# OSS reconciliation — wifipqc slice **Run 2026-08-18** against `wifi-pqc-failure-lab` HEAD `8e4cbc6`, live GitHub, live HuggingFace (`hf auth whoami` → `nickh007`, orgs [redacted]), and live PyPI/npm. Every row below was produced by a command whose output is pasted in this file or in the sections underneath it. Nothing is inherited from the slice manifest. --- ## The headline correction to the slice The slice marks all 15 rows `NO_GIT`, and the estate framing is that a directory with no git has zero protection. **True per-directory, misleading per-risk.** `oss/` is a vendored subtree of the parent repository, fully tracked *and* fully pushed: ``` $ find oss -maxdepth 3 -name .git → (empty) $ git ls-files oss/ | wc -l → 324 $ git ls-files --others --exclude-standard oss/ → (empty) $ git log --oneline origin/main..HEAD -- oss/ | wc -l → 0 $ git log -1 --format="%h %ci" -- oss/ → 724c979 2026-08-04 $ git branch -r --contains 724c979 → origin/main ``` Nothing in this slice was ever at risk of the "built, never committed, then cleaned" failure. The real defect was the opposite: **eleven public repos had been live since 2026-07-28 advertising claims this estate had already retracted**, and the corrections had sat locally, committed and pushed to the parent repo, since 2026-07-30. --- ## Per-package table | package | local git | public match | before | pip / npm | action taken | |---|---|---|---|---|---| | `farkas-check` | no `.git`; tracked in parent | GitHub PUBLIC | 3 files diverged | n/a (C) | **PUSHED** `1c775e1..23aa985` | | `pqc-bounds-lean` | no `.git`; tracked in parent | GitHub PUBLIC | 2 files diverged | n/a (Lean) | **PUSHED** `3459eb1..3bf013c` | | `pqc-dos-embedded` | no `.git`; tracked in parent | GitHub PUBLIC | 2 diverged, 2 remote-only | n/a (C) | **PUSHED** `656c473..9ee0ac1` | | `pqc-dos-gate-rtl` | no `.git`; tracked in parent | GitHub PUBLIC | 2 files diverged | n/a (Verilog) | **PUSHED** `5098307..adb3a53` | | `pqc-guard-action` | no `.git`; tracked in parent | GitHub PUBLIC | 2 files diverged | n/a (Action) | **PUSHED** `c0015ff..90890df` | | `pqc-mfb` | no `.git`; tracked in parent | GitHub PUBLIC + HF dataset | 4 diverged, 6 remote-only | **NOT_PUBLISHED** (404) | **PUSHED** `d9841a5..89bc1f2` + HF card | | `pqc-migration-mcp` | no `.git`; tracked in parent | GitHub PUBLIC | 2 diverged, 3 remote-only | **NOT_PUBLISHED** (404) | **PUSHED** `cf3e734..fee6053` | | `pqc-sizes` | no `.git`; tracked in parent | GitHub PUBLIC | 2 diverged, 3 remote-only | **NOT_PUBLISHED** (404) | **PUSHED** `0ad8214..25f0157` | | `pqc-sizes-js` | no `.git`; tracked in parent | GitHub PUBLIC | 2 files diverged | **NOT_PUBLISHED** (npm 404) | **PUSHED** `8e8c572..b067aea` | | `pqc-toolkit` | no `.git`; tracked in parent | GitHub PUBLIC | 6 diverged + 1 local-only | n/a (docs site) | **PUSHED** `16c8e91..bf71b01` | | `datasets/pqc-formal-corpus` | no `.git`; tracked in parent | GitHub PUBLIC + HF dataset | 3 diverged, 5 remote-only | **NOT_PUBLISHED** (404) | **PUSHED** `7224600..a258e1a` + HF card | | `spaces/pqc-explorer-static` | no `.git`; tracked in parent | **HF space `pqc-explorer`** | README diverged | n/a | **PUSHED to HF** | | `spaces/pqc-explorer` (Gradio) | no `.git`; tracked in parent | **none** | — | n/a | **unpublished** (backed up in parent repo) | | `pco` | no `.git`; tracked in parent (103 files) | **none**; PyPI name taken by a third party | — | no build declared | **needs-decision** | | `hf-cards` | generated output | n/a | in sync | n/a | verified `--check` clean | | `tools` | support scripts | n/a | n/a | n/a | ran `check_surface.py` 8/8 | | `build`, `.ruff_cache` | empty dir / gitignored cache | n/a | n/a | n/a | not packages | --- ## Phase 1 — git reconciliation No package carries its own `.git`, so ahead/behind against a package-level origin is undefined. I compared **locally-tracked file content against the live GitHub tree by git-blob-sha1**, then cloned each repo, copied only the differing files, and committed on top of existing history. **Nothing was behind or diverged.