01
Verified fact
A traceable observation supported by a verified or corroborated source. AI may organise or summarise it only after the factuality guard passes.
Evidence and publishing policy
How Whisper turns market signals into defensible CXO opportunity intelligence.
Whisper separates sourced observations, computed aggregates and AI-supported inference. Every public claim must carry an evidence label, an as-of date and a reproducible source trail. Claims that fail evidence, freshness, privacy or publishing checks are suppressed rather than presented as fact.
Claim separation
Four labels prevent signal from masquerading as certainty.
A corporate event can be factual while its possible implication for executive demand remains an inference. Whisper keeps those statements separate and never upgrades an inference into a confirmed role without qualifying evidence.
02
Evidence-backed aggregate
A calculation built from eligible observations and released only after sample, confidence, dominance and privacy tests pass.
03
Whisper inference
An interpretation of evidence, visibly separated from fact. It must state its boundary and what evidence would invalidate it.
04
Confirmed mandate
A role supported by an employer confirmation, public job posting or regulatory disclosure. A mandate is never confirmed by AI inference.
Source hierarchy
Evidence is graded before it supports a claim.
| Grade | Meaning | Examples | Publishing treatment |
|---|---|---|---|
| A | Primary or directly verified evidence | Regulatory filings, government records, audited reports, company disclosures and direct verified records. | Eligible when the record and observation pass verification, date and claim checks. |
| B | Strong professional secondary evidence | Licensed datasets, methodological research and reputable news reporting. | Eligible when verified or corroborated and used within the claim boundary. |
| C | Structured but less authoritative evidence | Structured secondary sources and analyst estimates. | May support a labelled inference or broader evidence mix; it cannot support a confirmed-mandate claim. |
| D | Unverified or generated material | Anonymous assertions, unverified aggregations and AI-generated material. | Never treated as verified fact. A source mix is suppressed when Grade D exceeds 25% of evidence weight, and an all-D mix is always suppressed. |
A source grade describes source quality, not the truth of every statement it contains. Verification state, observation dates, claim scope and corroboration are evaluated separately.
Freshness windows
Evidence expires at the speed of the claim.
Whisper evaluates the claim date and every supporting observation against its content class. Future-dated, invalid or stale evidence is suppressed instead of being quietly carried forward.
Breaking signal
Fast-moving events and early company signals
7 days
Leadership movement
Appointments, departures and leadership changes
30 days
Company intelligence
Operating, capital, expansion and governance context
90 days
Compensation
Benchmark observations and computed compensation cells
365 days
Evergreen methodology
Methods, definitions and stable explanatory material
730 days
AI boundary
AI can organise evidence. It cannot manufacture proof.
- AI output records the model, model version and prompt version used.
- A factuality guard must pass before an AI transformation can restate evidence-backed facts.
- AI-authored inference is labelled “Whisper inference” and includes a visible limitation statement.
- AI inference can never create a “Confirmed mandate” claim.
Aggregate privacy
Small or identifying cohorts do not become public benchmarks.
Aggregate evidence must pass every applicable release control. Passing one threshold does not compensate for failing another.
At least 12 distinct subjects
The minimum public cohort protects against direct identification.
Effective sample of at least 40
Weighted observations below this level receive Grade D and are suppressed.
No represented company above 35%
For company-based cohorts, one organisation cannot dominate the released cell.
No adjacent-cell residual from 1 to 11
Differencing checks prevent small residual groups being inferred by subtraction.
At least 70% input coverage
Incomplete calculations do not receive a publishable confidence grade.
Confidence is reproducible
Completeness 35%, reference depth 30%, recency 20% and validation 15%.
Correction lifecycle
Corrections preserve the record instead of rewriting history.
Every public claim is linked to the observations that support it. That lineage makes a correction targeted, reviewable and reversible without silently changing unrelated material.
- 01
Detect and record
A member report, source change, automated check or internal review opens a correction record linked to the affected claim or observation.
- 02
Fail closed
Disputed, withdrawn, stale or contract-invalid evidence stops supporting public publication while the issue is assessed.
- 03
Reconstruct
The original evidence, dates, transformation trace and affected calculations are reviewed. Material changes are not silently overwritten.
- 04
Correct or withdraw
A supported replacement is linked to the prior record, or the claim remains withdrawn when the evidence cannot be restored.
- 05
Revalidate before release
Freshness, evidence, privacy and publishing checks run again. A page returns to search only when the complete contract passes.
Found a source, date or factual error in Whisper material? Include the page URL, the disputed statement and the supporting record so the claim can be reconstructed.
Submit a correctionProduct boundary
Whisper provides private decision context—not recruitment representation.
Whisper is not a job board, recruitment agency, executive-search mandate or placement service. It does not create a public candidate profile, contact employers, submit applications, guarantee interviews or promise an offer. A company signal is not an open role unless the evidence supports the “Confirmed mandate” label.