#Discipline‑CHR - Field Health & Structure

Pattern C.21 · Stable Part C - Kernel Extension Specifications

Purpose. Give FPF a typed, auditable way to speak about the health, maturity, and structure of a scientific/engineering discipline, without collapsing into taste, anecdotes, or single-number scores. The pattern defines a portable set of Characteristics and guards (legality, freshness, scope) that any Context can specialize.

This pattern supplies the CHR “vocabulary of health” for disciplines. C.20 composes the discipline; C.21 measures its health; Part G (G.2, G.12) harvests SoTA and operationalizes dashboards; Bridges keep meaning honest; penalties touch R only.

Status & placement. Part C (Kernel Extention Specifications) → Cluster C.I (Core CHRs/CALs). Depends on: MM-CHR (C.16), KD-CAL (C.2), USM/Scope (A.2.6), Trust & Assurance (B.3), E.10 (LEX‑BUNDLE). Coordinates with: C.20 Discipline‑CAL (what a U.Discipline is), G.2 (SoTA palette), G.12 (dashboard), G.0 (CG‑Spec registry).

FPF treats disciplines as first-class holons (see C.20): they aggregate epistemes, practices, standards, institutions, and observed Work. Teams routinely say “the field is fragmented,” “standards are converging,” or “replication is improving,” but these claims are rarely typed (scale/unit/polarity) or auditable (evidence lanes, freshness, scope). C.21 supplies the CHR layer—named Characteristics with CSLC typing—so disciplines can be compared lawfully (CG‑Spec) and monitored through time (G.12). Each published value MUST declare ReferencePlane ∈ {world|concept|episteme} and DisciplineId (U.Discipline@UTS); cross‑plane use applies CL^plane in Assurance (penalty to R_eff only).

#Keywords

• discipline
• field health
• reproducibility
• standardisation
• alignment
• disruption.

#Relations

#Content

#Problem Frame

FPF treats disciplines as first-class holons (see C.20): they aggregate epistemes, practices, standards, institutions, and observed Work. Teams routinely say “the field is fragmented,” “standards are converging,” or “replication is improving,” but these claims are rarely typed (scale/unit/polarity) or auditable (evidence lanes, freshness, scope). C.21 supplies the CHR layer—named Characteristics with CSLC typing—so disciplines can be compared lawfully (CG‑Spec) and monitored through time (G.12). Each published value MUST declare ReferencePlane ∈ {world|concept|episteme} and DisciplineId (U.Discipline@UTS); cross‑plane use applies CL^plane in Assurance (penalty to R_eff only).

#Problem

Narrative health claims cause three recurrent failure modes:

1. Illegality. Averaging ordinals, mixing units, or comparing incommensurate Contexts ⇒ nonsense roll-ups.
2. Staleness. Health “scores” rarely declare freshness windows or evidence lanes (TA/VA/LA).
3. Scope slippage. “The field” is left implicit; cross-Context reuse lacks Bridges & CL, leading to silent semantic loss. Any numeric comparison/aggregation MUST cite a CG‑Spec row (characteristics, lawful ScaleComplianceProfile (SCP), Γ‑fold, MinimalEvidence) before computation.

#Forces

#Solution — Discipline Health Characterisation (DHC)

#Ontology quick sheet (normative, clarifying)

What “DHC” is. DHC is a CHR vocabulary pack (intensional) that defines Characteristics + Scales/Units/Polarity for discipline health; it is not a document or a run. Artifacts. • U.DHCPack (I‑layer name; published as an episteme): the slot set (Characteristic/Scale declarations) for a Context.
• U.DHCMethodSpec (S‑layer): the computational specification(s) for deriving each DHC slot (e.g., replication‑window definition, CD‑index class), table‑backed; multiple per slot allowed, editioned separately.
• U.DHCSeries (episteme w/ EditionSeries): a time‑indexed publication of computed DHC readings for a Discipline×Context, each value bound to …Ref.edition for every referenced method/metric/distance. Edition subjects.
(i) DHCPack.edition — when the slot semantics (Characteristic/Scale) change.
(ii) DHCMethodSpecRef.edition — when a computation method (formula/class/policy) changes.
(iii) DHCSeries.edition — when the published series changes its content (not carriers).
Publication. Releases are Work on Carriers; no edition change unless content changes per U.EditionSeries.
Ref discipline. All bindings to packs/methods/distances SHALL use …Ref.edition (dot on the Ref).

Define a portable minimal set of CHR slots. Each slot is CHR-typed (Characteristic, Scale/Unit/Polarity per A.17–A.18), Context-local, and guarded by USM (Claim scope G), freshness windows, and evidence lanes (TA/VA/LA). Contexts MAY extend the set; MUST NOT alter scale types illegally.

