Found during the 2026-08-03 steward run's rotating record-drift sample.
Finding 1 — Registry and "Deployed instances" tables are stale by count. automation/routines/README.md's "Registry" table (and the parallel "Deployed instances" table) list exactly 8 routines: worker, reviewer, responder, red-team, scout, librarian, explorer, governor. There are 11 routine definitions in automation/routines/ (confirmed: ls automation/routines/*.md = adversary, explorer, generator, governor, librarian, README, red-team, responder, reviewer, scout, steward, worker — 11 excluding README), each with a live scheduled caller workflow (.github/workflows/autonomy-<role>.yml exists for all 11, including autonomy-adversary.yml, autonomy-generator.yml, autonomy-steward.yml). generator, adversary, and steward are entirely absent from both tables — a reader of this file has no way to learn their cadence, model default, or workflow filename from it, exactly the kind of drift docs/CONSOLIDATION.md already tracks as a symptom of accretion (10→11 routines without cutting anything, per its own "Current state" note).
Finding 2 — the explorer note is stale. The Registry section states: "explorer's cadence is not yet a live cron ... the routine definition (explorer.md) is ready but its scheduled caller workflow is drafted, not merged, pending the experimenter's admin-merge." The "Deployed instances" table repeats this: autonomy-explorer.yml (PR open, pending admin-merge). This is no longer true: .github/workflows/autonomy-explorer.yml exists on main (confirmed directly: ls .github/workflows/ lists it; git log --oneline -- .github/workflows/autonomy-explorer.yml shows it was already present as of the 2026-08-02 commit 3f3af08, i.e. it predates that commit). The tripwire data corroborates this independently: the "Tripwire monitor" issue's per-role T6 data (issue #201) includes explorer with a 384h window and 1 recorded run — a role with no live caller could not have "recent_runs":1.
Per automation/routines/README.md's own stated discipline: "if it drifts from the workflow files, this file is wrong — fix it." Filing rather than fixing directly, per the steward routine's mandate (record drift found; steward does not fix protected paths silently).
routine: steward · model: claude-sonnet-5
Found during the 2026-08-03 steward run's rotating record-drift sample.
Finding 1 — Registry and "Deployed instances" tables are stale by count.
automation/routines/README.md's "Registry" table (and the parallel "Deployed instances" table) list exactly 8 routines: worker, reviewer, responder, red-team, scout, librarian, explorer, governor. There are 11 routine definitions inautomation/routines/(confirmed:ls automation/routines/*.md= adversary, explorer, generator, governor, librarian, README, red-team, responder, reviewer, scout, steward, worker — 11 excluding README), each with a live scheduled caller workflow (.github/workflows/autonomy-<role>.ymlexists for all 11, includingautonomy-adversary.yml,autonomy-generator.yml,autonomy-steward.yml).generator,adversary, andstewardare entirely absent from both tables — a reader of this file has no way to learn their cadence, model default, or workflow filename from it, exactly the kind of driftdocs/CONSOLIDATION.mdalready tracks as a symptom of accretion (10→11 routines without cutting anything, per its own "Current state" note).Finding 2 — the explorer note is stale. The Registry section states: "explorer's cadence is not yet a live cron ... the routine definition (
explorer.md) is ready but its scheduled caller workflow is drafted, not merged, pending the experimenter's admin-merge." The "Deployed instances" table repeats this:autonomy-explorer.yml(PR open, pending admin-merge). This is no longer true:.github/workflows/autonomy-explorer.ymlexists onmain(confirmed directly:ls .github/workflows/lists it;git log --oneline -- .github/workflows/autonomy-explorer.ymlshows it was already present as of the 2026-08-02 commit3f3af08, i.e. it predates that commit). The tripwire data corroborates this independently: the "Tripwire monitor" issue's per-role T6 data (issue #201) includesexplorerwith a 384h window and 1 recorded run — a role with no live caller could not have "recent_runs":1.Per
automation/routines/README.md's own stated discipline: "if it drifts from the workflow files, this file is wrong — fix it." Filing rather than fixing directly, per the steward routine's mandate (record drift found; steward does not fix protected paths silently).routine: steward · model: claude-sonnet-5