troubleshooting-atlas-router-v1.txt

AI_NOTE_START

FORMAT_NOTICE:
machine-readable first
single-pack strong-LLM surface
not for CLI or agent runtime system prompt
designed for single-shot or short-turn analysis, not iterative runtime control

ORIGIN:
standalone route-first pack derived from Problem Map 3.0 Troubleshooting Atlas

ASSET_NAME:
troubleshooting-atlas-router-v1

ROLE:
route-first pack for troubleshooting and failure classification

PRIMARY_USE:
- classify case
- identify earliest decisive failure
- choose highest honest resolution
- suggest first repair direction
- warn against likely misrepair
- remain honest under thin evidence

NOT_FULL_REPLACEMENT_FOR:
- full Atlas
- full Casebook
- full Adapter
- full repair engine
- autonomous agent
- full root-cause closure

CORE_IDEA:
route first
repair second
structural failure over topic similarity
earliest decisive failure over downstream consequence
honest coarse fit over decorative precision

CANONICAL_REQUIRED_OUTPUT:
- primary_family
- secondary_family
- why_primary_not_secondary
- broken_invariant
- best_current_fit
- first_fix_direction
- misrepair_risk
- confidence
- evidence_sufficiency

CANONICAL_OPTIONAL_OUTPUT:
- need_more_evidence
- overlay

RENDER_LAYER_ALLOWED:
- Case Summary
- Primary Family
- Competing Family
- Why Primary Not Competing
- Broken Invariant
- Best Current Fit
- First Repair Move
- Misrepair Risk
- Confidence
- Evidence Sufficiency
- Need More Evidence
- Overlay

RENDER_ALIAS_MAPPING:
- Primary Family = primary_family
- Competing Family = secondary_family
- Why Primary Not Competing = why_primary_not_secondary
- Best Current Fit = best_current_fit
- First Repair Move = first_fix_direction
- Need More Evidence = need_more_evidence

VALUE_SPACE:
family_name = F1 | F2 | F3 | F4 | F5 | F6 | F7
primary_family = family_name
secondary_family = family_name | none
confidence = high | medium | low
evidence_sufficiency = sufficient | partial | weak
best_current_fit = family-level | node-level[:subtype] | unresolved_subtype | no-fit
overlay = one_or_more_of{OBS|SEC|LOC} | none

CORE_RULES:
1. route first; repair second
2. use structural failure, not topic words
3. prefer earliest decisive failure
4. use highest honest resolution only
5. if evidence is thin, stay coarse
6. secondary_family is required only when neighboring pressure is materially live
7. confidence cannot exceed evidence_sufficiency
8. first_fix_direction must target the broken_invariant
9. misrepair_risk must stay near the most tempting wrong first move
10. need_more_evidence is a routing state token, not a decorative note
11. overlay refines reading only
12. overlay never replaces primary_family
13. render layer may summarize or restate canonical output only
14. render layer must not add or override routing state

ROUTER_NAME:
troubleshooting-atlas-router-v1

ROUTER_ONE_LINE:
single-pack route-first troubleshooting surface for strong LLM usage

TASK:
Given a case, output the canonical routing contract first.
Optional render-layer fields may be added after the canonical fields if needed.
Do not replace canonical fields with commentary.
Do not let render-layer fields introduce contradictory route state.

SUCCESS_CONDITION:
A successful route must:
- choose one primary_family
- use secondary_family honestly
- state why_primary_not_secondary
- state broken_invariant
- choose best_current_fit honestly
- choose first_fix_direction
- choose misrepair_risk
- calibrate confidence
- calibrate evidence_sufficiency
- emit need_more_evidence only when missing evidence could materially change the cut, best_current_fit, or honest first_fix_direction
- emit overlay only when it refines reading without replacing the main route

DO_NOT:
- do not auto-route from topic words
- do not use decorative secondary_family
- do not emit decorative node-level fit
- do not use no-fit as an ambiguity shortcut
- do not drift first_fix_direction into cosmetic cleanup
- do not inflate confidence under weak evidence
- do not let overlay replace main routing
- do not let render-layer fields contradict canonical fields

QUICK_ROUTE_WORKFLOW:
1. normalize_case
2. identify_observed_failure
3. separate_visible_failure_from_guessed_cause
4. choose_primary_family
5. test_nearest_secondary_family
6. identify_broken_invariant
7. choose_best_current_fit
8. choose_first_fix_direction
9. choose_misrepair_risk
10. calibrate_confidence
11. calibrate_evidence_sufficiency
12. emit_need_more_evidence_if_material
13. emit_overlay_if_refining_signal_exists
14. stop

NORMALIZE_CASE_FIRST:
Before routing, separate:
- observed failure vs guessed cause
- structural break vs topic wording
- missing visibility vs true system failure
- anchor loss vs carrier distortion
- execution deadlock vs continuity drift
- boundary erosion vs generic confusion
- downstream symptom vs earliest decisive break
- explanation style vs route quality

SEPARATE:
Observed failure must describe what actually happens.
Guessed cause must not be treated as route evidence by default.
A strong label is not a route.
A popular label is not a route.
A fashionable label is not a route.
A route must be earned by structural evidence.