“Health” is a vector of CHR‑typed coordinates; no single scalar is implied. Lawful scalarization lives in Acceptance (G.4) under an explicit CG‑Spec ScaleComplianceProfile (SCP) and Γ‑fold rules, and is never embedded in CHR.

#Core Characteristics (kernel-portable names)

1. ReproducibilityRate (ratio ∈ [0,1]; polarity ↑; ReferencePlane=episteme; CG‑Spec‑bound) Fraction of tested claims/benchmarks that independent teams replicate under a declared ContextSlice and Γ_time window. Lane tags: LA (validation) with TA (typing) for protocols.

2. StandardisationLevel (ordinal; polarity ↑; ReferencePlane=episteme) {none, emerging, de facto, de jure}. No mean. Use medoid/mode; legal comparisons are ≤/=/> only. Tracks convergence on vocabularies, interfaces, or procedures.

3. AlignmentDensity (ratio; polarity ↑; ReferencePlane=episteme; CG‑Spec‑bound)
   Density of Substitution Bridges (same senseFamily, CL≥2) between major U.Traditions per 100 DHC‑SenseCells (G.2 F‑hooks) in the SoTA palette. Substitution rule: free substitution permitted at CL=3; at CL=2 substitute only with extra‑guard (count in reporting, but this is not «free substitution») Units: bridges_per_100_cells. Cross‑Context use requires Bridge+CL; penalties → R_eff only.

4. DisruptionBalance (interval; polarity = target band; ReferencePlane=episteme; CG‑Spec‑bound)
   Relative share of disruptive vs consolidating works within Γ_time using a registered CD‑index class (editioned; cite method id in UTS). Default plane: episteme. Publish the target band via Acceptance (G.4); not in CHR.

5. EvidenceGranularity (Context-declared: ordinal|ratio; polarity ↑; ReferencePlane=episteme)
   If ratio: units = claims_per_artifact or anchors_per_claim (declare). If ordinal: publish level names and ORD_COMPARE_ONLY. Fineness of evidential units and declared envelopes (experiment cards, benchmark tasks, audit granules). Encourages smaller, well-scoped claims over monoliths.

6. MetaDiversity (portfolio dispersion; polarity ↑ to band; ReferencePlane=episteme; CG‑Spec‑bound)
   Use entropy/HHI over MethodFamily/Tradition shares (method edition id in UTS); publish guard‑band as Acceptance binding; cross‑ordinal scalarisation is forbidden. Entropy/Herfindahl-type dispersion across U.Traditions, method families, or data regimes, bounded by a Context-declared guard-band (too low ⇒ monoculture; too high ⇒ incoherence).

Typing & legality. Each slot MUST declare Scale/Unit/Polarity; illegal ops (e.g., mean on ordinals; unit mixing) are fail-fast per A.18/MM-CHR.

#Guard Macros (normative)

• ORD_COMPARE_ONLY(x) — for StandardisationLevel (ordinal).
• UNIT_CHECK(x) — forbid cross-unit aggregation (AlignmentDensity, ReproducibilityRate).
• POLARITY_CHECK(x) — enforce declared polarity (↑/↓/target-band) per MM‑CHR.
• FRESHNESS(x; window) — ensure values come from evidence within declared Γ_time; record valid_until; stale ⇒ {degrade|abstain} at Acceptance.
• PLANE_NOTE(x) — record ReferencePlane; compute CL^plane on crossings; penalties → R_eff only.
• LANE_TAGS(x; {TA|VA|LA}) — annotate contribution lanes.
• SCOPE_COVERS(x; TargetSlice) — enforce USM coverage of the computation.
• BRIDGE_CL(x; id, CL≥k) — on cross‑Context roll‑ups, require Bridge with CL; penalties route to R only. If planes differ, apply CL^plane and cite Φ_plane policy id. Hint: for AlignmentDensity reporting, set k=2 (CL≥2); CL=3 counts as free substitution.

#Legality Matrix (extract)

Note: For MetaDiversity/EvidenceGranularity (ordinal) use median/mode; forbid affine ops; unit mix always fails.

#Interfaces & Data Paths

• Inputs. U.Discipline from C.20 (composition), SoTA Palette/BridgeMatrix from G.2 (DHC‑SenseCells included), EvidenceProfiles from G.4/G.6.
• Outputs. Per‑Context DHC rows (these six slots), UTS Name Cards with twin labels (E.5/F.17–F.18), Registry/RSCR hooks on method edition changes; feeds G.12 (time‑series).
• Cross-Context reuse. Only via F.9 Bridges with CL and loss notes; Φ(CL) penalties applied to R (never F/G).

