#Problem Typing & TaskSignature Assignment (Problem‑CHR)

Pattern C.22 · Stable Part C - Kernel Extension Specifications

Purpose. Give FPF a lawful, minimal, and portable way to speak about “the problem we face” so that the selector (G.5) can legally admit/abstain without prose or guesswork. We do this by (i) typing problems with CHR‑grounded traits and (ii) attaching them to a TaskSignature (S2) record that downstream patterns can consume. The Standard is Context‑local, evidence‑anchored, tri‑state‑aware, and bridge‑savvy. TaskSignature is minimal but sufficient for eligibility, acceptance, and policy‑governed choice.

Status & placement. Part C (Kernel Extentions Specifications) → Cluster C.I (Core CHRs/CALs). Depends on: C.16 MM‑CHR (measurement legality), G.5 (selector S2/S3), G.0 (CG‑Spec invariants). Coordinates with: G.4 (Acceptance/Evidence profiles), C.23 (MethodFamily admissibility & maturity), C.18 NQD‑CAL (QD/illumination), C.19 E/E‑LOG (emitters/policies), E.10 (LEX).

Operationalise No‑Free‑Lunch discipline in selection by ensuring every run‑time decision sees a typed TaskSignature (S2), not a paragraph, and that “problem” (method unknown) is cleanly separated from “task” (method known; signature bound). The signature is the smallest CHR‑typed set sufficient to drive Eligibility → Acceptance → policy‑governed selection without illegal arithmetic or silent coercions; crossings are auditable (Bridge+CL → R_eff only).

#Keywords

• Problem-CHR
• TaskSignature
• TaskKind
• ScopeSlice(G)
• unknown handling
• specialization anchor.

#Relations

#Content

#Intent

Operationalise No‑Free‑Lunch discipline in selection by ensuring every run‑time decision sees a typed TaskSignature (S2), not a paragraph, and that “problem” (method unknown) is cleanly separated from “task” (method known; signature bound). The signature is the smallest CHR‑typed set sufficient to drive Eligibility → Acceptance → policy‑governed selection without illegal arithmetic or silent coercions; crossings are auditable (Bridge+CL → R_eff only).

#Term split used in this pattern

• TaskSignature attachment means attaching one typed problem record to one declared task family or work target; it does not pre-bind a method.
• ScopeSlice(G) means the claim-bounding scope cut over describedEntity/scope; it is not an evidence-path slice and not a baseline-set slice.
• threshold is not one undifferentiated family here:
  • articulation and closure thresholds stay with cue/prompt owners such as B.4.1 and B.5.2.0
  • acceptance-gate thresholds stay with G.4
  • the work-measure threshold target used in specialization claims is only the declared success mark for the current task family or work target

#Problem Frame (design/run split; crossing-visible)

method‑first stance In FPF a Problem exists when a Holder or external Transformer cannot cite a known Method (or specialisation thereof) that satisfies the current TaskSignature under the declared ScopeSlice(G). Problem‑solving therefore entails strategizing (selecting or synthesising a method). The resulting strategy/policy is a composition under G.5/E/E‑LOG and is not a new kernel type.
Unknown‑first discipline. Author S2 with unknown traits rather than coercions; SoS‑LOG branches MUST specify {admit|degrade|abstain|sandbox} handling for unknown via closed enums registered at UTS.

Un‑typed "problems" collapse into informal prose; selectors cannot filter/abstain lawfully; acceptance-gate thresholds leak into scoring; cross‑Context reuse is by name, not Bridge. We need a Context‑local descriptor that (i) obeys MM‑CHR legality (Scale/Unit/Polarity proven before any aggregation), (ii) records Assurance lanes (TA/VA/LA) per A.10 and ReferencePlane, (iii) carries tri‑state unknowns explicitly, and (iv) publishes crossing attestations (BridgeCard + UTS row) with Φ(CL)/Φ_plane policy‑ids.

#Problem

Without typed descriptors, Eligibility/Acceptance degenerate into prose; illegal ops creep in (ordinal means; unit mixing); cross‑plane comparisons lose CL/Φ routing (penalties to R_eff only).

#Forces

#Solution — Problem‑CHR (fields) + TaskSignature (S2) attachment (normative)

#Minimal CHR fields (tri‑state aware).

