Record work-mode baseline and real-provider verification

doc/plans/2026-09-22-explicit-work-mode-authority.md

@@ -32,5 +32,13 @@ attempts and final results before declaring verification complete.
 
 ## Verification
 
-In progress. The historical remaining-findings inventory is retained in
+Deterministic baseline: 990/992 pass, with only the two previously recorded native
+compatibility probes failing. All work-mode assertions pass; E2E support,
+typechecks and build pass. Both explicit planning controls passed in campaign
+35805477715; its four new wording cases exposed a Markdown/browser fixture
+mismatch, with one additional duplicate model response. Those failures remain
+recorded. After a fixture-only plain-text correction, both wording pairs passed
+in campaign 35806360797 (4/4). See the
+[measurement report](../../tests/lifecycle-baseline/EXPLICIT-WORK-MODE-2026-09-22.md).
+The historical remaining-findings inventory is retained in
 [the continuation plan](2026-09-22-legacy-continuation-authority.md).

doc/plans/2026-09-22-legacy-continuation-authority.md

@@ -118,13 +118,14 @@ Keep this section even if this slice incidentally resolves some assertions.
 
 | Finding | Recorded count | Disposition |
 |---|---:|---|
-| Narrative changes continuation | 36 | Current focus |
-| Title/description words select work mode | 4 | Retain for explicit work-mode follow-up; no title/body field in the new authority contract |
+| Narrative changes continuation | 36 | Fixed by the persisted-disposition authority slice; see verification above |
+| Title/description words select work mode | 4 | Heuristic removed in [explicit work-mode follow-up](2026-09-22-explicit-work-mode-authority.md). Clarification: this was a liveness diagnostic exemption, not a stored mode mutation. Deterministic checks and 4/4 corrected live wording cases pass; both explicit planning controls pass |
 | Commentary counts as progress | 1 | Retain for progress-evidence audit; no comment-count budget resets in this slice |
 | Wording selects liveness versus handoff path in heartbeat | 1 | Retain separately in results; must be covered while closing the current shared authority boundary |
 | Native autonomous continuation compatibility probes | 2 | Replace with reachable public-contract tests; `same_agent` injection is not a current model-facing defect |
 | Productive continuation versus repair/failure budgets | Coverage boundary | Separate audit after this slice; do not infer progress from text/event counts |
 | Blank task route after reload | New live observation | Investigate browser bootstrap/rendering; seen after successful persisted lifecycle in both runtimes; campaigns 35744074887 and 35747200170 |
+| Duplicate legacy response | One later live observation | Campaign 35805477715 legacy work-mode neutral: provider issued two successful PATCH calls, second escaping an underscore. Retain as model behavior evidence; exact-once matcher rejected it |
 | Missing-comment policy ownership | Coverage boundary | Existing typed retry policy remains; review separately if consolidating all post-run compliance into one disposition contract |
 | Maintain CI and prepare App/Evals review | Pending | Promote fixed invariants into maintained cheap gates; paid suites remain explicit |
 

