Verdicts and Pathway Adequacy: The Result-Free Audit Layer

The verdict layer is the formal mechanism by which Orthemology separates two questions that are often conflated: was the answer right? and was the process sound? These are independent axes. A system can reach the correct answer through a defective process (the stopped-clock validator), and a well-run process can produce an incorrect answer on a rare case (the justified miss). The verdict layer makes all four combinations representable — and auditable — on a single episode record.

Two Independent Axes

Result Axis

V1 (RESULT_CORRECT) and V2b-T (TOKEN_TRUTH_LINKED) assess whether the placed profile agrees with the ground truth O*(m; A(e)) and whether the correctness was achieved through the truth-relevant mechanism. Both are factive. V1 is the primary result verdict; V2b-T is a claim-wise annotation for stopped-clock and Gettier diagnoses.

Pathway Axis

The result-free pathway core CorePath = {V2a, V2b-P, V2c, V3a, V3b, V3c, V3d, V3e, V4a, V5, V6} assesses whether the process was sound, without presupposing any particular result. PathwayAdequate is defined over this core — never over the result axis.

PathwayAdequate(e) does not entail V1(e). A reliable process used correctly may produce a rare error — this is the AdequatePathError cell. Conversely, V1(e) does not entail PathwayAdequate(e) — a correct answer through a defective process is the stopped-clock cell. Neither entailment holds, and both directions are witnessed by mandatory deterministic fixtures.

The Full Verdict Table

Semantic IDs are authoritative in machine-readable records. Display aliases (V1, V2a, etc.) are prose forms only. The registry is normative: docs/verdict-registry.yaml (Decision 0004).

Semantic ID	Alias	Side	Factive	Meaning
`RESULT_CORRECT`	V1	result	yes	Placed profile agrees with `O*(m; A(e))` at the governed level
`EVIDENCE_SUPPORT`	V2a	pathway	no	Typed evidence, within declared scopes, meets the declared support standard per placed claim
`PROCEDURE_RELIABLE`	V2b-P	pathway	no	Configured-procedure truth-conduciveness over the predeclared `RelSpec` reference class (non-factive)
`TOKEN_TRUTH_LINKED`	V2b-T	result	yes	Factive, claim-wise: this claim was correct through the truth-relevant mechanism
`EVIDENCE_CURRENT`	V2c	pathway	no	Load-bearing evidence current for `(κ, v)` with admissible provenance
`GOV_CONFIG_ADEQUATE`	V3a	pathway	no	Governing configuration adequate for the case’s risk class
`GOV_POLICY_ADEQUATE`	V3b	pathway	no	Meta-policy well-formed for the configuration it runs under
`GOV_TOKEN_ADEQUATE`	V3c	pathway	no	Every applicable metaorthemma correctly instantiated, compatible, anchored, scoped, current, provenanced, and authorized
`EXECUTION_FAITHFUL`	V3d	pathway	no	The actor executed the procedure as written
`EX_ANTE_JUSTIFIED`	V3e	pathway	no	Decision-time reasonableness on exactly the evidence then available
`ROUTE_ADMISSIBLE`	V4a	pathway	no	Route safe and admissible under the constraints in force
`ROUTE_QUALITY`	V4b	advisory	no	Route near-optimality (safe ≠ best; advisory only, not part of pathway adequacy)
`CLOSURE_TRUTHFUL`	V5	pathway	no	Closure matches the burden ledger — no disposition collapse
`ROBUST_NEIGHBORHOOD`	V6	pathway	no	Robustness under the declared perturbation family

V4b (ROUTE_QUALITY) is advisory and is explicitly excluded from the pathway core. A safe but suboptimal route fails V4b only, not V4a, and does not make the pathway defective. Safe does not mean optimal; the distinction routes to a different remedy (improve the routing policy, not the placement standard).

