Skip to content

Changelog

Unreleased

Changed

  • Citation metadata: Zenodo DOI of 0.6.0 (10.5281/zenodo.24291830), a DOI badge (concept DOI 10.5281/zenodo.24021995) and an abstract that describes the ledger core.

0.6.0 - 2026-11-16

Changed

  • Renamed to Protocolith (was ClaimBound Evidence). Package and module protocolith, command protocolith with the short alias plith. Releases up to 0.5.0 remain available under the old name.
  • Evidence ids use the PLITH- prefix. Every card keeps its old id in legacy_evidence_id; docs/MIGRATION_ID_MAP.json maps old to new ids with file hashes. Hash-pinned artifacts keep their original text.
  • The core uses only the standard library. numpy moved to the runners extra and the PDF libraries to the campaign extra; commands that need them explain what to install.
  • Python 3.11 to 3.15 are supported (was 3.12 and newer).
  • Core specification draft: docs/spec/CORE_SPEC.md.

  • evidence_id is unique per record: PLITH-<protocol_id>-<date>-<seq>, where seq is the ledger seq of the record (six digits), so two records of one protocol on one day can both be issued. Ledger format v stays 1; this is a 0.6.0 pre-release change; a ledger written by an earlier development build reports BAD_CARD on its issue entries.

  • log accepts --protocol, --type, --status and --tail; protocolith_log takes the same arguments.
  • Errors print a one-line hint (CODE): ... (and error.hint in JSON) where advice is cheap. Exit codes are unchanged.

Added

  • protocolith atlas (also core.build_atlas and the MCP tool protocolith_atlas): one self-contained, offline HTML atlas of experiments from a ledger and/or evidence card directories, with summary tiles, status chart and timeline, search, filters, a sortable paged table, a detail drawer with permalinks and keyboard navigation, light and dark themes and a print style. English UI catalog atlas_i18n/en.json; further languages only from user files (--i18n DIR). A ledger that fails verification still produces the atlas with a red banner and exit code 1. See the atlas guide.
  • protocolith summary [--protocol ID] (also core.summary and the MCP tool protocolith_summary): verdict, head, counts by entry type and result status, a per-protocol table and the problems found, in one fast call.
  • protocolith show SEQ (core.show, protocolith_show): one entry with its protocol context and references.
  • protocolith capture [--file PATH ...] [--env] (core.capture, protocolith_capture): a JSON fragment with environment and input file hashes for a record outcome; runs nothing, reads no environment values, confines files to the working directory (BAD_PATH).
  • protocolith schema [NAME] (core.schema_for, protocolith_schema): JSON Schema (draft 2020-12) generated from the verifier's field rules.
  • MCP resources protocolith://summary, protocolith://ledger and protocolith://schema/<name> (resources/list, resources/read).
  • protocolith init: create an empty store (idempotent).
  • Guides: Use Protocolith from an AI agent and Query the ledger quickly; reference pages for MCP and the Python API; SUPPORT.md.

Fixed

  • A test fixture no longer depends on the developer's global git signing configuration.

Removed

  • The peer-comparison fair-race material was removed from the tree at the owner's decision (2026-11-14): the card with registry sequence 8073, its protocol charter, artifacts, two runner scripts, the atlas link and the Pages workflow entries. The registry now holds 8072 cards and its statistics and static views were regenerated. The removed files remain in git history before this change; history was not rewritten.

