PERSPICAZITY · PUBLIC-EVIDENCE INSTRUMENTS

Intelligence collects the dots. Perspicacity knows which dots are painted by the game to waste your time.

Perspicazity is a truth-reconciliation engine. It renders institutions as their behavior, not their marketing: independent public-evidence streams, fused into confidence-weighted signals that trace to archived primary sources, with the industry’s own metrics graded for how gameable they are.

SEC 01 / THE PROBLEMFILED UNDER: SELF-REPORT DRIFT

The industry is graded on its own paperwork.

Most institutional metrics in circulation are self-reported: disclosed by the measured party, on its schedule, in its format. A metric that is cheap to optimize and expensive to audit will, over time, measure the optimization rather than the behavior.

The stated number and the observed record can drift apart indefinitely, and every ranking, dashboard, and procurement decision built on the stated number inherits the drift. The market does not lack data. It lacks data that can be checked.

SEC 02 / THE ENGINEEVIDENCE MODEL · REV 2026.06

Behavior, reconciled from evidence that does not share an author.

M-01 / FUSIONIndependent streams, reconciled
The engine fuses public evidence streams with different authors and different incentives: regulator adjudications, mandated disclosures, audit findings, arbitration records. Weight follows independence: a signal corroborated across streams that cannot coordinate carries more confidence than any single account.
M-02 / PROVENANCEEvery signal carries its receipt
Each value traces to an archived primary-source document: SHA-256 digest, archive date, retrieval endpoint. The receipt is part of the output, not a footnote. Anyone downstream can retrieve the document and check the work.
M-03 / GRADINGThe metric itself is graded
Every axis carries a gaming-resistance grade, A through F: how expensive the underlying metric is to manipulate. Adjudicated outcomes grade high. Self-attested figures grade low. The grade travels with the number, so the number is never mistaken for more than it is.
M-04 / PARALLELNever one number
Axes are returned side-by-side, each under an explicit comparability class. A composite score would launder incommensurable evidence into false precision. The engine does not composite by design, at any tier, for any customer.
M-05 / REFUSALBelow threshold, silence
Where the evidence does not meet the confidence threshold, the engine returns insufficient_evidence instead of a number. A missing value is information. A fabricated one is debt.
SEC 04 / DELIVERYWHAT ACCESS RETURNS

What a subscription actually returns.

Every instrument delivers the same evidence base in four forms.

F-01 / CONSOLEA working console
A logged-in instrument: watchlists that open with receipt-backed changes since your last visit, and full dossiers with a how-to-cite strip on every value.
F-02 / EXHIBITSExhibits you can quote
Any datum exports as a citation-ready exhibit card: value, measurement period, source filing ID, retrieval date, SHA-256 digest. Printable, ready for a letter, memo, or committee file.
F-03 / ARCHIVEA standing record
Exhibits and memos generated under the account are kept in an append-only archive: a dated record of the review work performed, retrievable later.
F-04 / APIThe same records, programmatically
Every number the console shows is available over a versioned read API. The console is the API’s first customer: if it renders, the API returned it.
SEC 05 / PRINCIPLESSTANDING RULES OF THE ENGINE

The rules do not bend for the conclusion.

  1. P-01

    Receipts or silence

    No signal is published without an archived primary source attached.

  2. P-02

    Method on the record

    The evidence model and its weights are published and versioned. Changes are logged, never silent.

  3. P-03

    Correlation only

    Signals describe observed records. They do not establish cause, and they never impute intent.

  4. P-04

    Never composited

    Axes are reported side-by-side with an explicit comparability model. There is no single score.

  5. P-05

    Refusal below threshold

    Insufficient evidence returns as insufficient evidence, not as a softer number.

  6. P-06

    Corrections on the record

    Errors are corrected visibly and version-logged. The prior state remains retrievable.

  7. P-07

    Right of response

    Any graded party may submit a response. Responses are archived and attached to the record they address.

ACCESSREVIEWED MANUALLY · NOT SELF-SERVE

Request access.

The instruments are in limited release. Requests are reviewed manually; design-partner inquiries are prioritized.

Tier (Antaeus)