the-custodian/workplans/CUST-WP-0075-dependency-graph-driven-unblocking.md
codex 786b082a19
All checks were successful
CI Smoke / host-smoke (push) Successful in 0s
CI Smoke / container-smoke (push) Successful in 1s
Python Tests / pytest (push) Successful in 23s
CUST-WP-0075: reconcile owner waits and decision routing
2026-10-04 01:17:18 +02:00

13 KiB
Raw Blame History

id type title domain repo status owner topic_slug flavor created updated related origin origin_ref state_hub_workstream_id
CUST-WP-0075 workplan Use the dependency graph to shrink the blocked set and route human gates to decisions infotech the-custodian finished the-custodian custodian planning 2026-10-03 2026-10-04
CUST-WP-0072
CUST-WP-0074
STATE-WP-0092
COORDINATION-WP-0005
analysis the-custodian/docs/evidence/2026-10-03-dependency-graph-triage/README.md 6dac666d-b2f8-5d50-b060-272713b6e8a1

Use the dependency graph to shrink the blocked set

Why

CUST-WP-0072 and CUST-WP-0074 built the facility: flavor, workplan and task depends_on, needs_human, blocked_on, wait_until, decision_id, indexed by C-20 and checked by C-39/C-40/C-41. Nobody has yet read the resulting graph. On 2026-10-03 the open set is 108 workplans (100 blocked) with 232 wait tasks. The security concepts (credential custody, attended OpenBao provisioning, identity, access engine) are the dominant blockers, and every one of their 69 human gates is free text: zero carry a decision_id, so Bernd has no single list to resolve and C-40 rule 4 never fires.

This workplan uses the graph in three ways: clear what the graph says is already satisfied, correct what the graph says is mis-statused, and collapse the human gates into a handful of hub decisions that one attended session can resolve. Evidence and scripts: docs/evidence/2026-10-03-dependency-graph-triage/.

Rules: owner repos are edited directly (Bernd's 2026-09-30 direction on CUST-WP-0074-T06), one commit per repo, synced by fix-consistency; no hub writes except recording and resolving decisions; residual-flavor workplans are left alone except to finish one whose tasks are all done.

Capture the graph and publish the triage

id: CUST-WP-0075-T01
status: done
priority: high
state_hub_task_id: "9d9e994d-0b48-5945-ab13-f301f3343aae"

File-view and hub-view scans committed under docs/evidence/2026-10-03-dependency-graph-triage/ with the headline numbers in its README. Done 2026-10-03.

Clear satisfied waits and correct mis-statused workplans

id: CUST-WP-0075-T02
status: done
priority: high
state_hub_task_id: "0e524c1f-3fd7-5736-9fb2-883f4e6df70a"

Graph says satisfied (every depends_on target terminal): ACTIVITY-WP-0032-T05, COORDINATION-WP-0004-T02, TD-WP-0003-T07, USER-WP-0027-T04, USER-WP-0028-T03, RCLK-WP-0005-T04 (keeps its blocked_on), RAPPS-WP-0014-T03 and MARKITECT-WP-0002-T07 (keep their human gate; drop the satisfied depends_on). Mis-statused: UPC-WP-0002/0003/0004 blocked without a wait task; MARKITECT-WP-0002 and RBT-WP-0006 active while every open task waits; ACTIVITY-WP-0042 proposed with three unqualified waits; RAIL-FAB-WP-0031 blocked with all tasks done; FLUID-WP-0007 carries the non-canonical status done; NK-WP-0011 depends on NK-WP-0003/0004/0006, which no file carries. Done when each owner repo is committed and synced and a re-run of the file scan lists none of the above.

Route the human gates to decisions

id: CUST-WP-0075-T03
status: done
priority: high
state_hub_task_id: "a363639a-d688-5c72-9224-63d06dec670e"

Cluster the 69 needs_human waits by the person's actual act (one attended OpenBao provisioning window, one ArgoCD go-ahead, one Telegram operator session, Vergabe two-user acceptance, trust-root design, and so on). For each cluster record one hub decision whose description lists the task ids, then set decision_id on those tasks in the owner files. Done when every needs_human wait in a non-residual open workplan carries a decision_id and Bernd has one ranked decision list.

Record the security-infrastructure lessons

id: CUST-WP-0075-T04
status: done
priority: medium
state_hub_task_id: "b861e0af-43ad-5052-b224-9bd95762ddfd"

Write history/20261003-security-gate-packets.md: what the graph shows about how credential, identity and access work gets blocked, and the improvements that make it easier to maintain and extend (decision-linked gates, batched attended windows, delegated-applier scope, catalog lanes as the unit of provisioning). Proposals only; no canon change here.

