measure-recursive-cell

name	measure/recursive-cell
description	A recursive coherence methodology for cnos cell architectures and cell executions. It measures the three-cell system, each cell's CCNF kernel, typed contract/evidence surfaces, cell and wave FSMs, and an actual receipt stream without allowing a strong local layer to average away a broken cross-level projection.
governing_question	Do the declared cell properties, their relations, their evolution, their internal kernels, their state machines, and the observed execution receipts still describe one executable coherence system?
artifact_class	measurement
scope	global
kata_surface	external
triggers	["cnos cell architecture measurement","WC PC CC ratification","recursive cell coherence","cell FSM coherence"]
inputs	["Immutable target revision and ordered L0-L4 bundle","Pinned TSC Core/Operational and scoring instruction","Cell/wave contracts, schemas, FSM tables, implementation seams, and receipts","Control-plane evidence snapshotted or content-hashed"]
outputs	["One immutable emission/ with six TSC prompts, one invariant-assessment prompt, and prompt-digests.json","One immutable publication/ snapshot with replay inputs and six validated canonical TSC reports","One publication/recursive-cell-run.json plus publication-success.json binding every exact path and digest"]
visibility	public
methodology	{"registry":"src/packages/cnos.cdd/skills/cdd/measure/recursive-cell/calibration/662/registry.tsc","calibration_preflight":"src/packages/cnos.cdd/skills/cdd/measure/recursive-cell/calibration/662/verify-target.sh","targets":["cc662-system","cc662-l0","cc662-l1","cc662-l2","cc662-l3","cc662-l4"],"cross_target":true,"instruction":"src/packages/cnos.cdd/skills/cdd/measure/recursive-cell/INSTRUCTION.md","path_base":"repository-root","activation":"source-checkout-only","target_root_contract":"explicit-coh-root","target_revision":"a0d39293a27cfe57b49dacff696345b1ee2cdb40","output_root":".tsc/cnos-recursive-cell","default_mode":"hybrid","execution":{"state":"State-A","runner":"src/packages/cnos.cdd/skills/cdd/measure/recursive-cell/runner/recursive-cell-runner.py","output_schema":"src/packages/cnos.cdd/skills/cdd/measure/recursive-cell/runner/recursive-cell-run.schema.cue","invariant_assessment_template":"src/packages/cnos.cdd/skills/cdd/measure/recursive-cell/runner/invariant-assessment-template.md","emits":["emission/prompts/cc662-system.md","emission/prompts/cc662-l0.md","emission/prompts/cc662-l1.md","emission/prompts/cc662-l2.md","emission/prompts/cc662-l3.md","emission/prompts/cc662-l4.md","emission/invariant-assessment-prompt.md","emission/prompt-digests.json"],"ingests":["cc662-system.json","cc662-l0.json","cc662-l1.json","cc662-l2.json","cc662-l3.json","cc662-l4.json","invariant-assessment.json"],"produces":["publication/emission/","publication/inputs/responses/","publication/inputs/invariant-assessment.json","publication/reports/cc662-system.json","publication/reports/cc662-l0.json","publication/reports/cc662-l1.json","publication/reports/cc662-l2.json","publication/reports/cc662-l3.json","publication/reports/cc662-l4.json","publication/recursive-cell-run.json","publication/publication-success.json"],"state_b":"unshipped"},"consistency":{"mechanical":"identical","llm_repeats":3,"llm_spread":"For each target, compute max absolute pairwise spread over alpha, beta, gamma, the three delta fields, and confidence; map max spread through phi(d)=d/(1-d), Coh_consistency=exp(-phi(d)). Report both max and mean-pairwise forms. A single semantic sample has no standing."},"standing":{"scope":"house-authored-public-commons","admissibility":"public-only","heldout_status":"none","external_anchor_count":0,"llm_consistency_gate":"reported-not-gating","llm_consistency_floor":0.9},"mechanical":{"backend":"usurobor/tsc@26aab5023f03dc7d0abf82e5fdba20134fc6adad:engine/ocaml/lib/mechanical_scoring.ml","determinism":"Identical ordered file bundle, target manifest, TSC revision, and parameters must produce identical numeric scores and axis-detail signals over at least three runs.","signals":{"alpha":["alpha.terminology_consistency","alpha.repeated_structure","alpha.duplicate_definition_tension","alpha.naming_drift"],"beta":["beta.cross_reference_consistency","beta.authority_alignment","beta.source_of_truth_alignment","beta.target_file_fit"],"gamma":["gamma.canonical_generated_distinction","gamma.version_surface_consistency","gamma.traceability_presence","gamma.authority_evolution_consistency"]}},"llm":{"estimates":["target","alpha","beta","gamma","delta_alpha_beta","delta_beta_gamma","delta_gamma_alpha","bottleneck_axis","confidence","summary","axis_evidence","defect_cards","unresolved_ambiguity","next_fixes"],"must_not":["compute Coh or C_sigma; the engine owns barrier and aggregate computation","infer files outside the frozen target bundle","treat specified or illustrative behavior as shipped","average away a failed hard invariant or cross-level projection","use the CM being created or modified as its own independent warrant"],"validation":"Validate the standard TSC v3.2.4 witness contract, the fixed checklist and defect-card correspondence, L0-L4 completeness, hard-invariant coverage, FSM totality coverage, and target/prompt digest equality.","providers":{"local":"TSC external-response route using a frozen emitted prompt","ci":"Any pinned witness CLI restricted to the emitted prompt and one JSON response"},"ci_prompt":"Read only the frozen prompt for the named L0-L4 target. Apply INSTRUCTION.md, emit exactly one standard TSC v3.2.4 JSON witness, and write no other artifact. Do not compute Coh or C-sigma."}}