tests/lifecycle-baseline/EXPLICIT-WORK-MODE-2026-09-22.md

@@ -0,0 +1,91 @@
+# Explicit work-mode authority — 2026-09-22
+
+Product source: `d17c92eae10e949643a3b9d1be829061c30e77d6`.
+Fixture correction: `2e9ee319379ef263a511ef31717ee1185b53abdf`.
+Isolated worktree: `/private/tmp/paperclip-lifecycle-baseline-20260921`.
+
+## Behavior
+
+The task creation/update and native/legacy runner mode paths already use explicit
+fields. The remaining title/description regexes were a legacy liveness diagnostic
+exemption, not a write to `issues.work_mode`. They are now removed. Heartbeat and
+activity-ledger backfill pass the persisted work mode to the classifier.
+
+A standard task can deliver a requested plan in its thread or canonical plan
+document and complete. It does not become a planning-mode task. Explicit planning
+mode and revision-bound approval transitions retain their existing contract.
+No schema migration or inferred rewrite of existing tasks is needed.
+
+This is a bounded work-mode change. Other prose diagnostics, productive-progress
+accounting, the native compatibility probes and the browser reload finding remain
+tracked in [the backlog](../../doc/plans/2026-09-22-legacy-continuation-authority.md).
+
+## Deterministic verification
+
+The [retained baseline](baselines/2026-09-22-explicit-work-mode.json) records
+**990 passed / 992 total**: unit 393/393, runner 182/182, database integration
+361/363, grading 54/54. The same two unsupported native `same_agent` probes
+remain failed and visible; no new failure was accepted or skipped. The original
+baseline reports remain unchanged. Measurement finished September 22 in
+America/Los_Angeles (September 23 UTC).
+
+Five direct comparison probes also reject the previous classifier: the four
+historical title/body triggers incorrectly received the planning exemption,
+while a neutral-title task explicitly in planning mode did not. The corrected
+classifier passes all five. These comparisons are supplemental to the counted
+Vitest baseline.
+
+Coverage includes all former title/description trigger words, missing/default
+mode, explicit standard/planning/ask/skill-test modes, full/resumed prompt
+directives, native execution input, database create/edit/explicit-mode updates,
+canonical plan creation in standard mode, activity backfill, and real heartbeat
+repair pairs. The heartbeat comparison uses durable effects because successor
+completion can precede a predecessor's diagnostic projection.
+
+All 443 Product E2E support tests and all four inventory/report checks pass.
+Repository recursive typecheck, Product E2E typecheck, lifecycle-baseline typecheck,
+and repository build pass. The full general repository test command was not
+rerun for this small classifier change; the relevant suites are included above.
+
+## Real-provider verification
+
+Campaign [35805477715](https://github.com/paperclipai/paperclip/actions/runs/35805477715)
+finished **2/6 passed** on the product source SHA above. It selected six local Codex cells, using
+the existing `gpt-5.6-sol` profiles:
+
+- `lifecycle-work-mode-neutral` and `lifecycle-work-mode-challenge` on both
+  legacy and native: one provider turn per cell, a requested two-step plan,
+  exact visible output, standard mode before/after execution, Done, no extra
+  recovery or pending interaction.
+- `core-compatibility` / `plan-revise-accept` on both runtimes: explicit planning
+  mode, initial and revised plan, exact revision-bound approval, then completion.
+
+Both explicit planning controls passed. All four wording cases preserved
+standard mode and completed, but failed the browser's literal full-text match:
+Markdown renders ordered-list numbers separately from the DOM text. The retained
+legacy challenge screenshot shows the correct plan and Done. Three cases passed
+all persisted-state/output matchers. Legacy-neutral additionally posted the
+response twice: the provider log records two successful PATCH calls, the second
+escaping an underscore. That is a model behavior failure, not a duplicate server
+projection. The original machine grades remain unchanged.
+
+The fixture correction uses an equivalent plain-text two-step plan so the raw
+exact-output and browser exact-text contracts agree. All 443 support tests and
+the E2E typecheck pass again. Campaign
+[35806360797](https://github.com/paperclipai/paperclip/actions/runs/35806360797)
+is a fresh four-cell measurement of both wording pairs on that corrected source;
+**all four live cells passed**. Each runtime delivered the exact two-step plan
+once, preserved standard mode, completed in one run and left no extra work or
+pending interaction. All evidence validated and cleanup passed, with no retries
+or incomplete cells. The two explicit planning/revision/acceptance controls
+passed in the first campaign on identical product code. The corrected campaign
+does not regrade the first campaign or establish a statistical reliability rate.
+
+These are real browser/server/database/provider journeys. The full 46-cell
+lifecycle catalog is not selected by this follow-up. Both campaigns and their
+catalog hashes, source revisions, cleanup and billing coverage are retained in
+the [live inventory](baselines/2026-09-22-explicit-work-mode-live.json).
+Missing cost evidence must not be interpreted as free.
+
+Published reports: [original six-cell attempt](https://d1p6rlowie26tp.cloudfront.net/runner-e2e/campaigns/gha-35805477715-1/index.html),
+[corrected four-cell campaign](https://d1p6rlowie26tp.cloudfront.net/runner-e2e/campaigns/gha-35806360797-1/index.html).