INTAKE_QUESTIONS:
- what failed first
- what can actually be observed
- what is only being guessed
- what evidence supports the cut
- what nearby family is still materially plausible
- what missing evidence could change the route
- what first fix would be most dangerous if chosen wrongly

USE_SECONDARY_ONLY_IF:
- a nearby family is genuinely plausible
- wrong first repair would likely come from nearby confusion
- the visible case surface can mislead the route
- evidence favors one family but another still has live pressure

DO_NOT_USE_SECONDARY_IF:
- neighboring pressure is weak
- adding a secondary family would only sound smart
- the primary cut is already stable enough
- no nearby family is materially live

DEFAULT:
secondary_family = none when no nearby family remains materially live
overlay = none when no overlay pressure dominates

BROKEN_INVARIANT_RULE:
Broken invariant must name the structural rule that broke first.
Broken invariant must be more valuable than family naming alone.
Broken invariant must constrain first_fix_direction.
Broken invariant must constrain misrepair_risk.

WHY_PRIMARY_NOT_SECONDARY_RULE:
Use decisive cut language.
Do not explain with adjectives alone.
Use:
- what arrives first
- what is structurally supported
- what is still unsupported
- why the nearby family is not yet primary

TIE_BREAKS:
- earliest decisive failure beats later consequence
- structural evidence beats vocabulary cue
- visible carrier failure beats higher-order reasoning claims when the carrier is clearly broken
- real anchor loss beats formal resemblance
- hidden failure path means observability may come first
- explicit execution closure failure beats vague continuity language when readiness or ordering evidence is concrete
- real boundary erosion beats danger wording
- if evidence is thin, prefer honest family-level fit over fake subtype precision
- if nearby family remains live, do not oversell certainty

CONFIDENCE_HIGH_IF:
- primary cut is explicit
- secondary_family is materially rejected
- broken_invariant is stable under current evidence
- first_fix_direction follows directly from the cut
- evidence_sufficiency is sufficient

CONFIDENCE_MEDIUM_IF:
- primary cut is plausible
- evidence is partial
- neighboring pressure still has limited life
- route is useful, but not fully locked

CONFIDENCE_LOW_IF:
- only family-level fit is honestly defensible
- decisive evidence is weak or partial
- neighboring confusion remains materially live
- route is provisional and must stay humble

CONFIDENCE_ALIGNMENT_RULE:
- confidence high requires evidence_sufficiency sufficient
- confidence medium usually pairs with partial
- confidence low usually pairs with partial or weak
- confidence can never outrun evidence_sufficiency
- weak evidence forbids precision theater

EVIDENCE_SUFFICIENCY_SUFFICIENT_IF:
- route is supported by direct evidence
- decisive cut is materially visible
- nearby confusion is materially rejected
- first_fix_direction is stable under current evidence

EVIDENCE_SUFFICIENCY_PARTIAL_IF:
- route is plausible
- important evidence is still missing
- nearby confusion still has limited life
- route is usable but not fully locked

EVIDENCE_SUFFICIENCY_WEAK_IF:
- decisive path is still thin
- route depends heavily on indirect signal
- family-level fit is possible but finer precision is unsafe
- next decisive check matters more than deeper naming

THIN_EVIDENCE_GATE:
Under thin evidence:
- prefer family-level over fake node-level
- prefer partial or weak over false sufficiency
- prefer low confidence over decorative certainty
- prefer need_more_evidence over fake closure
- prefer honest coarse routing over no-fit shortcut

SECTION_06_SEVEN_FAMILY_OPERATIONAL_CORE

F1_NAME:
Grounding & Evidence Integrity

F1_WHAT_FAILS_FIRST:
outputs, claims, or actions lose reliable tie to anchors, referents, or evidence

F1_BROKEN_INVARIANT:
anchor_to_claim_coupling_broken

F1_NOT:
not primary if the real first break is carrier distortion, hidden failure path, or execution closure

F1_CONFUSION:
F7 | F5

F1_TRIGGER:
anchor drift | wrong grounding | referent mismatch | evidence detachment | truth-anchor break

F1_FIRST_FIX_PATTERN:
re_grounding | anchor_verification | claim_to_evidence_trace

F1_MISREPAIR:
style_or_prompt_polish_before_anchor_restoration

F1_MIN_EVIDENCE:
at least one visible sign of anchor/referent/evidence tie failure

F2_NAME:
Reasoning & Progression Integrity

F2_WHAT_FAILS_FIRST:
inferential path, decomposition path, or recovery path breaks first

F2_BROKEN_INVARIANT:
progression_continuity_broken

F2_NOT:
not primary if the main issue is container distortion, anchor loss, or hidden failure path