cnos Recursive Cell Coherence Methodology v0.3.0

0. Status and epistemic boundary

This is a candidate CM derived from cnos#662. That cycle is its calibration source, not a held-out anchor. The CM may describe and reproduce the #662 measurement, but that result gives this CM no independent standing. Standing requires at minimum:

1. three validated same-route semantic samples over a frozen prompt;
2. Coh_consistency >= 0.90;
3. a public calibration commons containing positive and adversarial cases;
4. at least one held-out cnos cell architecture not used to design this CM;
5. CM-of-CMs/admissibility measurement of this artifact.

The current TSC CLI cannot consume this arbitrary CM declaration as a first-class instrument. The shipped deterministic State-A runner invokes the six registered #662 targets through one materialized INSTRUCTION.md, validates the six standard TSC witnesses, emits canonical TSC reports, validates a separate typed H01-H13 assessment, and produces one CUE-validated aggregate. Arbitrary State-B declaration and target loading remains explicitly unshipped.

All declared authority paths use path_base: repository-root; --root names the explicit cnos target checkout. This v0.3 candidate is therefore source-checkout-only: an installed cnos.cdd package must refuse activation because TSC does not yet expose separate authority-root and target-root bases. resolve-authority.sh enforces that boundary and the package smoke tests the refusal. A new application must materialize its own immutable registry by specializing these manifests; the bundled registry is calibration evidence, not a wildcard production target.

State-A also treats filesystem path identity as evidence custody. The output root and every existing component in its lexical path must be nonsymlinked before the root is created. Each of the six response paths and the separate invariant-assessment path must be a regular file with no symlink component before staging and immediately before snapshot. Internal emission, publication, staging, and lock paths retain their independent symlink guards.

I5 parses the 0.1 registry/manifests with the pinned TSC semantics, checks exact target/name/path equality, expands each target to a non-empty bundle at the declared historical revision, and checks the exact v3.2.4 top-level semantic output field list. The mechanical.signals entries below are a cnos taxonomy, not a claim that I5 proves an unversioned cross-repository engine signal schema.

1. Measurement object and recursion

The object is not one Markdown file. It is a recursive system observed at five levels. Every level is articulated symmetrically as α/pattern, β/relation, and γ/process. “Dominant telos” never exempts a level from either other axis.

Level	System under measurement	α — properties/pattern	β — relations	γ — process/evolution
L0	Three-cell agent	WC/PC/CC definitions and output types	handoffs, authority, next-MCA, ε scope lift	intent→plan→work→judgment loop and system evolution
L1	One WC, PC, or CC	contract, α/β/γ/V/δ components, role boundaries	signatures, evidence flow, firebreaks, scope projection	production→review→closure→validation→decision and repair recursion
L2	Typed surfaces	schemas, tags, locators, receipt forms, CM result	contract↔matter↔review↔measurement↔receipt↔V↔δ	versioning, binding, provisional/final boundaries, migrations
L3	Runtime/FSM	state/event/guard/action vocabularies	transition authority, command/table parity, class/domain routing	complete cell lifecycle, wave lifecycle, recovery, replan, termination
L4	Concrete execution	actual matter, receipts, review artifacts, claimed identities	exact-SHA and role/evidence links	round history, monotonic repair, mutability, exit behavior

The recursion closes only if both conditions hold:

• within-level coherence: each Lk is coherent under α/β/γ;
• cross-level coherence: every authoritative Lk claim has a lossless, non-contradictory realization or explicitly owned migration at Lk+1.

