Updates

What moved since the pin. Every one of the 44 assessments was pinned September 22, 2026 (the assessed commit), and says nothing about what happened next. The suite kept moving. This page records the movement without rewriting a single verdict: a read-only census of every repository two days later, and a dated re-check wherever a verdict cell could change.

Found a census row or a re-check that is wrong? Send it through the correction form, a GitHub issue that asks for the claim, your evidence and the pin you checked.

Six definitions. The pin is the exact commit an assessment froze on; every claim is evaluated at the pin, not at a later HEAD (the newest commit). TRL is technology readiness level, scored 1 to 9. NODUS is the program's four-ring verdict scale (Invest, Pilot, Explore, Monitor). CI is continuous integration: automated checks that run on every push. The rider is the license clause withholding all rights, including benchmarking and testing, from OpenAI, Anthropic, their affiliates, and anyone acting for them. Bus factor is how many people would have to leave before a project stalls.

Live watchChecked every day, filed as issues

Watched daily by a GitHub Action: see the latest census and the changes open for triage.

Every day at 11:23 UTC a scheduled workflow reads all 44 repositories through the GitHub API and compares each with its pin and with the previous day. It writes a census, and opens an issue labelled watch for any change that could move a matrix cell: a new release or tag, a license edit, workflow files removed or appearing where there were none, archiving, a rename, a deletion, a rewritten pin, or a new franken or Rust repository. New commits and added workflow files are counted in the census but never filed.

Each issue then goes through the same loop: triage (could this move a cell?), a dated re-check under updates/ if it could, a review of that re-check by a separate agent session, any new practice added to the rigor table, and a close that links all of it. The watch reports and never edits a verdict. The rules are in updates/METHOD.md, and what the watch reads is in watch/README.md.

Why pinsA verdict needs a fixed object

Some of these repositories take dozens of commits a day. A finding about "the repo" with no commit attached is out of date before anyone reads it, and nobody can check it. So every packet names one commit and its date, and every claim in it is about that commit. When the code moves, the packet stays true about what it read; it just stops being the newest reading.

The response to movement is a new, dated file beside the old one, never an edit to the old one. A re-check pins its own commit, uses the same evidence tiers, and says for each matrix cell whether it would change and why. The published suite counts stay as they were on September 22; if a re-check would move one, it says so in its own text and leaves the count alone. The full rules are in updates/METHOD.md.

Census, 2026-09-24 UTCWhat moved in two days

For each repository we took the pin from its packet, then read the default branch today through the GitHub API: commits past the pin, releases and tags since, workflow files now against the pin, and the first line of the license now. Nothing was cloned for this table and nothing was run.

  • 32moved past the pin
  • 12did not move
  • 2could change a matrix cell
  • 1new GitHub Release
  • 0license changes
  • 0missing, renamed, or archived

Commits alone never change a cell here: readiness and ring need someone to read the new code, which is what a re-check is for. The two flagged rows come first below; the rest are ordered by how far they moved. Repository names link to the brief written at the pin. "Workflow files" counts files in .github/workflows, which shows whether CI exists but not whether it passes. The raw census, with head commits, run snapshots, commit authors, and the exact commands, is movement-2026-09-24.tsv.

