PUBLIC SPECIFICATION

CIRE separates structure, publication, and reliability.

CIRE 2.0 requires every model to declare what its probabilities mean, computes structural signals only where they are defined, gates publication on every risk dimension without letting one offset another, and states reliability only from confirmed outcomes. Every example on this page is generated by the reference engine and can be reproduced with the published formula.

Overview CIRE Standard Model Cards Developers Passive Signals Network Learning Population Intelligence PetPass Edge Box

JSON specification Reference formula (.py)

Version

2.0.0

Specification

Public

Reference implementation

MIT

Generated examples

14

Three states that must never be merged

structural_state

CPS band. Structure only.

NOMINAL

CPS < 0.25.

No structural floor.

REVIEW

0.25 ≤ CPS < 0.5.

Structural floor of REVIEW.

CAUTION

0.5 ≤ CPS < 0.75.

Structural floor of CAUTION.

SUPPRESSED

CPS ≥ 0.75.

Structural floor of SUPPRESSED.

UNSUPPORTED

Concentration is not defined under the declared probability semantics.

No structural value is published; the semantics floor applies.

publication_state

What the runtime may do with the output.

NOMINAL

No risk dimension set a floor.

Published as decision support; clinician confirmation is still required and the reliability_state is shown with it.

REVIEW

At least one dimension requires clinician review.

Published for clinician review only; treatment guidance is withheld.

CAUTION

At least one dimension (for example demonstrated miscalibration or a physiological contradiction) requires caution.

Evidence-only output; treatment guidance is withheld.

SUPPRESSED

Invalid probability semantics or a dimension at its suppression threshold.

The clinical answer is not published as actionable; human review or better input is required.

reliability_state

What confirmed outcomes support.

UNVALIDATED

No confirmed outcomes exist for the clinical slice.

Shown with every output from the slice. No reliability claim may be made.

INSUFFICIENT_EVIDENCE

Some confirmed outcomes exist, but support, coverage, recency, or metric availability is below threshold.

No reliability claim may be made; blockers name each unmet threshold.

REVIEW_REQUIRED

Evidence is sufficient in volume but calibration, mechanistic consistency, or semantics fail a threshold.

Demonstrated miscalibration also sets a CAUTION publication floor.

VALIDATED_FOR_SLICE

Every reliability threshold is satisfied for this tenant × model version × species slice.

The only state that supports a slice-scoped reliability statement. It is not external or clinical validation.

Probability semantics (required)

categorical_exhaustive

Mutually exclusive hypotheses whose values sum to 1 (±0.02). Otherwise the output is invalid and is suppressed.

categorical_topk_with_other

The top-k slice of a categorical distribution. The unreported remainder is an explicit other = 1 − sum(top_k) bucket; it is never renormalized away. Mass above 1 + 0.02 is invalid.

independent_multilabel

Independent marginals. Categorical concentration is undefined, so no structural signal is published and publication is capped at REVIEW.

rank_scores

Ordinal scores. Never treated as probabilities; no structural signal; publication capped at REVIEW.

unknown

Undeclared. Treated conservatively: no structural signal; publication capped at REVIEW.

Formulas

differential_concentration

0.0 to 1.0, or null when the declared semantics are not categorical or are violated

Differential concentration (field alias: phi_hat)

s = top 5 named masses (descending) + one residual bucket (explicit other + all lower-ranked mass); concentration = 1 − H(s) / ln(6), H(s) = −Σ s_i ln s_i

How concentrated the ranked hypothesis distribution is. Structural only: it does not measure correctness, calibration, evidence quality, or clinical reliability. The normalizer is fixed, so appending negligible hypotheses cannot inflate it.

input_m_hat

0.0 to 1.0

Input integrity risk

input_m_hat = 0.40 × missingness + 0.35 × contradiction_rate + 0.25 × out_of_envelope_rate

Higher values mean the clinical input is incomplete, internally contradictory, or outside the supported operating envelope.

delta_rolling

real number

Rolling concentration drift

delta_rolling = EMA(concentration_t − concentration_(t−1), alpha = 0.1); state computed under another concentration policy is discarded, never mixed

Short-horizon movement of the structural signal.

sigma_delta

0.0 and above

Rolling concentration volatility

sigma_delta = population standard deviation of the last 50 concentration deltas

Higher values mean less stable runtime structure.

cps

0.0 to 1.0, or null when concentration is null

Runtime perturbation score (CPS)

cps = clamp01(0.40 × (1 − concentration / phi0) + 0.35 × max(0, −delta_rolling) / phi0 + 0.25 × sigma_delta / phi0)

Places the output in a structural band. It does not measure diagnostic correctness.