F2_CONFUSION:
F7

F2_TRIGGER:
inferential stall | decomposition break | recursive loop | failed recovery path | progression collapse

F2_FIRST_FIX_PATTERN:
decomposition_reset | progression_scaffold | recovery_path_rebuild

F2_MISREPAIR:
surface_cleanup_before_progression_repair

F2_MIN_EVIDENCE:
at least one visible sign of reasoning progression failure rather than carrier failure

F3_NAME:
State & Continuity Integrity

F3_WHAT_FAILS_FIRST:
memory, role, provenance, handoff, ownership, or continuity thread destabilizes first

F3_BROKEN_INVARIANT:
state_continuity_broken

F3_NOT:
not primary if the main issue is execution closure, pure observability darkness, or primary boundary erosion

F3_CONFUSION:
F4 | F6

F3_TRIGGER:
memory drift | role instability | ownership confusion | handoff fracture | continuity break | provenance instability

F3_FIRST_FIX_PATTERN:
continuity_restoration | role_fencing | ownership_reisolation

F3_MISREPAIR:
patching downstream behavior while continuity state remains unstable

F3_MIN_EVIDENCE:
at least one visible sign of continuity persistence failure

F4_NAME:
Execution & Contract Integrity

F4_WHAT_FAILS_FIRST:
ordering, readiness, bridge, liveness, or closure breaks first

F4_BROKEN_INVARIANT:
execution_skeleton_closure_broken

F4_NOT:
not primary if the main issue is only continuity drift, pure observability darkness, or pure boundary erosion

F4_CONFUSION:
F3 | F5 | F6

F4_TRIGGER:
ordering failure | readiness gap | bridge break | liveness failure | closure break | operational action incompletion

F4_FIRST_FIX_PATTERN:
readiness_audit | ordering_audit | bridge_audit | liveness_audit

F4_MISREPAIR:
memory_or_style_interpretation_before_execution_closure_check

F4_MIN_EVIDENCE:
at least one visible sign of operational closure failure

F5_NAME:
Observability & Diagnosability Integrity

F5_WHAT_FAILS_FIRST:
the true failure path is not yet visible enough to justify deeper repair

F5_BROKEN_INVARIANT:
failure_path_visibility_broken

F5_NOT:
not primary if clear anchor collapse, clear execution closure failure, or clear boundary erosion is already explicit

F5_CONFUSION:
F6 | F4 | F1

F5_TRIGGER:
hidden failure path | under-instrumented path | weak trace | weak auditability | poor coherence exposure

F5_FIRST_FIX_PATTERN:
trace_exposure | observability_insertion | audit_route_exposure

F5_MISREPAIR:
deep_repair_before_path_exposure

F5_MIN_EVIDENCE:
at least one sign that the decisive failure path remains insufficiently observable

F6_NAME:
Boundary & Safety Integrity

F6_WHAT_FAILS_FIRST:
goal boundary, control boundary, incentive corridor, or safety regime erodes first

F6_BROKEN_INVARIANT:
boundary_integrity_broken

F6_NOT:
not primary if the issue is only missing logs, generic confusion, or plain execution closure

F6_CONFUSION:
F5 | F4 | F3

F6_TRIGGER:
control-path breach | regime instability | safety-boundary collapse | incentive corridor erosion | unsafe authority shift

F6_FIRST_FIX_PATTERN:
boundary_guard | control_path_audit | regime_stability_check

F6_MISREPAIR:
generic_observability_treatment_before_boundary_confirmation

F6_MIN_EVIDENCE:
at least one visible sign of boundary or control-path erosion

F7_NAME:
Representation & Localization Integrity

F7_WHAT_FAILS_FIRST:
descriptor, carrier, local structure, symbolic shell, or representation container breaks first

F7_BROKEN_INVARIANT:
representation_container_fidelity_broken

F7_NOT:
not primary if the main issue is anchor loss, pure reasoning failure, or pure observability deficit

F7_CONFUSION:
F1 | F2 | F5

F7_TRIGGER:
descriptor corruption | malformed carrier | localization break | structural shell failure | representation distortion

F7_FIRST_FIX_PATTERN:
descriptor_audit | carrier_audit | structural_preservation

F7_MISREPAIR:
reasoning_rewrite_before_carrier_repair

F7_MIN_EVIDENCE:
at least one visible sign of representational or carrier distortion

MINIMAL_EVIDENCE_CHECKLIST:
F1 requires anchor/referent/evidence tie signs
F2 requires progression failure signs
F3 requires continuity or persistence instability signs
F4 requires ordering/readiness/bridge/closure signs
F5 requires hidden-path or under-instrumented-path signs
F6 requires boundary or regime erosion signs
F7 requires descriptor/carrier/representation distortion signs

