v0.2.1 — P3 author≠approver enforcement

frontmatter-schema (check #1) now requires approver on DecisionRecords with status: ratified created on/after 2026-07-03 — the enforcement companion of the spec v1.1.0 P3 ratification (DR-2026-07-02-author-approver-separation). Legacy DECs are exempt (fix-on-touch back-fill, P12). 57 tests, green with and without the optional deps.

v0.2.0 — evidence-schema check #10

Check #10: evidence-schema

Validates evidence.v1 records (the canonical SliceOps evidence record format, ratified in spec DR-2026-07-02) against the canonical JSON Schema.

• Discovery: files ending .evidence.json / .evidence.v1.json under --root; filenames containing .example. are golden fixtures and are never picked up. Zero records → SKIPPED (green).
• Full JSON Schema draft 2020-12 validation when jsonschema is importable; otherwise a documented stdlib subset (required fields, schemaVersion const, enums, ID/hash patterns, top-level additionalProperties, and the P6 slice-merge completeness rule) — same optional-dependency policy as PyYAML. Enums/patterns are read from the vendored schema at run time.
• The vendored schema (templates/consistency-validators/schemas/evidence.v1.schema.json) is byte-compared against the spec canonical on every CI run; drift fails the build.
• 20 new tests (54 total), green with and without the optional deps.

The toolkit now ships 10 checks.

SliceOps Toolkit v0.1.1

Hardening release — fixes an external review (2026-06-19), each fix with regression tests (33 total; green with PyYAML and the stdlib fallback CI uses).

validators.py

• Windows/Linux portability — path filtering is now OS-agnostic (os.sep-aware), so .git, frozen dirs, and .github/workflows are detected on Windows too.
• topic-tags semantics — an unconfigured --topic-taxonomy reports SKIPPED (green); a configured-but-missing file is a clear hard error (no more "skipped" that failed the build).
• Robust frontmatter — uses PyYAML automatically when installed (block scalars, nesting, comments); documented stdlib subset + inline-comment handling as fallback. No dependency required.
• llm-ci-cost — strips YAML comments before the endpoint heuristic (no false-positive on comments/docs); structural concurrency / on.pull_request checks via PyYAML when available.

calibrate.py

• Honest bands — propose_bands (which ignored its percentile inputs) is split into canonical_bands() (fixed reference) + observed_bands() (genuinely data-driven from p25/p50/p75/p90); both are printed.
• Stable percentiles — inclusive method + clamp to the observed range + n==1 handling + a small-sample (<8) advisory warning, so small corpora no longer extrapolate past the observed max.

The counter-atomicity date-slug false-positive was already fixed in v0.1.0.

SliceOps Toolkit v0.1.0

First public release of the SliceOps™ toolkit — reference CI templates + tooling for adopters. Stdlib-only Python 3.9+, no dependencies. MIT.

Included

• templates/consistency-validators/ — deterministic validators.py (9 checks: frontmatter-schema, no-orphan-decs, cross-references-bidirectional, topic-tags, counter-atomicity, principle/entity-count-coherence, band-unit, llm-ci-cost) + a tested regression suite + a ready CI wireup. Calibrated for real corpora (DR↔DR bidirectional, frozen-skip, --entity-key).
• templates/ci-guardrails/ — 5 Layer B.2 cost-economy levers (concurrency-cancel, change-scoped-gating, aggregation-required-gate, draft-skip, dependency-cache).
• templates/llm-ci-economy/ + templates/cost-ledger/ — LLM-inference cost economy workflow + 3-dimension cost ledger.
• calibration/calibrate.py — token/context bands from session logs.

Dogfooded: this repo runs its own validators + tests + gitleaks in CI on a protected main. See the README "Use it" for the quickstart.

SliceOps™ trademark pending (EUIPO #019381071); usage governed by TRADEMARK.md in sliceops-spec.