Monitor: keep repo and hub in step

id: CUST-WP-0075-T05
status: done
priority: medium
state_hub_task_id: "c78b104d-cdf0-5aca-81a4-b5569cc90479"

After every slice: commit, statehub fix-consistency for each touched repo, one progress event on this workplan, and the headline numbers appended below so progress is visible from the hub.

2026-10-03 Codex continuation: repeatable read-only monitor implemented in scripts/monitor_dependency_triage.py, with seven regression checks covering YAML lists/quoting, pagination, missing projections, decision drift, residual exclusion and independent human gates. Run it with --output <scratch-dir>; review monitor.json before acting. Current live findings are retained in docs/evidence/2026-10-03-dependency-graph-triage/monitor-2026-10-03.json: four unrouted human gates, sixteen unqualified waits and twelve qualifier differences. This task remains in progress until those findings are reconciled; the earlier T02/T03 results are point-in-time, not a current clean scan.

Hand the infrastructure proposals to their owners

id: CUST-WP-0075-T06
status: done
priority: high
state_hub_task_id: "91fd64c1-2dfd-57d1-8930-51477c530202"

history/20261003-security-gate-packets.md proposes P1 (unrouted human gate is a WARN), P2 (standing attended provisioning window with a generated checklist), P3 (decide the dual-control value path), P4 (catalog lane as the unit of extension), P5 (daily owner-return sweep) and P6a (CLI qualifier sync gap, already reported to state-hub as message 70d08f35). 2026-10-03: sent state-hub b4825539 (P1, P5, P6), railiance-platform f84637fa (P2, P3), ops-warden 79549660 (P4). Send each owner one message with the proposal and the evidence; record their workplan ids here. Do not write their workplans for them.

2026-10-03 Codex continuation: all three original proposal messages remain unread; no acceptance, owner workplan ID or decline found. Three local derived cross_owner_wait observations are attached under handoffs in the monitor snapshot, with stable case IDs and original message IDs. Obligor workplan IDs remain unknown pending owner identification. Newer platform/warden replies concern D1 readiness and the fetch-policy bug, respectively; neither answers these proposals. No worker was woken. T06 remains wait.

2026-10-03 worker continuation (founder explicitly requested repo workers): state-hub, railiance-platform and ops-warden workers dispatched first, with further affected repos queued in docs/evidence/2026-10-03-dependency-graph-triage/worker-dispatch-2026-10-03.json. ops-warden accepted P4 through reply cda45346-67d1-4c1c-a0b1-4a13b5032551: WARDEN-WP-0041 owns catalog/template source delivery, with T02 implementation and T03 platform-seam acceptance still open. Source review and handoff are done in ac112d8. The proposed contract is ops-warden/wiki/CatalogLaneScaffoldContract.md; platform owns its rendering schema. This resolves that owner handoff, not the downstream implementation. Reviewed dispositions live in owner-handoff-receipts.json in the evidence folder.

Progress log

  • 2026-10-03 start: 108 open / 100 blocked / 232 wait / 69 human gates / 0 with decision_id.

  • 2026-10-03 T02: satisfied or mis-statused records fixed in 11 repos. COORDINATION-WP-0004, RAIL-FAB-WP-0031 and FLUID-WP-0007 finished. ACTIVITY-WP-0032-T05, COORDINATION-WP-0004-T02 and TD-WP-0003-T07 had prose-only blockers; the real edge was encoded (or, for the canon registration, found already delivered and closed).

  • 2026-10-03 T03: eleven hub decisions recorded (D1 attended OpenBao window 62aa95e4 … D11 operator setups b6a44b55). 70 human gates carry a decision_id in their owner files across 29 repos. WARDEN-WP-0037-T03 has no stated gate and stays with its owner.

  • 2026-10-03 T04: history/20261003-security-gate-packets.md.

  • 2026-10-03 sweep: 17 owner-return waits had an unread reply from the very agent they wait on. Repo agents dispatched for ops-mason (5) and gate-house (5); user-engine (3), activity-core, railiance-platform and rapp-policy-nexus follow.

  • 2026-10-03 sweep results: COORDINATION-WP-0004 finished (owner return sat unread a week). RCLK-WP-0002-T01 done: railiance-master ruled 2026-09-21, gate-house reviewed today, repo-manager registered RCLK-WP (03869ff). GH-WP-0004-T07/T08 and MASON-WP-0007-T03 done; MASON-WP-0005-T02 and RPF-WP-0043-T02 to todo (the latter broke a mutual wait with rapp-policy-nexus). Other replies were questions back (answered) or hand-offs to a founder decision (encoded).