0.5.0 - 2026-11-06

Added

  • SourceProbe v1: claimbound probe URL [--marker TEXT] [--out FILE] [--json] records HTTP status, final URL, content type, byte size, SHA-256 (only for fully read bodies under a size cap), marker checks, licence/terms link hints and a safe_to_continue | needs_human_review | blocked recommendation. Metadata only: no raw body is stored and a probe never produces a result status (docs/SOURCE_PROBE.md).
  • Run-root deviation log and source-rights checklist: claimbound run-root deviation add|list and run-root rights answer (docs/DEVIATION_LOG.md).
  • Rerun report: claimbound rerun report classifies reproduction status and source-byte drift from two hash files without modifying the original card.
  • Static registry views: scripts/claimbound_build_registry_view.py (with --check) generates deterministic Markdown and a self-contained HTML index under docs/registry/views/ (docs/REGISTRY_VIEWS.md).
  • Public AI claim protocol: docs/PUBLIC_AI_CLAIM_PROTOCOL.md, a JSON schema and validator for model/API metadata and prompt/transcript hash policy.
  • Scaffold hardening: claimbound new prepares request, charter, playbook, checklist, operator declaration, draft card and registry-patch scaffolds; --dry-run; refuses to overwrite or escape --out.
  • Validator hardening: non-breaking warning checks (warn_evidence_card, registry consistency); existing cards are unchanged.
  • Accessibility: the SVG renderer now emits role="img", <title> and <desc> and meets 4.5:1 contrast for declared text colours; scripts/claimbound_lint_svg_a11y.py reports gaps (docs/ACCESSIBILITY.md). No WCAG conformance is claimed.
  • scripts/release_gate.sh: resumable local release gate.

Changed

  • conda-forge template converted to the v1 recipe.yaml format, which staged-recipes now requires for new recipes.
  • DCO check exempts the "Update branch" merge commit in Dependabot pull requests.

Known limits

  • The ~8000 committed SVG previews were not regenerated, so most still lack the new accessibility attributes.
  • No new evidence cards were added in this release, and no independent external rerun has happened yet.

0.4.8 - 2026-09-30

Added

  • claimbound --version.
  • PyPI project page: long description from the README (links and images rewritten to absolute GitHub URLs at build time), author, keywords, classifiers, license files and Discussions/Citation links.
  • AGENTS.md/CLAUDE.md contributor and coding-assistant rules, CITATION.cff, .zenodo.json and the pseudonym policy.

  • Brand assets: light and dark logo without background, square mark (SVG and PNG), 1280x640 social preview and a terminal demo (docs/assets/demo/, regenerated from real CLI output by make_demo.py).

  • Lint and coverage: ruff configuration and CI step, coverage report artifact, .pre-commit-config.yaml, CODEOWNERS, release-notes categories and an OpenSSF Scorecard workflow; SECURITY.md documents the release supply chain.

  • Documentation site (MkDocs Material) published under /docs/ next to the atlases on GitHub Pages; sources in docs_site/, assembled by scripts/build_docs_site.py. Adds a generated command reference (docs/CLI.md), a glossary, a scripts index and "I want to" paths in the documentation index.

  • packaging/: unsubmitted recipes for conda-forge, a Homebrew tap and a Docker image, plus packaging/render.py which fills them from a published PyPI release and uv.lock.

Fixed

  • NasaPowerMockConfig without an explicit seed raised NameError (missing hashlib import); covered by a regression test.

Changed

  • README first screen: theme-aware logo, badges (PyPI, downloads, tests, license, release, card count), demo recording, install in two lines and an "is / is not" table.
  • claimbound doctor recognises an installed package: mode=installed-package and ready=yes for card-level commands instead of repo_layout=FAIL.
  • validate-all, demo, rerun, drift and verify print how to clone the repository and exit with code 2 when run from a pip install, instead of a traceback.
  • Relative paths given to validate-card, inspect, hash and new --out resolve against the current directory when there is no repository clone.
  • claimbound inspect card FILE prints a short summary when --keys is omitted.
  • The source distribution no longer ships the whole repository (about 7 MB to about 60 KB).
  • Publication guard tests now check the files git would commit and ignore local, git-ignored outputs such as _site/.
  • docs/PLANNED_NOT_SHIPPED.md reflects that the package is published on PyPI.

0.4.7.4 - 2026-09-24

Changed

  • Release tooling: version bump after PyPI publishing workflow fixes.

0.4.7.3 - 2026-09-24

Changed

  • Release tooling: TestPyPI environment renamed from pypi to testpypi in the publish workflow; README surfaces the PyPI package (#364, #365).