Each field is CHR‑typed (Characteristic/Scale/Unit/Polarity; MM‑CHR discipline). Every predicate admits unknown (tri‑state). Unknowns must propagate to {degrade|abstain|sandbox} per Acceptance/EvidenceProfile policy (recorded in SCR). (G.4/G.6 alignment)

• DataShape — data regime & admissible transforms (e.g., tabular, sequence, graph; density; stationarity claims).
• NoiseModel — uncertainty class / robustness envelope (e.g., iid Gaussian; heavy‑tailed; adversarial budget).
• ObjectiveProfile — objective heads (Scale/Unit/Polarity and ReferencePlane declared), polarity, and lawful orders (lexicographic, Pareto, medoid/median where legal). Weighted sums across mixed scale types are forbidden; ordinal heads use order‑only guards. For QD tasks, explicitly enumerate Q (quality), D (diversity), and QD‑score heads; see DominanceRegime below.
• RegularityTraits — method‑relevant structure (convexity/differentiability/separability/monotonicity) as CHR‑typed predicates with guard‑macros (e.g., ORD_COMPARE_ONLY, UNIT_CHECK, POLARITY_CHECK). Include ConditionClass (e.g., stiffness/κ‑proxies) where applicable.
• Constraints — explicit hard/soft constraint classes (feasibility predicates; ResourceEnvelope/RiskEnvelope). Acceptance-gate thresholds live in G.4 only; never inside CHR or code paths.
• ShiftClass/Stationarity — CHR‑typed claims about regime stability (iid | covariate‑shift | concept‑drift | adversarial). Default=unknown. Acceptance/Flows MUST honor this in gating or abstain.
• EvidenceGraphRef (A.10) — carriers & lane tags (TA/VA/LA) with freshness windows; no self‑evidence; default Γ‑fold = weakest‑link unless CAL proves an alternative.
• ScopeSlice(G) — the USM claim-bounding scope cut over describedEntity/scope (discipline governance in CG‑Spec; Domain is a catalog mark only).
• Size/Scale — size/condition proxies (n, m, κ, sparsity) with declared units; unit mismatch ⇒ {sandbox|refuse}.
• Freshness — validity window for descriptors.
• Missingness — MCAR/MAR/MNAR (or mapped equivalents) per CHR.Missingness; MUST be honoured by Acceptance/Flows.
• KindSet — U.Kind[] of objects‑of‑talk addressed by the TaskKind; separates describedEntity (Kind) from Scope (USM).

QD / Illumination extensions (normative; ties to C.18/C.19).

• CharacteristicSpaceRef — reference to U.CharacteristicSpace, with declared d≥2; characteristics are CHR‑typed; ReferencePlane per characteristic; pin edition via CharacteristicSpaceRef.edition.
• ArchiveConfig — archive topology (grid/CVT/graph), resolution (bins/centroids), K‑capacity, InsertionPolicyRef (elite replacement/dedup/novelty), and DistanceDefRef.edition (declare metric/pseudometric status and invariances; any normalisation MUST cite lawful scale transforms in CG‑Spec); legality follows CG‑Spec.
• EmitterPolicyRef — pointer to emitter/governor policy (C.19) applicable to this TaskSignature; edition id recorded.
• DominanceRegime — {ParetoOnly | ParetoPlusIllumination}. Default = ParetoOnly (illumination remains report‑only telemetry unless CAL explicitly authorises ParetoPlusIllumination, policy‑id cited).
• IlluminationSummary — a telemetry summary over Diversity_P; published by default; excluded from dominance unless a CAL enables ParetoPlusIllumination (policy‑id cited).
• IlluminationMap (parity‑run) — required publication artefact (grid/CVT/graph per ArchiveConfig) recording coverage per niche/cell with DescriptorMapRef/DistanceDefRef.edition. Leaderboards as single‑score lists are forbidden; comparisons MUST be under CG‑frames.
• PortfolioMode — {Pareto | Archive}. Default = Archive: selectors publish portfolios (QD archives) rather than a single “best” set; ε‑fronts remain admissible for local decisions under CG‑Spec.
• Budgeting — evaluation/time/batch budgets, including E/E‑LOG exploration budget id; units declared (CG‑Spec).
• TelemetryHooks — PathSliceId, decay/refresh policy ids, and edition counters to record U.DescriptorMap and policy‑id updates upon illumination gains.
• GeneratorIntent (OEE) — optional intent to invoke a GeneratorFamily (G.5) with pointers to EnvironmentValidityRegion, TransferRulesRef, and coverage/regret reporting expectations.

