decision ledger

{
  "schema_version": "actionist.decision-ledger.v1",
  "part": "P15",
  "run_id": "2026-08-27-sprint-1-fable",
  "recorded": "2026-08-27",
  "owner": "ACTIONIST-S1-L5-RUNTIME",
  "boundary": {
    "research_only": true,
    "execution_status": "UNEXECUTED",
    "admission_status": "NOT_ADMITTED",
    "admitted_blocks": 0,
    "implementation_authorized": false,
    "promotion": "unpromoted"
  },
  "decisions": [
    {
      "id": "D-P15-01",
      "statement": "Ranking is min-gated on the weakest evidence family, never a weighted average",
      "state": "accepted",
      "support": "Local anti-averaging rule: a single T1 family pins the block at T1 regardless of eight T4s; averaging would let unresolved rights be compensated",
      "falsifier": "Min-gating proves degenerate (all assets pin to one tier) AND averaging demonstrably does not admit rights-compromised assets",
      "evidence_class": "first_party_docs"
    },
    {
      "id": "D-P15-02",
      "statement": "Rank by observed adaptation cost, not popularity metadata",
      "state": "accepted",
      "support": "A22 stars-predict-quality is unproven/weak; A07 adaptation cost unknown",
      "falsifier": "Metadata ranking matches outcome ranking over N builds",
      "evidence_class": "first_party_docs"
    },
    {
      "id": "D-P15-03",
      "statement": "Select estimators honest at n=10; frequentist A/B is disqualified",
      "state": "accepted",
      "support": "Regime is tens of expensive observations, not millions of free ones",
      "falsifier": "Posteriors remain too diffuse to separate candidates even with correct estimators",
      "evidence_class": "observed_behavior"
    },
    {
      "id": "D-P15-04",
      "statement": "Cold start is the only case that currently exists",
      "state": "accepted",
      "support": "Zero accepted plans, edits, incidents, maintenance outcomes, admitted blocks",
      "falsifier": "Production data appears that makes warm-start the dominant case",
      "evidence_class": "first_party_docs"
    },
    {
      "id": "D-P15-05",
      "statement": "Registry-health scorers, not recommenders, supply the scoring shape",
      "state": "accepted",
      "support": "OpenSSF Scorecard, ecosyste.ms, SourceRank, Renovate merge-confidence score assets on evidence; recommenders score users on preferences",
      "falsifier": "A recommender formulation outperforms an evidence-scoring formulation on asset selection",
      "evidence_class": "observed_behavior"
    },
    {
      "id": "D-P15-06",
      "statement": "Falsified-belief records are a first-class output",
      "state": "accepted",
      "support": "AutoSaaS lists failed assumptions as an update target; absent from the entire external census",
      "falsifier": "Falsified-belief records prove unusable or unmaintained in practice",
      "evidence_class": "first_party_docs"
    },
    {
      "id": "D-P15-07",
      "statement": "Exploration is permitted over quality, never over safety",
      "state": "accepted",
      "support": "Exploration means using a less-proven capability on a paying client's project",
      "falsifier": "A safety-relevant exploration proves necessary and acceptable",
      "evidence_class": "inference"
    },
    {
      "id": "D-P15-08",
      "statement": "UNDERDETERMINED aggregates are contract defects, not model failures",
      "state": "accepted",
      "support": "Composition architecture names this the single most valuable output for the contract lane",
      "falsifier": "UNDERDETERMINED results trace to model error rather than missing contract fields",
      "evidence_class": "first_party_docs"
    },
    {
      "id": "D-P15-09",
      "statement": "Catalogue size must never be a ranking signal",
      "state": "accepted",
      "support": "OpenConnector: 1,445 providers and 15,156 actions with a tenancy-unsafe storage model would have ranked top",
      "falsifier": "Catalogue breadth correlates with production reuse quality",
      "evidence_class": "first_party_docs"
    },
    {
      "id": "D-P15-10",
      "statement": "Cross-client aggregation availability is unknown and gates the premise",
      "state": "open",
      "support": "Client-private work; no contractual determination exists",
      "falsifier": "A client owner confirms or refuses anonymised cross-client aggregation",
      "evidence_class": "first_party_docs"
    },
    {
      "id": "D-P15-11",
      "statement": "Commercial ~100 target is not_applicable; there is no coherent market of production-learning-loop vendors",
      "state": "accepted",
      "support": "Threefold: Azure AI Personalizer (the category's flagship managed service) retired 25 Aug 2026 with the migration path being the microsoft/learning-loop OSS repo, not a successor; every genuine loop found is a feature inside another product category (Renovate merge confidence, v0 autofixer, Algolia Dynamic Re-Ranking) with no analyst category or competitive set; and the nearest true analogue is deliberately unbuilt — deps.dev and OpenSSF Scorecard contain no outcome feedback at all, while Renovate's is private. Built 22 records across 7 weighted categories instead.",
      "falsifier": "A set of ~100 vendors can be assembled that position against each other on production-feedback-loop quality",
      "evidence_class": "observed_behavior"
    },
    {
      "id": "D-P15-12",
      "statement": "Package-health scoring is the closest analogue and it contains no outcome feedback — the gap Actionist would fill is real",
      "state": "accepted",
      "support": "deps.dev and OpenSSF Scorecard are static analysis of repository state with no production-outcome signal; Mend's Renovate merge confidence is the only surveyed system pooling real cross-project outcomes and its algorithm is private. So outcome-based reranking of software assets is rare rather than standard.",
      "falsifier": "An existing package-health service is shown to ingest production outcomes and rerank on them",
      "evidence_class": "first_party_docs",
      "limitations": "absence established across the surveyed set, not proven across all vendors"
    }
  ]
}