{ "requiredPairs": 10, "currentPairs": 10, "onPositive": 0, "neutralOrMixed": 0, "offPositive": 10, "countBasis": "measured-evidence", "acceptedMeasurement": { "label": "local-current-acceptance-ab-20260702", "archivePath": "/Users/yongtae/Desktop/persona-harness-artifacts/archive/2026-06-24-desktop-persona-runs/local-current-acceptance-ab-20260702-20260702-153213", "head": "1563a25ca5bbddcaf3d63e7f5e7e73d61b9b718d", "packageVersion": "0.4.1-rc.2", "acceptanceChecklist": { "pass": 49, "notApplicable": 5, "fail": 0 }, "pairedOpenCodeSessions": { "pairs": 10, "totalSessions": 20, "pairControls": [ "same README inside each pair", "same task prompt inside each pair", "same starting state inside each pair", "counterbalanced OFF/ON order" ], "offSuccess": "10/10", "onSuccess": "10/10", "allProviderTokenTotalDeltasFavoredOff": true, "allReadCharDeltasFavoredOff": true, "allToolCallDeltasFavoredOff": true, "allElapsedDeltasFavoredOff": true, "meanProviderTotal": { "off": 119320.7, "on": 712935.8 }, "meanReadChars": { "off": 1152.7, "on": 20650 }, "meanToolCalls": { "off": 15.4, "on": 38.9 }, "meanElapsedMs": { "off": 51261.7, "on": 152525.5 } }, "pminus": { "surface": "ph-runtime-injection", "outcome": "worse", "decisionHint": "remove-candidate", "automaticRemoval": false }, "scope": "bounded local-current OpenCode app-generation fixture set", "claimBoundary": "This is not a universal token-saving, provider-token-saving, product-efficacy, navigation-benefit, or generated-app-quality claim." }, "legacySelfRatedCounts": { "currentPairs": 3, "onPositive": 3, "neutralOrMixed": 0, "offPositive": 0, "basis": "self-rated qualitative review only; retained for audit, not measured evidence" }, "evalProgram": { "status": "stopped", "stoppedAt": "2026-06-28", "aggregatePath": "experiments/eval-signal/aggregate.json", "diagnosisPath": "experiments/eval-signal/diagnosis.json", "aggregateCommit": "fa19539 test(eval): preserve aggregate signal", "diagnosisCommit": "75a16ac test(eval): diagnose frozen signal", "totalResultCount": 30, "originalResultCount": 12, "replayResultCount": 18, "unknownResultCount": 0, "originalRunCount": 80, "diskRecovery": { "experimentsBefore": "4.0GB", "experimentsAfter": "132MB", "evalRunsAfter": "1.1MB", "evalSignalAfter": "131MB" }, "phOnWorkflowPassCount": 10, "phOnProviderLimitedWorkflowFailureCount": 7, "phOnWorkflowDeadEndFailureCount": 3, "phOnAmbiguousWorkflowFailureCount": 0, "runtimeSmokePh": 0.8235294117647058, "runtimeSmokePlain": 0.8823529411764706, "stackAlignmentPh": 0.7875, "stackAlignmentPlain": 0.625, "workflowFinishPh": 0.5, "workflowFinishPlain": 0, "deconfoundedRuntimeSmokePhMinusPlain": -0.05882352941176472, "deconfoundedStackAlignmentPhMinusPlain": 0.16249999999999998, "deconfoundedWorkflowFinishPhMinusPlain": 0.5, "aggregationBasis": "original-only; replay results excluded from aggregate signal", "historicalCompileTestCaveat": "Historical compile/test aggregates are toolchain-confounded and are not apples-to-apples PH win rates.", "completionSemantics": "Provider/tool completion was separated from generated app external outcome before the final frozen step.", "diagnosisInterpretation": "Workflow/stack positive signal exists. R1 CLI-only reproduction found and fixed a workflow continuation UX gap: finish surfaced raw final verification/bearshell blockers, but workflow continue did not carry that blocker into its next-action prompt. Remaining historical failures still cannot be fully classified from pruned raw evidence.", "workflowContinuationUxFix": { "commit": "19ae341 fix(cli): surface final verification continuation", "verification": "cli-only reproduction", "freshFinishRateEval": "not-rerun", "effectOnFinishRate": "unverified", "finishGate": "not-weakened", "enforcementAdded": false }, "closureRecommendation": "No further eval or analysis loop is recommended without a new acceptance question and explicit HQ approval.", "stoppingReason": "Deconfounded runtime smoke does not show PH clearly beating the plain baseline; keep injection effect unproven." }, "decision": "freeze-expansion", "decisionRationale": "Accepted 10-pair local-current OpenCode PH OFF/ON evidence found PH runtime injection worse for provider-token total, read chars, tool calls, and elapsed time in all pairs. Runtime guidance is downgraded to explicit opt-in preview while gate/evidence CLI surfaces remain the product center." }