COMPO · 10.04 · TOLERANCE / OPTICAL / SCORING

Explicit
alignment
fit.

Final local evaluation layer for Topic 10. Raw geometry stays upstream. 10.04 applies only explicit hard/soft targets, explicit visual tolerance and optional owner-qualified optical target shifts. Hard mismatch rejects before scoring. Required N/A remains N/A. The sole independent Topic10 numeric vote is alignment.system.localAlignmentFit; per-target fits are non-independent components of that composite.

ModuleCompo 10.04
Scopeexplicit target evaluation
Hardfilter before score
Opticalexplicit hooks only
Vote1 independent
NextTopic10 audit

No target, no score

00 / explicit targets

10.04 is the first Topic10 module allowed to score, and it scores only explicit local alignment targets.

HARD

reject, never numeric penalty

SOFT

explicit fit component

1 VOTE

localAlignmentFit composite

Signed offset or maximum spread

01 / target models

Targets bind to an exact upstream evidence ref.

ModelObserved evidenceTarget
SIGNED_OFFSET_TARGET10.01/10.02 measurementRefexplicit signed delta
MAX_SPREAD_TARGET10.03 sharedAxisRefexplicit maximum spread

Deadband + falloff

02 / soft fit

Visual tolerance is explicit and separate from numerical EPS.

SIGNED: errorRaw = |observedDelta - effectiveTargetDelta| if errorRaw <= deadbandRaw: fit = 1 else: fit = 1 - clamp( (errorRaw-deadbandRaw) / falloffRaw, 0, 1 ) SPREAD: excessRaw = max( 0, spreadRaw-maxSpreadRaw ) fit = 1 - clamp( excessRaw/falloffRaw, 0, 1 )

Hard mismatch is not score 0

03 / hard filter

HARD constraints are evaluated before local soft aggregation.

SIGNED HARD: errorRaw <= deadbandRaw → KEEP else → REJECT SPREAD HARD: spreadRaw <= maxSpreadRaw → KEEP else → REJECT REJECT: numeric component = NONE not 0

Explicit evidence only

04 / optical hooks

10.04 may consume an optical target adjustment but never invent one.

OpticalTargetHook {
  opticalHookRef,
  ownerRef,
  targetDeltaAdjustmentRaw,
  sourceUnit,
  confidence,
  provenance,
  mutationAllowed: false
}
opticalMode = NONE: effectiveTargetDelta = baseTargetDelta opticalMode = APPLY_EXPLICIT: effectiveTargetDelta = baseTargetDelta + targetDeltaAdjustmentRaw

Required vs optional is explicit

05 / profile

Optional N/A can leave the denominator only because the profile explicitly says optional.

AlignmentEvaluationProfile {
  profileRef,
  requiredTargetRefs[],
  optionalTargetRefs[],
  weightsByTargetRef,
  confidencePolicy
}
NO HIDDEN OPTIONALITY. A missing required target cannot quietly disappear from the score.

One independent Topic10 vote

06 / metric ownership

Target components are already part of the composite.

alignment.system.localAlignmentFit owner = Compo10.04 composite = true scoringGroup = alignment-local independentVoteAllowed = true alignment.target.componentFit.<targetRef> owner = Compo10.04 componentOf = alignment.system.localAlignmentFit scoringGroup = alignment-local independentVoteAllowed = false

INVALID > REJECT > N_A > KEEP

07 / precedence

Status is not inferred from a numeric score.

INVALID malformed/incompatible evidence REJECT explicit HARD target fails N_A required evidence unavailable or no soft components remain KEEP hard filter passed and evaluation context valid

Same context only

08 / comparison

Ranking is only meaningful inside one explicit profile and target set.

compare allowed only if: profileRef A == profileRef B AND targetSetFingerprint A == targetSetFingerprint B rankable: status = KEEP composite.status = OK tie: same rank lexical resultRef: display order only

Local conservative policy, not universal law

09 / confidence

10.04 records an explicit local policy rather than silently claiming a project-wide confidence theorem.

component confidence: MIN_RELEVANT_INPUTS_LOCAL composite confidence: MIN_INCLUDED_COMPONENTS_LOCAL provenance: IMPLEMENTATION_HEURISTIC scope: Compo10.04 only future replacement: requires reviewed change

Request, never mutate

10 / corrections

Scoring may identify mismatch but cannot move foreign-owned geometry.