SECTION_08_BOUNDARY_DECISION_MATRIX

F1_VS_F7:
choose_F1_if = anchor / referent / evidence tie breaks first
choose_F7_if = carrier / descriptor / structural container breaks first
not_F1_too_early_if = anchor status is still unresolved and carrier distortion is explicit
not_F7_too_early_if = carrier looks intact but anchor mismatch is already visible
evidence_threshold = need visible support for whichever first break is claimed
default_tie = coarse_family_level_until_anchor_vs_carrier_order_is_clear

F5_VS_F6:
choose_F5_if = failure path is still not visible enough to justify deeper repair
choose_F6_if = a real control or safety boundary is already clearly eroding
not_F5_too_early_if = explicit regime or control-path breach is already visible
not_F6_too_early_if = boundary language is dramatic but path visibility remains the first bottleneck
evidence_threshold = choose F6 only with visible boundary erosion, not fear words alone
default_tie = F5_if_visibility_is_first_else_coarse

F3_VS_F4:
choose_F3_if = continuity of role / memory / provenance breaks first
choose_F4_if = readiness / ordering / bridge / closure breaks first
not_F3_too_early_if = operational closure failure is explicit
not_F4_too_early_if = continuity persistence is failing before execution path is actually tested
evidence_threshold = need visible continuity evidence for F3 or explicit closure evidence for F4
default_tie = coarse_until_continuity_vs_closure_order_is_clear

F2_VS_F7:
choose_F2_if = reasoning progression breaks while the carrier remains usable
choose_F7_if = the carrier itself is malformed and cannot reliably carry reasoning
not_F2_too_early_if = malformed carrier already blocks progression judgment
not_F7_too_early_if = carrier is intact and inferential collapse is visible
evidence_threshold = do not name F2 subtype without visible progression pattern
default_tie = coarse_family_level_if_carrier_quality_is_uncertain

F1_VS_F5:
choose_F1_if = anchor failure is already visible
choose_F5_if = the main issue is that the real failure path is still too dark
not_F1_too_early_if = anchor status is guessed, not observed
not_F5_too_early_if = anchor mismatch is already explicit
evidence_threshold = F1 needs visible anchor tie failure; F5 needs path-opacity evidence
default_tie = F5_if_anchor_status_is_thin_else_coarse

F4_VS_F5:
choose_F4_if = execution closure failure is explicit
choose_F5_if = workflow is still too dark to justify execution-first repair
not_F4_too_early_if = execution failure is inferred from opacity alone
not_F5_too_early_if = closure failure is already visible and concrete
evidence_threshold = F4 needs explicit readiness/order/bridge/closure evidence
default_tie = F5_until_execution_failure_is_exposed

F3_VS_F6:
choose_F3_if = continuity state breaks first without clear regime breach
choose_F6_if = control or goal boundary erosion is primary
not_F3_too_early_if = real control-path or regime erosion is already explicit
not_F6_too_early_if = boundary language is present but continuity instability arrives first
evidence_threshold = F6 requires visible boundary evidence beyond general instability
default_tie = coarse_until_state_vs_boundary_order_is_clear

F4_VS_F6:
choose_F4_if = main break is execution closure
choose_F6_if = main break is boundary or control-path integrity
not_F4_too_early_if = closure trouble is only guessed from control tension
not_F6_too_early_if = failure is operational closure without visible boundary erosion
evidence_threshold = separate bridge/closure evidence from control/regime evidence
default_tie = coarse_if_both_closure_and_boundary_pressure_remain_live

SECTION_09_RESOLUTION_DISCIPLINE

BEST_CURRENT_FIT_RULE:
best_current_fit must be chosen honestly.
It is not a decorative label.
It is the highest honest resolution under available evidence.

family-level:
- use when family fit is stable enough
- subtype is not yet honest
- thin evidence is acceptable

node-level:
- use only when narrower subtype is directly supported
- do not jump to node-level from thematic resemblance alone

unresolved_subtype:
- use when family is stable
- subtype is still unresolved
- do not force subtype naming

no-fit:
- use only when family-level fit is not honestly defensible
- do not use as an ambiguity shortcut
- no-fit is rare by design

need_more_evidence:
- use when one or more missing observations could materially change the route
- this is a routing state token
- this is not a decorative note
- use it before fake precision

THIN_EVIDENCE_DOWNGRADE:
- node-level -> family-level
- clear confidence claims -> lower confidence
- sufficient -> partial or weak
- false closure -> need_more_evidence

SECTION_10_AMBIGUOUS_NOFIT_DISCIPLINE

AMBIGUITY_RULE:
If two nearby families remain plausible under thin evidence:
- prefer honest coarse routing
- prefer family-level over decorative precision
- preserve the live competing family
- do not claim closure too early

NO_FIT_RESTRAINT:
Do not choose no-fit while family-level fit remains plausible.
No-fit must be earned by genuine non-fit, not by discomfort.