Overall coherence is the geometric mean of per-level results only after every hard invariant passes. A failed hard invariant forces a non-accepting disposition regardless of a high mean.

2. L0 — three-cell system

2.1 Classifier

Class is determined by the canonical requested output and its operational consumer, never by filename, prose topic, or matter domain:

• WC: canonical output is a repository/product artifact.
• PC: canonical output is executable relational structure—a D0 relation contract or wave graph whose consumers can derive work/dependencies.
• CC: canonical output is an evidence-backed state/process judgment with one disposition and, when required, a next-MCA.

A Markdown file may project any class for humans. If the Markdown itself is the normative merged deliverable, it is an artifact and therefore WC. Mixed outputs must split into cells or declare one canonical typed output and mark every other surface as a derived projection.

2.2 Relations

Measure all edges, including:

intent → PC-D0? → WC/PC-Wave → CC judgment → V → δ/FSM → next cell/human gate
closed cells/receipts → ε observation → CC at n+1

Every edge must name source type, target type, immutable identity, authority, and failure behavior. A narrative arrow is not an executable relation.

2.3 Process

Check that system evolution cannot silently change class, domain, authority, or measurement instrument. A CM must be selected and pinned before its result is used as warrant. A CC may request creation or repair of a CM; it may not invent the CM that retroactively justifies the same judgment.

3. L1 — internal cell kernel

For every WC/PC/CC instance, measure:

matter := α(contract)
review := β(contract, matter)
measurement := CM(pinned bundle)
receipt := γ.bind(contract, matter, review, measurement, evidence)
verdict := V(contract, receipt)
decision := δ(receipt, verdict)

This is the intended non-circular form. If source doctrine instead types a receipt before measurement/V/δ while requiring those outputs inside that same receipt, record a blocking type cycle. A conforming design may use a typed receipt_core followed by a final receipt, but both forms and their hash edge must be explicit.

Hard firebreaks:

• α and β are distinct for the matter under review.
• κ is outside the cell and is not equal to α/β/γ/δ.
• shared account/model hosting is disclosed separately from functional-role identity; authorization does not turn unequal role types into equality.
• CC does not implement its recommendation.

4. L2 — contracts, CM, evidence, and receipts

The cell contract must pin:

measurement_policy:
  cm_ref: { immutable_revision, path_or_id, content_digest }
  target_projection_ref: { immutable_revision, content_digest }
  mode: mechanical | llm | hybrid | auto
  thresholds_ref: { immutable_revision, content_digest }
  standing_required: <typed policy>

The measurement result must bind:

measurement:
  cm_ref: <same immutable CM>
  target_bundle_hash: <exact ordered bytes>
  prompt_or_config_hash: <exact instrument invocation>
  scores: { alpha, beta, gamma }
  pair_deltas: { alpha_beta, beta_gamma, gamma_alpha }
  bottleneck: <one axis>
  witnesses: <TSC operational witnesses>
  consistency: <sample count and spread>
  provenance: <engine/model/version/time>

The receipt binds the measurement artifact by immutable digest. V reads the measurement only through the receipt. CM measures; V gates; δ effects.

External issue/comment identities never bind content by themselves. Every control-plane input needs a content hash plus retrievable snapshot or an immutable repository artifact.

5. L3 — FSM measurement

FSM coherence is triadic at two scopes.

5.1 Cell FSM

• α: one state/event/guard/action vocabulary; terminal and exceptional states are typed; no state is simultaneously “not a state” and a reachable state without a distinct terminal type.
• β: exactly one authority owns each transition. Commands submit typed requests to the table/controller; they do not implement competing transition paths. Guards match contracts, V, δ, labels, and receipts.
• γ: every nonterminal state has an owned exit or explicit wait condition; repair, recovery, rejection, override, and resumption preserve history.

Required mechanical checks:

1. state closure: every target state is declared or typed terminal;
2. totality: every declared state/event has an explicit result;
3. determinism: at most one rule wins for a fact snapshot, or priority is normative and overlap-tested;
4. reachability and no accidental dead states;
5. authority uniqueness per transition;
6. positive/negative guard fixtures;
7. command↔table parity;
8. class/domain orthogonality;
9. State-A→State-B migration preserving live cycles.

