innovation register

{
  "id": "inn-p01-001",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Precompute-first discovery ladder: rank every ClientContext field by acquisition cost (public scrape < enrichment API < integration read < question) and never ask what a cheaper rung can supply",
  "evidence_class": "hypothesis",
  "source": "lane reasoning over parts.json P01 thesis",
  "limitations": "Cheaper rungs can be stale or wrong; needs freshness/confidence gating",
  "disposition": "candidate"
}

{
  "id": "inn-p01-002",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Industry-prior Bayesian warm start: initialize the client model from the 17-industry prior (entities, systems, atoms, pain distribution) so the first question already discriminates within-industry variance rather than re-deriving the industry",
  "evidence_class": "hypothesis",
  "source": "industry-discovery-priors.jsonl",
  "limitations": "Prior wrong for atypical clients; needs explicit prior-override path",
  "disposition": "candidate"
}

{
  "id": "inn-p01-003",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Expected-value-of-information (EVOI) question selector: score each candidate question by how much its answer is expected to change the recommended workflow set, not by template order",
  "evidence_class": "hypothesis",
  "source": "lane reasoning; active-learning literature (survey pending)",
  "limitations": "EVOI needs a decision model to compute value against; cold-start hard",
  "disposition": "candidate"
}

{
  "id": "inn-p01-004",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Stop asking when decision-invariant: stopping rule fires when the top-ranked CandidateOutcome is stable under all plausible answers to remaining questions",
  "evidence_class": "hypothesis",
  "source": "lane reasoning",
  "limitations": "Requires calibrated answer distributions; may stop early on overconfident priors",
  "disposition": "candidate"
}

{
  "id": "inn-p01-005",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Wants-vs-pain dual ledger: record stated wants and observed/derived pain as separate typed streams; a recommendation must cite at least one pain evidence item, not only wants",
  "evidence_class": "hypothesis",
  "source": "parts.json P01 open question; phase-2 contradiction fields",
  "limitations": "Pain evidence is scarce pre-engagement; risk of dismissing valid wants",
  "disposition": "candidate"
}

{
  "id": "inn-p01-006",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Contradiction as first-class object: when scraped evidence conflicts with a stated answer (e.g., stale listings vs 'all current'), persist a Contradiction record with both sources and surface it as a clarifying question rather than silently preferring either",
  "evidence_class": "hypothesis",
  "source": "phase-2 industry contradiction risks",
  "limitations": "Confronting clients with contradictions needs tactful UX",
  "disposition": "candidate"
}

{
  "id": "inn-p01-007",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Stack fingerprinting from public surfaces: detect storefront/PMS/ATS/CMS/analytics from the client's public web presence (headers, scripts, job posts, DNS) to pre-fill systems-of-record before any question",
  "evidence_class": "hypothesis",
  "source": "ecommerce/saas/recruiting priors note public detectability",
  "limitations": "Coverage varies by industry; law/healthcare nearly opaque publicly",
  "disposition": "candidate"
}

{
  "id": "inn-p01-008",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Job-posting mining as workflow evidence: a client's own job ads describe their processes, tools and volumes; parse them into WorkflowEvidence with provenance",
  "evidence_class": "hypothesis",
  "source": "lane reasoning",
  "limitations": "Postings lag reality; small firms post rarely",
  "disposition": "candidate"
}

{
  "id": "inn-p01-009",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Question budget as explicit client-effort contract: cap discovery at N minutes/questions per session, chosen from client seniority and deal size; the selector optimizes within budget rather than asking until confident",
  "evidence_class": "hypothesis",
  "source": "lane reasoning; parts.json 'client effort' analysis axis",
  "limitations": "Budget may truncate below decision threshold; needs graceful 'known unknowns' output",
  "disposition": "candidate"
}

{
  "id": "inn-p01-010",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Answer-reuse across engagements: tenant-scoped memory so a returning client is never asked anything already answered; deltas only (what changed since last time)",
  "evidence_class": "hypothesis",
  "source": "parts.json P01 owns conversation history",
  "limitations": "Staleness detection needed; org changes invalidate old answers",
  "disposition": "candidate"
}