Verdict Semantics in Depth

V1 — RESULT_CORRECT

correct(e): p̂(e) agrees with O*(m; A(e)) at the governed level. The weaker route-correct(e) holds when p̂(e) is route-sufficient and every placed claim is in O*(m; A(e)).V1 is claim-wise at bottom: V1_q(e) holds iff placed claim q is true of (m; A(e)). The episode-level verdict aggregates by conjunction over the placed claims at the governed level — no averaging or majority rule.V1 is never a conjunct of PathwayAdequate. Result correctness and pathway adequacy are independent axes by construction.

V2a — EVIDENCE_SUPPORT

The typed evidence H, within its declared scopes, meets the declared standard for each placed claim. “Within declared scopes” is load-bearing: a mis-scoped pass (pass ∧ scope ∩ claim = ∅) is no evidence for that claim. A green check whose scope does not intersect the release claim contributes nothing to V2a.Separated from V2c (EVIDENCE_CURRENT) because evidence can meet the strength standard while being the wrong vintage: stale evidence about the old version is not weak support for the current version’s claim — it is evidence about a different occurrence.

V2b-P — PROCEDURE_RELIABLE (non-factive)

The procedure family actually instantiated in this episode — under its governing configuration μ⃗, applicable metaorthemmata MetaTok(e), and execution mode — satisfies the predeclared reliability criterion over its declared reference class.Per-claim RelSpec_q carries: declared reference class; relevant case/risk stratum; reliability metric; threshold or tolerance; perturbation or comparison family; evaluation protocol; evidence used to establish reliability; version and validity conditions.The reference class and threshold must be fixed independently of this episode’s outcome — never selected after seeing the result to rescue or condemn it. One current correct result cannot by itself establish V2b-P.V2b-P does not entail V1: a reliable configured procedure may produce a rare error (the AdequatePathError cell).

V2b-T — TOKEN_TRUTH_LINKED (factive, result-side)

Factive and claim-wise: V2b-T_q(e) holds when this placed claim q was correct through the truth-relevant evidential mechanism rather than merely correct alongside it.By definition: V2b-T_q(e) → V1_q(e). This is the only declared definitional entailment in the verdict layer.V2b-T is excluded from the pathway core precisely because its factivity would re-import result correctness into PathwayAdequate. It is reportable for stopped-clock and Gettier diagnoses on the result side. Non-factive token-local procedural defects belong under V2a, V2b-P, V2c, V3a, V3b, V3c, V3d, or V6 — never under V2b-T alone.

V3c — GOV_TOKEN_ADEQUATE

V3c(e) ⟺ ∧_{μ̄ ∈ MetaTok(e)} TokenAdequate(μ̄, e)where TokenAdequate(μ̄, e) requires: MetaInst(μ̄, μ) ∧ Compatible(μ̄, A(e)) ∧ Anchored(μ̄, κ(e), v(e)) ∧ ScopeCorrect(μ̄, 𝒬(e)) ∧ Current(μ̄, t(e)) ∧ Provenanced(μ̄) ∧ AuthorizedBinding(μ̄).This verdict isolates a failure mode the earlier vector could not name: correct standard (V3a pass) + sound policy (V3b pass) + faithful execution (V3d pass) + defective case-specific binding — wrong reference plane, wrong-role tolerance, expired calibration, wrong fixture or success surface.Per-token statuses are preserved alongside the episode-level conjunction; auditors see which specific token failed and why.Zero-burden rule: MetaTok(e) = ∅ ⟹ V3c ∉ ReqPath(e) (status recorded not-applicable, with reason).

V6 — ROBUST_NEIGHBORHOOD

Robustness is always relative to a declared perturbation specification PerturbSpec(e):

varied fields: input version bumped, marker string removed or spoofed, evidence reordered or withheld, near-identical sibling substituted
invariant fields: policy π, governing configuration μ⃗, analysis A and version
size or measure: how the neighborhood is generated

robust(e) holds iff either (i) the declared neighborhood is finite and enumerated and the empirical proportion of V1 failures is at or below the declared tolerance, OR (ii) the specification declares a probability measure over the perturbation family and the failure probability is at or below tolerance.An undeclared “failure rate” over an unspecified infinite neighborhood is not a well-formed V6 claim.V6 is not V2b-P. V2b-P asks whether the configured procedure is truth-conducive over its declared reference class. V6 asks whether the episode is stable under a declared neighboring perturbation family. One test suite may supply evidence for both; the questions differ, and no entailment holds in either direction.

