_by: null
+related: [2026-09-02-release-0.1.0-parity-milestone, 2026-09-09-near-zero-gram-documented-not-guarded]
+---
+
+## Status
+
+Released 2026-09-10. Tag `acir-v0.2.0` is annotated, sits on `6461892` (the
+merge of the release pull request), and its tree is byte-identical to the
+candidate that returned `Status: 1 NOTE` from `R CMD check --as-cran` locally,
+where the single note is `New submission`. Ten checks green on that commit, read
+back from the host API rather than inferred. `main` then opened at `0.2.0.9000`
+and branch protection was enabled and read back the same way.
+
+## Alternatives considered
+
+| Option | One-line description |
+|---|---|
+| A | Patch release 0.1.1, treating the work as documentation corrections |
+| B (CHOSEN) | Minor release 0.2.0, on the ground that three changes stop working calls from working |
+| C | Hold the corrections unreleased until the observation-scale guard closes the near-zero case |
+| D | Remove the retired `T` argument at 0.2.0, as the expiring deprecation notice promised |
+
+## Rationale for rejection
+
+### Option A
+
+`API_STABILITY.md` states that at 0.x the interface may change with the change
+announced under a "Breaking changes" heading. Three of this release's items make
+previously working calls fail: the declared R floor moved to 4.1.0 so
+installation on 4.0.x is refused; a degenerate observation-noise Gram is refused
+wherever it occurs, where before it returned numbers; and overflow in a
+recursion raises rather than returning a silent non-finite value. A patch
+release asserts that nothing a caller wrote stops working, which would be false.
+
+### Option C
+
+The near-zero case cannot be closed by a threshold chosen under release
+pressure -- the reasoning is in the companion cairn. Holding an otherwise
+finished release for it would have kept the corrected claims off the public
+record for the sake of a limitation that is now stated in the help, in NEWS, in
+five tests and in a work package. Correct claims shipped beat correct claims
+withheld.
+
+### Option D
+
+The notice said `T` would be accepted "until 0.2.0", so arriving here forced a
+choice. Removing it would break callers at the same release that already carries
+three breaks, for an alias that costs nothing to keep. The promise was retired
+instead of the argument, and NEWS records that no removal version is scheduled.
+The cost is a deprecation with no end date, which the 1.0 planning inherits.
+
+## Implemented option (B)
+
+Version, citation metadata and the software-metadata record all read 0.2.0 and
+R 4.1.0. `NAMESPACE` is byte-identical to the `acir-v0.1.0` tag, so no exported
+surface moved. The fixture and evidence subtrees hash identically to the
+pre-release `main`, and the only R file the release touched carries a
+deprecation message and its roxygen. No graded quantity changed.
+
+## Forward cost
+
+- The release ships a documented route on which the implicit stepper returns a
+  collapsed posterior from a near-degenerate observation model, and a flooring
+  policy that can return a value of order 1e21 with a warning. Both are stated;
+  neither is closed.
+- The `T` alias now has no scheduled removal. Semantic versioning is promised
+  from 1.0, so the removal needs a home in the 1.0 plan or the promise erodes.
+- Branch protection binds the ordinary workflow from this release onward: each
+  author now needs the other's approval, and administrators may bypass. The
+  process record was corrected to describe what the host actually enforces.
+- The evidence register carries 68 rows of which 40 are behavioural-only, and
+  the fixture producer route remains static and unexercised. Both are recorded
+  gaps that a reader of the register will meet, and the next release is when
+  they are most likely to be read closely.
+
+## References
+
+### Methodological
+
+- Preston-Werner, T. *Semantic Versioning 2.0.0*. https://semver.org/ -- the 0.x clause under which a minor bump carries breaking change before 1.0.
+
+### Empirical
+
+- `acir/NEWS.md`, `# acir 0.2.0` -- the Breaking changes heading and its three items.
+- `acir/API_STABILITY.md` -- the 0.x interface promise this release is measured against.
+- Local release-candidate check, 2026-09-10: `R CMD build` then `R CMD check --as-cran` returned `Status: 1 NOTE` (`New submission`); tests, examples with `--run-donttest`, and all three vignette rebuilds passed; the installed candidate reported version 0.2.0 and `citation("acir")` followed it.
+- Subtree hashes against the pre-release `main`: fixtures and evidence register identical; all of `R/` except `aci-model.R` identical.
+
+### Operational
+
+- Cairn: `2026-09-09-near-zero-gram-documented-not-guarded.cairn.md` -- why the near-zero limitation ships documented.
+- Cairn: `2026-09-02-release-0.1.0-parity-milestone.cairn.md` -- the release this one succeeds.
+- Work package: `dev/work/2026-09-10-release-0.2.0/plan.md`, and `dev/work/2026-09-09-observation-scale-guard/intent.md` for the deferred guard.