Help us improve
Share bugs, ideas, or general feedback.
From spotlight
Classifies findings by evidence strength using a grounding ladder. Use for fact-checking claims, assigning confidence, and diagnosing weak support in investigations.
npx claudepluginhub buriedsignals/spotlight --plugin spotlightHow this skill is triggered — by the user, by Claude, or both
Slash command
/spotlight:epistemic-groundingThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use this skill whenever Spotlight turns source material into claims or evaluates whether a claim is ready for editorial use.
Evaluates claims by triangulating sources, rating evidence quality, and reaching confidence-rated conclusions. Use for fact-checking, due diligence, or resolving conflicting evidence.
Entry point for the investigation toolkit. Routes to the right skill for tracing sources, decomposing claims, auditing evidence, generating counter-hypotheses, or triangulating across independent sources.
Extracts and verifies factual claims from PR copy or journalistic drafts, providing citations and warning on low certainty. Use before sending pitches or press releases.
Share bugs, ideas, or general feedback.
Use this skill whenever Spotlight turns source material into claims or evaluates whether a claim is ready for editorial use.
The core question is:
Does this exact evidence justify believing this exact claim?
A source is only an anchor. Grounding is the relationship between the claim and the anchor.
Classify each candidate finding on this ladder:
Only levels 4-5 can be high confidence. Levels 1-2 are leads, not findings. Level 3 is at most medium confidence and often low.
For every finding or claim:
primary — original record, document, direct statement, data source, filing, archived page, or observable artifact.secondary — reporting, analysis, database aggregation, third-party summary.contextual — useful background, but not evidence for the claim.direct — evidence states the claim elements plainly.indirect — evidence supports the claim through a short, explicit inference.inferred — evidence requires unstated assumptions or synthesis across sources.contradicted — reliable evidence conflicts with the claim.insufficient — source does not support the claim.Use these caps even if the claim sounds plausible:
| Condition | Maximum Confidence |
|---|---|
| No scraped local file | low |
| Search snippet only | low |
| Source is contextual or adjacent | low |
| Evidence is inaccessible, abstract-only, or materially redacted | low |
| Claim requires unstated assumptions | medium |
| Only secondary sources support the claim | medium |
| Single primary source, no contradiction search yet | medium |
| Direct primary source plus independent corroboration | high |
| Credible unresolved contradiction | low or disputed |
Never upgrade a claim beyond the weakest material element. If the amount is directly supported but the date is inferred, the whole claim is only partially grounded.
Every findings.json finding must include:
"grounding": {
"support_type": "direct|indirect|inferred|contradicted|insufficient",
"source_role": "primary|secondary|contextual",
"claim_elements_supported": ["actor", "action", "date"],
"missing_assumptions": [],
"confidence_cap": "high|medium|low",
"misgrounding_risk": "short risk statement",
"grounding_rationale": "why the evidence does or does not ground the claim; include the contradiction-search outcome here"
}
Every fact-check claim must include a grounding_assessment explaining whether the cited evidence actually grounds the claim.
When a finding feels wrong, do not patch the wording first. Diagnose the grounding failure:
Read references/failure-router.md for deeper failure classes. Read references/grounding-theory.md when designing or revising grounding policy.