{
  "id": "inn-p01-011",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Multi-stakeholder discovery routing: model the client as several actors (owner, ops lead, admin) and route each question to the cheapest credible answerer instead of one contact answering everything",
  "evidence_class": "hypothesis",
  "source": "phase-2 terminal-owner fields",
  "limitations": "Coordination overhead; small clients have one person anyway",
  "disposition": "candidate"
}

{
  "id": "inn-p01-012",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Show-your-model discovery UI: display the evolving ClientContext (systems, workflows, pains, confidence) to the client during discovery so they correct the model instead of answering a quiz",
  "evidence_class": "hypothesis",
  "source": "lane reasoning",
  "limitations": "Exposes model errors early (also a feature); UI investment",
  "disposition": "candidate"
}

{
  "id": "inn-p01-013",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Confidence-typed ClientContext fields: every field carries {value, source, evidence_class, confidence, observed_at}; downstream P02 consumes confidence, not bare values",
  "evidence_class": "hypothesis",
  "source": "AGENTS.md evidence discipline applied to product",
  "limitations": "Schema weight; calibration of confidence numbers is itself unproven",
  "disposition": "candidate"
}

{
  "id": "inn-p01-014",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Discovery from the tool the client already opens: importable evidence bundles (CSV exports, mailbox stats, calendar density) requested as optional uploads, each mapped to which questions they eliminate",
  "evidence_class": "hypothesis",
  "source": "lane reasoning; workflow-mining survey pending",
  "limitations": "Upload friction; privacy/authority review per bundle type",
  "disposition": "candidate"
}

{
  "id": "inn-p01-015",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Screen-recording workflow walkthrough: client records a 5-minute 'show me how you do X today'; system extracts steps/systems/handoffs as WorkflowEvidence — task-mining without installing anything",
  "evidence_class": "hypothesis",
  "source": "task-mining segment of OSS survey (pending)",
  "limitations": "PII exposure in recordings; consent and redaction required",
  "disposition": "candidate"
}

{
  "id": "inn-p01-016",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Volume denominators before value claims: discovery must capture at least one countable denominator (tickets/week, invoices/month) per candidate workflow; a workflow without a denominator cannot be ranked",
  "evidence_class": "hypothesis",
  "source": "project evidence standard (denominator discipline) applied to discovery",
  "limitations": "Clients often don't know their volumes; estimation protocol needed",
  "disposition": "candidate"
}

{
  "id": "inn-p01-017",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Pain-cost triangulation: estimate workflow cost from (volume denominator × time-per-item × who does it) captured as three separate low-precision answers, rather than asking 'how much does this cost you'",
  "evidence_class": "hypothesis",
  "source": "lane reasoning",
  "limitations": "Compounding estimate error; anchoring bias",
  "disposition": "candidate"
}

{
  "id": "inn-p01-018",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Negative-space probe: explicitly ask one question designed to falsify the current best hypothesis (disconfirmation quota) so discovery doesn't only confirm the industry prior",
  "evidence_class": "hypothesis",
  "source": "falsifiable-investigation discipline",
  "limitations": "Costs question budget; needs careful phrasing",
  "disposition": "candidate"
}

{
  "id": "inn-p01-019",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Authority-boundary discovery as its own track: capture who may approve sends/writes/money moves early, because authority shape changes which workflows are even proposable",
  "evidence_class": "hypothesis",
  "source": "phase-2 authority_boundary fields; atom contract authority family",
  "limitations": "Clients under-report approval reality",
  "disposition": "candidate"
}

{
  "id": "inn-p01-020",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "System-of-record conflict detector: when two named systems both claim an entity (CRM and spreadsheet both 'own' leads), emit a SourceOfTruthConflict requiring resolution before spec, since it predicts adaptation cost",
  "evidence_class": "hypothesis",
  "source": "phase-2 source_of_truth fields; SISOCRM one-owner rule",
  "limitations": "Resolution may need org change, not software",
  "disposition": "candidate"
}

