Report the settled usage and review census for 2026-10-19 to 2026-10-23 (Refs #4823)

.agents/skills/atlas/assets/seed/records/2d46924a94c85f4275a449dd3d9eb6c602cf8a9104db19d224e3aa469b858d06.json

@@ -0,0 +1 @@
+{"citations":[{"path":"docs/reports/2026-10-23-usage-census-4823.md","repository":"FS-GG/.github","revision":"529198a4c108260d0959ecd9d71955fb49527b26","sha256":"3f3b87cb0a3100d284f6deb992768042dd56b6fdbca12da577529dd4041d1d8e"},{"path":"docs/reports/evidence/2026-10-23-usage-census-4823.json","repository":"FS-GG/.github","revision":"529198a4c108260d0959ecd9d71955fb49527b26","sha256":"d152db3b8dc4a6b3a6d876bcaabd3b3ceec562784352cd2cb970c5bda565b580"},{"path":"src/FS.GG.Coord.Cli/TelemetryApplication.fs","repository":"FS-GG/.github","revision":"61526e947ff595c5eb416efd810ce3c1348bedc5","sha256":"04e5c25cf898524108e7b862409a4a04fb21927cac048a4679cfe22604d14555"},{"path":"src/FS.GG.Coord.Core/Telemetry.fs","repository":"FS-GG/.github","revision":"61526e947ff595c5eb416efd810ce3c1348bedc5","sha256":"dba776ca57f820450d4430a30b06166adb18111624e47afea2f98727abb47375"}],"conflictsWith":[],"id":"github/settled-usage-review-census-method","kind":"lesson","recordedAt":"2026-10-23T15:13:32Z","repository":"FS-GG/.github","schema":"fsgg.atlas/1","status":"recorded","summary":"The finite census for FS-GG/.github#4823 (window [2026-10-19T18:19:36Z, 2026-10-23T00:00Z), fdev rescope of #4639) is reported in docs/reports/2026-10-23-usage-census-4823.md with its compact per-member evidence JSON (column/row tables; paths containing \"usage\" are capped at 64 KiB by telemetry-git-safety). It is a sample at its own scope; whole-programme coverage and pre-window history stay Unknown under #4639.\n\nReusable method findings:\n- Population: enumerate every org repository through GraphQL pullRequests(states: MERGED), not only search; the org search count agreed (134). Freeze members with a digest over full owner/repo lines in mergedAt order.\n- Review evidence: no PR in the window had a submitted GitHub review or a machine-readable verdict marker, and one account authored, merged and posted every record, so independence is only as asserted. An exact-head verdict must bind the final head SHA in the same statement or dedicated review comment; many records say \"independent exact-head review passed\" without a SHA, or record the verdict at an earlier head. Paragraph-level regex matching gave false negatives; every review-mentioning PR needs hand adjudication.\n- Actions time: the runs listing caps filtered queries at 1000 rows, so busy days need hourly slices. Merged PR runs carried no native pull_requests links. Stacked PRs share commits, so assign each run to one member by head branch, then by which PR was open; exclude other-PR runs. Count only push runs on a merge SHA; issues/schedule/dispatch runs on the default branch are not caused by the merge. Force-pushed heads drop out of the PR commit list, so their runs stay uncounted. A bulk Jobs API pass (about 9,600 calls) exhausts the shared single-account REST quota; pace it and keep headroom.\n- Required checks: GraphQL isRequired is not readable with the fine-grained PAT; use base-branch rules plus classic protection, read at census time. For the private repository both reads returned 403 with the census credential. Private members must be redacted from a public report, including timestamps: a public member digest plus a published close time let the item number be recovered by brute force.\n- Native usage through the strict reader at 61526e94: the usage receipt store must lie outside the current directory, any repository and the system temporary directory. A resumed Claude Code session copies earlier records into its new transcript (313 responses in this window), so totals must be deduplicated by provider response ID across files. Subagent transcripts with a response lacking a completed record are withheld (9 files), and 2 subagent files mixing session identities are refused; both stay Unknown. The reader emits no branch or cwd, so branch joins need a separate allowlisted key parser. Codex rollouts record only the session-start branch and cwd (main), so coordinator-run Codex usage cannot be attributed; only Claude records carrying a PR head branch joined (3 PRs).\n\nUnknown: other hosts including Home, billed cost, pruned transcripts, window-time branch rules, reviews held privately or only in linked issues.","supersedes":[],"tags":["census","usage","review","telemetry","claude","codex","actions","attribution","unknown"],"title":"Settled usage and review census: method, joins and Unknowns"}