AlignmentCorrectionRequest {
  correctionRequestRef,
  resultRef,
  targetRef,
  reasonCode,
  targetOwnerRef,
  evidenceRefs[],
  suggestedDeltaRaw?,
  mutationAllowed: false
}

Evaluation context becomes dataset evidence

11 / IDEAS + DATA

Profile, optical evidence and human disagreement remain replayable.

DATA_CAPTURE: profile + target set raw evidence refs hard decisions tolerance parameters optical hook lineage component fits composite score N_A exclusions comparison context repairs human reasons SEMANTIC: separate evidence channel PRIORITY: module order != priority

Disagreement is useful data

12 / human feedback

Human feedback can challenge target, tolerance, optical hook or ranking without auto-changing CANON.

ALIGNMENT_TARGET_WRONG TOLERANCE_TOO_TIGHT TOLERANCE_TOO_LOOSE OPTICAL_HOOK_WRONG OPTICAL_HOOK_MISSING INTENTIONAL_MISALIGNMENT SEMANTIC_CONFLICT WRONG_RANKING_FOR_CONTEXT OTHER

Same geometry, different explicit targets

13 / Visual Lab

Zero offset is only one possible target.

A · ZERO TARGET explicit target delta = 0 B · OPTICAL TARGET explicit optical hook shifts target C · INTENTIONAL OFFSET non-zero target can be explicit numerical EPS != design tolerance HARD mismatch → REJECT · never score 0 required N/A remains N/A · optional N/A may explicitly drop denominator one independent Topic10 vote: alignment.system.localAlignmentFit same profile + same targetSetFingerprint required for comparison local rank != beauty/global quality/final acceptance

10.04 scoring regression

14 / deterministic tests

Tests cover explicit targets, hard/soft separation, optical hooks, N/A semantics, component ownership and comparison guards.

Tests not run yet.

DATA FOR LAYOUT ENGINE

15 / machine contract

Machine-readable 10.04 contract.