#Archetypal Grounding (three fields)

#Computer Vision (Benchmarks 2015→)

• ReproducibilityRate. Ratio of independently reproduced results on ImageNet-style tasks within rolling 24 mo (LA lane).
• StandardisationLevel. de facto for dataset specs and metrics in Vision_2024; emerging for robustness protocols.
• DisruptionBalance. Use an editioned CD‑index class (e.g., Wu‑style disruption family) with method id; publish target band via Acceptance; annotate ReferencePlane=episteme.
• AlignmentDensity. Bridges with CL≥2 across sub-traditions (supervised vs self-supervised).
• MetaDiversity. Entropy across method families (CNN/ViT/Hybrid) kept within guard-band to avoid monoculture.

#Biomedicine (Gene–Disease Associations)

• ReproducibilityRate. Fraction of associations replicated in independent cohorts within Γ_time(36 mo); LA lane with TA (typing of protocols).
• StandardisationLevel. de jure for certain reporting guidelines; emerging for pre-registration norms.
• EvidenceGranularity. Move from “paper-level” to claim-level units (Context raises the score).
• DisruptionBalance. Target band discourages sustained “novelty spikes” unbacked by replication.

#Software Performance Engineering (SPE)

• StandardisationLevel. emerging → de facto for SLO taxonomies and trace schemas across vendors.
• AlignmentDensity. CL-rated Bridges between tracing ecosystems.
• ReproducibilityRate. Share of publicly replicable perf claims in rolling windows.
• MetaDiversity. Balance across load models, failure modes, and toolchains.

#Decision‑Making (2015→)

• ReproducibilityRate — share of causal effect estimates replicated across independent datasets within Γ_time; LA lane. • StandardisationLevel — emerging for identification checklists; de facto for SCM notation in leading stacks (ordinal; no means). • AlignmentDensity — CL‑rated Bridges between SCM/DoWhy‑style and RL/BO traditions per 100 DHC‑SenseCells. • MetaDiversity — dispersion across method families (SCM/RL/BO/DT) within guard‑band; entropy/HHI (units declared in CG‑Spec).

#Evolutionary Architecture (software)

• ReproducibilityRate — fraction of architecture fitness results reproduced on independent workloads (rolling 18–24 mo; LA lane). • StandardisationLevel — de facto for ADR/ATAM patterns; emerging for continuous fitness protocols. • AlignmentDensity — Bridges across ATAM/SAAM/ADR style guides (CL≥2) normalised per 100 DHC‑SenseCells. • MetaDiversity — portfolio dispersion across patterns (microservices, event‑driven, layered) with guard‑bands; no ordinal arithmetic.

#Measurement & Publication Procedure (authoring harness)

1. Declare Context & TargetSlice. (USM) Name editions, Standards, env params, Γ_time.
2. Collect evidence. Bind sources via G.6 EvidenceGraph; tag lanes and freshness.
3. Compute DHC slots. Enforce Legality Matrix and Guard Macros.
4. Bridge (if needed). Map via F.9; attach CL and loss notes; apply R penalties.
5. Publish to UTS. Name Cards (Tech/Plain), twin labels; bind DHCMethodSpecRef.edition, DistanceDefRef.edition, and, where templates are used, DHCMethodRef.edition; register RSCR triggers (method change, ScoringMethod/NormalizationMethod edits).
6. Dashboard. Feed G.12 with time-series and guard-bands (disruption, diversity).

#Bias-Annotation (E-cluster lenses)

• Didactic. Plain names + twin labels; one-screen tables for managers.
• Architectural. No ordinals averaged; all cross-Context movement goes through Bridges+CL; penalties never touch F/G.
• Pragmatic. Freshness-aware; unknowns tri-state; values are decision-support, not trophies.
• Epistemic. Evidence lanes explicit; reproducibility is LA, typing is TA; validation distinct from verification in dashboards.

#Conformance Checklist (normative)