tests/lifecycle-baseline/README.md

@@ -10,6 +10,8 @@ failures, representative observations and unmeasured coverage. The separate
 [live Actions baseline](LIVE-BASELINE-2026-09-21.md) records actual LLM/browser runs,
 assertion corrections and retained behavior failures. The [fix report](LIVE-FIXES-2026-09-21.md)
 records the later cancellation fix, corrected fixtures and fresh campaigns.
+The [explicit work-mode report](EXPLICIT-WORK-MODE-2026-09-22.md) records the
+title/description heuristic removal and its layered verification.
 
 ## Run and inspect
 

tests/lifecycle-baseline/baselines/2026-09-22-explicit-work-mode-live.json

@@ -0,0 +1,286 @@
+{
+  "schema": "paperclip.explicit-work-mode-verification/v1",
+  "measurementDateLocal": "2026-09-22",
+  "timezone": "America/Los_Angeles",
+  "campaigns": [
+    {
+      "source": {
+        "sha": "d17c92eae10e949643a3b9d1be829061c30e77d6",
+        "ref": "refs/heads/codex/lifecycle-behavior-baseline-20260921",
+        "workflowRunUrl": "https://github.com/paperclipai/paperclip/actions/runs/35805477715",
+        "eventName": "workflow_dispatch"
+      },
+      "campaignId": "gha-35805477715-1",
+      "artifactSha256": "921c157c55769bd3febe3a29d1aaa64c6c8c2c5e7b349c4e865964425c2b0c2f",
+      "totals": {
+        "selected": 6,
+        "executed": 6,
+        "passed": 2,
+        "failed": 4,
+        "incomplete": 0,
+        "retries": 0,
+        "cleanupPassed": true
+      },
+      "billing": {
+        "testCount": 6,
+        "agentRunDurationMs": 119026,
+        "leaseDurationMs": 0,
+        "llm": {
+          "runCount": 10,
+          "runsWithTokenUsage": 10,
+          "runsWithReportedCost": 5,
+          "inputTokens": 641012,
+          "outputTokens": 7375,
+          "cachedInputTokens": 410079,
+          "totalTokens": 1058466,
+          "reportedCostUsd": 0,
+          "costStatus": "partial"
+        },
+        "reportedLlmCostUsd": 0,
+        "estimatedRuntimeCostUsd": 0,
+        "observedAndEstimatedCostUsd": 0,
+        "testsWithCompleteBilling": 3
+      },
+      "definitionHashes": [
+        "b38be17e21fd6b4f81ae4a1ffc0488559940c65dec016ffbcf41107195775249",
+        "f82ddc73601921ce113ace60db788053b1c93ed11cc5add75fa237ea4c3e0393"
+      ],
+      "cells": [
+        {
+          "executionId": "lifecycle-baseline.legacy-codex.local.lifecycle-work-mode-neutral",
+          "caseId": "lifecycle-work-mode-neutral",
+          "profileId": "legacy-codex",
+          "environmentId": "local",
+          "runtimeMode": "legacy",
+          "model": "gpt-5.6-sol",
+          "status": "failed",
+          "failureClass": "candidate_failure",
+          "attempt": 1,
+          "durationMs": 62305,
+          "evidenceValid": true,
+          "evidenceErrors": [],
+          "cleanup": "passed",
+          "runCount": 1,
+          "failedMatchers": [
+            {
+              "matcher": {
+                "kind": "message_exact",
+                "expected": "ROLLOUT_7ddf65563aed-1\n1. Inspect the exporter.\n2. Implement and verify the change."
+              },
+              "passed": false,
+              "detail": "expected {\"kind\":\"message_exact\",\"expected\":\"ROLLOUT_7ddf65563aed-1\\n1. Inspect the exporter.\\n2. Implement and verify the change.\"}; observed \"ROLLOUT_7ddf65563aed-1\\n1. Inspect the exporter.\\n2. Implement and verify the change.\\nROLLOUT_7ddf65563aed-1\\n1. Inspect the exporter.\\n2. Implement and verify the change.\""
+            },
+            {
+              "matcher": {
+                "kind": "message_occurrences",
+                "expected": "ROLLOUT_7ddf65563aed-1\n1. Inspect the exporter.\n2. Implement and verify the change.",
+                "count": 1
+              },