Legality. Before any numeric comparison/aggregation, prove CSLC legality (Scale/Unit/Polarity) and cite CG‑Spec.Characteristics; publish ReferencePlane. Unknowns propagate as {degrade|abstain|sandbox}; no unknown→0/false coercions.

#TaskSignature (S2) — attachment definition (design‑time + run‑time).

A TaskSignature is a minimal typed record the selector consumes: ⟨Context, TaskKind, TaskFamilyRef?, KindSet:U.Kind[], DataShape, NoiseModel, ObjectiveProfile, Constraints{incl. Resource/Risk Envelopes}, ScopeSlice(G), EvidenceGraphRef, Size/Scale, Freshness, Missingness, ShiftClass?, BehaviorSpaceRef?, ArchiveConfig?, EmitterPolicyRef?, DominanceRegime?, PortfolioMode?, Budgeting?, TelemetryHooks?, GeneratorIntent?⟩

Minimality rule. S2 carries only fields required for Eligibility→Acceptance→lawful selection; any additional traits derived at design‑time are published as provenance (UTS) but do not expand S2.

Values are CHR‑typed with provenance; traits may be inferred from CHR/CAL bindings (e.g., convexity known? differentiable? ordinal vs interval scales?) and from USM scope metadata. Unknowns are tri‑state; Missingness semantics MUST align with CHR.Missingness and be honored by Acceptance/Flows.

TaskKind names the governing kind of work or problem under this context. TaskFamilyRef? names one comparison-relevant family inside that task kind when specialization, transfer, or parity burden is live. TaskSignature is the context-bound typed attachment record for one current case under that kind and scope cut. KindSet continues to name the objects-of-talk governed by the task kind; it is not a substitute for the task family anchor.

Design/Run hygiene. Do not mix DesignRunTag in one signature; publish GateCrossings as CrossingBundles (E.18; Bridge+UTS A.27; BridgeCard F.9) when importing design‑time traits into run‑time.

#Specialization-claim anchoring (normative)

Any claim that one holder, dyad, team, or explicitly scoped specialist portfolio acquired usable specialization SHALL anchor that claim to one declared TaskFamilyRef or TaskSignature, one named work-measure threshold target, an adaptation budget, and the freshness or provenance basis for reuse. A method may be selected, refined, or retired as part of that story, but the method is not the bearer of the specialization claim. The attached task-family record should stay rich enough for C.22.1 adaptation signatures, G.5 specialization profiles, and G.9 adaptation parity to attach to the same task family and work target without reconstructing the claim from narrative prose.

Low-human-overlap or newly discovered task families remain admissible when those anchors are explicit by value.

#Provenance & planes.

Record Context, ReferencePlane for each value; on any cross‑Context/plane reuse, attach BridgeDescription + UTS row, apply CL (and, if planes differ, CL^plane) penalties to R_eff only; both Φ(CL) and (if used) Φ_plane MUST be monotone, bounded, and table‑backed; no “distance” language; penalties never mutate F/G. Publish policy‑ids in SCR and cite Bridge ids on crossings.

#Attachment & use.

• Eligibility gates read TaskSignature against each MethodFamily.Eligibility (C.23) and CG‑Spec.MinimalEvidence for referenced characteristics.
• Acceptance clauses (G.4) use these fields for acceptance-gate threshold predicates (acceptance-gate thresholds live in Acceptance only).
• Selection kernel (G.5.S3) applies a lawful order (often partial); weighted sums across mixed scale types are forbidden. If only a partial order remains, return a Pareto (non‑dominated) set with tie notes. If PortfolioMode=Archive, the selector may return a QD archive (per ArchiveConfig) in addition to or instead of a Pareto set. Illumination enters dominance only if DominanceRegime=ParetoPlusIllumination is enabled by CAL (policy id cited); otherwise, QD metrics are reported but excluded from dominance.
• When GeneratorIntent is present, G.5 may dispatch to a registered GeneratorFamily (POET‑class); the selection surface becomes pairs {environment, method}, with Environment guarded by EnvironmentValidityRegion and TransferRulesRef (C.23 wiring). Report IlluminationSummary as a telemetry summary over Diversity_P (report‑only by default) in telemetry; dominance remains unaffected unless policy changes as above.

