/* global React */ const { Card, Chip, ClaimCeiling, GradientText, DomainTile } = window.OregonCoastAIDesignSystem_e04d97; function EvidenceScreen({ onNav }) { const Section = window.OCASection, Wrap = window.OCAWrap, Kicker = window.OCAKicker, SH = window.OCASH, Sub = window.OCASub; const deliverables = [ ["Feasibility report", "Scoped assessment of data, baselines, and whether the subgroup claim you want is reachable — with explicit limits.", ["Fixed scope", "Auditable"]], ["Versioned results package", "Structured results (JSON), human-readable report, and checksum — immutable experiment ID; prior runs never overwritten.", ["Versioned", "Reproducible"]], ["Methods packet", "Insert-ready methods, governance, and risk language for regulatory, publication, or grant review.", ["Grant-ready", "Human-verified"]], ["Claim-ceiling memo", "Exactly what each artifact is — and is not — allowed to claim, written to survive a reviewer or auditor.", ["Evidence-bound", "Gated emission"]], ]; const principles = [ ["Typed study graph", "Cohorts, hypotheses, experiments, and claims are named entities with explicit relations — not ad hoc folders or one-off prompts."], ["Evidence bindings", "Every outward claim has a ceiling and a receipt path. Upgrades require versioned results — not model confidence."], ["Gated emission", "AI assists literature, drafting, and formalization. Public statements and deliverables pass human-verified gates."], ]; return (
Results with receipts

Artifacts before adjectives.

Serious results ship as structured JSON, a human report, and a checksum — not slide-deck superlatives. Every metric carries its dataset, its claim ceiling, and an experiment ID you can trace.

{/* Receipt */}
One receipt, in full A methods demonstration on public data The spectral-entropy engine was first exercised on public whole-blood gene-expression cohorts. This is the kind of receipt every sponsored engagement returns — bounded, versioned, and explicit about what it does not establish.
methods-only · public benchmark · does not establish a clinical or diagnostic claim}> On a held-out public cohort, the trace-normalized local-covariance spectral-entropy readout separates molecular-state groups against a platform-matched healthy-control reference, under permutation and bootstrap controls.

entropy-deficit separation AUC 0.91 (95% CI 0.87–0.94) · permutation p < 0.001 · bootstrap-stable across 2,000 resamples

{[["Data", "Public whole-blood gene-expression cohort"], ["Records", "N=210 · n≪p regime"], ["Controls", "split-before-selection · label-null · complexity-penalty"]].map(([k, v]) => (
{k}
{v}
))}
// results-receipt.v1.json
{`{
  "experiment_id": "EXP-IE-DEMIX-0142",
  "status": "hypothesis-generating",
  "metric": "entropy_deficit_auc",
  "value": 0.9104,
  "ci95": [0.871, 0.938],
  "permutation_p": 0.0007,
  "abstained": 12,
  "claim_ceiling": "methods-only",
  "reference": "platform-matched HC",
  "sha256": "9f2c…a7e1",
  "prior_runs": "immutable · not overwritten"
}`}
{/* Epistemic principles */}
Epistemic technology Why the output is defensible Models generate faster than judgment. We separate the map of what we study from the machinery of inference — so a justified claim stays traceable to the run that earned it.
{principles.map(([t, d]) => {d})}
{/* Deliverables */}
What you receive Concrete artifacts — not promises
{deliverables.map(([t, d, chips]) => ( {chips.map((c) => {c})}
}>{d} ))}
); } window.EvidenceScreen = EvidenceScreen;