Compellingness

Compellingness answers: how strongly does this case bear on the two public questions — anomalous technology/biologics, and a non-mundane origin? It reads the claims in the observableAssessments and origin slots and reduces them to a transparent reference score.

Usage

import { score } from '@disclosureos/scoring';

const result = score(observation);

The result

interface ScoreResult {
  score: number;                  // 0–1 consensus point estimate
  range: { low: number; high: number };  // spread across competing claims
  contested: boolean;             // claims disagree on direction
  components: {                   // per-signal contributions, before weighting
    technology: number;
    biologics: number;
    origin: number;
  };
  weights: CompellingnessWeights; // what produced this score
  scoringVersion: string;         // methodology version (e.g. '2.0.0')
}

Read all three headline values together: score: 0.71 with range: { low: 0.68, high: 0.74 } is a strong consensus; score: 0.4 with range: { low: 0, high: 0.8 }, contested: true is a live dispute — and presenting it as "0.4" alone would be a lie of omission.

How the signal is computed

Observable claims convert to anomaly magnitude via evidentiary tier × confidence:

Within each domain (technology, biologics) the scorer takes the strongest established observable — the "is there any anomaly?" reduction. Six weak signals don't outscore one confirmed one.

Origin claims contribute their confidence — unless the hypothesis is prosaic (the OCS 1.1.1 branch by default), in which case the magnitude is zero. A confident "it was a balloon" classification doesn't just fail to add anomaly signal; it pulls the consensus down and can flag the case contested against anomaly claims.

Combination is a weighted average of the three signals:

import { DEFAULT_WEIGHTS } from '@disclosureos/scoring';
// { technology: 0.45, biologics: 0.35, origin: 0.2 }

The default values are a reference methodology (marked @experimental) — pin your own weights if you need stable numbers across releases.

Options

const result = score(observation, {
  weights: { technology: 0.5, biologics: 0.3, origin: 0.2 },
  prosaicPrefixes: ['1.1.1', '2'],   // also treat psychosocial as explained
  evaluatorWeight: (claim) =>
    claim.evidenceRefs?.length ? 1 : 0.5,  // trust hook: discount uncited claims
});

The evaluatorWeight hook answers who gets the most weight: trust policies (weighting verified institutions higher, discounting anonymous claims) plug in here without touching the formula. It affects the consensus point only — range and contested always report the honest spread of what evaluators actually asserted.

Ranking a corpus

import { rankByCompellingness } from '@disclosureos/scoring';

const ranked = rankByCompellingness(observations);
for (const { observation, result } of ranked.slice(0, 10)) {
  console.log(observation.id, result.score, result.contested ? '⚠ contested' : '');
}

This is the "one case that matters" workflow: across a large ingest, the multi-sensor, multi-witness cases float to the top — with their disputes visible.

Reading scores responsibly

• A score is a function of the claims, not the event. No claims → score 0 — "unevaluated", not "mundane".
• Always display contested and range alongside score.
• Cite scoringVersion when publishing numbers; methodology versions matter.
• Completeness and compellingness travel together: a compelling-but-thin record is an investigation lead, not a conclusion.