.claude/skills/atlas/assets/seed/records/2d46924a94c85f4275a449dd3d9eb6c602cf8a9104db19d224e3aa469b858d06.json

@@ -0,0 +1 @@
+{"citations":[{"path":"docs/reports/2026-10-23-usage-census-4823.md","repository":"FS-GG/.github","revision":"529198a4c108260d0959ecd9d71955fb49527b26","sha256":"3f3b87cb0a3100d284f6deb992768042dd56b6fdbca12da577529dd4041d1d8e"},{"path":"docs/reports/evidence/2026-10-23-usage-census-4823.json","repository":"FS-GG/.github","revision":"529198a4c108260d0959ecd9d71955fb49527b26","sha256":"d152db3b8dc4a6b3a6d876bcaabd3b3ceec562784352cd2cb970c5bda565b580"},{"path":"src/FS.GG.Coord.Cli/TelemetryApplication.fs","repository":"FS-GG/.github","revision":"61526e947ff595c5eb416efd810ce3c1348bedc5","sha256":"04e5c25cf898524108e7b862409a4a04fb21927cac048a4679cfe22604d14555"},{"path":"src/FS.GG.Coord.Core/Telemetry.fs","repository":"FS-GG/.github","revision":"61526e947ff595c5eb416efd810ce3c1348bedc5","sha256":"dba776ca57f820450d4430a30b06166adb18111624e47afea2f98727abb47375"}],"conflictsWith":[],"id":"github/settled-usage-review-census-method","kind":"lesson","recordedAt":"2026-10-23T15:13:32Z","repository":"FS-GG/.github","schema":"fsgg.atlas/1","status":"recorded","summary":"The finite census for FS-GG/.github#4823 (window [2026-10-19T18:19:36Z, 2026-10-23T00:00Z), fdev rescope of #4639) is reported in docs/reports/2026-10-23-usage-census-4823.md with its compact per-member evidence JSON (column/row tables; paths containing \"usage\" are capped at 64 KiB by telemetry-git-safety). It is a sample at its own scope; whole-programme coverage and pre-window history stay Unknown under #4639.\n\nReusable method findings:\n- Population: enumerate every org repository through GraphQL pullRequests(states: MERGED), not only search; the org search count agreed (134). Freeze members with a digest over full owner/repo lines in mergedAt order.\n- Review evidence: no PR in the window had a submitted GitHub review or a machine-readable verdict marker, and one account authored, merged and posted every record, so independence is only as asserted. An exact-head verdict must bind the final head SHA in the same statement or dedicated review comment; many records say \"independent exact-head review passed\" without a SHA, or record the verdict at an earlier head. Paragraph-level regex matching gave false negatives; every review-mentioning PR needs hand adjudication.\n- Actions time: the runs listing caps filtered queries at 1000 rows, so busy days need hourly slices. Merged PR runs carried no native pull_requests links. Stacked PRs share commits, so assign each run to one member by head branch, then by which PR was open; exclude other-PR runs. Count only push runs on a merge SHA; issues/schedule/dispatch runs on the default branch are not caused by the merge. Force-pushed heads drop out of the PR commit list, so their runs stay uncounted. A bulk Jobs API pass (about 9,600 calls) exhausts the shared single-account REST quota; pace it and keep headroom.\n- Required checks: GraphQL isRequired is not readable with the fine-grained PAT; use base-branch rules plus classic protection, read at census time. For the private repository both reads returned 403 with the census credential. Private members must be redacted from a public report, including timestamps: a public member digest plus a published close time let the item number be recovered by brute force.\n- Native usage through the strict reader at 61526e94: the usage receipt store must lie outside the current directory, any repository and the system temporary directory. A resumed Claude Code session copies earlier records into its new transcript (313 responses in this window), so totals must be deduplicated by provider response ID across files. Subagent transcripts with a response lacking a completed record are withheld (9 files), and 2 subagent files mixing session identities are refused; both stay Unknown. The reader emits no branch or cwd, so branch joins need a separate allowlisted key parser. Codex rollouts record only the session-start branch and cwd (main), so coordinator-run Codex usage cannot be attributed; only Claude records carrying a PR head branch joined (3 PRs).\n\nUnknown: other hosts including Home, billed cost, pruned transcripts, window-time branch rules, reviews held privately or only in linked issues.","supersedes":[],"tags":["census","usage","review","telemetry","claude","codex","actions","attribution","unknown"],"title":"Settled usage and review census: method, joins and Unknowns"}

docs/atlas/records/2d46924a94c85f4275a449dd3d9eb6c602cf8a9104db19d224e3aa469b858d06.json

