lane state

{
  "schema": "actionist.lane-state.v1",
  "part": "P06",
  "run_id": "2026-08-27-sprint-1-fable",
  "lane": "S1-L3",
  "agent": "ACTIONIST-S1-L3-EXPERIENCE",
  "authored_by": "opus_subagent_s1l3-p06 (denominators, report, first-principles, decision-ledger v2, registers); lane_owner (v1 placeholder packet, preserved and superseded)",
  "model_policy": "opus_only_subagents",
  "parent_model_at_write": "opus-5-1m",
  "counts": {
    "source_register": 26,
    "top_companies": 113,
    "top_companies_ranked": 10,
    "top_repos": 119,
    "top_repos_ranked": 10,
    "innovation_register": 110,
    "innovation_ranked": 10,
    "decisions": 10
  },
  "evidence_quality": {
    "source_register": {
      "observed": 25,
      "inferred": 1,
      "hypothesis": 0
    },
    "top_companies": {
      "observed": 13,
      "inferred": 33,
      "hypothesis": 57,
      "note": "counts exclude the 10 ranked rows, which duplicate their parent record"
    },
    "top_repos": {
      "observed": 108,
      "inferred": 0,
      "hypothesis": 1,
      "note": "counts exclude the 10 ranked rows, which duplicate their parent record"
    },
    "innovation_register": {
      "observed": 42,
      "inferred": 31,
      "hypothesis": 27,
      "note": "counts exclude the 10 ranked rows, which duplicate their parent record"
    }
  },
  "denominator_targets": {
    "commercial": 100,
    "oss_literature": 100,
    "innovations": 100
  },
  "denominator_actual": {
    "commercial": 103,
    "oss_literature": 109,
    "innovations": 100
  },
  "depth_contract": "MET ON COUNT; UNEVEN ON VERIFICATION. OSS/literature is strong: 109 base rows, 108 observed — repos verified via authenticated gh api with LICENCE BODIES read rather than badges, papers via arXiv/Crossref with several full-text PDF extractions. Commercial is the weak leg: 103 surfaces enumerated but only 13 observed, because consumer style quizzes are client-side JS behind bot protection and are not fetchable.",
  "resolved_questions": {
    "comparison_format": "4-up gallery with explicit outside option as PRIMARY; 4-way best-worst pre-registered as A/B arm B. This DIVERGES from the lane owner's v1 ledger, which chose best-worst; the disagreement is recorded in decision-ledger.json P06-D01 and is resolvable by Gate 4.",
    "how_many_choices": "8-12 rounds, hard floor 5, hard ceiling 15. Derived from a channel-capacity argument (36,000 packs = 15.14 bits; +/-1-level tolerance leaves 4.04 bits to acquire; ~0.91 effective bits per screen at p=0.15) and independently corroborated by five empirical sources all landing in the 5-15 band. 'Is it 7 or 14' is not well-posed as asked; 7 is approximately right by coincidence, 14 has no derivation.",
    "stopping_rule": "Three conditions, any of: per-knob posterior within +/-1 level AND expected information gain < ~0.25 bits; ceiling of 15 rounds; user presses an always-available 'this is right'. Floor of 5. Confidence reported PER KNOB, never as one scalar.",
    "taste_profile_to_tokens": "Continuous preference vector learned in knob space; closed pre-authored gate-passing pack selected by nearest neighbour. Never interpolate tokens — an interpolated pack has passed none of gates A-J, and WCAG luminance is non-linear so the midpoint of two AA-passing palettes can fail AA. Closed-pack snapping IS the anti-overfitting mechanism.",
    "screens_vs_fragments": "Whole screens re-rendered from our own spec, with only the targeted knobs varied and content held constant within a round. Fragments reserved for a targeted disambiguation phase only (falsifier F4)."
  },
  "checkpoints": [
    "prior-evidence audit (4 local priors read in full)",
    "live demo inspected — resolves v1 shortfall SF04",
    "dispatch of 2 Opus research workers",
    "OSS/literature denominator returned and verified (119 rows)",
    "commercial worker hit a credit limit at ~8 surfaces; lane re-ran the sweep directly",
    "information-theoretic derivation + sensitivity analysis",
    "packet authored; v1 owner packet preserved and explicitly superseded"
  ],
  "blockers": [
    "COMMERCIAL DENOMINATOR THINLY VERIFIED — 13 of 103 observed. Consumer style quizzes are client-side JS behind bot protection (403 from Stitch Fix, Warby Parker, Looka, Behr, Hinge; JS shell from Function of Beauty; 404 from 1000minds, Conjointly). Needs a real browser session, not more fetching.",
    "NO SUFFICIENCY GUARANTEE EXISTS for our setting. The 8-12 recommendation is a derivation plus five converging empirical sources, not a theorem. Gate 3 (simulate-to-size) must convert it into a measured number before it is quoted to the client as anything firmer.",
    "DUAL-AUDIENCE PROBLEM UNSOLVED — for the 6 of 17 industries where `portal` is the secondary archetype, the person choosing (operator) is not the person using (end customer). No prior art found anywhere in the denominator. This is a design gap, not a research gap.",
    "KNOB INDEPENDENCE ASSUMED, NOT MEASURED (Gate 1). The whole few-rounds argument rests on effective dimensionality being below 7.",
    "1000minds PAPRIKA UNVERIFIED (404 on both attempted paths). Its transitivity-based pair elimination is potentially directly applicable and should be verified before the format decision is frozen.",
    "FORMAT DIVERGENCE from the lane owner's v1 ledger is live and unresolved by argument (P06-D01)."
  ],
  "falsifiers": [
    {
      "id": "F1",
      "claim": "+/-1-level tolerance is acceptable",
      "gate": "Gate 0",
      "priority": "run first — most load-bearing and cheapest"
    },
    {
      "id": "F2",
      "claim": "8-12 rounds suffice",
      "gate": "Gate 3"
    },
    {
      "id": "F3",
      "claim": "knobs are correlated so effective d < 7",
      "gate": "Gate 1"
    },
    {
      "id": "F4",
      "claim": "whole-screen controlled stimuli beat fragments",
      "gate": "deferred"
    },
    {
      "id": "F5",
      "claim": "closed-pack nearest-neighbour output is right",
      "gate": "Gate 5"
    },
    {
      "id": "F6",
      "claim": "4-up + outside option beats best-worst",
      "gate": "Gate 4"
    },
    {
      "id": "F7",
      "claim": "preference is stable enough to be worth learning",
      "gate": "Gate 2",
      "priority": "would kill the part rather than adjust it"
    }
  ],
  "smoke": "pending",
  "boundaries": {
    "research_only": true,
    "implementation_authorized": false,
    "execution_status": "UNEXECUTED",
    "admission_status": "NOT_ADMITTED",
    "admitted_blocks": 0,
    "promotion_status": "unpromoted"
  },
  "note_on_supersession": "The lane owner authored a placeholder packet at ~16:35 while the first delegated worker had produced no output, then corrected lane-state to match files on disk. This run's packet supersedes research-report.md, first-principles.md, decision-ledger.json, top-repos.jsonl, top-companies.jsonl and innovation-register.jsonl. The owner's source-register.jsonl rows (SR-001..SR-014) are PRESERVED with SR-012/013/014 updated in place; SR-015..SR-026 are new. The owner's D01 best-worst choice is preserved as a recorded divergence rather than overwritten."
}