Verdict Status Values

Statuses are four-valued — not Boolean:

Status_i(e) ∈ { pass, fail, undetermined, not-applicable }

A missing assessment is undetermined — never silently counted as pass, never silently removed as not-applicable. Every verdict excluded from ReqPath(e) carries a recorded not-applicable reason. “Not tested” is never “not applicable.”

PathwayAdequate, PathwayDefective, PathwayUndetermined

The three pathway states are defined over ReqPath(e) — the governance-derived required verdict set. In the verdict-record.schema.json, these are recorded as the string enum values "adequate", "defective", and "undetermined" in the pathway_state field:

PathwayAdequate(e)      iff every V_i ∈ ReqPath(e) has status pass
                        (pathway_state: "adequate")

PathwayDefective(e)     iff some  V_i ∈ ReqPath(e) has status fail
                        (pathway_state: "defective")

PathwayUndetermined(e)  iff no required verdict fails and at least
                        one required verdict is undetermined
                        (pathway_state: "undetermined")

PathwayAdequate ↔ ¬PathwayDefective is not asserted — undetermined is a third state, not a tie-breaking default.

ReqPath(e): Governance-Derived Required Path

ReqPath(e) = CorePath ∩ RequiredBy(A(e), episode-shape(e),
               risk-class(e), claims(e), governance(e))

Applicability is governance-derived, never a discretionary post-hoc list — neither the executor nor the auditor sets ReqPath(e); it is derived from the declared analysis, episode shape, risk class, claims, and governance. An executor cannot exempt a failing or untested verdict by omitting it from their own record. Minimum derivation rules:

V3c is required exactly when MetaTok(e) ≠ ∅
V4a is required when the episode selects or authorizes a route
V5 is required when the episode makes a closure claim or dispositions residuals
V6 is required when A(e) or the risk class predeclares a robustness obligation

The Result × Pathway Matrix

The matrix applies only after the pathway status resolves as adequate or defective. PathwayUndetermined episodes remain outside the four cells until audited.

F1 — Nominal (correct + adequate)
F2 — Stopped Clock (correct + defective)
F3 — Justified Rare Miss (incorrect + adequate)
F4 — Defective Token, Lucky Result (correct + defective)

A current behavioral validator, correctly bound to the current artifact (V3c pass where applicable), under a sufficiently reliable configured procedure, returns the correct result.

V1: pass
All ReqPath(e) verdicts: pass
Pathway state: adequate

Nothing to fix. This is the base case.

A scan validator’s rule: pass iff the output log contains SCAN-CLEAN. The artifact is genuinely compliant today. The procedure fails its declared reliability and perturbation criteria.

V1: pass (the answer is right)
V2b-P: fail (the configured procedure is not truth-conducive over its reference class)
V6: fail (compliant-without-marker is rejected; non-compliant-with-marker is passed)
V2b-T: fail as a result-side diagnosis (correct, but not through the truth-relevant mechanism)
Pathway state: defective

Remedy targets the procedure or binding — not the verdict. The defect is statable on this episode record; a system tracking only result correctness has no object on which it is even expressible.

A properly configured, faithfully executed procedure meets its declared reference-class reliability threshold but errs on this specific case. Every available signal pointed to “infrastructure flake”; the true cause was a rare race condition in new code.

V1: fail (the placed claim is false)
V2b-P: pass (the procedure met its declared reference-class reliability threshold)
V3a, V3b, V3c, V3d, V3e: pass where applicable
V2b-T: fail for the false claim (a false claim’s truth-linkage fails by factivity — never recorded “unavailable” absent an explicit applicability rule)
Pathway state: adequate

AdequatePathError(e) := ¬V1(e) ∧ PathwayAdequate(e). Remedy: a new discriminating evidence source. No rule or executor blame.