+              "passed": false,
+              "detail": "expected {\"kind\":\"message_occurrences\",\"expected\":\"ROLLOUT_7ddf65563aed-1\\n1. Inspect the exporter.\\n2. Implement and verify the change.\",\"count\":1}; observed 2"
+            }
+          ]
+        },
+        {
+          "executionId": "lifecycle-baseline.legacy-codex.local.lifecycle-work-mode-challenge",
+          "caseId": "lifecycle-work-mode-challenge",
+          "profileId": "legacy-codex",
+          "environmentId": "local",
+          "runtimeMode": "legacy",
+          "model": "gpt-5.6-sol",
+          "status": "failed",
+          "failureClass": "candidate_failure",
+          "attempt": 1,
+          "durationMs": 56471,
+          "evidenceValid": true,
+          "evidenceErrors": [],
+          "cleanup": "passed",
+          "runCount": 1,
+          "failedMatchers": []
+        },
+        {
+          "executionId": "lifecycle-baseline.runner-codex.local.lifecycle-work-mode-neutral",
+          "caseId": "lifecycle-work-mode-neutral",
+          "profileId": "runner-codex",
+          "environmentId": "local",
+          "runtimeMode": "native",
+          "model": "gpt-5.6-sol",
+          "status": "failed",
+          "failureClass": "candidate_failure",
+          "attempt": 1,
+          "durationMs": 57976,
+          "evidenceValid": true,
+          "evidenceErrors": [],
+          "cleanup": "passed",
+          "runCount": 1,
+          "failedMatchers": []
+        },
+        {
+          "executionId": "lifecycle-baseline.runner-codex.local.lifecycle-work-mode-challenge",
+          "caseId": "lifecycle-work-mode-challenge",
+          "profileId": "runner-codex",
+          "environmentId": "local",
+          "runtimeMode": "native",
+          "model": "gpt-5.6-sol",
+          "status": "failed",
+          "failureClass": "candidate_failure",
+          "attempt": 1,
+          "durationMs": 51867,
+          "evidenceValid": true,
+          "evidenceErrors": [],
+          "cleanup": "passed",
+          "runCount": 1,
+          "failedMatchers": []
+        },
+        {
+          "executionId": "core-compatibility.legacy-codex.local.plan-revise-accept",
+          "caseId": "plan-revise-accept",
+          "profileId": "legacy-codex",
+          "environmentId": "local",
+          "runtimeMode": "legacy",
+          "model": "gpt-5.6-sol",
+          "status": "passed",
+          "failureClass": null,
+          "attempt": 1,
+          "durationMs": 62650,
+          "evidenceValid": true,
+          "evidenceErrors": [],
+          "cleanup": "passed",
+          "runCount": 3,
+          "failedMatchers": []
+        },
+        {
+          "executionId": "core-compatibility.runner-codex.local.plan-revise-accept",
+          "caseId": "plan-revise-accept",
+          "profileId": "runner-codex",
+          "environmentId": "local",
+          "runtimeMode": "native",
+          "model": "gpt-5.6-sol",
+          "status": "passed",
+          "failureClass": null,
+          "attempt": 1,
+          "durationMs": 60071,
+          "evidenceValid": true,
+          "evidenceErrors": [],
+          "cleanup": "passed",
+          "runCount": 3,
+          "failedMatchers": []
+        }
+      ],
+      "interpretation": "Both planning controls passed. Four new wording cases failed literal browser matching of Markdown lists. Legacy-neutral also issued two successful comment PATCH calls; model behavior evidence is retained. No regrading."
+    },
+    {
+      "source": {
+        "sha": "2e9ee319379ef263a511ef31717ee1185b53abdf",
+        "ref": "refs/heads/codex/lifecycle-behavior-baseline-20260921",
+        "workflowRunUrl": "https://github.com/paperclipai/paperclip/actions/runs/35806360797",