Publication policy (cire-publication-policy-v2)

Each dimension independently sets a minimum publication state; the result is the most severe one. Invalid probability semantics suppress the output; unsupported semantics cap it at REVIEW. Demonstrated miscalibration sets a CAUTION floor. Missing outcome evidence sets no publication floor and is disclosed through reliability_state instead.

Dimension	REVIEW at	CAUTION at	SUPPRESSED at
structural_risk	≥ 0.25	≥ 0.5	≥ 0.75
input_integrity_risk	≥ 0.25	≥ 0.5	≥ 0.75
mechanistic_risk	≥ 0.2	≥ 0.4	≥ 0.8
domain_safety_risk	≥ 0.25	≥ 0.5	≥ 0.75

Reliability thresholds (cire-reliability-thresholds-v1)

VALIDATED_FOR_SLICE is reachable only when every threshold holds for one tenant × model version × species slice.

min_confirmed_outcomes: 200
min_outcome_coverage: 0.25
max_brier_score: 0.18
max_calibration_error: 0.08
max_mechanistic_risk: 0.39
max_evidence_age_days: 180

Generated examples, including adversarial cases

Every value below was computed by the reference engine from a committed fixture. The tail-invariance examples append negligible hypotheses to the same distribution; CIRE 1.0 flipped its badge from REVIEW to HIGH on the five 1e-12 tails, and CIRE 2.0 does not move.

degraded-inputSynthetic

Contradictory input

A feline patient recorded with a dog breed, an impossible weight, and an unsupported region. Input integrity sets the publication floor.

differential = [0.9, 0.06, 0.04]

probability_semantics = categorical_exhaustive

semantics_status = valid

differential_concentration = 0.781006

cps = 0.087598

structural_state = NOMINAL

publication_state = REVIEW

reliability_state = UNVALIDATED

publication.reasons = input_integrity_risk:REVIEW:0.3425

generated by @vetios/cire-engine · CIRE 2.0.0 · sha256 2a789f07a5c2ba494b9d5c865e119ba3a5ced2bd36acd7fc478a50ecdb6a6268

demonstrated-miscalibrationSynthetic

Concentrated differential, demonstrated miscalibration

400 confirmed outcomes show a Brier score of 0.26. The highly concentrated differential cannot compensate: publication is CAUTION and reliability is REVIEW_REQUIRED. Outcome evidence in this example is synthetic.

differential = [0.95, 0.03, 0.02]

probability_semantics = categorical_exhaustive

semantics_status = valid

differential_concentration = 0.870426

cps = 0.05183

structural_state = NOMINAL

publication_state = CAUTION

reliability_state = REVIEW_REQUIRED

publication.reasons = outcome_calibration_risk:CAUTION:1

generated by @vetios/cire-engine · CIRE 2.0.0 · sha256 6e25a9b7596b520cbbd632cf7cc5c028c23fc156ebc435dfdb0aa0231bd3c0e0

independent-multilabel-unsupportedSynthetic

Independent multilabel marginals

Independent marginals are not a categorical distribution; categorical concentration is undefined, so no structural signal is published and publication is capped at REVIEW.

differential = [0.82, 0.49, 0.17]

probability_semantics = independent_multilabel

semantics_status = unsupported

differential_concentration = null (not defined under the declared semantics)

cps = null (not defined under the declared semantics)

structural_state = UNSUPPORTED

publication_state = REVIEW

reliability_state = UNVALIDATED

publication.reasons = probability_semantics:REVIEW:unsupported

generated by @vetios/cire-engine · CIRE 2.0.0 · sha256 1bd8405bf2d02012bc921969b7a3d8c07227ec93cb88c2908a2ad4b78334eb8d

insufficient-outcome-evidenceSynthetic

Some outcomes, below threshold

40 confirmed outcomes is below the 200 required; the slice is INSUFFICIENT_EVIDENCE regardless of structure. Outcome evidence in this example is synthetic.