{
  "id": "inn-p01-021",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Discovery-session compiler: after each conversation, compile transcript → typed ClientContext delta + unresolved DiscoveryGapList, so the artifact (not the chat log) is the state",
  "evidence_class": "hypothesis",
  "source": "conversation-intelligence segment (survey pending)",
  "limitations": "Extraction errors propagate; needs review surface",
  "disposition": "candidate"
}

{
  "id": "inn-p01-022",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Two-sided confidence thresholds: a follow-up question triggers when confidence is low AND the field is decision-relevant for the current candidate set; low-confidence irrelevant fields are left unknown deliberately",
  "evidence_class": "hypothesis",
  "source": "parts.json P01 open question on threshold",
  "limitations": "Decision-relevance itself shifts as candidates change",
  "disposition": "candidate"
}

{
  "id": "inn-p01-023",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Industry-prior drift telemetry: log where clients contradict the industry prior; periodically re-fit priors from accumulated engagements (P15 feedback loop applied to discovery)",
  "evidence_class": "hypothesis",
  "source": "P15 thesis joined to P01",
  "limitations": "Needs engagement volume that doesn't exist yet",
  "disposition": "candidate"
}

{
  "id": "inn-p01-024",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Pre-meeting client brief auto-draft: before the first call, generate a one-page 'what we think we know' brief (stack, size, likely pains from prior) — turning discovery into confirmation and correction",
  "evidence_class": "hypothesis",
  "source": "pre_meeting_account_briefs use case turned inward",
  "limitations": "Wrong guesses can anchor or embarrass; tone matters",
  "disposition": "candidate"
}

{
  "id": "inn-p01-025",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Effort-aware modality mixing: same information gathered via form field, chat question, call snippet or upload; selector picks modality by client persona (desk worker vs field operator) not one-size chat",
  "evidence_class": "hypothesis",
  "source": "lane reasoning",
  "limitations": "Multi-modal state sync complexity",
  "disposition": "candidate"
}

{
  "id": "inn-p01-026",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Regulated-industry discovery profiles: healthcare/law/mortgage get a distinct discovery mode that never ingests regulated content, asks about categories not records, and records the exclusion as evidence",
  "evidence_class": "hypothesis",
  "source": "healthcare/law priors; discovery-must-avoid-clinical-data note",
  "limitations": "Less evidence available → wider spec uncertainty; must be priced in",
  "disposition": "candidate"
}

{
  "id": "inn-p01-027",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "CandidateOutcome portfolio, not single answer: discovery outputs 2–4 ranked candidate outcomes with the evidence for each, and the client picks; picking is itself high-value evidence",
  "evidence_class": "hypothesis",
  "source": "parts.json P01 outputs CandidateOutcomes (plural)",
  "limitations": "Choice overload if candidates poorly differentiated",
  "disposition": "candidate"
}

{
  "id": "inn-p01-028",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Time-boxed async discovery: send the client 3 questions/day over chat channels they already use rather than a 45-minute session; measure completion lift",
  "evidence_class": "hypothesis",
  "source": "lane reasoning on client effort",
  "limitations": "Slower wall-clock; channel consent",
  "disposition": "candidate"
}

{
  "id": "inn-p01-029",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Question phrasing bound to industry vocabulary: render each canonical question in the industry's own terms from the priors file (e.g., 'matters' not 'projects' for law) — measured effect on answer quality",
  "evidence_class": "hypothesis",
  "source": "industry priors vocabulary layer",
  "limitations": "Wrong vocabulary is worse than neutral; needs per-industry review",
  "disposition": "candidate"
}

{
  "id": "inn-p01-030",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Explicit unknowns ship with the handoff: DiscoveryGapList is a typed artifact consumed by P02, which must convert each gap into either an assumption (labelled) or an acceptance-test question — no silent gap loss",
  "evidence_class": "hypothesis",
  "source": "parts.json P01→P02 output contract",
  "limitations": "Discipline cost; requires P02 cooperation contract",
  "disposition": "candidate"
}

{
  "id": "inn-p01-031",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Client-effort ROI display: show the client 'answering these 4 questions removes 2 weeks of build risk' — making question cost/value explicit to raise completion of high-value questions",
  "evidence_class": "hypothesis",
  "source": "lane reasoning",
  "limitations": "Risk quantification is initially hand-tuned",
  "disposition": "candidate"
}