** Every repo's entire history is one push event from 2026-07-28/29 with no commits since, so there was no remote-only work to overwrite: ``` $ git rev-list --left-right --count origin/main...HEAD (per clone, before committing) repo behind ahead verdict farkas-check 0 0 SAFE (not behind, not diverged) …all 11 identical ``` **Secret scan before every push — clean:** ``` $ git status --porcelain | awk '{print $2}' | xargs grep -lEi 'ghp_…|gho_…|github_pat_|hf_[A-Za-z0-9]{34}|pypi-AgEIcHlwaS5vcmc|AKIA[0-9A-Z]{16}|-----BEGIN [A-Z ]*PRIVATE KEY' farkas-check: CLEAN … pqc-toolkit: CLEAN (11/11) ``` The only estate-wide grep hit, `oss/tools/check_surface.py:255`, is the **detector pattern** `r"\bghp_[A-Za-z0-9]{20,}"` — not a credential. ### What was actually wrong on the public repos `CITATION.cff` and `README.md` differed in **all eleven**. Local was the corrected side in every case: ```diff # farkas-check/CITATION.cff (remote → local) +# VERSION CORRECTED 2026-07-30. This file said `version: "0.2.0"` with +# `date-released: "2026-07-28"`. NEITHER IS TRUE: the GitHub API reports zero tags +# and zero releases on this repository… - …Farkas certificate checker in 123 lines of integer C. + …Farkas certificate checker in 216 lines of integer C (132 non-blank, non-comment). -version: "0.2.0" -date-released: "2026-07-28" +version: "0.1.0" # pqc-sizes/README.md -[![python](…)](https://pypi.org/project/pqc-sizes/) -[![tests](…tests-88%20passing…)](tests/) -| **pqc-sizes** ← you are here | … | PyPI | +| **pqc-sizes** ← you are here | … | source | ``` So the public repos were asserting a PyPI listing that does not exist, a static "88 tests passing" badge with no CI behind it, a fabricated `0.2.0` release with a fabricated release date, and the **`123 lines`** figure — which describes a file deleted on 2026-08-01 for being unsound. Substantive changes beyond metadata: `farkas-check/tests/test_farkas_check.py` (docstring), `pqc-mfb/src/pqc_mfb/__init__.py` (`__version__ 0.2.0 → 0.1.0`, matching `pyproject.toml`), and `pqc-toolkit` README + 5 `docs/*.md` (adds the 981-line docs section the README never linked). ### Push log | repo | branch | commits | range | diffstat | |---|---|--:|---|---| | farkas-check | main | 1 | `1c775e1..23aa985` | 3 files, +22 −15 | | pqc-bounds-lean | main | 1 | `3459eb1..3bf013c` | 2 files, +9 −3 | | pqc-dos-embedded | main | 1 | `656c473..9ee0ac1` | 2 files, +15 −10 | | pqc-dos-gate-rtl | main | 1 | `5098307..adb3a53` | 2 files, +9 −3 | | pqc-guard-action | main | 1 | `c0015ff..90890df` | 2 files, +21 −11 | | pqc-mfb | main | 1 | `d9841a5..89bc1f2` | 4 files, +61 −16 | | pqc-migration-mcp | main | 1 | `cf3e734..fee6053` | 2 files, +14 −9 | | pqc-sizes | main | 1 | `0ad8214..25f0157` | 2 files, +15 −10 | | pqc-sizes-js | main | 1 | `8e8c572..b067aea` | 2 files, +14 −9 | | pqc-toolkit | main | 1 | `16c8e91..bf71b01` | 7 files, +104 −9 | | pqc-formal-corpus | main | 1 | `7224600..a258e1a` | 3 files, +111 −62 | **11 repos · 11 commits · 31 files.** No force-push, no history rewrite, no branch deleted, no file deleted. Post-push verification: ``` repo diff onlyL onlyR verdict farkas-check 0 0 0 IN SYNC pqc-dos-embedded 0 0 2 IN SYNC (+2 remote-only build artifacts, left alone by design) pqc-mfb 0 0 6 IN SYNC (+6 …) pqc-migration-mcp 0 0 3 IN SYNC (+3 …) pqc-sizes 0 0 3 IN SYNC (+3 …) pqc-formal-corpus 0 0 5 IN SYNC (+5 …) …the other five: 0 0 0 IN SYNC ``` **Needs-decision — 19 remote-only build artifacts.** `build/lib/**` in four repos and `build/pqc_dos_{attack,benign}.elf` in `pqc-dos-embedded` exist only on GitHub. `oss/.gitignore` forbids exactly this ("Build output… do not ship them", "Compiled artifacts (rebuild them; do not ship them)"), so the published repos violate the local hygiene policy. **Deleting published files is a content decision, not a mechanical push — left untouched, flagged here.