{
  "module": "Compo 10.04",
  "name": "tolerance_optical_hooks_scoring_tests",
  "version": "1.0-canonical",
  "canonical_main_title": "Tolerance, optical hooks, scoring/tests",
  "topic": "10_БАЗОВЫЕ_ЛИНИИ_И_ВЫРАВНИВАНИЕ",
  "scope": "EXPLICIT_TOPIC10_TARGET_TOLERANCE_OPTICAL_HOOK_HARD_FILTER_LOCAL_SCORING_REPAIR_AND_COMPARISON",
  "review_basis": {
    "previous_modules": [
      "Compo10.00 Alignment taxonomy",
      "Compo10.01 pairwise geometric relations",
      "Compo10.02 baseline/text alignment",
      "Compo10.03 multi-object shared axes"
    ],
    "cross_cutting_instruction": "COMPO project instruction v2.0"
  },
  "consumes": [
    "Compo10.01/10.02 pairwise measurement records",
    "Compo10.03 shared-axis evaluation records",
    "AlignmentEvaluationProfile",
    "AlignmentTargetSpec[]",
    "optional owner-qualified OpticalTargetHook[]",
    "optional opaque semanticConstraintRefs[]"
  ],
  "provides": [
    "AlignmentTargetEvaluation[]",
    "AlignmentHardFilterResult",
    "alignment.system.localAlignmentFit",
    "AlignmentEvaluationResult",
    "AlignmentCorrectionRequest",
    "Topic10 comparison/ranking"
  ],
  "owns_metrics": [
    "alignment.system.localAlignmentFit",
    "alignment.target.componentFit.<targetRef>"
  ],
  "descriptors_only": [
    "alignment.target.errorRaw",
    "alignment.target.excessRaw",
    "alignment.target.effectiveTargetDeltaRaw",
    "alignment.target.hardDecision",
    "alignment.evaluation.status",
    "alignment.evaluation.targetSetFingerprint"
  ],
  "target_models": {
    "SIGNED_OFFSET_TARGET": {
      "subject": "exact pairwise measurementRef",
      "fields": [
        "baseTargetDeltaRaw",
        "deadbandRaw",
        "falloffRaw for SOFT",
        "opticalMode",
        "opticalHookRef?"
      ],
      "hard_pass": "abs(observedSignedDelta-effectiveTargetDelta)<=deadbandRaw"
    },
    "MAX_SPREAD_TARGET": {
      "subject": "exact sharedAxisRef",
      "fields": [
        "maxSpreadRaw",
        "falloffRaw for SOFT"
      ],
      "hard_pass": "observedSpreadRaw<=maxSpreadRaw"
    }
  },
  "optical_hook_schema": {
    "opticalHookRef": "stable ref",
    "ownerRef": "explicit evidence owner",
    "targetDeltaAdjustmentRaw": "finite signed value",
    "sourceUnit": "must match measurement sourceUnit",
    "confidence": "0..1",
    "provenance": "DERIVED|EMPIRICAL|INITIAL_HEURISTIC|STYLE_PRESET",
    "mutationAllowed": false
  },
  "profile_schema": {
    "profileRef": "stable explicit profile ref",
    "requiredTargetRefs": "set-like unique refs",
    "optionalTargetRefs": "set-like unique refs",
    "weightsByTargetRef": "positive weights for SOFT targets only",
    "confidencePolicy": "MIN_INCLUDED_COMPONENTS_LOCAL",
    "semanticConstraintRefs": "opaque optional refs"
  },
  "metric_contract": {
    "composite": {
      "key": "alignment.system.localAlignmentFit",
      "owner": "Compo10.04",
      "composite": true,
      "componentOf": null,
      "scoringGroup": "alignment-local",
      "independentVoteAllowed": true
    },
    "component": {
      "key": "alignment.target.componentFit.<targetRef>",
      "owner": "Compo10.04",
      "composite": false,
      "componentOf": "alignment.system.localAlignmentFit",
      "scoringGroup": "alignment-local",
      "independentVoteAllowed": false
    }
  },
  "status_precedence": [
    "INVALID",
    "REJECT",
    "N_A",
    "KEEP"
  ],
  "comparison_policy": {
    "requires_same_profileRef": true,
    "requires_same_targetSetFingerprint": true,
    "rankable": "KEEP + composite MetricResult status=OK only",
    "ties": "same numeric rank",
    "lexical_order": "presentation only",
    "meaning": "EXPLICIT_TOPIC10_TARGET_FIT_ONLY"
  },
  "confidence_policy": {
    "component": "MIN_RELEVANT_INPUTS_LOCAL",
    "composite": "MIN_INCLUDED_COMPONENTS_LOCAL",
    "provenance": "IMPLEMENTATION_HEURISTIC",
    "scope": "Compo10.04 local evaluation only; not a universal COMPO confidence law"