@@ -0,0 +1 @@
+{"citations":[{"path":"docs/reports/2026-10-23-usage-census-4823.md","repository":"FS-GG/.github","revision":"529198a4c108260d0959ecd9d71955fb49527b26","sha256":"3f3b87cb0a3100d284f6deb992768042dd56b6fdbca12da577529dd4041d1d8e"},{"path":"docs/reports/evidence/2026-10-23-usage-census-4823.json","repository":"FS-GG/.github","revision":"529198a4c108260d0959ecd9d71955fb49527b26","sha256":"d152db3b8dc4a6b3a6d876bcaabd3b3ceec562784352cd2cb970c5bda565b580"},{"path":"src/FS.GG.Coord.Cli/TelemetryApplication.fs","repository":"FS-GG/.github","revision":"61526e947ff595c5eb416efd810ce3c1348bedc5","sha256":"04e5c25cf898524108e7b862409a4a04fb21927cac048a4679cfe22604d14555"},{"path":"src/FS.GG.Coord.Core/Telemetry.fs","repository":"FS-GG/.github","revision":"61526e947ff595c5eb416efd810ce3c1348bedc5","sha256":"dba776ca57f820450d4430a30b06166adb18111624e47afea2f98727abb47375"}],"conflictsWith":[],"id":"github/settled-usage-review-census-method","kind":"lesson","recordedAt":"2026-10-23T15:13:32Z","repository":"FS-GG/.github","schema":"fsgg.atlas/1","status":"recorded","summary":"The finite census for FS-GG/.github#4823 (window [2026-10-19T18:19:36Z, 2026-10-23T00:00Z), fdev rescope of #4639) is reported in docs/reports/2026-10-23-usage-census-4823.md with its compact per-member evidence JSON (column/row tables; paths containing \"usage\" are capped at 64 KiB by telemetry-git-safety). It is a sample at its own scope; whole-programme coverage and pre-window history stay Unknown under #4639.\n\nReusable method findings:\n- Population: enumerate every org repository through GraphQL pullRequests(states: MERGED), not only search; the org search count agreed (134). Freeze members with a digest over full owner/repo lines in mergedAt order.\n- Review evidence: no PR in the window had a submitted GitHub review or a machine-readable verdict marker, and one account authored, merged and posted every record, so independence is only as asserted. An exact-head verdict must bind the final head SHA in the same statement or dedicated review comment; many records say \"independent exact-head review passed\" without a SHA, or record the verdict at an earlier head. Paragraph-level regex matching gave false negatives; every review-mentioning PR needs hand adjudication.\n- Actions time: the runs listing caps filtered queries at 1000 rows, so busy days need hourly slices. Merged PR runs carried no native pull_requests links. Stacked PRs share commits, so assign each run to one member by head branch, then by which PR was open; exclude other-PR runs. Count only push runs on a merge SHA; issues/schedule/dispatch runs on the default branch are not caused by the merge. Force-pushed heads drop out of the PR commit list, so their runs stay uncounted. A bulk Jobs API pass (about 9,600 calls) exhausts the shared single-account REST quota; pace it and keep headroom.\n- Required checks: GraphQL isRequired is not readable with the fine-grained PAT; use base-branch rules plus classic protection, read at census time. For the private repository both reads returned 403 with the census credential. Private members must be redacted from a public report, including timestamps: a public member digest plus a published close time let the item number be recovered by brute force.\n- Native usage through the strict reader at 61526e94: the usage receipt store must lie outside the current directory, any repository and the system temporary directory. A resumed Claude Code session copies earlier records into its new transcript (313 responses in this window), so totals must be deduplicated by provider response ID across files. Subagent transcripts with a response lacking a completed record are withheld (9 files), and 2 subagent files mixing session identities are refused; both stay Unknown. The reader emits no branch or cwd, so branch joins need a separate allowlisted key parser. Codex rollouts record only the session-start branch and cwd (main), so coordinator-run Codex usage cannot be attributed; only Claude records carrying a PR head branch joined (3 PRs).\n\nUnknown: other hosts including Home, billed cost, pruned transcripts, window-time branch rules, reviews held privately or only in linked issues.","supersedes":[],"tags":["census","usage","review","telemetry","claude","codex","actions","attribution","unknown"],"title":"Settled usage and review census: method, joins and Unknowns"}