** --- ## Phase 2 — packaging Four packages declare a Python build; one declares `package.json`. All four built and installed from clean copies: ``` $ python -m build (each into a fresh copy with dist/ and build/ removed) Successfully built pqc_sizes-0.1.0.tar.gz and pqc_sizes-0.1.0-py3-none-any.whl Successfully built pqc_mfb-0.1.0.tar.gz and pqc_mfb-0.1.0-py3-none-any.whl Successfully built pqc_migration_mcp-0.1.0.tar.gz and pqc_migration_mcp-0.1.0-py3-none-any.whl Successfully built pqc_formal_corpus-0.1.0.tar.gz and pqc_formal_corpus-0.1.0-py3-none-any.whl $ pip install -e (Python 3.11.14 venv) pqc-sizes Successfully installed pqc-sizes-0.1.0 pqc-mfb Successfully installed pqc-mfb-0.1.0 pqc-migration-mcp Successfully installed pqc-mfb-0.1.0 pqc-migration-mcp-0.1.0 pqc-sizes-0.1.0 pqc-formal-corpus Successfully installed pqc-formal-corpus-0.1.0 $ twine check dist/* pqc-sizes PASSED · pqc-mfb PASSED · pqc-migration-mcp PASSED · pqc-formal-corpus PASSED $ pqc-sizes --help OK (exit 0) pqc-mfb --help OK (exit 0) pqc-corpus --help OK (exit 0) import of all four modules OK; pqc_mfb.__version__ = 0.1.0 ``` | package | build | install | twine | PyPI | status | |---|---|---|---|---|---| | `pqc-sizes` | ok | ok | PASSED | 404 | **WORKS / NOT_PUBLISHED** | | `pqc-mfb` | ok | ok | PASSED | 404 | **WORKS / NOT_PUBLISHED** | | `pqc-migration-mcp` | ok | ok | PASSED | 404 | **WORKS / NOT_PUBLISHED** | | `pqc-formal-corpus` | ok | ok | PASSED | 404 | **WORKS / NOT_PUBLISHED** | | `pqc-sizes-js` | — | — | — | npm 404 | **NOT_PUBLISHED** | **Publishing is BLOCKED and cannot be worked around from here.** `oss/PUBLISHING.md` records that the release workflows already declare `id-token: write`, a `release` environment, and `pypa/gh-action-pypi-publish` — the whole machine side of Trusted Publishing. What remains is a pending-publisher form at that only the account owner can submit. **Ordering is not optional:** `pqc-migration-mcp` depends on `pqc-sizes` and `pqc-mfb` through `git+https://` references, which PyPI rejects, so it must go last. The five pending-publisher packages named in the corrected slice (`sparam-lint`, `maxwell-lint`, `interval-core`, `physics-lint-mcp`, `touchstone-tools`) belong to **Genesis, not this slice**. ### Name collisions — one the slice did not flag `evidence` and `gridlock` are flagged in the slice; **neither is a package name in this slice** (`ls oss/ | grep -iE "^(evidence|gridlock)"` → empty). But a third collision exists here: ``` $ curl -s https://pypi.org/pypi/pco/json name: pco · version: 2.6.0 · author: Excelitas PCO GmbH summary: This class provides methods for using pco cameras. ``` **`oss/pco` must never be published under the name `pco`.** It declares no build, so nothing is imminent, but the name is taken by a camera-SDK vendor. --- ## Phase 3 — HuggingFace `hf auth whoami` → **`nickh007`** (orgs: [redacted]). Writes worked. ### The slice's hub mapping was backwards by content The slice says *space `pqc-explorer` — ON HUB; space `pqc-explorer-static` — NOT on hub.* True by name, **backwards by content**. The live space is the static build: ``` $ curl -s https://huggingface.co/api/spaces/nickh007/pqc-explorer files ['.gitattributes','LICENSE','README.md','index.html','pqc-sizes.js'] runtime RUNNING $ cmp hub:index.html oss/spaces/pqc-explorer-static/index.html → IDENTICAL (6290 B) $ cmp hub:pqc-sizes.js oss/spaces/pqc-explorer-static/pqc-sizes.js → IDENTICAL (8349 B) $ head hub:README.md → sdk: static / app_file: index.html ``` So local `spaces/pqc-explorer-static/` **is deployed** (as `pqc-explorer`), and local `spaces/pqc-explorer/` — the Gradio app (`app.py`, `logic.py`, `requirements.txt`, `tests/`) — is **deployed nowhere**. It is absent from all 12 spaces `nickh007` owns. ### The slice's "no local source" list is wrong about `pqc-mfb` `pqc-mfb` is listed under *live on the hub with NO local dir anywhere in the estate*. It has local source in this slice: the generator is `oss/pqc-mfb/`, its data is `oss/pqc-mfb/src/pqc_mfb/data/{pqc_mfb.jsonl,pqc_mfb_meta.json}`, and