NEED_MORE_EVIDENCE_RULE:
need_more_evidence may appear only when:
- missing evidence could materially change primary vs secondary cut
- missing evidence could change best_current_fit
- missing evidence blocks honest first_fix_direction
need_more_evidence must not be emitted as a generic safety blanket.

SECTION_11_FIRST_FIX_AND_MISREPAIR

FIRST_FIX_RULE:
first_fix_direction must:
- be one primary move or one tightly coupled first bundle
- match broken_invariant
- target earliest decisive break
- not become a long repair tree
- not drift into cosmetic cleanup

MISREPAIR_RULE:
misrepair_risk must:
- identify the most tempting wrong first move
- stay near the nearest plausible wrong cut
- warn against the wrong-first-fix most likely to waste time

PAIRING_FORMAT:
broken_invariant -> first_fix_direction -> misrepair_risk

SECTION_12_EXEMPLAR_COVERAGE_INDEX

EX1 = F1 | F5
EX2 = F7 | F2
EX3 = F5 | F4
EX4 = F4 | none
EX5 = F3 | F4
EX6 = F6 | F5
EX7 = F1 | F7
EX8 = F5 | F6 | thin_evidence
EX9 = F4 | F6
EX10 = F3 | F6
EX11 = family-level | unresolved_subtype | need_more_evidence
EX12 = overlay_OBS | overlay_SEC | overlay_LOC

SECTION_13_GOLD_MINI_EXEMPLARS

EX1_F1_GROUNDING_FIRST:
case = answer cites retrieved chunks but chosen chunks do not support actual claim
primary_family = F1
secondary_family = F5
why_primary_not_secondary = evidence-anchor mismatch arrives before opacity
broken_invariant = anchor_to_claim_coupling_broken
best_current_fit = family-level
first_fix_direction = re_grounding | claim_to_chunk_trace
misrepair_risk = explanation_polish_before_anchor_restoration
confidence = high
evidence_sufficiency = sufficient

EX2_F7_CONTAINER_FIRST:
case = OCR output preserves topic hints but text carrier is malformed and local structure is broken
primary_family = F7
secondary_family = F2
why_primary_not_secondary = malformed carrier breaks reasoning reliability before progression can be judged
broken_invariant = representation_container_fidelity_broken
best_current_fit = family-level
first_fix_direction = descriptor_and_carrier_audit
misrepair_risk = reasoning_rewrite_before_carrier_repair
confidence = medium
evidence_sufficiency = partial

EX3_F5_OBSERVABILITY_FIRST:
case = workflow fails repeatedly but logs do not expose where the path actually breaks
primary_family = F5
secondary_family = F4
why_primary_not_secondary = path invisibility arrives before confirmed execution closure failure
broken_invariant = failure_path_visibility_broken
best_current_fit = family-level
first_fix_direction = trace_exposure | observability_insertion
misrepair_risk = execution_surgery_before_path_exposure
confidence = medium
evidence_sufficiency = partial

EX4_F4_EXECUTION_FIRST:
case = pipeline step order is wrong and the action bridge never completes
primary_family = F4
secondary_family = none
why_primary_not_secondary = explicit ordering and closure failure is already visible
broken_invariant = execution_skeleton_closure_broken
best_current_fit = node-level
first_fix_direction = ordering_and_bridge_audit
misrepair_risk = continuity_or_prompt_blame_before_execution_check
confidence = high
evidence_sufficiency = sufficient

EX5_F3_CONTINUITY_FIRST:
case = agent handoff loses role ownership and long-task continuity
primary_family = F3
secondary_family = F4
why_primary_not_secondary = continuity persistence breaks before execution closure is the primary bottleneck
broken_invariant = state_continuity_broken
best_current_fit = family-level
first_fix_direction = continuity_restoration | role_fencing
misrepair_risk = bridge_repair_without_role_reisolation
confidence = medium
evidence_sufficiency = partial

EX6_F6_BOUNDARY_FIRST:
case = system increasingly follows exploitative objectives despite intact traces
primary_family = F6
secondary_family = F5
why_primary_not_secondary = control-path erosion is explicit; observability remains secondary
broken_invariant = boundary_integrity_broken
best_current_fit = family-level
first_fix_direction = boundary_guard | control_path_audit
misrepair_risk = observability_patch_before_boundary_control
confidence = medium
evidence_sufficiency = sufficient

EX7_F1_VS_F7_ANCHOR_WINS:
case = format looks slightly messy, but wrong targets and wrong referents are the real failure
primary_family = F1
secondary_family = F7
why_primary_not_secondary = anchor mismatch arrives first; carrier damage is secondary
broken_invariant = anchor_to_claim_coupling_broken
best_current_fit = family-level
first_fix_direction = target_reference_re_grounding
misrepair_risk = container_cleanup_before_anchor_fix
confidence = medium
evidence_sufficiency = partial