RepositoryCommits past the pinReleases and tags sinceWorkflow files nowLicense nowWould a matrix cell change?
franken_code_browser 33 v0.1.0 Release 2026-09-23 none MIT + rider Yes. Release class: no release at the pin, a signed, notarized release now (R1 to R3 at the release commit). Readiness re-assessed from TRL 2–3 to 4–5. CI, license, and bus factor unchanged. Read the dated re-check.
frankengit 109 none 8 files; +1 / -71 vs the pin (78) MIT + rider Yes. CI class: CI with no verdict for the pinned commit (C3) at the pin; CI switched off (C5) at the re-check commit. 71 files were deleted, the 7 kept run only when started by hand, and no workflow has run since 2026-09-22. Readiness, ring, release, license, and bus factor unchanged. Read the dated re-check.
franken_surveillance_system 611 none 2 files, same set as at the pin MIT + rider No. The 611 include 373 commits dated before the pin: older history made reachable after it, not 611 new commits in a day and a half.
frankenterm 141 v0.15.12 tag 2026-09-22
v0.15.11 tag 2026-09-22
v0.15.9 tag 2026-09-22
none MIT + rider No. New tags only, no new Release object; release class already R3.
frankengraphdb 106 none 2 files, same set as at the pin MIT + rider No. Commits only.
frankensearch 104 none 8 files, same set as at the pin MIT + rider No. Commits only.
frankensim 90 none 4 files; +1 / -0 vs the pin (3) MIT + rider No. One workflow file added; C6 rests on the private DSR/RCH lanes, which did not change.
frankensqlite 64 none 9 files, same set as at the pin MIT + rider No. Commits only.
franken_manim 63 none 95 files; +8 / -0 vs the pin (87) MIT + rider No. Workflow set grew (87 to 95); C3 stands (no HEAD verdict yet).
franken_lean 50 none 97 files; +7 / -0 vs the pin (90) MIT + rider No. Workflow set grew (90 to 97) but runs at HEAD still fail, so C2 stands.
asupersync 44 none 12 files, same set as at the pin MIT + rider No. Commits only.
franken_markdown 41 none 1 file, same set as at the pin MIT + rider No. Commits only.
frankenlibc 41 none 8 files; +6 / -0 vs the pin (2) MIT + rider No. Workflow set grew (2 to 8) but runs at HEAD still fail, so C2 stands.
frankenfs 38 none 6 files; +2 / -0 vs the pin (4) MIT + rider No. Workflow set grew (4 to 6) but runs at HEAD still fail, so C2 stands.
franken_remote 37 none 36 files, same set as at the pin MIT + rider No. Commits only.
frankentui 31 none 4 files, same set as at the pin MIT + rider No. Commits only.
franken_engine 30 none 23 files, same set as at the pin MIT + rider No. Commits only.
franken_alignment 24 none none MIT + rider No. Commits only.
franken_node 24 none 51 file, same set as at the pin MIT + rider No. Commits only.
franken_threed 18 none 1 file, same set as at the pin MIT + rider No. Commits only.
frankenpandas 15 none 7 files, same set as at the pin MIT + rider No. Commits only.
frankensympy 13 none 2 files, same set as at the pin MIT + rider No. Commits only.
frankenscipy 11 none 3 files, same set as at the pin MIT + rider No. Commits only.
franken_numpy 7 none 10 files, same set as at the pin MIT + rider No. Commits only.
franken_snowflake 5 none 1 file, same set as at the pin MIT + rider No. Commits only.
frankenredis 2 none 2 files, same set as at the pin MIT + rider No. Commits only.
franken_networkx 1 none 4 files, same set as at the pin MIT + rider No. Commits only.
franken_ocr 1 none none MIT + rider No. Commits only.
frankenjax 1 none none MIT + rider No. Commits only.
frankensqlite_website 1 none none no LICENSE file No. Commits only.
frankentorch 1 none 2 files, same set as at the pin MIT + rider No. Commits only.
frankentui_website 1 none none MIT + rider No. Commits only.
beads-for-frankentui 0 none 1 file, same set as at the pin no LICENSE file No. Nothing new since the pin.
beads_for_franken_engine 0 none 1 file, same set as at the pin no LICENSE file No. Nothing new since the pin.
franken_agent_detection 0 none 5 files, same set as at the pin MIT, no rider No. Nothing new since the pin.
franken_drone_geometry_reconstruction 0 none 1 file, same set as at the pin MIT + rider No. Nothing new since the pin.
franken_markdown_website 0 none none MIT + rider No. Nothing new since the pin.
franken_native_capsule 0 none none no LICENSE file No. Nothing new since the pin.
franken_nlp 0 none 1 file, same set as at the pin MIT + rider No. Nothing new since the pin.
franken_overlap 0 none none MIT + rider No. Nothing new since the pin.
franken_tts 0 none none MIT + rider No. Nothing new since the pin.
franken_whisper 0 none none MIT + rider No. Nothing new since the pin.
frankenmermaid 0 none none MIT + rider No. Nothing new since the pin.
frankensim_website 0 none none no LICENSE file No. Nothing new since the pin.
Reading the commit countsA large count is not a large change. franken_surveillance_system's 611 are mostly older commits that became reachable after the pin. Two repositories show commits under new author names, BlackThrush and BoldGorge, not linked to any GitHub account. They look like the maintainer's agent personas, which the packets saw elsewhere in the suite, but git metadata cannot prove who is behind a name, so bus factor is not re-scored on this evidence. [Inference, Medium]