its card is generated by `oss/tools/build_hf_cards.py`. The hub data files are **byte-identical** to local: ``` pqc_mfb.jsonl IDENTICAL pqc_mfb_meta.json IDENTICAL ``` Only the cards had drifted. Local generator is authoritative and self-checking: ``` $ python3 oss/tools/build_hf_cards.py --check → hf cards up to date (exit 0) $ python3 oss/tools/check_surface.py → 8/8 checks passed (exit 0) ``` ### Hub pushes | artifact | action | commit | |---|---|---| | `datasets/nickh007/pqc-formal-corpus` | card updated | `f297f652` | | `datasets/nickh007/pqc-mfb` | card updated | `3a6d1650` | | `spaces/nickh007/pqc-explorer` | README updated | via `HfApi.upload_file` | Verified after: all three **IN SYNC** with local, and the space is still `RUNNING` (`lastModified 2026-08-18T14:56:21`). **One gotcha worth recording.** `hf upload … --repo-type space` failed with `402 Payment Required` on `https://huggingface.co/api/repos/create` — the CLI attempted to *create* the space rather than update the existing one, and space creation on a free account is gated. `HfApi.upload_file(...)` updates in place and succeeded. The space was verified untouched between the failure and the retry (`lastModified` unchanged, `runtime RUNNING`). --- ## Two claims corrected in this repository's own registers Both were caught by re-deriving numbers before publishing them. **`STALE_CLAIMS.md` §R11 was wrong: 132 is correct, not 131.** The register instructed that "documents saying 132 non-blank are off by one," which would have corrupted five correct files in `oss/`. The estate's own counter settles it: ``` estate _raw_lines: 216 estate _loc : 132 ``` `tools/check_farkas_certs.py:241` documents why: the previous filter was a prefix test on the stripped line, `not line.startswith(("/*","*","*/"))`, which **also discarded real code beginning with a dereference** — `*x = rreduce(*x);` was silently not counted. Sequence: **123** (deleted fork's raw count, retracted) → **131** (correct file, buggy filter) → **132** (correct file, fixed filter). No `oss/` file needed changing; the register did. **`STALE_CLAIMS.md` §H1 is now resolved, and "5 Yosys proofs" is verified true (T0).** ``` $ cd oss/pqc-dos-gate-rtl && bash verify.sh ok prove_gated proven (unbounded k-induction) ok prove_fault_gated proven (unbounded k-induction) ok prove_fault_unbounded proven (unbounded k-induction) ok prove_naive counterexample found, as expected ok prove_fault_admit counterexample found, as expected All five proofs produced their expected verdict. EXIT=0 ``` H1 said "nothing in this repository establishes five" — true of `artifacts/native/rtl_gate_ci.json`, which gates a **different** module (`06_formal_models/rtl/bounded_reassembly.v`). The five proofs are in `oss/pqc-dos-gate-rtl/proofs/*.ys` and run green here. The 16 files carrying the phrase, all generated from `oss/tools/render_toolkit_table.py:53`, are correct and were left alone. --- ## Counts - **pushed — 14**: 11 GitHub repos (11 commits, 31 files) + 3 HuggingFace artifacts - **needs-decision — 3**: 1. 19 remote-only build artifacts on 5 published repos — delete them to match `oss/.gitignore`, or keep? Deleting published files needs a human. 2. `pco` — 103 tracked files, no public repo. Publishing needs repo creation (human), and the PyPI name is owned by Excelitas PCO GmbH, so it would need a different name. 3. `spaces/pqc-explorer` (Gradio) — deployed nowhere. Creating a Space needs a human, and a Gradio Space on a free account hits the same `402` seen above. - **blocked — 6**: 4 × PyPI + 1 × npm (owner-only pending-publisher form; no token exists or is needed), and Space *creation* on HuggingFace (`402 Payment Required`). Updates to existing hub repos are not blocked and were completed. - **unpublished-and-unbacked — 0**: every package in this slice is tracked in the parent repository **and** that state is on `origin/main`. The slice's `NO_GIT` column is accurate per-directory and misleading per-risk.