  },
  "metric_policy": {
    "no_profile": "composite N_A",
    "hard_mismatch": "REJECT; no numeric target vote",
    "required_N_A": "composite N_A unless INVALID/REJECT",
    "optional_N_A": "drop denominator only because optional is explicit",
    "no_soft_components": "composite N_A",
    "beautyScore": "FORBIDDEN"
  },
  "hard_constraints": [
    "10.04 owns explicit Topic10 target/tolerance interpretation, owner-qualified optical target hooks, hard filtering, local alignment component fits, the single independent Topic10 composite score, correction requests and comparison tests.",
    "10.04 consumes raw pairwise measurements from 10.01, baseline measurements from 10.02 and shared-axis evaluations from 10.03; it does not recompute their geometry.",
    "10.04 does not own source object/text/grid geometry.",
    "10.04 does not own baseline extraction.",
    "10.04 does not own shared-axis membership generation.",
    "No AlignmentEvaluationProfile means local alignment composite is N_A.",
    "Targets are explicit; 10.04 has no hidden assumption that zero offset or shared-axis coincidence is always preferred.",
    "Supported target models are SIGNED_OFFSET_TARGET and MAX_SPREAD_TARGET.",
    "SIGNED_OFFSET_TARGET requires an exact measurementRef and finite baseTargetDeltaRaw.",
    "MAX_SPREAD_TARGET requires an exact sharedAxisRef and finite maxSpreadRaw>=0.",
    "Each target declares HARD or SOFT mode.",
    "HARD mismatch produces REJECT and no numeric component vote for that target.",
    "SOFT target requires deadbandRaw>=0 and falloffRaw>0.",
    "For SIGNED_OFFSET_TARGET, errorRaw = abs(observedSignedDeltaRaw-effectiveTargetDeltaRaw).",
    "For MAX_SPREAD_TARGET, excessRaw = max(0, observedSpreadRaw-maxSpreadRaw).",
    "SOFT signed fit: 1 inside deadband; outside, 1-clamp((errorRaw-deadbandRaw)/falloffRaw,0,1).",
    "SOFT spread fit: 1 when excessRaw=0; otherwise 1-clamp(excessRaw/falloffRaw,0,1).",
    "HARD signed pass requires errorRaw<=deadbandRaw.",
    "HARD spread pass requires spreadRaw<=maxSpreadRaw.",
    "Numerical EPS from 10.01/10.02/10.03 is not design tolerance and is never reused as deadbandRaw unless an explicit target literally supplies that same numeric value.",
    "Optical shifts are never inferred by 10.04.",
    "An optical hook must be explicit owner-qualified evidence.",
    "Optical hook application requires opticalMode=APPLY_EXPLICIT and exact opticalHookRef.",
    "effectiveTargetDeltaRaw = baseTargetDeltaRaw + targetDeltaAdjustmentRaw when explicit optical hook is applied.",
    "Absent optical hook means adjustment 0 only because opticalMode=NONE is explicit, not because a hidden optical model was run.",
    "Optical hook provenance must be a canonical provenance class.",
    "Optical hook sourceUnit must match the consumed signed-offset measurement unit.",
    "Optical hook does not mutate geometry; it only changes the explicit evaluation target.",
    "Optical hook confidence remains evidence metadata and cannot silently upgrade source measurement confidence.",
    "Target applicability is exact by measurementRef/sharedAxisRef; a target cannot evaluate another subject by kind similarity.",
    "Target input order is irrelevant; targetSetFingerprint canonicalizes targetRef order.",
    "Profile requiredTargetRefs and optionalTargetRefs are set-like, unique, disjoint and canonicalized.",
    "Profile must explicitly include every evaluated targetRef in required or optional sets.",
    "SOFT target weights are explicit positive finite values.",
    "HARD targets do not receive numeric voting weight.",
    "Required target N_A makes the composite N_A unless a higher-precedence INVALID or REJECT exists.",
    "Optional target N_A may be excluded from the soft-score denominator only because it is explicitly optional.",
    "INVALID > REJECT > N_A > KEEP is the status precedence.",
