Methods and Research
Use the minimum-sufficient method; do not mistake the ecosystem for the Standard
A Foundation method is not a HEART requirement unless the applicable verification plan selects it under current authority. Existing sufficient evidence and external methods remain first-class.
Current portfolio
| Method | Purpose and use | Do not use when | Evidence / validation | Normative status and claim ceiling |
|---|---|---|---|---|
| RCTA v2.0 | Structure a bounded behavioral-governance measurement problem around Recognition, Calibration, Transparency, and Accountability. | The governing source supplies a better direct test or the four dimensions are not valid for the claim. | Research-bounded construct requiring valid operationalization and comparison with alternatives. | Conditional Method. Does not define HEART conformance or exhaust all governance. |
| BGF v2.0 | Aggregate valid, commensurable RCTA measurements where a decision rule justifies aggregation. | Dimensions are unvalidated, non-commensurable, missing, or the source requires independent determinations. | Formula behavior is known; construct, threshold, and criterion validity are claim-specific. | Conditional Method. No universal threshold, tier, HVC, compliance, or market meaning. |
| GTE v2.0 | Increase confidence that selected governance controls executed and evidence was produced under declared conditions. | Ordinary logs, tests, attestations, or existing assurance are sufficient. | Technical assurance depends on implementation, threat model, keys, coverage, and independent testing. | Conditional Technical Standard. Does not prove policy adequacy or legal compliance. |
| MAP-States v2.0 | Capture structured model-assisted process states where the method fits the research or evidence question. | Ordinary logs or other evidence are sufficient, or internal-state self-report would be treated as ground truth. | Cross-model feasibility research does not prove hidden-state fidelity or universal validity. | Conditional Method. Not the universal HEART evidence format. |
| Behavioral Oracle v2.0 | Bind selected evidence to provenance, attestation, integrity, and verification rules. | Cryptography adds no material assurance or the evidence source remains untrustworthy. | Assurance is bounded by source, collection, threat model, coverage, keys, and appraisal policy. | Conditional Method / research vehicle. Not a universal trust layer or automatic truth oracle. |
| Behavioral Evidence Review Toolkit | Produce a bounded preliminary evidence packet from already available artifacts. | Formal forensic, legal, clinical, or conformity conclusions are required. | Informative review workflow; validation and calibration remain method-specific. | Informative Method. No certification, legal chain of custody, or formal conformity. |
| Agentic Workflow Forensics / SCOPA | Reconstruct objectives, scope, state, verification, stop conditions, evidence, reliance, and recovery in agentic work. | A simpler operational record answers the question. | Open research and prospective validation program. | Research / Conditional Forensic Method. No universal legal or conformity claim. |
| ABTF / TRACE | Reconstruct bounded AI behavioral trajectories and evidence lineage. | The available evidence cannot support trajectory reconstruction or a simpler incident review is sufficient. | Research and method-development program with claim-specific validation needs. | Research / Conditional Forensic Method. No automatic causation or legal conclusion. |
Selection record
For every selected method, record:
- the requirement and verifiable condition it serves;
- why the method is necessary and fit for purpose;
- evidence inputs and assumptions;
- competence and independence needs;
- validation state and known limitations;
- failure and indeterminate conditions;
- what simpler or external alternatives were considered; and
- the maximum claim the output can support.
Forensics is one method family
Forensic methods matter when governance state, behavior, decision lineage, or evidence must be reconstructed. Routine verification may instead use ordinary records, deterministic checks, technical tests, external standards, or qualified professional assessment.
HEART is broader than forensics.