Provenance: the index records the checker that swept, not the exporter - #15
Merged
Conversation
…rted published-runs.json said all four canonical runs were schema 7, checker 0.6.0. The database — served verbatim by /api/runs — says schema 5, checker 0.5.0 for all four, and the site displayed both at once: the homepage quoted the API while /data quoted the index, the same run with two different checkers. The commits were right everywhere; the version and schema numbers belonged to the 2026-08-01 rebuild that produced the archives, not to the sweeps. The index now carries the sweep-time values plus each archive's rebuilt_at, verified against the four downloaded archives (sha256 re-checked against the published checksums in the process). Manifests gain exporter_version/exporter_commit so the writer and the judge are never again one field — on a live sweep the two are one build, on a rebuild they are not, and a resume writes the resuming binary's identity rather than echoing the sweep-time commit it read back from the database. publish-run.sh carries the new fields into the index and tolerates their absence from pre-2026-08 manifests. The archives are immutable and keep their overstated internal manifests. METHODOLOGY §5 now states the precedence — where an archive manifest and the database disagree, the database is authoritative — and adopts the standing policy this incident earns: canonical runs are swept from clean commits only, from the next sweep onward. Three of the four July runs are -dirty builds; the stamp is honest and they stay published, but a -dirty build cannot be checked out and rerun by anyone, which is a poor foundation for a census whose claim is recomputability. Both corrections are logged in CHANGELOG-METHODOLOGY.md, including the false "four largest chains" H1 this repo's ledger is the home for. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This was referenced Aug 2, 2026
Merged
NemboKid
added a commit
to AgentCount/agentcount-web
that referenced
this pull request
Aug 2, 2026
#20) The fallback copy of published-runs.json said schema 7, checker 0.6.0 for all four canonical runs; the database — and therefore the homepage's provenance panel — says schema 5, checker 0.5.0. Same run, two checkers, on one site. The wrong values belong to the 2026-08-01 rebuild that produced the archives, not to the sweeps. Mirrors the core repo's index correction (AgentCount/agentcount#15). The checker line also stops truncating the -dirty marker off the commit: shortening 987ce…-dirty to twelve characters made this page silently claim a clean build the homepage was honestly reporting as dirty. The marker stays displayed, a note explains what it means and states the new standing policy (canonical runs sweep from clean commits only), and runs rebuilt from the database now show their rebuild date. Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
NemboKid
added a commit
to AgentCount/agentcount-web
that referenced
this pull request
Aug 2, 2026
…rs (#21) * fix: /data agrees with the homepage about which checker swept each run The fallback copy of published-runs.json said schema 7, checker 0.6.0 for all four canonical runs; the database — and therefore the homepage's provenance panel — says schema 5, checker 0.5.0. Same run, two checkers, on one site. The wrong values belong to the 2026-08-01 rebuild that produced the archives, not to the sweeps. Mirrors the core repo's index correction (AgentCount/agentcount#15). The checker line also stops truncating the -dirty marker off the commit: shortening 987ce…-dirty to twelve characters made this page silently claim a clean build the homepage was honestly reporting as dirty. The marker stays displayed, a note explains what it means and states the new standing policy (canonical runs sweep from clean commits only), and runs rebuilt from the database now show their rebuild date. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix: the payments framing meets the site's own sourcing standard The census demands recomputability and named sources of everyone it measures, and its own x402 framing fell short of that in five ways, all found by the 2026-08 review: * "x402, the ecosystem's own payment protocol" claimed a relationship the spec disclaims — ERC-8004 says payments are orthogonal to it. The epithet is now the verifiable one: the payment protocol the spec itself name-checks (x402Support). * Nothing reader-facing said why x402 is the protocol-level signal or what the flag detects. Both the report (§9, beside the definition of "paid") and the linkage page now carry the passage: plain transfers are the primary measure, the x402-style flag is EIP-3009 co-occurrence, that signal is broader than x402 itself, and that is why it was cross-checked. * The cross-check's result — 18 of 20 addresses corroborated, 9 exact — was in the report but never on the site. The linkage page states it. * x402.org and x402scan.com are now linked at first mention; the x402scan-has-no-disclaimer sentence carries its as-of date. * The headline's "138 largest x402 sellers" dropped the EVM qualifier that §§8-9 state correctly; 21 of the actual largest sellers are Solana addresses nobody tested. Restored. Also from the review's claim audit: §8 no longer says Solana carries "the same CREATE2 registries" (its ERC-8004 presence is a third-party port); "provably Morpho vault yield" drops the "provably" no published trace backs; the directory intro drops "no other tool in the ecosystem can answer"; and the methodology page's contract-level-invariant claim for rung 7 is now stated as verified rather than assumed — the deployed Reputation Registry's source and a live eth_call simulation both revert owner and operator feedback, with the upgradeability and second-wallet limits stated beside it. The page also says, for the first time on the site, that ERC-8004 is a Draft. The core repo's copy of the report gets the identical edits in its own PR, so the two files stay byte-identical. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Phase 1, item 4 (F-12) of the post-review work order, plus the corrections-ledger entries for it and for the false H1.
The bug.
published-runs.jsonrecorded all four canonical runs as schema 7 / checker 0.6.0; the DB (and/api/runs, and the homepage) says schema 5 / checker 0.5.0. Same run, two checkers, one site. Verified by downloading all four archives (sha256s match the published checksums; the internal manifests carry the rebuild era's values).Changes
published-runs.json: the four entries corrected to sweep-time values, plus each archive'srebuilt_at.RunManifest/RebuiltManifestgainexporter_version+exporter_commit— the writer and the judge are separate fields from now on. The sweeper stamps its own build identity (not the resumed run's commit);export-runstamps the rebuilder.publish-run.shcarries the new fields into the index, tolerating old manifests.METHODOLOGY.md§5: precedence rule (DB is authoritative over archive manifests) + standing policy: canonical runs are swept from clean commits only, from the next sweep onward (3 of 4 July runs are-dirtybuilds; they stay published, stamped honestly).CHANGELOG-METHODOLOGY.md: two entries — the provenance correction (no semantic effect) and the false "four largest chains" H1 (fixed in agentcount-web#18/Capture the minter consistently: backfill existing rows, survive a failed registration scan #19), both dated 2026-08-02.The web repo's fallback copy of the index and the
/datapage display get the mirror fix in a separate PR there.cargo build -p sweeperclean;bash -non the script.🤖 Generated with Claude Code