+        "eventName": "workflow_dispatch"
+      },
+      "campaignId": "gha-35806360797-1",
+      "artifactSha256": "6927541ae792c0f4e7013672de4e56135eeada52d5976b78bae467ba35b61647",
+      "totals": {
+        "selected": 4,
+        "executed": 4,
+        "passed": 4,
+        "failed": 0,
+        "incomplete": 0,
+        "retries": 0,
+        "cleanupPassed": true
+      },
+      "billing": {
+        "testCount": 4,
+        "agentRunDurationMs": 53034,
+        "leaseDurationMs": 0,
+        "llm": {
+          "runCount": 4,
+          "runsWithTokenUsage": 4,
+          "runsWithReportedCost": 2,
+          "inputTokens": 202266,
+          "outputTokens": 1767,
+          "cachedInputTokens": 124958,
+          "totalTokens": 328991,
+          "reportedCostUsd": 0,
+          "costStatus": "partial"
+        },
+        "reportedLlmCostUsd": 0,
+        "estimatedRuntimeCostUsd": 0,
+        "observedAndEstimatedCostUsd": 0,
+        "testsWithCompleteBilling": 2
+      },
+      "definitionHashes": [
+        "6c1a13b14f9f15e6eac6e557f39337ce4396b40a732f37a6fa1096b6b5ab2868"
+      ],
+      "cells": [
+        {
+          "executionId": "lifecycle-baseline.legacy-codex.local.lifecycle-work-mode-neutral",
+          "caseId": "lifecycle-work-mode-neutral",
+          "profileId": "legacy-codex",
+          "environmentId": "local",
+          "runtimeMode": "legacy",
+          "model": "gpt-5.6-sol",
+          "status": "passed",
+          "failureClass": null,
+          "attempt": 1,
+          "durationMs": 37511,
+          "evidenceValid": true,
+          "evidenceErrors": [],
+          "cleanup": "passed",
+          "runCount": 1,
+          "failedMatchers": []
+        },
+        {
+          "executionId": "lifecycle-baseline.legacy-codex.local.lifecycle-work-mode-challenge",
+          "caseId": "lifecycle-work-mode-challenge",
+          "profileId": "legacy-codex",
+          "environmentId": "local",
+          "runtimeMode": "legacy",
+          "model": "gpt-5.6-sol",
+          "status": "passed",
+          "failureClass": null,
+          "attempt": 1,
+          "durationMs": 34480,
+          "evidenceValid": true,
+          "evidenceErrors": [],
+          "cleanup": "passed",
+          "runCount": 1,
+          "failedMatchers": []
+        },
+        {
+          "executionId": "lifecycle-baseline.runner-codex.local.lifecycle-work-mode-neutral",
+          "caseId": "lifecycle-work-mode-neutral",
+          "profileId": "runner-codex",
+          "environmentId": "local",
+          "runtimeMode": "native",
+          "model": "gpt-5.6-sol",
+          "status": "passed",
+          "failureClass": null,
+          "attempt": 1,
+          "durationMs": 39179,
+          "evidenceValid": true,
+          "evidenceErrors": [],
+          "cleanup": "passed",
+          "runCount": 1,
+          "failedMatchers": []
+        },
+        {
+          "executionId": "lifecycle-baseline.runner-codex.local.lifecycle-work-mode-challenge",
+          "caseId": "lifecycle-work-mode-challenge",
+          "profileId": "runner-codex",
+          "environmentId": "local",
+          "runtimeMode": "native",
+          "model": "gpt-5.6-sol",
+          "status": "passed",
+          "failureClass": null,
+          "attempt": 1,
+          "durationMs": 33067,
+          "evidenceValid": true,
+          "evidenceErrors": [],
+          "cleanup": "passed",
+          "runCount": 1,
+          "failedMatchers": []
+        }
+      ],
+      "interpretation": "All four corrected standard-mode wording cells passed exact output, visible browser state, persisted mode and completion. One provider run per cell, no retries, valid evidence and successful cleanup."
+    }
+  ]
+}