#Unknowns.

Every field supports unknown; downstream degrade/abstain/sandbox behavior is explicit per Acceptance/EvidenceProfile; no implicit coercions.

#Publication.

Emit a ProblemProfile (…Description) that carries the bound TaskSignature, UTS Name Cards for any minted values (twin labels), and SCR‑visible provenance (A.10 anchors, lane tags, freshness, ReferencePlane). Mark any vendor/tool examples as Plain‑register only (LEX V‑4); they are non‑normative.

#Open‑Ended tasks (GeneratorFamily) (normative).

If the problem requires open‑ended generation of tasks/environments, S2 SHALL include GeneratorIntent with pointers to EnvironmentValidityRegion (lawful support of generated environments), TransferRulesRef (cross‑environment transfer constraints), and coverage/regret telemetry expectations. Selector outputs are then portfolios over {environment, method}; coverage/regret are telemetry metrics (reported) and IlluminationSummary is a telemetry summary (reported), excluded from dominance unless a CAL policy promotes them (policy‑id recorded in SCR; see DominanceRegime). Edition increments of CharacteristicSpaceRef.edition/DescriptorMapRef.edition/DistanceDefRef.edition and (OEE) TransferRulesRef.edition, and the policy‑id that caused illumination increases SHALL be recorded in SCR.

#Archetypal Grounding (Tell–Show–Show)

Tell–Show–Show hook (per E.8): label examples as Show‑1 (continuous ODE) and Show‑2 (MIP) and cite CHR guard‑macros in‑line so engineers can see which field drove which gate. Explicitly annotate which S2 fields triggered each Eligibility/Acceptance decision (e.g., service_level@ordinal → ORD_COMPARE_ONLY, budget@ratio → unit alignment check).

A. Differential equations (continuous systems, solver choice). ProblemProfile. DataShape=ODE, stiff?=unknown, Size/Scale={n≈10^3}, ObjectiveProfile={↓error@ratio, ↑throughput@ratio}, Constraints={budget≤X, safety_gate@ordinal}, RegularityTraits={Lipschitz known?=unknown, Jacobian sparsity=high}, Missingness=MAR. Attachment. Selector reads TaskSignature; eligibility filters MethodFamilies that require known stiffness or differentiability (unknown ⇒ degrade/abstain per family); Acceptance enforces safety_gate as ordinal predicate, not averaged (ORD_COMPARE_ONLY), and budgets with unit‑aligned sums (ratio). The selector returns a Pareto set; no cross‑ordinal weighting.

B. Mixed‑integer optimisation (planning/scheduling). ProblemProfile. DataShape=MIP, NoiseModel=deterministic, ObjectiveProfile={↓cost@ratio, ↑service_level@ordinal}, Constraints={SLA hard, workforce soft}, RegularityTraits={convex_relaxation=available}, Size/Scale={vars~10^5}, Missingness=MCAR. Attachment. CG‑Spec forbids means over service_level (ordinal); Acceptance holds acceptance-gate thresholds; Eligibility checks convex‑relaxation availability; Selection applies lexicographic guard (assumption‑fit ≻ evidence‑fit ≻ resource), compute R_eff with Γ‑fold, route CL to R only; if partial order remains, return a Pareto set.

Contemporary anchors (informative): modern Julia ecosystems illustrate the “general call outside, specialised implementations inside” ethos (e.g., DifferentialEquations.jl, JuMP), aligning with C.22→G.5 multi‑method dispatch under NFL.

C. Quality‑Diversity portfolio (illumination). ProblemProfile. DataShape=policy‑search; ObjectiveProfile={↑reward@ratio, ↑coverage@ratio (report‑only)}, DominanceRegime=ParetoOnly, PortfolioMode=Archive, CharacteristicSpaceRef(d=3, characteristics=CHR‑typed), ArchiveConfig(grid, res=32×32×16, K=1, InsertionPolicyRef=elite‑replace, DistanceDefRef.edition=v1), EmitterPolicyRef=v2, Budgeting{eval=1e6}, TelemetryHooks{PathSliceId=…}. Binding. Selector may return an archive; coverage/illumination are reported but excluded from dominance (default). Any change of DistanceDefRef.edition/Emitter policy is editioned and logged in SCR.