    "Rejected, invalid and N_A evaluation bundles are not converted to numeric zero.",
    "If no applicable SOFT components remain after explicit optional N_A exclusion, local composite score is N_A.",
    "The only independent numeric Topic10 vote is alignment.system.localAlignmentFit.",
    "alignment.system.localAlignmentFit is a composite owned by Compo10.04 with scoringGroup alignment-local.",
    "Per-target component metrics use alignment.target.componentFit.<targetRef> and have componentOf=alignment.system.localAlignmentFit and independentVoteAllowed=false.",
    "Raw offsets, spreads, numerical coincidence flags, relation classes and shared-axis statuses do not vote independently.",
    "No universal beautyScore is introduced.",
    "More alignment is not automatically better.",
    "More zero offsets are not automatically better.",
    "More grid adherence is not automatically better.",
    "Intentional misalignment can be represented by an explicit non-zero SIGNED_OFFSET_TARGET.",
    "Comparison/ranking requires same profileRef and same targetSetFingerprint.",
    "Only KEEP results with score status OK participate in numeric ranking.",
    "Ties share rank; lexical resultRef may be used only for deterministic display order.",
    "Ranking means EXPLICIT_TOPIC10_TARGET_FIT_ONLY, not beauty, global quality or final acceptance.",
    "Correction requests are mutationAllowed=false and owner-qualified.",
    "10.04 never moves objects/text/grid itself.",
    "SemanticConstraintRefs remain separate evidence; 10.04 cannot invent semantic priority.",
    "Module order is not rule priority.",
    "IDEA != CANON.",
    "EXPERIMENT != CANON.",
    "All profile/target/optical/scoring/filter/repair evidence must be serializable.",
    "N/A != 0."
  ],
  "hard_rules": [
    "10.04 owns explicit Topic10 target/tolerance interpretation, owner-qualified optical target hooks, hard filtering, local alignment component fits, the single independent Topic10 composite score, correction requests and comparison tests.",
    "10.04 consumes raw pairwise measurements from 10.01, baseline measurements from 10.02 and shared-axis evaluations from 10.03; it does not recompute their geometry.",
    "10.04 does not own source object/text/grid geometry.",
    "10.04 does not own baseline extraction.",
    "10.04 does not own shared-axis membership generation.",
    "No AlignmentEvaluationProfile means local alignment composite is N_A.",
    "Targets are explicit; 10.04 has no hidden assumption that zero offset or shared-axis coincidence is always preferred.",
    "Supported target models are SIGNED_OFFSET_TARGET and MAX_SPREAD_TARGET.",
    "SIGNED_OFFSET_TARGET requires an exact measurementRef and finite baseTargetDeltaRaw.",
    "MAX_SPREAD_TARGET requires an exact sharedAxisRef and finite maxSpreadRaw>=0.",
    "Each target declares HARD or SOFT mode.",
    "HARD mismatch produces REJECT and no numeric component vote for that target.",
    "SOFT target requires deadbandRaw>=0 and falloffRaw>0.",
    "For SIGNED_OFFSET_TARGET, errorRaw = abs(observedSignedDeltaRaw-effectiveTargetDeltaRaw).",
    "For MAX_SPREAD_TARGET, excessRaw = max(0, observedSpreadRaw-maxSpreadRaw).",
    "SOFT signed fit: 1 inside deadband; outside, 1-clamp((errorRaw-deadbandRaw)/falloffRaw,0,1).",
    "SOFT spread fit: 1 when excessRaw=0; otherwise 1-clamp(excessRaw/falloffRaw,0,1).",
    "HARD signed pass requires errorRaw<=deadbandRaw.",
    "HARD spread pass requires spreadRaw<=maxSpreadRaw.",
    "Numerical EPS from 10.01/10.02/10.03 is not design tolerance and is never reused as deadbandRaw unless an explicit target literally supplies that same numeric value.",
    "Optical shifts are never inferred by 10.04.",
    "An optical hook must be explicit owner-qualified evidence.",
    "Optical hook application requires opticalMode=APPLY_EXPLICIT and exact opticalHookRef.",