Orthemic adequacy does not establish moral, legal, institutional, or theological blamelessness. “Blameless-at-the-time” is EX_ANTE_JUSTIFIED (V3e), a formal predicate about decision-time reasonableness indexed to a specific actor and information state — nothing more.

Sound standard (V3a pass), sound policy (V3b pass), faithful execution (V3d pass). The concrete metaorthemma μ̄ binds the wrong reference plane and wrong-role tolerance. The result happens to be correct because the wrong reference plane locally coincides with the right one today.

V1: pass
V3a: pass (correct standard)
V3b: pass (sound policy)
V3d: pass (faithful execution)
V3c: fail (defective case-specific binding)
Pathway state: defective

Remedy targets the binding — not the verdict. This is the “governing-side luck” cell. Without V3c as a separate verdict, this defect is invisible.

Claim-Relative Reasoning Paths

Per-claim reasoning adequacy is distinct from the episode-level pathway state:

ReqReason_q(e) = the projection of ReqPath(e) onto the verdicts that bear on claim q
reasoning_path_adequate_q = conjunction over ReqReason_q(e) only

A downstream routing or closure failure must not flip a soundly reached claim. The verdict record carries claim_reasoning_paths separately from pathway_state. Collapsing one into the other is a schema-level error. Verdict records include a full index block — analysis_id, analysis_version, actor, decision_time, information_state — because pathway adequacy is objective given a full index. Dropping any component makes “objective” an overstatement.

Declared Implication Table

The only declared definitional entailment is:

V2b-T_q(e) → V1_q(e)   (factive, claim-wise)

All other pairs are declared non-entailments:

Pair	Status
`PROCEDURE_RELIABLE → RESULT_CORRECT`	No — non-factive by construction
`EX_ANTE_JUSTIFIED → RESULT_CORRECT`	No — decision-time reasonableness is result-free
`PathwayAdequate → RESULT_CORRECT`	No — the core excludes V1 and V2b-T
`RESULT_CORRECT → PathwayAdequate`	No — stopped-clock and defective-binding cases
`ROBUST_NEIGHBORHOOD → PROCEDURE_RELIABLE`	No — distinct questions; shared evidence is possible
`PROCEDURE_RELIABLE → ROBUST_NEIGHBORHOOD`	No — same
All other pairs	No definitional implication; empirical correlation is never entailment

Introduction

Core Concepts

Worked Examples

Schemas & Validation

Experiments & Terminology

Verdicts and Pathway Adequacy: The Result-Free Audit Layer

Two Independent Axes

Result Axis

Pathway Axis

The Full Verdict Table

Verdict Semantics in Depth

Verdict Status Values

PathwayAdequate, PathwayDefective, PathwayUndetermined

ReqPath(e): Governance-Derived Required Path

The Result × Pathway Matrix

Claim-Relative Reasoning Paths

Declared Implication Table

Build docs developers (and LLMs) love

Introduction

Core Concepts

Worked Examples

Schemas & Validation

Experiments & Terminology

Documentation Index

​Two Independent Axes

Result Axis

Pathway Axis

​The Full Verdict Table

​Verdict Semantics in Depth

​Verdict Status Values

​PathwayAdequate, PathwayDefective, PathwayUndetermined

​ReqPath(e): Governance-Derived Required Path

​The Result × Pathway Matrix

​Claim-Relative Reasoning Paths

​Declared Implication Table

Build docs developers (and LLMs) love

Two Independent Axes

The Full Verdict Table

Verdict Semantics in Depth

Verdict Status Values

PathwayAdequate, PathwayDefective, PathwayUndetermined

ReqPath(e): Governance-Derived Required Path

The Result × Pathway Matrix

Claim-Relative Reasoning Paths

Declared Implication Table