EX8_THIN_EVIDENCE_COARSE:
case = both observability and boundary pressure look possible, but trace and regime evidence are thin
primary_family = F5
secondary_family = F6
why_primary_not_secondary = opacity is slightly better supported, but boundary pressure remains live
broken_invariant = failure_path_visibility_broken
best_current_fit = family-level
first_fix_direction = collect_visibility_and_boundary_disambiguation_evidence
misrepair_risk = locking_F6_too_early
confidence = low
evidence_sufficiency = weak
need_more_evidence = determine_whether_visibility_bottleneck_or_boundary_erosion_arrives_first

EX9_F4_VS_F6_CLOSURE_WINS:
case = action chain never completes even though no explicit boundary breach is visible
primary_family = F4
secondary_family = F6
why_primary_not_secondary = execution closure failure is explicit while boundary erosion remains unproven
broken_invariant = execution_skeleton_closure_broken
best_current_fit = family-level
first_fix_direction = readiness_ordering_bridge_audit
misrepair_risk = boundary_theory_before_execution_verification
confidence = medium
evidence_sufficiency = partial

EX10_F3_VS_F6_CONTINUITY_WINS:
case = role identity degrades across sessions without explicit control-path breach
primary_family = F3
secondary_family = F6
why_primary_not_secondary = continuity persistence fails before boundary erosion is visible
broken_invariant = state_continuity_broken
best_current_fit = family-level
first_fix_direction = role_and_memory_reisolation
misrepair_risk = boundary_hardening_before_continuity_stabilization
confidence = medium
evidence_sufficiency = partial

EX11_UNRESOLVED_SUBTYPE:
case = family F2 is stable, but exact subtype of progression failure is not yet defensible
primary_family = F2
secondary_family = none
why_primary_not_secondary = progression failure is primary and nearby family pressure is weak
broken_invariant = progression_continuity_broken
best_current_fit = unresolved_subtype
first_fix_direction = progression_trace_exposure_then_decomposition_reset
misrepair_risk = naming_specific_subtype_without_visible_inferential_pattern
confidence = low
evidence_sufficiency = partial
need_more_evidence = isolate_visible_inferential_break_pattern

EX12_OVERLAY_CASE:
case = main route is F5, but layout and OCR anchor pressure also matters
primary_family = F5
secondary_family = F1
why_primary_not_secondary = path opacity remains primary; local text-anchor stress refines reading only
broken_invariant = failure_path_visibility_broken
best_current_fit = family-level
first_fix_direction = trace_exposure_before_layout_specific_intervention
misrepair_risk = treating_overlay_as_primary_family_replacement
confidence = low
evidence_sufficiency = partial
overlay = OBS | LOC

SECTION_14_GOLD_OUTPUTS

OUT1_HIGH:
primary_family=F1
secondary_family=F5
why_primary_not_secondary=decisive failure is evidence-anchor mismatch; opacity exists but does not arrive first
broken_invariant=anchor_to_claim_coupling_broken
best_current_fit=node-level:Retrieval Anchor Drift
first_fix_direction=re_grounding|chunk_to_target_trace
misrepair_risk=explanation_or_style_refinement_before_anchor_restoration
confidence=high
evidence_sufficiency=sufficient

OUT2_MEDIUM:
primary_family=F5
secondary_family=F6
why_primary_not_secondary=coherence visibility bottleneck appears first; boundary pressure may emerge later
broken_invariant=failure_path_visibility_broken
best_current_fit=family-level
first_fix_direction=coherence_visibility_uplift|audit_route_exposure
misrepair_risk=boundary_stabilization_before_structure_is_inspectable
confidence=medium
evidence_sufficiency=partial

OUT3_LOW_NEED_MORE:
primary_family=F4
secondary_family=F3
why_primary_not_secondary=execution skeleton stress visible but continuity loss remains plausible under thin trace
broken_invariant=possible_execution_closure_instability
best_current_fit=family-level
first_fix_direction=collect_more_readiness_ordering_continuity_evidence
misrepair_risk=forcing_bridge_repair_interpretation_too_early
confidence=low
evidence_sufficiency=weak
need_more_evidence=clarify_whether_ordering_bridge_failure_or_continuity_thread_loss_arrives_first

OUT4_FAMILY_ONLY:
primary_family=F7
secondary_family=F2
why_primary_not_secondary=carrier stress is clear but subtype evidence insufficient
broken_invariant=representation_container_fidelity_broken
best_current_fit=family-level
first_fix_direction=descriptor_and_carrier_audit
misrepair_risk=progression_first_reading_before_carrier_check
confidence=medium
evidence_sufficiency=partial

OUT5_TWO_FAMILY_COMPETITION:
primary_family=F5
secondary_family=F6
why_primary_not_secondary=both diagnosability pressure and boundary pressure exist, but current evidence supports opacity first
broken_invariant=failure_path_visibility_broken
best_current_fit=family-level
first_fix_direction=trace_exposure|coherence_visibility_uplift
misrepair_risk=using_F6_stabilization_before_boundary_failure_is_explicit
confidence=low
evidence_sufficiency=partial