    "effectiveTargetDeltaRaw = baseTargetDeltaRaw + targetDeltaAdjustmentRaw when explicit optical hook is applied.",
    "Absent optical hook means adjustment 0 only because opticalMode=NONE is explicit, not because a hidden optical model was run.",
    "Optical hook provenance must be a canonical provenance class.",
    "Optical hook sourceUnit must match the consumed signed-offset measurement unit.",
    "Optical hook does not mutate geometry; it only changes the explicit evaluation target.",
    "Optical hook confidence remains evidence metadata and cannot silently upgrade source measurement confidence.",
    "Target applicability is exact by measurementRef/sharedAxisRef; a target cannot evaluate another subject by kind similarity.",
    "Target input order is irrelevant; targetSetFingerprint canonicalizes targetRef order.",
    "Profile requiredTargetRefs and optionalTargetRefs are set-like, unique, disjoint and canonicalized.",
    "Profile must explicitly include every evaluated targetRef in required or optional sets.",
    "SOFT target weights are explicit positive finite values.",
    "HARD targets do not receive numeric voting weight.",
    "Required target N_A makes the composite N_A unless a higher-precedence INVALID or REJECT exists.",
    "Optional target N_A may be excluded from the soft-score denominator only because it is explicitly optional.",
    "INVALID > REJECT > N_A > KEEP is the status precedence.",
    "Rejected, invalid and N_A evaluation bundles are not converted to numeric zero.",
    "If no applicable SOFT components remain after explicit optional N_A exclusion, local composite score is N_A.",
    "The only independent numeric Topic10 vote is alignment.system.localAlignmentFit.",
    "alignment.system.localAlignmentFit is a composite owned by Compo10.04 with scoringGroup alignment-local.",
    "Per-target component metrics use alignment.target.componentFit.<targetRef> and have componentOf=alignment.system.localAlignmentFit and independentVoteAllowed=false.",
    "Raw offsets, spreads, numerical coincidence flags, relation classes and shared-axis statuses do not vote independently.",
    "No universal beautyScore is introduced.",
    "More alignment is not automatically better.",
    "More zero offsets are not automatically better.",
    "More grid adherence is not automatically better.",
    "Intentional misalignment can be represented by an explicit non-zero SIGNED_OFFSET_TARGET.",
    "Comparison/ranking requires same profileRef and same targetSetFingerprint.",
    "Only KEEP results with score status OK participate in numeric ranking.",
    "Ties share rank; lexical resultRef may be used only for deterministic display order.",
    "Ranking means EXPLICIT_TOPIC10_TARGET_FIT_ONLY, not beauty, global quality or final acceptance.",
    "Correction requests are mutationAllowed=false and owner-qualified.",
    "10.04 never moves objects/text/grid itself.",
    "SemanticConstraintRefs remain separate evidence; 10.04 cannot invent semantic priority.",
    "Module order is not rule priority.",
    "IDEA != CANON.",
    "EXPERIMENT != CANON.",
    "All profile/target/optical/scoring/filter/repair evidence must be serializable.",
    "N/A != 0."
  ],
  "soft_constraints": [],
  "preferences": [],
  "overrides_allowed_by": [
    "a different explicit profile/target set creates a different evaluation context",
    "new upstream measurement/shared-axis evidence creates a new result lineage",
    "new explicit optical hook creates new target evaluation lineage",
    "future context/global owners may use the local metric but do not change 10.04 ownership"
  ],
  "initial_heuristics": {
    "confidence_policy": {
      "value": "MIN_RELEVANT_INPUTS_LOCAL / MIN_INCLUDED_COMPONENTS_LOCAL",
      "provenance": "IMPLEMENTATION_HEURISTIC",
      "warning": "local conservative policy only; not universal confidence propagation"
    }
  },
  "style_preset_values": [],
  "candidate_generation": {