D. Open‑ended environment generation (POET‑class). ProblemProfile. GeneratorIntent{GeneratorFamilyRef=…, EnvironmentValidityRegion=… (CHR‑typed), TransferRulesRef=…, CoverageMetric=…}, PortfolioMode=Archive. Binding. Selector outputs {environment, method} pairs that pass Eligibility; TransferRules govern cross‑environment policy reuse; telemetry reports coverage/regret and IlluminationSummary with edition/policy‑id when improved.

#Bias‑Annotation (LEX/discipline guards)

• No minted “Strategy” head. “Strategy/policy” are roles/lenses inside G.5; do not introduce a new U.Type “Strategy”.
• No minted U.Type “Strategy”. Strategy/policy are roles/lenses inside G.5 Compose under E/E‑LOG; keep “strategy” as Plain where pedagogically needed.
• Transdiscipline vs domain. Comparability flows through U.Discipline CG‑Spec; “Domain” is a catalog mark stitched to D.CTX + UTS; do not attach norms to Domain labels.
• Plain twins & head‑anchoring. Use Description/Spec morphology correctly (I/D/S; E.10.D2).

#Anti‑patterns (normative):

• AP‑1 Pre‑binding a Method into S2 (“problem as if task”); Remedy: keep S2 method‑agnostic; bind only lawful traits.
• AP‑2 Silent unknown→false/0 in Eligibility/Acceptance.
• AP‑3 Cross‑ordinal averaging / ordinal–interval scalar mixes.
• AP‑4 Design/run chimera signatures (mixing stances).
• AP‑5 Domain treated as governance (attach governance to U.Discipline/CG‑Spec, not Domain).
• AP‑6 Implicit handling of data‑shift (assume iid); Remedy: declare ShiftClass (or unknown) and gate via Acceptance.
• AP‑7 Tool/vendor tokens in normative text; Remedy: move to Plain‑register note; keep Tech anchors on CHR/CAL ids (LEX V‑4). Remedies: tri-state predicates; lawful orders (lexi/Pareto/median/medoid); GateCrossing visibility via Bridge+UTS+CL/CL^plane (penalties → R only); Domain stitched to D.CTX + UTS only. Remedies: tri‑state predicates; lawful orders (lexi/Pareto/median/medoid); explicit GateCrossing publication via CrossingBundle (BridgeCard + UTS row + CL/Φ policy‑ids; E.18/A.27/F.9); Domain stitched to D.CTX + UTS only.

#Conformance Checklist (normative)

0. Minimal S2. S2 contains only fields necessary for Eligibility/Acceptance/selection; any extra derived traits remain provenance.

1. TaskSignature present (S2). All TaskKinds publish a TaskSignature with all fields declared and CHR‑typed; unknown supported for each.

2. CHR legality proven. Any numeric comparison/aggregation cites CG‑Spec by Characteristic id and proves CSLC legality; no mean on ordinals; no unit mixing.

3. Unknowns propagate. Unknowns must map to {pass|degrade|abstain} in Acceptance/Eligibility; no implicit coercions; behavior recorded in SCR.

4. Evidence lanes. A.10 anchors + Assurance lanes (TA/VA/LA) + freshness windows recorded; Γ‑fold default=weakest‑link unless proved otherwise.

5. ReferencePlane guarded. ReferencePlane noted per value and per ObjectiveProfile head; on crossings apply CL (and CL^plane if planes differ); Φ(CL)/Φ_plane are monotone, bounded, table‑backed and documented in the CG‑Spec; penalties → R_eff only (F/G invariant).

6. Acceptance thresholds live in CAL. No acceptance-gate thresholds in CHR or code paths; only in G.4 AcceptanceClauses.

7. Selector legality. Selection uses admissible (possibly partial) orders; weighted sums across mixed scale types are forbidden; return a Pareto set when appropriate.

8. Crossings published (visibility). Any cross-stance/cross-Context reuse emits BridgeCard/BridgeDescription + UTS row with CL notes and (if planes differ) CL^plane + Φ_plane.

9. UTS twin labels. All exported cards publish Name Cards with twin labels; Bridges carry loss notes.

10. GateCrossing checks. Published TaskSignature and any referenced crossings satisfy: (i) stance tagging (if used; informative only), (ii) CrossingBundle presence/consistency (E.18; A.27; F.9), (iii) LanePurity (CL→R only; F/G invariant; Φ tables present), and (iv) Lexical SD (E.10). Failures are blocking under the active GateProfile / GateChecks (A.21).