OUT6_HIGH_RISK_SOUNDING_THIN:
primary_family=F6
secondary_family=F5
why_primary_not_secondary=control drift evidence is explicit, but the exact boundary subtype is not yet honest while observability pressure remains live
broken_invariant=control_boundary_stability_degraded
best_current_fit=unresolved_subtype
first_fix_direction=boundary_guard|control_path_audit|regime_disambiguation
misrepair_risk=claiming_full_boundary_certainty_or_treating_remaining_opacity_as_only_observability_noise
confidence=medium
evidence_sufficiency=partial

SECTION_15_OVERLAY_RULE

OBS:
use_if=visibility|traceability|auditability|interpretability pressure dominates
type=overlay_not_primary_family

SEC:
use_if=boundary_abuse|adversarial_manipulation|exploitative_misuse|attack_pressure clear
type=overlay_not_primary_family

LOC:
use_if=layout|OCR|placement|spatial_text_anchor|local_structural_alignment failure dominates
type=overlay_not_primary_family

MAIN_FAMILY_FIRST:
decide_main_family_first
overlay_refines_reading
overlay_does_not_replace_routing
overlay_does_not_become_primary_family

SECTION_16_FIT_CANDIDATE_REGISTRY

F1_FITS:
stable_enough=Retrieval Anchor Drift|Semantic Grounding Mismatch|Synthetic Truth Grounding|OOD World Grounding Failure|Train-Deploy Distribution Grounding Gap
use_with_caution=Policy-to-World Grounding Failure
family_level_only_if_thin=general_grounding_failure|evidence_anchor_mismatch|target_reference_drift
do_not_overreach=no_specific_grounding_subtype_without_clear_anchor_target_referent_world_evidence

F2_FITS:
stable_enough=Interpretation Collapse|Long-Chain Reasoning Continuity Failure|Collapse-Recovery Failure
use_with_caution=Recursive Horizon Instability|Symbolic Progression Breakdown|Philosophical Recursion Pressure
family_level_only_if_thin=reasoning_progression_failure|recursive_instability|decomposition_failure
do_not_overreach=no_named_reasoning_subtype_without_visible_inferential_pattern

F3_FITS:
stable_enough=Memory Continuity Failure|Role / Ownership Contamination|Viable State-Space Collapse|Multi-Agent Continuity Instability
use_with_caution=Ownership Continuity Break|Interaction-Thread Drift
family_level_only_if_thin=continuity_failure|persistence_break|role_thread_instability
do_not_overreach=no_specific_continuity_subtype_without_isolation_of_role_ownership_memory_interaction

F4_FITS:
stable_enough=Bootstrap Ordering Failure|Deployment Deadlock|Pre-Readiness Execution Failure|Bridge Integrity Failure|Cross-Layer Liveness Degradation
use_with_caution=Institutional Enforcement Drift|Rule-to-Action Closure Failure|Accountability Path Thinning|Fallback Realism Gap
family_level_only_if_thin=execution_skeleton_failure|closure_path_failure|operational_dependency_collapse
do_not_overreach=no_detailed_F4_fit_if_ordering_bridge_liveness_enforcement_only_guessed_from_symptom

F5_FITS:
stable_enough=Failure Path Opacity|Oversight Coverage Failure|Coherence Observability Deficit|Meaning-Profile Auditability Failure
use_with_caution=Scalable Interpretability Pressure|Early Warning Deficit|Fragility Signature Blindness
family_level_only_if_thin=diagnosability_failure|auditability_failure|warning_visibility_failure
do_not_overreach=no_specific_observability_subtype_without_identifiable_missing_path_or_warning_deficit

F6_FITS:
stable_enough=Alignment Boundary Drift|Control Boundary Erosion|Incentive Boundary Distortion|Incentive Misalignment Drift|Polarization Incentive Amplification
use_with_caution=Collective Boundary Drift|Collective Legitimacy Erosion|Coordination-Boundary Friction
family_level_only_if_thin=boundary_failure|control_drift|incentive_corridor_erosion
do_not_overreach=no_specific_boundary_subtype_without_visible_regime_or_control_path_evidence

F7_FITS:
stable_enough=OCR / Layout Structural Failure|Representation Carrier Distortion|Semantic Localization Failure|Anchor-Container Mismatch
use_with_caution=Symbolic Shell Instability|Mixed Carrier / Anchor Confusion
family_level_only_if_thin=representation_failure|carrier_failure|localization_stress
do_not_overreach=no_specific_F7_subtype_without_visible_descriptor_carrier_local_structure_evidence

SECTION_17_AUDIT_COVERAGE_APPENDIX

AUDIT_STRUCTURE:
must_preserve=
product_identity_and_scope|
core_routing_contract|
case_intake_normalization|
secondary_family_discipline|
tie_break_principles|
confidence_and_evidence_discipline|
seven_family_operational_core|
minimal_evidence_checklist|
boundary_decision_matrix|
resolution_discipline|
first_fix_and_misrepair|
overlay_rule|
fit_candidate_registry|
exemplar_support|
gold_output_support