0.4.7.2 - 2026-06-19

Added

  • Volunteer one-pager — single page with baseline, verify commands, closing comment and troubleshooting for external operators.
  • GitHub outreach templates — copy-paste DM, issue and Discussion invites for maintainers.

0.4.7.1 - 2026-06-19

Changed

  • External verification docs: honest time estimates (15 min first setup, ~2 min per verify pack after baseline), explicit why per VERIFY pack, and claimbound verify starter-pack as the primary #85 path.
  • VERIFY issue bodies #85–#87 synced to 86 tests and one-command verify shortcuts.
  • GitHub release notes for v0.4.7 corrected (GIF guide removed in #115).

Removed

  • Motion-demo placeholder guide and references from documentation (#115).

0.4.7 - 2026-06-19

Added

Changed

  • Documentation sync: 86 tests, Tier C wording, legacy optional shell sections in runbooks (#113).
  • VERIFY mirrors #85–#87 reopened for independent external operator closure.
  • Synchronized pyproject.toml, specs/repo_version.yaml, README release badge and GitHub release tag to v0.4.7.

Fixed

  • Audit gaps after cross-platform rollout: verify eea-drift, nasa-rerun, noaa-rerun shortcuts and CONTRIBUTING platform section (#113).

0.4.6 - 2026-06-15

Added

  • NOAA CO-OPS D-131 frozen gate runner (claimbound_run_noaa_coops_prereg.py) and maintainer negative reproduction_attempt card (closes #56).
  • NASA POWER D-103 maintainer reproduction_attempt card and rerun summary (closes #55).
  • API_PARITY_D001 software-development evidence card for registry validation parity (closes #59).
  • Operator runbooks under docs/runbooks/ for issues #55–#59.
  • EEA source drift check report and operator note (closes #57).
  • AI source-audit boundary review record (closes #58).

Fixed

  • Synchronized pyproject.toml, specs/repo_version.yaml, README release badge and GitHub release tag to v0.4.6 (resolves 0.4.5 / 0.4.4 / 0.4.3 mismatch).

Changed

  • Registry: 24 validated cards including NASA and API parity reruns.
  • Original NASA POWER D-103 card verification metadata updated after maintainer rerun.

0.4.5 - 2026-06-11

Added

  • Three European public-data source-audit cards (sequences 19–21): EU Data Portal, EEA content reuse FAQ and Eurostat API detailed guidelines.
  • scripts/claimbound_run_eu_public_source_audit.py runner with frozen profiles.
  • scripts/claimbound_build_eu_evidence_cards.py card builder.
  • tests/test_eu_public_source_audit.py with mocked HTTP coverage.

Changed

  • Registry: 21 validated cards; European Dimension page with five EU examples, SVG embeds, blocked-track narrative, operator rerun commands and explicit non-endorsement note.
  • EU card builder now rebuilds from saved sanitized summaries by default; live source refresh is explicit with --refresh-reports.
  • Updated current evidence tracks, reviewer summary and future applications index.

0.4.4 - 2026-06-11

Fixed

  • Evidence-card SVG renderer now uses inline text attributes so cards render in GitHub README <img> embeds (GitHub strips internal <style> blocks).
  • Clarified SOFTWARE_DEV_D001 allowed claim: required-field regression gate passed for execution_mode enforcement (no failure-style headline wording).

Changed

  • Moved SVG render logic to claimbound_evidence.card_svg_render; regenerated all committed evidence-card SVG files.

0.4.3 - 2026-06-11

Added

  • Software development workflow: where/how ClaimBound was used, strengths, and counterfactual without ClaimBound; track pattern examples split to companion doc.

Changed

  • docs/SOFTWARE_DEVELOPMENT_WORKFLOW.md expanded with completed in-repo and external case-study usage narrative.

0.4.2 - 2026-06-11

Changed

  • Embed the SOFTWARE_DEV_D001 evidence card SVG in README, software development workflow, case study page and current tracks summary.

0.4.1 - 2026-06-11

Changed

  • Cross-linked the public software-rnd-evidence-case-study repository (v1.0.0) from README and case study page.
  • Removed staging / manual public-flip wording from the external case study docs.