Dated re-checkfranken_markdown shipped v0.5.0; no cell moves

At the pin, franken_markdown was at 0.4.5: a zero-dependency Rust Markdown renderer with a claim-discipline gate, TRL 7, held at Explore by the absence of any independent benchmark or adoption evidence. On 2026-10-02 its maintainer shipped v0.5.0: 135 commits, 443 files changed, multi-chapter books (PDF, EPUB, offline sites), breaking public-API changes, and a fix for highlighted code leaking live markup into the default safe HTML output, a bug present since 0.4.4. The daily census flagged the release; this is the first re-check worked end to end under the drift pipeline in docs/landscape-rigor.md.

We re-checked it at the release commit, cc5196d, from a fresh clone: the full commit range, both Cargo.toml versions, and byte-diffs of the license, workflow set, and CommonMark conformance floor, plus the GitHub release record and the crates.io registry. Nothing was compiled or run.

CellAt the pinAt the re-check
TRL77, unchanged. A large feature release is consistent with a released, iterating product; 8 would need production use by someone else. The safe-HTML security fix cuts the other way: the pin’s safe-by-default reading was more generous than the code deserved. [Inference, Medium]
NODUS ringExploreExplore, unchanged. Stars moved 104 to 105 and crates.io downloads 151 to 176; no independent benchmark, review, or deployment appeared. [Inference, Medium]
ReleaseR3R3, unchanged. v0.5.0 targets the re-check commit itself, with 30 assets: five platform archives with sha256 sidecars, a SHA256SUMS file, a manifest, and the npm tarball. [Verified, High]
CIC5C5, unchanged. The same single, explicitly disabled workflow file at both pins; enforcement still runs on the maintainer’s own hosts, unobservable externally. [Verified, High]
LicenseMIT + riderUnchanged. The LICENSE file is byte-identical to the pin. [Verified, High]
Bus factor11, unchanged. All 135 commits in the range are the maintainer’s. [Git-observed, High]

The pinned suite counts are not changed by this: every cell is unchanged, so nothing would move under a re-pin at cc5196d either. The census flag did its job; the reading is the finding. One packet revisit trigger was tested and failed: npm still sits at 0.4.4, so the release-pipeline lag the packet named stands. The full re-check, with every command and its output, is updates/franken_markdown-2026-10-03.md. The brief still describes the pin and now says so at the top.

Dated re-checkfranken_code_browser shipped a notarized developer preview

At the pin, franken_code_browser was ten days old, had no release, no CI, and a README that disavowed any application. The next day, 2026-09-23, its maintainer published v0.1.0, an Apple Silicon disk image containing a signed, Apple-notarized Mac app, alongside a checksum-verifying install script and a Homebrew cask. He reports submitting a sandboxed Mac App Store build for review the same evening. Shipping a notarized app eleven days after creating the repository is a real event, and it is the first revisit trigger the packet named. The app source was not written after the pin, though: it was imported in one commit from a private history, which the packet had noted was missing.

We re-checked it at the release commit, c7c5310, with the release artifact itself: download, hash, signature, Gatekeeper, and notarization ticket, all read-only. We did not launch the app.

CellAt the pinAt the re-check
TRL2–34–5. A signed, notarized build of the whole app exists and matches its published hash; that much we verified. The maintainer's records of launching it on real repositories would support 5 if an outside launch reproduced them. Clean-machine installs and zoom performance are still open by his own account. [Inference, Medium]
NODUS ringExploreExplore, unchanged. A Pilot ring needs a release artifact and a bounded, real workload fit. The release now exists; the workload fit has not been shown to anyone outside the project. [Inference, Medium]
ReleaseR1, no release or tagR3 at the release commit. HEAD at the re-check (623340b) was five commits later (App Store edition source and docs; more commits have landed since), so a pin at HEAD would read as a release targeting an earlier commit. [Verified, High]
CIC4, no test CIC4, unchanged. .github still holds only issue templates. Every signing and packaging check in the commit messages ran on the maintainer's own Mac. [Verified, High]
LicenseMIT + riderUnchanged. The LICENSE file is byte-identical to the pin. [Verified, High]
Bus factor11, unchanged. All 33 commits since the pin are by the maintainer, and the README still refuses outside contributions. [Verified, High]
One correction to the release notesThe notes say the disk image is "signed with Jeffrey Emanuel's Developer ID". The image itself carries no code signature (codesign: "code object is not signed at all"); the app inside is signed with that Developer ID, Gatekeeper accepts it as "Notarized Developer ID", and the notarization ticket is stapled to the image. The packaging script is written that way, so this reads as a wording slip, not a missing step. The App Store review status is the maintainer's claim; Apple's public lookup returned no listing when we checked.