AUDIT_FAMILY_OPERATIONALIZATION:
each_family_must_have=
what_fails_first|
broken_invariant|
what_this_family_is_not|
confusion_family|
boundary_trigger|
first_fix_pattern|
misrepair_trap|
minimum_evidence

AUDIT_BOUNDARY_COVERAGE:
required_boundaries=
F1/F7|F5/F6|F3/F4|F2/F7|F1/F5|F4/F5|F3/F6|F4/F6
each_boundary_must_have=
choose_A|
choose_B|
not_A_too_early|
not_B_too_early|
evidence_threshold|
default_tie

AUDIT_RESOLUTION_COVERAGE:
must_define=
family-level|
node-level|
unresolved_subtype|
need_more_evidence|
no-fit|
downgrade_rules|
thin_evidence_gate

AUDIT_OUTPUT_COVERAGE:
canonical_fields_must_be_supported_by_examples_or_gold_outputs=
primary_family|
secondary_family|
why_primary_not_secondary|
broken_invariant|
best_current_fit|
first_fix_direction|
misrepair_risk|
confidence|
evidence_sufficiency

optional_tokens_must_have_at_least_one_explicit_support_case=
need_more_evidence|
overlay

OVERLAY_SUPPORT_NOTE:
overlay_token_support_is_required_not_every_overlay_value_needs_its_own_exemplar

SECTION_18_NO_LAZINESS_AUDIT

FAIL_IF_TOPIC_SHORTCUT:
- safety_words_auto_F6
- OCR_logic_schema_words_auto_F7
- memory_words_auto_F3
- execution_words_auto_F4
- abstract_language_auto_F5

FAIL_IF_DECORATIVE_SECONDARY:
secondary_family_used_only_to_sound_intelligent

FAIL_IF_DECORATIVE_NODE_FIT:
node_level_from_thematic_resemblance_without_direct_structural_evidence

FAIL_IF_AMBIGUITY_ESCAPE:
ambiguous_or_no-fit_used_when_honest_coarse_fit_exists

FAIL_IF_COSMETIC_FIX_FIRST:
first_fix_direction_drifted_to_style_prompt_surface_cleanup_before_broken_invariant

FAIL_IF_FAKE_PRECISION:
high_confidence_or_sufficient_evidence_claimed_under_thin_or_ambiguous_case

FAIL_IF_CANONICAL_RENDER_CONFLICT:
render_layer_adds_or_overrides_route_state_not_present_in_canonical_output

SECTION_19_PRESSURE_TESTS

TEST_1_GROUNDING_VS_CONTAINER:
can_router_distinguish_external_anchor_loss_from_structural_carrier_failure_without_collapsing_both_into_F1_or_F7

TEST_2_OBSERVABILITY_VS_BOUNDARY:
can_router_avoid_choosing_F6_too_early_when_first_bottleneck_is_missing_visibility_or_coherence_exposure

TEST_3_CONTINUITY_VS_EXECUTION:
can_router_distinguish_role_ownership_persistence_drift_from_ordering_readiness_bridge_closure_failure

TEST_4_FORMAL_LOOKING_BUT_GROUNDING_FIRST:
can_router_avoid_overreading_F7_when_the_real_failure_is_anchor_loss_with_presentational_noise

TEST_5_THIN_EVIDENCE_DOWNGRADE:
can_router_hold_family-level_or_need_more_evidence_without_fake_node_precision

TEST_6_SECONDARY_DISCIPLINE:
can_router_use_secondary_family_honestly_without_turning_it_into_decorative_hedging

TEST_7_MODE_STABILITY:
can_router_preserve_same_primary_family_same_broken_invariant_and_same_best_current_fit_level_under_canonical_only_vs_canonical_plus_render_output

SECTION_20_RELEASE_GATE

PASS_IF:
1. all_7_families_operationalized
2. all_8_required_boundaries_decisionized
3. best_current_fit_resolution_rules_explicit
4. fit_candidate_registry_present
5. first_fix_and_misrepair_paired
6. at_least_12_mini_exemplars_present
7. at_least_6_gold_outputs_present
8. overlay_rule_present
9. audit_coverage_appendix_present
10. pressure_tests_present
11. optional_tokens_have_explicit_support_case
12. mode_stability_test_present

FAIL_IF:
- family_names_without_operational_definitions
- decorative_node_fit_common
- F6_or_F7_absorb_cases_from_topic_words_alone
- ambiguous_or_no-fit_used_as_uncertainty_shortcut
- first_fix_direction_cosmetic
- output_fields_exist_without_support_case
- render_layer_contradicts_canonical_route_state

FINAL_RULE:
Prioritize the best first cut, not decorative explanation.

AI_NOTE_END