differential = [0.9, 0.06, 0.04]

probability_semantics = categorical_exhaustive

semantics_status = valid

differential_concentration = 0.781006

cps = 0.087598

structural_state = NOMINAL

publication_state = NOMINAL

reliability_state = INSUFFICIENT_EVIDENCE

generated by @vetios/cire-engine · CIRE 2.0.0 · sha256 1d3b1c59ece47d5ee5e9c741fff2f04841b134899e0381fbe52353bf3ac9e542

invalid-exhaustive-massSynthetic

Declared exhaustive, mass is 1.48

The producer declared categorical_exhaustive but the values sum to 1.48. The declaration is violated, so the output is suppressed instead of being silently renormalized.

differential = [0.82, 0.49, 0.17]

probability_semantics = categorical_exhaustive

semantics_status = invalid

differential_concentration = null (not defined under the declared semantics)

cps = null (not defined under the declared semantics)

structural_state = UNSUPPORTED

publication_state = SUPPRESSED

reliability_state = UNVALIDATED

publication.reasons = probability_semantics:SUPPRESSED:invalid

generated by @vetios/cire-engine · CIRE 2.0.0 · sha256 bc041771e65b828ae79991e68cc58721ce085c9e9844893f43262bc1faabde49

mechanistic-contradictionSynthetic

Confident but mechanistically contradicted

Anticoagulant rodenticide (brodifacoum) toxicity is proposed with 95% of the mass; the remaining 5% is unassigned. Observed thrombocytopenia and petechiae are platelet (primary hemostasis) findings that a pure coagulation-factor disorder does not produce. Physiology contradicts the leading hypothesis, and concentration cannot compensate.

differential = [0.95]

probability_semantics = categorical_topk_with_other

semantics_status = valid

differential_concentration = 0.889207

cps = 0.044317

structural_state = NOMINAL

publication_state = SUPPRESSED

reliability_state = UNVALIDATED

publication.reasons = mechanistic_risk:SUPPRESSED:0.8

generated by @vetios/cire-engine · CIRE 2.0.0 · sha256 af966cf2a70589fc935077efaffa4beddad46b1cd1616893f8b0494a6b4016dc

nominal-structure-no-outcomesSynthetic

Concentrated differential, no outcome evidence

Structure is nominal, so publication may be NOMINAL; with zero confirmed outcomes the reliability state is UNVALIDATED.

differential = [0.9, 0.06, 0.04]

probability_semantics = categorical_exhaustive

semantics_status = valid

differential_concentration = 0.781006

cps = 0.087598

structural_state = NOMINAL

publication_state = NOMINAL

reliability_state = UNVALIDATED

generated by @vetios/cire-engine · CIRE 2.0.0 · sha256 c1ae352bbf79952842f8c3069fa8cc446fb1c853f1f58da5795f3e7b4d6955b3

rank-scores-unsupportedSynthetic

Rank scores

Ordinal scores are never treated as probabilities.

differential = [3, 2, 1]

probability_semantics = rank_scores

semantics_status = unsupported

differential_concentration = null (not defined under the declared semantics)

cps = null (not defined under the declared semantics)

structural_state = UNSUPPORTED

publication_state = REVIEW

reliability_state = UNVALIDATED

publication.reasons = probability_semantics:REVIEW:unsupported

generated by @vetios/cire-engine · CIRE 2.0.0 · sha256 98ef69324b2d2be52de035e3fe321bb6d3698679eba744d3385f67d961bc0378

spec-worked-exampleSynthetic

Specification worked example

Top-4 slice of a categorical distribution. The unreported 3% is an explicit residual bucket, not renormalized away.

differential = [0.62, 0.21, 0.09, 0.05]

probability_semantics = categorical_topk_with_other

semantics_status = valid

differential_concentration = 0.388413

cps = 0.244635

structural_state = NOMINAL

publication_state = NOMINAL

reliability_state = UNVALIDATED

generated by @vetios/cire-engine · CIRE 2.0.0 · sha256 8bdfd10c8209b38eadb01b492225fe91124dd312170fb7c48aa9e8a2febef7d5

tail-invariance-appended-1e-12Synthetic

Tail invariance: five 1e-12 hypotheses appended

Same distribution with five negligible hypotheses appended. Concentration and publication state are unchanged (CIRE 1.0 flipped REVIEW to HIGH here).