11. QD fields (when QD is in scope). If PortfolioMode=Archive or QD heads are present, CharacteristicSpaceRef (d>=2), ArchiveConfig (topology, resolution, K, InsertionPolicyRef, DistanceDefRef.edition), and EmitterPolicyRef SHALL be present and CHR-typed; characteristics declare ReferencePlane.

12. DominanceRegime default. DominanceRegime defaults to ParetoOnly; inclusion of illumination in dominance MUST be enabled by a CAL.Acceptance policy; the policy id SHALL be published in SCR.

13. Telemetry. PathSliceId, refresh/decay policies, and edition counters for CharacteristicSpaceRef/DistanceDefRef/EmitterPolicyRef SHALL be recorded; any illumination increase SHALL log the policy-id that triggered it.

14. GeneratorIntent (when OEE is in scope). GeneratorIntent SHALL cite EnvironmentValidityRegion and TransferRulesRef (ids resolvable in G.5/C.23); absence => Abstain for OEE dispatch.

15. Budgets. Budgeting (eval/time/batch) SHALL declare units and E/E-LOG exploration budget id when used.

16. Archive legality. DistanceDefRef.edition and any novelty measures SHALL be CSLC-lawful and editioned; illegal ops => Abstain.

17. Planes. ReferencePlane SHALL be declared for all QD heads/axes; plane crossings apply Phi_plane (penalty to R only).

18. Unknowns. Unknown QD fields map to {degrade|abstain|sandbox}; no coercions.

19. Specialization claims anchored. Any declared specialization on this TaskSignature SHALL name the task family/work target, named work-measure threshold target, adaptation budget, freshness or provenance basis for reuse, and enough attachment detail for C.22.1, G.5, and G.9 to consume the same claim lawfully.

#Interfaces & Data Paths

Inputs. ProblemProfile (…Description), CG-Spec ids, Evidence Graph Ref (A.10), D.CTX; (if QD) CharacteristicSpaceRef/ArchiveConfig/EmitterPolicyRef configs; (if OEE) GeneratorIntent. Produces. TaskSignature@Context (S2) with provenance; SCR-visible fields; UTS Name Cards for any minted traits; (if QD/OEE) archive/portfolio semantics and telemetry hooks declared. Consumed by. G.5 (Eligibility/Selection kernel), G.4 (Acceptance/Evidence), C.23 (admit/degrade/abstain rules; maturity ladder).

#Consequences (informative)

• Lawful selection. Dispatch is explainable and audit-ready; every reason in/out cites TaskSignature fields, CG-Spec rows, and Gamma-fold contributors.
• Local first, portable later. Context-local semantics are primary; Bridges make portability deliberate and costed (penalties to R only).
• Frictionless downstream. G.1-G.5 consume a single, typed Standard; thresholds are cleanly separated into Acceptance; unknowns are not guessed.
• QD/OEE-ready. Typed QD and GeneratorIntent fields make portfolio and open-ended workflows first-class, with lawful dominance, editioned distances, and policy-aware illumination.

#Relations

Builds on: C.16 MM-CHR, G.0 CG-Spec. Coordinates with: G.4 Acceptance, G.5 Selector, C.18 NQD-CAL, C.19 E/E-LOG, C.23 Method-SoS-LOG. Constrained by: E.10 (LEX/I/D/S), E.18 (GateCrossing visibility / publication gating).

#Practical reading checks

• If two candidate approaches are answering different TaskKinds or different ScopeSlice(G) cuts, this pattern does not yet license a direct comparison.
• If specialization is the live burden, the task-family anchor, threshold target, adaptation budget, and provenance basis should already be recoverable from the attached TaskSignature.
• If crossing, normalization, or missingness changes what comparison means, publish that in the signature and its cited refs rather than hiding it in code, local memory, or later prose.
• If QD or OEE heads are in scope, archive and generator fields belong in the same typed signature rather than in a detached explanatory appendix.

#Goldilocks hook (design‑time)

When generating candidate solutions for a TaskKind, target “goldilocks” slots (feasible‑but‑hard) so that the TaskSignature is informative (neither trivial nor impossible); this aligns with G.1 (target goldilocks, abductive provenance) and ensures the TaskSignature is informative (neither trivial nor impossible) for G.5 selection.

#C.22:End