Part of CNS 7.1 / GCTS: Grounded Chiral Tensor Synthesis

Adversarial Evidence And Access Modeling

GCTS is designed for environments where evidence is limited, contradictory, controlled, or strategically curated. The central discipline is simple: absence has structure.

Absence States

StateMeaning
Absence of evidenceNo available supporting evidence has been found
Evidence of absenceAn expected record or observation exists and affirmatively negates the claim
Inaccessible evidenceThe record may exist outside the current access path
Sealed evidenceThe record exists or plausibly exists under restricted access
Withheld evidenceNon-production is more likely under a withholding world than under benign missingness
Destroyed evidenceThe record existed or was expected and is no longer available
Not-generated evidenceThe record should not be expected to exist
Unknown accessCurrent evidence cannot classify the access state

Only evidence of absence can directly penalize a claim as absent. Other states usually create access uncertainty, record contingencies, or competing worlds.

Access Features

For each expected record, GCTS models:

  • who owns it;
  • who controls production;
  • whether ordinary procedure would generate it;
  • whether the event should be observable by that record system;
  • whether the record was requested, produced, refused, partially produced, contradicted, destroyed, sealed, delayed, or unavailable;
  • confidence in the access-state classification.

Incentive Features

Institutional incentive profiles model:

  • control over records or testimony;
  • reputational, legal, financial, operational, or political exposure;
  • incentive to disclose;
  • incentive to conceal, delay, narrow, or frame evidence;
  • expected penalty if concealment is detected;
  • prior source reliability.

Incentives affect missingness likelihood, source quality, and world energy while leaving proof to evidence and rules. Claims still require evidence and rules.

Suppression Discipline

The system should infer strategic withholding only when several conditions line up:

  • a record was expected to exist;
  • a responsible actor plausibly controlled it;
  • the access path was legitimate or ordinary;
  • non-production is less likely under benign missingness;
  • the hypothesis reduces contradiction or explains access asymmetry without excessive unsupported complexity.

Unsupported suppression hypotheses should increase parsimony penalty.

Selective Production

Adversarial environments often produce some records while withholding, narrowing, delaying, or reframing others. GCTS should treat partial production as an observed production state with remaining access limits.

Examples:

  • A roster is produced but the incident report is not.
  • A policy is produced but the compliance log is not.
  • Metadata is produced but content is withheld.
  • A summary is produced but source records are not.
  • A record appears only after an initial nonresponsive response.

Selective production can support some claims while increasing access uncertainty around others.

Output Requirements

Any report involving missing or controlled evidence should state:

  • which records matter;
  • expected generation duty;
  • observed access state;
  • production response;
  • confidence in the classification;
  • whether the claim is record_contingent;
  • what evidence would raise, lower, or resolve the claim ranking.
Step 6 of 11 in CNS 7.1 / GCTS: Grounded Chiral Tensor Synthesis