5.2 Wave FSM

Apply the same checks plus:

1. total mapping from CC dispositions to wave transition requests;
2. child receipt/completion aggregation rule;
3. dependency/gate prerequisites for dispatchability;
4. holding and replanning exits;
5. completion and complete-with-residuals predicates;
6. operator authority only at declared irreversible/scope-expanding gates;
7. no CC mutation of wave state—the FSM owns transition.

A sequence of state names is not an FSM. It must define events, guards, actions, authority, invalid transitions, and terminal semantics.

6. L4 — execution and receipt stream

Bind the exact matter bytes, complete receipt directory, review-content snapshots, PR/issue history snapshots, checks, and state timeline. Check:

• the instance's output type matches its class;
• role claims match actual authorship/review;
• every review names the exact matter SHA;
• γ binds both matter and immutable review bytes;
• repairs narrow or explain any widening;
• historical artifacts are explicitly superseded;
• State-A limitations are true and do not redefine target invariants;
• the emitted disposition is within CC authority.

7. Cross-level hard invariants

These are gating, not averaged. Their stable identifiers are part of the State-A output contract:

1. H01: one CCNF kernel for all classes;
2. H02: class determined by canonical output telos;
3. H03: class and matter domain orthogonal;
4. H04: α≠β;
5. H05: κ outside and κ≠α/β/γ/δ;
6. H06: ε is an observation/projection, not a cell role or CC itself;
7. H07: immutable matter, review, CM, target-bundle, measurement, and receipt links;
8. H08: CM/V/δ separation with a type-correct data path;
9. H09: single dependency authority;
10. H10: single transition authority per FSM edge;
11. H11: complete wave-FSM semantics before “executable” claims;
12. H12: honest shipped/specified/illustrative partition;
13. H13: CC judgment distinct from FSM transition and operator acceptance.

The six TSC witnesses remain exact, unextended TSC v3.2.4 objects. A separate invariant assessment contains exactly H01-H13 with pass, fail, or unknown, specific evidence, level, primary axis, and next MCA. It binds the generated assessment prompt, all six TSC prompt digests, and the exact six response digests. unknown is nonaccepting. A failed item must cross-reference a systemic defect card in the standard witness for its stated level and primary axis using [Hxx].

8. Aggregation and disposition

TSC computes canonical C_sigma for system and L0-L4. The declared cross-level aggregate is the unweighted geometric mean of L0-L4 canonical C_sigma; the system report is retained but not averaged into that value. The runner emits exactly one bottleneck: the lowest numeric L0-L4 C_sigma, ties resolved by level order L0→L4, with the axis taken directly from that level's canonical TSC bottleneck_axis. It does not re-derive the axis from raw axis scores.

State-A disposition rules are deterministic:

• any failed H09-H11 emits request_planning;
• failed H12 emits request_working unless the planning rule already applies;
• any other failure or any unknown emits hold;
• all thirteen passing still emits hold, because this calibration source has zero independent standing.

The broader methodology vocabulary remains reserved for a future route with standing:

• complete: every hard invariant passes and formal TSC Operational ACCEPT has standing.
• complete_with_residuals: invariants pass; only bounded non-gating debt remains and standing policy permits it.
• hold: target or assurance is frozen but one or more repairable hard invariants fail.
• request_planning: relations/FSM/migration require redesign before work.
• request_working: design is coherent but an implementation artifact is absent or defective.
• request_human: methodology/authority choice is genuinely undecidable.
• block: continuing would destroy evidence, violate authority, or act on a moved target.

For ratification, a high score cannot override a failed invariant. Name the invariant, evidence, level, and minimum next MCA.

9. Calibration commons required before standing

At minimum:

• positive: coherent WC artifact cycle;
• positive: coherent PC wave with complete typed child contracts;
• positive: coherent CC judgment with pinned CM and immutable receipt chain;
• negative: PC emits canonical artifact instead of relation graph;
• negative: κ/α or α/β role equality;
• negative: receipt requires its own downstream outputs;
• negative: mutable comment ID used as content revision;
• negative: wave state list without transitions/guards;
• negative: command and table own the same transition differently;
• comparative: each positive must outrank its paired negative.

Until these are independently authored or held out, the CM remains house-authored-public-commons with no off-diagonal standing.