# Answerworthy Standard 0.1 — Conformance

**Applies to:** `0.1.0-draft.1`  
**Canonical URI:** https://answerworthy.md/standard/0.1/conformance.md  
**Licence:** Apache-2.0 for test logic and fixtures

This document turns the normative requirements into a testable conformance contract.

## 1. Result states

A conformance result has one of four states:

- `pass` — every REQUIRED test for the claimed profile passed;
- `fail` — one or more REQUIRED tests failed;
- `incomplete` — the suite could not finish or required artifacts were unavailable;
- `not_claimed` — the profile was not claimed.

`incomplete` MUST NOT be presented as `pass` or `fail`.

## 2. Manifest Profile

### M-001 — Canonical path

The manifest is available at `/answerworthy.md` or a declared canonical equivalent permitted by the profile.

### M-002 — Front matter

YAML front matter parses and contains:

- `answerworthy`;
- `entity`;
- `canonical`;
- `last_verified`;
- `kind`.

### M-003 — Required sections

`## Identity` and at least one question under `## Answers` are present and parseable.

### M-004 — Source discipline

Every material fact claimed as sourced contains a basis, source class, resolvable source and verification date.

An owned source does not become independent corroboration.

### M-005 — Consumer safety

The manifest does not claim authority to override a consumer's policies, system instructions, evidence hierarchy or independently observed contradictions.

### M-006 — Corroboration boundary

M3 is claimed only when an independent re-fetch or equivalent evidence process records support, contradictions and unavailable states.

### M-007 — Freshness

M4 is claimed only when `last_verified` remains within the declared freshness policy and a dated maintenance process exists.

## 3. Core profile

### C-001 — Schema identity

The bundle declares:

- `standard.name = "Answerworthy Standard"`;
- the exact version;
- the canonical URI;
- the matching versioned schema URI.

### C-002 — Structural validation

The canonical bundle validates against the published JSON Schema Draft 2020-12 schema.

### C-003 — Unique identifiers

Evidence, finding, action, projection and provider-state identifiers are unique within their collections.

### C-004 — Reference resolution

Every internal reference resolves:

- finding `evidence_refs`;
- action `source_finding_refs`;
- action `dependencies`;
- score `finding_refs`, when present;
- projection `source_bundle_id`.

### C-005 — Epistemic integrity

Every finding and evidence item uses an allowed epistemic state.

A finding labelled `observed` cannot rely exclusively on `inferred`, `founder_supplied`, `provider_produced`, `manual_adjudication` or `unevaluated` evidence.

### C-006 — Evidence boundary

Every evaluated `pass`, `warn` or `fail` finding has at least one evidence reference.

A finding without evaluated evidence uses `manual`, `not_applicable` or `unavailable` and states why.

### C-007 — Provider truthfulness

Every configured or relevant provider is represented by a provider state.

A failed, blocked, not-configured or unavailable provider state does not produce a zero-valued performance conclusion unless independent evaluated evidence establishes that value.

### C-008 — Score sufficiency

When scores are present:

- scale, method and evidence coverage are declared;
- applicable and evaluated counts are coherent;
- an insufficient evidence state has a null value;
- caps and renormalisation are declared.

### C-009 — Limitations

The bundle contains at least one run-level limitation and every material provider or methodology limitation.

### C-010 — Projection declaration

Every shipped projection appears in `projections` with format, path or URI, source bundle, omissions and limitations.

## 4. Agentic profile

### A-001 — Action source

Every remediation or maintenance action references at least one source finding.

An evidence-collection or verification action without a source finding explains the missing-evidence condition.

### A-002 — Permission mode

Every action declares exactly one:

- `advisory`;
- `human_required`;
- `agent_ready`;
- `external_authorisation_required`.

### A-003 — Completion contract

Every action contains:

- definition of done;
- at least one verification instruction;
- expected outcome without a guaranteed external result.

### A-004 — Dependency graph

Every dependency resolves to another action in the bundle.