{
  "id": "inn-p01-032",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Integration-read consent ladder: request read-only OAuth scopes ordered by evidence value per intrusiveness (calendar metadata → mailbox counts → CRM objects), each unlocking precomputation that removes question blocks",
  "evidence_class": "hypothesis",
  "source": "introspection segment of OSS survey (pending); P11 connector plane",
  "limitations": "Consent friction; tenant-safe storage required (A26 lesson)",
  "disposition": "candidate"
}

{
  "id": "inn-p01-033",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Denominator-first industry scoring for pilot choice: S3-L1 needs all-17 scoring; P01 should define the score now as f(evidence availability, authority risk, workflow value density) computable from priors + discovery data",
  "evidence_class": "hypothesis",
  "source": "task-graph S3-L1 dependency",
  "limitations": "Score weights are judgment until pilot data exists",
  "disposition": "candidate"
}

{
  "id": "inn-p01-034",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Anti-questionnaire guarantee as product positioning: publicly commit 'we never ask more than N questions before showing you something' — a falsifiable UX contract that differentiates from agency discovery decks",
  "evidence_class": "hypothesis",
  "source": "lane reasoning; P02 prototype-vs-ask question",
  "limitations": "N must be honest across industries; regulated industries strain it",
  "disposition": "candidate"
}

{
  "id": "inn-p01-035",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Second-source rule for load-bearing fields: any field that changes the archetype choice (industry, primary system of record, volume order-of-magnitude) requires two independent sources or an explicit single-source flag",
  "evidence_class": "hypothesis",
  "source": "evidence discipline applied to ClientContext",
  "limitations": "Doubles acquisition cost on key fields",
  "disposition": "candidate"
}

{
  "id": "inn-p01-036",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Discovery replay determinism: same evidence in → same ClientContext out; model-extracted fields carry extraction-version so re-running with a better model is a diff, not a rewrite",
  "evidence_class": "hypothesis",
  "source": "deterministic-spine preference in composition plane applied to demand plane",
  "limitations": "LLM nondeterminism needs pinning/temperature control",
  "disposition": "candidate"
}

{
  "id": "inn-p01-037",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Competitor-app teardown as demand evidence: if the client names a tool they envy, teardown its public surface into feature atoms and add to CandidateOutcomes with 'envy' provenance class",
  "evidence_class": "hypothesis",
  "source": "lane reasoning",
  "limitations": "Envy ≠ need; must be weighed below pain evidence",
  "disposition": "candidate"
}

{
  "id": "inn-p01-038",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Cold-start question pack per industry, versioned: ship 17 curated 8-question opening packs derived from priors; every pack question must name which ClientContext fields it fills and which candidate outcomes it discriminates",
  "evidence_class": "hypothesis",
  "source": "industry priors must_ask fields",
  "limitations": "Curation effort; packs drift without telemetry",
  "disposition": "candidate"
}

{
  "id": "inn-p01-039",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Trust-graded evidence ingestion: client-uploaded documents are data not instructions (prompt-injection boundary); discovery pipeline strips/flags any imperative content before model exposure",
  "evidence_class": "hypothesis",
  "source": "C04 stop condition in first-principles framework",
  "limitations": "Detection imperfect; needs adversarial fixtures",
  "disposition": "candidate"
}

{
  "id": "inn-p01-040",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Discovery-quality score as a gate: an engagement may not proceed to ProductSpec until ClientContext reaches a minimum completeness/confidence profile per archetype — the P01/P02 seam becomes a measurable contract",
  "evidence_class": "hypothesis",
  "source": "lane reasoning on P01→P02 contract",
  "limitations": "Thresholds initially arbitrary; must not become bureaucracy",
  "disposition": "candidate"
}

{
  "id": "inn-p01-041",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Ask-the-system-not-the-person: where an integration read exists, compute the answer and ask the client only to confirm anomalies ('you seem to get ~40 tickets/week — right?') — confirmation is cheaper than recall",
  "evidence_class": "hypothesis",
  "source": "precompute ladder + conversation design",
  "limitations": "Wrong precomputed numbers erode trust fast",
  "disposition": "candidate"
}