The pinned suite counts are not changed by this. Under a re-pin at c7c5310, the release tally would go from 23 repos with no release to 22 and from 15 with a release artifact to 16; nothing else would move. The full re-check, with every command and its output, is updates/franken_code_browser-2026-09-24.md. The brief still describes the pin and now says so at the top.

Dated re-checkfrankengit deleted 71 of its 78 workflow files

At the pin, frankengit had 78 workflow files and no pass or fail on record for the pinned commit. Four hours later, commit 53776f9b deleted 71 of them and set the 7 it kept to run only when someone starts them by hand. Their check steps call the repository's own scripts/verify.sh; the other steps install the pinned toolchain, record the revision and compiler, or archive the source. An hour after that, a merge brought back one old single-use job that applies a patch to main, which makes 8. GitHub has recorded no workflow run of any kind since 2026-09-22 18:11 UTC.

The checks did not move somewhere new. Since 2026-08-20 the repository's own rules have said that workflow files are hand-started adapters and that the project "MUST NOT rely on GitHub-hosted Actions". At the pin all 78 files broke that rule. Nine could not even be parsed, and 6,828 of the repository's 7,204 Actions runs were instant failures from files like those; the 18 runs on the pinned commit were all of that kind, so no test ran against it. The repository has its own checker for the rule. We compiled it and ran it: 238 workflow errors at the pin, none at the removal commit, and 3 today, all in the file the merge brought back.

CellAt the pinAt the re-check
CIC3, CI exists but no verdict for the pinned commitC5, CI switched off. No workflow runs by itself on main; the one that could fire listens on a branch that no longer exists. No runs for the re-check commit, dfa5bb8. [Verified, High]
TRL44, unchanged. The README's status section, now dated 2026-09-23, reports real git clients working against the node at one commit, and also says "Main is not currently verifiable". Nothing is integrated or tested outside the lab. [Inference, Medium]
NODUS ringExploreExplore, unchanged. Still no release and no independent validation. [Inference, High]
ReleaseR1, no release or tagR1, unchanged. The releases and tags APIs both return none. [Verified, High]
LicenseMIT + riderUnchanged. The LICENSE file is byte-identical to the pin. [Verified, High]
Bus factor11, unchanged. All 137 commits since the pin are under the maintainer's names; the co-author lines in 16 of them name an AI model, not a person. [Verified, High]
Where the checks run nowOn the maintainer's machine, through scripts/verify.sh. Each run writes a replay record (commit, toolchain, exit code, output hashes) into a directory that git ignores, so no pass or fail for any commit is published. The maintainer's own docs check fails at today's HEAD; one of its errors is the registry demoting its own claim CLM-001 because the checker's source changed, which is the auto-demotion the packet described, seen working.

The pinned suite counts are not changed by this. Under a re-pin at dfa5bb8, the CI tally would go from 6 repos with CI but no pin verdict to 5 and from 8 with CI disabled or deleted to 9; nothing else would move. The full re-check, with the run history, the checker output, and the new rigor row, is updates/frankengit-2026-09-24.md. The brief still describes the pin and now says so at the top.

Help fix it

Each opens a GitHub issue form with this page’s address filled in. You need a GitHub account.

Made by Joshua Nowak (ZestStream) with AI coding agents. Not affiliated with or funded by Jeffrey Emanuel; Joshua uses his public tools and is a paying subscriber to jeffreys-skills.md.

Source on GitHub MIT licensed