The dependency graph is acyclic.

### A-005 — Agent-ready execution boundary

Every `agent_ready` action contains:

- preconditions;
- bounded steps;
- verification;
- rollback or irreversibility note.

The action does not embed credentials or claim authority beyond the bundle.

### A-006 — Evidence fabrication guardrail

An action that creates or strengthens a public claim, statistic, testimonial, citation, credential, review or independent source includes an evidence requirement or guardrail prohibiting fabrication.

### A-007 — Agent projection

`answerworthy.plan.md` is declared and contains the bundle ID, profile claims, provider states, permission modes, completion tests and verification tests.

## 5. Observed Web profile

### W-001 — Dated observation

At least one successful evidence item uses `browser_render`, `search_observation` or `external_public_source`.

### W-002 — Observation scope

Search observations declare engine or provider, query, timestamp, language or locale, and geography or a statement that geography was not controlled.

### W-003 — Parser state

Blocked, failed and layout-changed observations remain explicit and do not become absence-of-target findings.

### W-004 — Frozen provenance

A retained URL, artifact reference, digest or bounded parser record permits later review of every material observation.

## 6. Answer Probe profile

### P-001 — Prompt boundary

Every provider observation references a prompt with intent, language and market.

### P-002 — Provider identity

Every observation declares provider, model, timestamp and replicate number.

### P-003 — Replicate accounting

Requested and completed replicate counts are declared. Majority outcomes are null when the completed group is below its threshold.

### P-004 — Outcome separation

Mention, citation, recommendation and follow-through are independent fields or metrics.

### P-005 — Raw-output caution

Retained raw output is labelled provider-produced and unverified.

## 7. Outside Buyer profile

### B-001 — Respondent boundary

The run declares intended panel size, valid response count and validity threshold.

### B-002 — Suppression

Rates, consensus and synthesized conclusions are withheld when the validity threshold is not met.

### B-003 — Outcome separation

Comprehension, trust, relevance, recommendation and likely next action remain separate.

### B-004 — Priority provenance

Panel evidence may corroborate an existing priority. A panel-only allegation cannot become a factual technical finding without independent evidence.

## 8. Recheck profile

### R-001 — Baseline identity

The bundle identifies the prior bundle and date.

### R-002 — Comparable boundary

The recheck declares material changes in scope, market, language, prompt set, provider set and methodology.

### R-003 — Diff states

The recheck separates:

- implemented;
- unresolved;
- regressed;
- new;
- unverifiable.

### R-004 — Historical preservation

The previous bundle is not silently mutated or replaced.

## 9. Integrity profile

### I-001 — Canonical digest

`answerworthy.bundle.json` has a SHA-256 digest in the integrity manifest.

### I-002 — Projection digests

Every shipped projection has a SHA-256 digest.

### I-003 — Byte match

Recomputed digests match the manifest.

### I-004 — Claim boundary

The conformance statement says that checksums establish byte integrity only. They do not establish factual truth, authorship, legal validity or external outcomes.

## 10. Projection parity

A projection parity test fails when a projection:

- introduces a finding or action absent from the bundle;
- strengthens certainty;
- publishes a score withheld in the bundle;
- reports provider evidence that was unavailable;
- omits a material permission or safety guardrail from an action shown as executable.

## 11. Conformance record

`answerworthy.conformance.json` MUST include:

- standard version and conformance schema URI;
- bundle ID and digest when available;
- implementation name and version;
- test timestamp;
- claimed profiles;
- result for every claimed profile;
- individual test results;
- limitations;
- certification state.

The certification states are:

- `self_declared`;
- `answerworthy_certified`;
- `not_certified`.

A third-party implementation MUST use `self_declared` or `not_certified` unless authorised under the certification programme.

## 12. Minimum useful conformance

A bundle claiming only `core` is a portable evidence record.

A bundle claiming `core` and `agentic` is a supervised implementation handoff.

Additional profiles describe evidence actually obtained in the run. They do not describe latent code capability.