All notable changes to ZPE Video are documented here.
Format follows Keep a Changelog.
Versions follow MAJOR.MINOR.PATCH. The package version in
pyproject.toml is the authoritative version record.
Nothing outstanding after the v0.1.0 wedge selection. Follow-up work is
recorded in the per-phase ledger under .gpd/phases/ and the updated
docs/STATUS.md.
After a long research journey (see
docs/TRANSPARENCY_JOURNEY.md), one
narrow commercial wedge survived every falsification attempt: the ZPE
packet as a cross-writer bit-exact perception-receipt format for AI
video pipelines. v0.1.0 promotes that surface to a stable public API,
ships the supporting docs and tests, and preserves the full research
evidence alongside the chosen wedge.
- Public perception-receipt API at
zpe_video.receipt:PerceptionReceipt,Box,encode_receipt,decode_receipt,receipt_hash,verify_receipt,read_receipt,write_receipt, plus error typesReceiptCorruptedandCrossWriterMismatch, and constantsWIRE_MAGIC/WIRE_VERSION. Zero upstream runtime dependency (stdlibstruct/zlib/hashlibonly). tests/test_receipt.py— 20 tests covering round-trip fidelity, cross-writer bit-exactness against an independent from-spec encoder, frame-level CRC32 integrity, error paths (bad magic, wrong version, truncated header, trailing bytes, CRC mismatch), theverify_receipthelpers, I/O helpers, and the 255-box / signed-16-delta limits.docs/WEDGE.md— commercial-wedge rationale with the explicit list of buyer shapes (C2PA perception credentials, regulated chain-of-custody, video-LLM object-memory caches, cross-platform training-data provenance).docs/TRANSPARENCY_JOURNEY.md— phase-by-phase verdict ledger from Phase 1 through Phase 09.4.1.1.2.1, including every kill verdict and defend verdict.docs/WIRE_FORMAT.md— byte-level spec for third-party re-implementation.docs/QUICKSTART.md— five-minute onboarding with three runnable examples.docs/transparency/— reproducible snapshot of the research evidence: PLAN, SUMMARY,summary.json, harnesses, and the research ledger (VERDICT-SYNTHESIS, ranked_hypothesis_ladder, takeover assessment) for Phases 09.4.1.1.2 and 09.4.1.1.2.1.examples/— three stdlib-only runnable examples (01_round_trip,02_cross_writer,03_file_round_trip).Makefile—install,test,lint,format,build,verify-package,cleantargets..github/workflows/ci.yml— CI on Python 3.11 / 3.12 / 3.13 matrix; runs the full test suite and ruff lint..github/workflows/verify-package.yml— wheel build + import surface check; pins the public__all__symbol set; independently runs the cross-writer wedge test.- Optional install extras in
pyproject.toml:dev(pytest, ruff),producer(numpy + opencv + ultralytics + torch for ingesting raw video),research(Wave-1 pipeline deps),all.
- Package version
0.0.0->0.1.0. src/zpe_video/__init__.pynow exports the perception-receipt surface as the primary public API.Wave1Pipelineremains importable for backward compatibility with the legacy research harness.pyproject.toml— full[project]metadata (classifiers, keywords, URLs), strictrequires-python = ">=3.11,<3.14", and the new optional extras.docs/ARCHITECTURE.md— reorganized around the public perception-receipt surface with the Wave-1 research pipeline moved below as subordinate.docs/STATUS.md— refreshed for the 2026-04-17 wedge-selection posture. Sovereign-gate status staysred.
The kill verdicts (Phase 08 universal-codec failure, Phase 09.4.1.1.2
archive-query falsification, Phase 09.4.1.1.2.1-A ROI-sidecar kill) are
preserved in full — both in the outer research tree and in
docs/transparency/. They are the reason the defend verdict is
trustworthy.
- Repo hardening and public-surface discipline (Phase 09.4). Root governance spine (CITATION, CODE_OF_CONDUCT, GOVERNANCE, ROADMAP, SECURITY, SUPPORT, CONTRIBUTING), public docs shell, and proof-routing surface were added.
- Pre-0.1.0 posture used "staging_only" labels throughout the doc set.
v0.1.0 retires that framing in favor of the always-in-beta public
posture: useful now, improving continuously. See
docs/STATUS.md.