    "none": "10.04 evaluates explicit measurements/shared-axis evidence; it does not generate layout candidates"
  },
  "correction_policy": {
    "foreign_mutation": "FORBIDDEN",
    "signed_offset_mismatch": "may request owner/context to shift/reselect anchor, mutationAllowed=false",
    "spread_mismatch": "may request membership/geometry review, mutationAllowed=false",
    "optical_missing": "N_A/INVALID according explicit target requirement, never infer",
    "semantic_conflict": "defer; do not invent priority"
  },
  "deferred_inputs": [
    "future semantic/context priority resolution",
    "global composition aggregation outside Topic10"
  ],
  "ideas_refs": [
    "IDEA-01",
    "IDEA-02",
    "IDEA-03",
    "IDEA-04",
    "IDEA-05",
    "IDEA-06",
    "IDEA-07"
  ],
  "data_capture": [
    "profileRef",
    "targetSetFingerprint",
    "target refs/modes/models",
    "raw measurement/sharedAxis refs",
    "base/effective targets",
    "optical hook refs/provenance",
    "deadband/falloff",
    "hard decisions",
    "component fits",
    "composite metric",
    "required/optional N_A",
    "repair refs",
    "comparison context",
    "human reasons"
  ],
  "deferred_semantic_inputs": [
    "semanticConstraintRefs[]",
    "intentRef?",
    "semanticSensitiveZoneRef?",
    "readingOrderRef?"
  ],
  "human_feedback_hooks": [
    "ALIGNMENT_TARGET_WRONG",
    "TOLERANCE_TOO_TIGHT",
    "TOLERANCE_TOO_LOOSE",
    "OPTICAL_HOOK_WRONG",
    "OPTICAL_HOOK_MISSING",
    "INTENTIONAL_MISALIGNMENT",
    "SEMANTIC_CONFLICT",
    "WRONG_RANKING_FOR_CONTEXT",
    "OTHER"
  ],
  "conflict_priority_assumptions": [
    "module order is not priority",
    "hard target constraints precede local soft score only within this explicit Topic10 evaluation pipeline",
    "semantic constraints are not overridden by local alignment score",
    "optical hook does not override hard semantic/context constraints",
    "ranking is local explicit-target fit only"
  ],
  "lab_hooks": [
    "LAB: human calibration of deadband/falloff by relation type",
    "LAB: optical target shift experiments",
    "LAB: compare explicit non-zero intentional alignment targets",
    "LAB: ranking disagreement capture",
    "LAB: confidence-policy diagnostics",
    "LAB: downstream text-to-graphic scoring experiment as EXPERIMENT only"
  ],
  "recompute_when": [
    "upstream measurement/sharedAxis evidence changes",
    "profile changes",
    "target changes",
    "weight changes",
    "optical hook changes",
    "required/optional membership changes"
  ],
  "tests": [
    "target validation",
    "profile validation",
    "optical hook validation",
    "hard vs soft separation",
    "signed target fit",
    "spread target fit",
    "N_A precedence",
    "optional denominator",
    "one independent composite vote",
    "componentOf/no double counting",
    "target set fingerprint order invariance",
    "comparison guards",
    "tie ranks",
    "repair mutation false",
    "semantic/data/human/LAB hooks"
  ],
  "finality": "TOPIC10_EXPLICIT_LOCAL_ALIGNMENT_TARGET_EVALUATION_AND_SCORE_ONLY"
}
ARCHITECTURE CHECKPOINT · 10.04 OWNS EXPLICIT TOPIC10 TARGET/TOLERANCE INTERPRETATION, OPTICAL TARGET HOOKS, HARD FILTER, LOCAL COMPONENT FITS, ONE INDEPENDENT COMPOSITE alignment.system.localAlignmentFit, CORRECTION REQUESTS AND SAME-CONTEXT COMPARISON. RAW GEOMETRY IS CONSUMED, NOT RECOMPUTED. NO PROFILE → N/A. HARD MISMATCH → REJECT, NEVER SCORE 0. REQUIRED N/A REMAINS N/A. OPTIONAL N/A MAY DROP DENOMINATOR ONLY EXPLICITLY. NUMERICAL EPS != DESIGN TOLERANCE. OPTICAL SHIFT MUST BE OWNER-QUALIFIED EXPLICIT EVIDENCE. NON-ZERO TARGETS ARE ALLOWED. COMPONENTS ARE componentOf THE COMPOSITE AND independentVoteAllowed=false. COMPARISON REQUIRES SAME profileRef + targetSetFingerprint. RANKING = EXPLICIT_TOPIC10_TARGET_FIT_ONLY. SEMANTIC EVIDENCE REMAINS SEPARATE. MODULE ORDER != PRIORITY. NO FOREIGN MUTATION. N/A != 0. NO BEAUTYSCORE.
VISITS / ...