0.4.0 - 2026-06-11

Added

  • Reviewer surface release: European Dimension, rerun examples, software-dev card, external R&D case study page, sustainability plan.
  • CLAIMBOUND-SOFTWARE_DEV_D001-2026-06-11 validated software-development card.
  • docs/case_studies/SOFTWARE_RND_EVIDENCE_CASE_STUDY.md for external NO-GO R&D narrative (private staging repo until public flip).
  • docs/SUSTAINABILITY.md maintenance and scope boundaries.

Changed

  • README foreground: three primary workflows plus future applications tier.
  • Registry: 18 validated cards.

0.3.3 - 2026-06-11

Added

  • Added docs/EUROPEAN_DIMENSION.md for the European public-data and digital commons framing.
  • Added examples/rerun/README.md with copy-paste local verification commands.

Changed

  • Linked European Dimension from README and reviewer summary.
  • Independent rerun workflow now points to runnable example commands.

0.3.2 - 2026-06-11

Added

  • Added docs/future_applications/ guides for robotics, procurement/civic and education/open-science secondary domains.
  • Added docs/artifacts/README.md catalog for NYC TLC Phase 4 and CDC mirror artifact-only records.

Changed

  • README foreground now highlights three primary workflows: AI transparency, European/public open data and software development evidence.
  • Evidence card README now includes card versus artifact versus scaffold table.
  • Synchronized pyproject.toml version with release badge v0.3.2.

0.3.1 - 2026-06-05

Added

  • Added post-submit public hardening docs: reviewer summary, public roadmap 2026, external operator starter pack, governance, maintainer boundary and release process.
  • Added source-drift and card-boundary issue templates and external-operator labels/issues.

Changed

  • Clarified evidence cards versus non-card artifacts for NYC TLC and CDC mirror records.
  • Updated release metadata from v0.3.0 to v0.3.1.

Added

  • Added ClaimBound protocol v3 as a compatible tree overlay using protocol_version: "claimbound-tree-v3".
  • Added iron_claim, flow_claim, tombstone, claimbound-tree-v3, badge counts and branch-blocking validation rules.
  • Added src/claimbound_evidence/tree_overlay.py for deterministic v3 tree overlay validation.
  • Added CLI command claimbound validate-tree for docs/track_families/*_TREE.json overlays.
  • Extended claimbound validate-all so optional v3 tree overlays are validated when present.
  • Added script entrypoint scripts/claimbound_validate_family_tree.py.
  • Added protocol documentation in docs/PROTOCOL_V3_TREE_OVERLAY.md.
  • Added software developers and maintainers as a public audience category in README-facing documentation.
  • Added docs/PROTOCOL_LAYERS_V2_V3.md to explain the difference between evidence cards, v2 family/frontier ledgers and v3 tree overlays.
  • Added a software developers and maintainers scenario to docs/AUDIENCE_TESTIMONIAL_WORKFLOWS.md.
  • Added docs/PROTOCOL_USE_BY_LAYER_AND_AUDIENCE.md with practical guidance for choosing card-only, v2 and v3 stacks by audience.
  • Added docs/AI_RISK_CONTROL_WITH_CLAIMBOUND.md with practical guidance for using ClaimBound as an evidence-bound AI risk-control layer without claiming certification, runtime safety or complete risk removal.
  • Added a README ClaimBound In 30 Seconds overview, a standalone docs/CLAIMBOUND_IN_30_SECONDS.md guide and a program eligibility self-check runbook.
  • Added PROGRAM_FIT_D001 as a completed non-branded program-eligibility self-check card with explicit single-operator and no-endorsement boundaries.

Changed

  • Replaced README badges that depended on dynamic GitHub-backed shields endpoints with local static SVG badges.

Compatibility

  • Historical evidence cards are not migrated or reinterpreted.
  • Existing claimbound-rnd-family-v2 family and frontier ledgers remain supported.
  • v3 tree overlays are optional planning/preflight metadata; they are not result evidence by themselves.