CC-C.21-1 (CHR typing). Every DHC slot MUST declare Characteristic + Scale/Unit/Polarity, with CSLC legality proved before any aggregation. CC-C.21-2 (Freshness). Published values MUST carry Γ_time selector and freshness window; stale rows escalate to {degrade|abstain} in G.4 Acceptance. CC-C.21-3 (Plane). ReferencePlane declared; cross‑plane re‑use publishes CL^plane (policy id) alongside CL; both penalties route to R_eff. CC‑C.21‑4 (DesignRunTag). Every DHC row SHALL declare DesignRunTag ∈ {design, run}; design‑ and run‑characteristics not mixing in one value/aggregate. CC-C.21-5 (Lane tags). Each value MUST tag TA/VA/LA lanes of contributing evidence. CC-C.21-6 (Ordinal discipline). StandardisationLevel is ordinal; no means, no z-scores. CC-C.21-7 (Scope). All computations declare TargetSlice; USM membership is decidable and deterministic. CC-C.21-8 (Bridges). Cross-Context comparisons/publishers MUST cite Bridge id + CL; penalties route to R_eff, never to F/G. CC-C.21-9 (UTS). Publish DHC rows as UTS Name Cards with twin labels (Tech/Plain). CC‑C.21‑10 (Registry). DHC methods are table-backed; silent method changes are forbidden (bump DHCMethodSpecRef.edition + RSCR trigger). CC-C.21-11 (Unknowns). Unknown inputs propagate tri-state {pass|degrade|abstain} to Acceptance; no unknown→0 coercion. CC-C.21-12 (No tool/vendor tokens). Core narrative follows E.5.1 (Lexical Firewall). CC-C.21-13 (CG‑Spec citation). Any numeric operation (comparison/aggregation) in DHC MUST refer to CG‑Spec (characteristics, ScaleComplianceProfile (SCP), Γ‑fold, MinimalEvidence). CC-C.21-14 (Φ‑policies). Φ(CL) and Φ_plane — monotone and table‑backed; published by policy id. CC‑C.21‑15 (Ref discipline). Any edition pinning SHALL appear as …Ref.edition on the relevant reference field (DHCPack/MethodSpec/DistanceDef/DHCMethodRef); bare …Edition fields are non‑conformant. CC‑C.21‑16 (Role kit, informative). Use standard roles from F.4: DisciplineStewardRole (governs DHCPack), DHCMethodAuthorRole, DHCSeriesPublisherRole. Roles are design‑time; values are run‑ or design‑stance per slot and must declare ReferencePlane.

#Consequences

Benefits. Lawful comparisons; freshness-aware governance; explicit cross-tradition alignment; dashboards that don’t lie by averaging ranks. Costs. Some ceremony (scales, windows, lanes, bridges), offset by template macros and UTS automation. Risks avoided. “Phlogiston disciplines” (charisma-driven fields) fail DHC audits; No-Free-Lunch preserved by G.5 (selector returns sets, not universal scalars).

#Rationale (post-2015 signals & practice)

• Replication & credibility (2015→). Field-level health in SciSci emphasizes replicability, fresh evidence windows, and claim-level units—captured by ReproducibilityRate and EvidenceGranularity.
• Disruption vs consolidation (2019→). Empirical “disruption indices” distinguish papers that open new lines from those that refine—hence DisruptionBalance with target bands, not monotone “more is better.”
• Standardization waves (2016→). Package/ecosystem norms show ordinal trajectories (none→emerging→de facto→de jure); ordinal typing prevents illegal arithmetic.
• Plural traditions (ongoing). Mature fields maintain bridges with explicit loss notes; AlignmentDensity rewards CL-rated bridges without semantic collapse.

(Names are illustrative of contemporary practice; the CHR is notation-agnostic and tool-neutral.)

#Relations

• Builds on: A.17–A.18 (Characteristic/CSLC), A.2.6 (USM scopes), B.3 (assurance lanes), C.16 (MM-CHR templates).
• Coordinates with: C.20 (what a U.Discipline is), G.2 (SoTA palette and BridgeMatrix), G.12 (Dashboard operationalization), G.9 (parity harness for fair comparisons).
• Constrains: G.10 (pack ships DHC rows + method ids), G.11 (refresh windows/decay), G.5 (selector may reference DHC only via admissible predicates; no cross‑ordinal scalarisation). Coordinates: F.9 (Bridges for cross‑Tradition comparisons).

#Annex — Author’s quick template (copy-paste)

C.21.DHC(Context: <name/edition>; TargetSlice: <tuple>; Γ_time: <policy>)
  ReproducibilityRate:
    value: <0..1>   lane: LA   window: <…>   scope: <…>
  StandardisationLevel:
    value: {none|emerging|de_facto|de_jure}   compare_only: true
  AlignmentDensity:
    value: <ratio>   units: bridges_per_100_DHC_SenseCells   CL_min: 2   scope: <…>
  DisruptionBalance:
    value: <−1..1>   method: <CD-index class / edition>   target_band: [l,u]
  EvidenceGranularity:
    value: <ordinal|ratio per Context>   notes: <…>
  MetaDiversity:
    value: <entropy/HHI>   target_band: [l,u]
Guards: ORD_COMPARE_ONLY(StandardisationLevel), UNIT_CHECK(*), FRESHNESS(*), LANE_TAGS, SCOPE_COVERS, BRIDGE_CL(if x-Context)
Publish: UTS twin labels; RSCR triggers on method edition change.

#C.21:End