  • 2026-10-03 D12 recorded for the company-vessel founder actions; WARDEN-WP-0037-T03 routed to D3. All 83 human-gated waits in open non-residual workplans carry a decision_id. Ranked list: docs/decision-packets-2026-10-03.md.

  • 2026-10-03 finding: the CLI sync left the hub without the decision_id on 57 of 58 routed tasks; a direct checker run with --fix is pushing them (see the assessment, P6a).

  • 2026-10-03 close of day: every routed gate verified in the hub (84 of 84 carry the file's decision_id after a direct consistency_check.py --fix over 32 repos). File view: 107 open workplans, 232 wait tasks, 83 human gates in non-residual plans, all routed to 12 decisions; 0 satisfied waits, 0 unqualified waits, 0 blocked plans without a wait task. Owner-return waits 62 → 52. T02/T03 done; T05 continues as the monitor; T06 next.

  • 2026-10-03 later: CCR-2026-0031 live end to end (fetch gate fixed and deployed, core-hub rev 29, verify-publisher 121/2). Re-scan: 109 open, 0 satisfied or unqualified waits. Next candidates: 45 open todo tasks that agents can pick up (NK-WP-0011 T02–T09 active, NK-WP-0045 T02/T03, INFD-WP-0005 T04/T07, FEP-WP-0002-T03, GH-WP-0004-T12, ORGREF-WP-0001 T01/T03–T05, MASON-WP-0005-T02, RPF-WP-0043-T02, STATE-WP-0079-T04, REINAH-WP-0003-T05, CR-WP-0005 T01–T03), and two unread owner replies: GH-WP-0004-T06 (informed-decision) and NK-WP-0042-T02 (user-engine). Follow-up on the CCR-2026-0031 lane: ops-warden catalog entry plus fetch-pin regeneration (checklist row 6).

  • 2026-10-03 Codex continuation: monitored 146 registered repos; six checkouts unavailable. Visible file view: 107 open plans including residuals; actionable non-residual view: 100 open / 94 blocked / 232 wait / 79 human gates, four unrouted; 56 owner returns. Twelve qualifier differences and sixteen unqualified waits are recorded as T05 follow-up. Four waits have terminal predecessors, but separate human/owner gates and task prose still need review. T06's three proposals remain unanswered. The monitor and seven regression checks are implemented; no fleet cleanup or proposal acceptance is claimed.

  • 2026-10-04 owner handoffs complete: STATE-WP-0094, RPF-WP-0050 and WARDEN-WP-0041 are live implementation records. State Hub reply c17795db-b3f2-47e8-8be8-04d5d6dfecbf partially accepts P1/P5/P6a; P1 and owner-return detection are delivered, wakeup execution, causal diagnosis and graph freshness remain T03/T04/T05. Legacy remote nightly sweep declined. Platform reply 981a20fc-315b-4fd8-9783-9c3460a4840d accepts local checklist preparation; P3 remains an explicit human custody choice and P4 remains source scaffold delivery. Reviewed original-thread receipts close T06 only.

  • Residual handed off before closure: CUST-WP-0076-T01 owns the operator's provisioning scope/date/cadence choice, returning to RPF-WP-0050-T03. It remains residual and is not implemented or scheduled.

  • Activity Core current source gates are reconciled separately from the historical registrar defect. Request 6bf8ca98-ada1-43ab-83e7-207d4ab195fe and ACTIVITY-WP-0042-T08 retain that actual blocker. Read-only derivation proof found omitted legacy YAML task lists and a shared bound UUID; the prepared preflight-only patch is deliberately unapplied. No receiving preservation, consumer refresh or runtime acceptance is inferred.

  • 2026-10-04 closing verification: all 28 selected qualifier/dependency rows, six human decision routes and three Activity/User actual-wait records match source at their task UUIDs. Freedom Intelligence was fast-forwarded before its checker repair; warning-level PASS alone was not accepted. T05 completes this reconciliation and the repeatable read-only monitor. Continuing automation and receiving diagnosis remain STATE-WP-0094, owner source scaffolds remain RPF-WP-0050/WARDEN-WP-0041, and cadence remains the unpromoted live residual CUST-WP-0076. Validation: 11 monitor tests; 182 State Hub checker/CLI tests; 118 Warden routing/plan tests; 7 platform renderer tests. Platform existing-suite catalog assertion (52/53) and three historical C07 assessment failures remain explicitly reported; no broad estate-wide clean audit is claimed.

  • Closing monitor: monitor-workers-2026-10-04.json: 146 registered repos, 144 available checkouts; 106 open non-residual plans, 241 waits, 86 human gates; zero unrouted/unqualified/terminal-predecessor waits and zero qualifier drift. All three owner handoffs satisfied. Two missing checkouts and sixteen historical parse warnings remain explicit coverage limits.