{
  "id": "inn-p01-042",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Explicit business-loop capture for idea-type clients: when the client is starting a business (72-ideas surface), discovery must capture business loop + monetization event, not just feature wants — reusing the sweep-spec's idea decomposition rule",
  "evidence_class": "hypothesis",
  "source": "actionist-solutions-sweep-spec idea rule",
  "limitations": "Different question policy than established-business discovery",
  "disposition": "candidate"
}

{
  "id": "inn-p01-043",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Session-zero from Actionist account data: if the client already uses Actionist agents, their existing workflow/agent/tool usage is the richest pain evidence available; mine it (with consent) before anything external",
  "evidence_class": "hypothesis",
  "source": "parts.json P01 inputs 'Actionist account profile'; P11 dependency",
  "limitations": "Actionist account data contract is U (unknown) per prior research",
  "disposition": "candidate"
}

{
  "id": "inn-p01-044",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Calibration ledger for the question model: every precomputed guess later confirmed/corrected by the client is logged as a calibration pair; the discovery system reports its own hit rate as a trust metric",
  "evidence_class": "hypothesis",
  "source": "lane reasoning; prove-before-claim culture",
  "limitations": "Needs volume; gaming risk if hit rate becomes a KPI",
  "disposition": "candidate"
}

{
  "id": "inn-p01-045",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Workflow-evidence taxonomy with court-grade provenance: WorkflowEvidence: {stated, document, recording, integration_read, public_observation} each with defined reliability ordering used by ranking",
  "evidence_class": "hypothesis",
  "source": "evidence-class discipline specialized for discovery",
  "limitations": "Ordering is a prior, not a law; some stated evidence is excellent",
  "disposition": "candidate"
}

{
  "id": "inn-p01-046",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Dead-end insurance: every discovery flow keeps a 'what we'd do with zero further answers' fallback spec so a client who ghosts mid-discovery still yields a usable (widely-hedged) proposal",
  "evidence_class": "hypothesis",
  "source": "lane reasoning on client-effort reality",
  "limitations": "Hedged proposals may underwhelm; label uncertainty honestly",
  "disposition": "candidate"
}

{
  "id": "inn-p01-047",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Cross-industry analog suggestion: when within-industry supply is thin, discovery may propose an analog from a sibling archetype ('what mortgage brokers do for conditions works for your change-orders') with the analogy recorded as inference",
  "evidence_class": "hypothesis",
  "source": "archetype compression (A02) exploited at discovery time",
  "limitations": "Analogies can mislead; must remain visible as inference",
  "disposition": "candidate"
}

{
  "id": "inn-p01-048",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Zero-question fast lane for tiny scopes: below a size threshold (one workflow, one system), skip structured discovery and go straight to prototype-as-question (P02 seam), measuring whether skipping hurts acceptance",
  "evidence_class": "hypothesis",
  "source": "P02 open question 'when should a prototype answer a question'",
  "limitations": "Threshold judgment; scope creep risk",
  "disposition": "candidate"
}

{
  "id": "inn-p01-049",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Client-language mirror check: before handoff, render the captured ClientContext back in the client's own vocabulary for sign-off; signed context becomes the P02 input baseline and disputes become recorded contradictions",
  "evidence_class": "hypothesis",
  "source": "read-back verification pattern from atom contracts applied to discovery",
  "limitations": "Sign-off fatigue; keep it one page",
  "disposition": "candidate"
}

{
  "id": "inn-p01-050",
  "part": "P01",
  "observed": "2026-08-27",
  "claim": "Regulatable evidence budget: discovery records exactly which external sources were touched per engagement (scrape log with robots/consent state) making the 'what may be scraped automatically' question an auditable policy rather than an open debate",
  "evidence_class": "hypothesis",
  "source": "parts.json P01 open question on scraping boundary",
  "limitations": "Policy content still needs an owner; log ≠ permission",
  "disposition": "candidate"
}