differential = [0.554054, 0.331081, 0.114865, 1e-12, 1e-12, 1e-12, 1e-12, 1e-12]

probability_semantics = categorical_exhaustive

semantics_status = valid

differential_concentration = 0.474423

cps = 0.210231

structural_state = NOMINAL

publication_state = NOMINAL

reliability_state = UNVALIDATED

generated by @vetios/cire-engine · CIRE 2.0.0 · sha256 dec01dd08aae2d7af9f703978669c5278f2bbedab15d1a51e871e565924fb5ec

tail-invariance-appended-1e-6Synthetic

Tail invariance: five 1e-6 hypotheses appended

Five 1e-6 hypotheses move concentration by less than 0.0001 and do not change publication state.

differential = [0.554054, 0.331081, 0.114865, 0.000001, 0.000001, 0.000001, 0.000001, 0.000001]

probability_semantics = categorical_exhaustive

semantics_status = valid

differential_concentration = 0.474386

cps = 0.210246

structural_state = NOMINAL

publication_state = NOMINAL

reliability_state = UNVALIDATED

generated by @vetios/cire-engine · CIRE 2.0.0 · sha256 616a69207d8d65a4657922643b13db7990f9a54d6399bf735a5e1aa10a3a1a75

tail-invariance-baselineSynthetic

Tail invariance: baseline

[0.82, 0.49, 0.17] normalized to a categorical distribution.

differential = [0.554054, 0.331081, 0.114865]

probability_semantics = categorical_exhaustive

semantics_status = valid

differential_concentration = 0.474423

cps = 0.210231

structural_state = NOMINAL

publication_state = NOMINAL

reliability_state = UNVALIDATED

generated by @vetios/cire-engine · CIRE 2.0.0 · sha256 077ccfb880401b65b9af8b17d422215ab25c845a40b1f6480c99eed9e4b6c543

unknown-semantics-conservativeSynthetic

Undeclared semantics

Undeclared semantics are handled conservatively: no structural signal, publication capped at REVIEW.

differential = [0.7, 0.2, 0.1]

probability_semantics = unknown

semantics_status = unsupported

differential_concentration = null (not defined under the declared semantics)

cps = null (not defined under the declared semantics)

structural_state = UNSUPPORTED

publication_state = REVIEW

reliability_state = UNVALIDATED

publication.reasons = probability_semantics:REVIEW:unsupported

generated by @vetios/cire-engine · CIRE 2.0.0 · sha256 8a87ea25df8f6de9dd5b2db1d81db5e96e751b3cde1fe86e1365157363eb84ed

validated-slice-synthetic-evidenceSynthetic

All reliability thresholds met (synthetic evidence)

Demonstrates the only path to VALIDATED_FOR_SLICE. The evidence here is synthetic; no production slice is claimed to be validated by this example.

differential = [0.9, 0.06, 0.04]

probability_semantics = categorical_exhaustive

semantics_status = valid

differential_concentration = 0.781006

cps = 0.087598

structural_state = NOMINAL

publication_state = NOMINAL

reliability_state = VALIDATED_FOR_SLICE

generated by @vetios/cire-engine · CIRE 2.0.0 · sha256 3b65ecec6fbc2bb73274dcac3cea6e1a076eaff4eb69708e8eeebce6d1cf5556

Reproduce every number yourself

Download the examples: curl -o cire-examples.json https://www.vetios.tech/api/public/cire-examples
Download the independent reference formula (Python standard library only, no VetIOS code): curl -o reference-formula.py https://www.vetios.tech/api/public/cire-standard/reference-formula.py
Run python3 reference-formula.py cire-examples.json. It prints how many examples it reproduced and names any mismatch.
Validate any implementation against the conformance fixture: https://www.vetios.tech/api/public/cire-conformance

Verification levels

Conformance listings in the registry always state their level. No level is called certification.

SELF_CONFORMANCE

The implementer ran the published conformance fixture and it passed. Numerical compatibility only.

REVIEWER_VERIFIED

A named reviewer other than the implementer re-ran the fixture against the submitted implementation.

SIGNATURE_VERIFIED

The submitted report is signed and the signature verifies against a published key.

THIRD_PARTY_VERIFIED

An organization independent of VetIOS and of the implementer verified conformance.

CLINICALLY_VALIDATED

Outcome-linked clinical validation by an independent party. Conformance never implies this level.

Limitations

Differential concentration, CPS and structural_state are structural signals. None of them measures diagnostic correctness.
reliability_state is scoped to one tenant × model version × species slice and to the thresholds version stamped on it.
VALIDATED_FOR_SLICE is internal outcome evidence, not external validation, regulatory clearance, or clinical endorsement.
The mechanistic model currently covers a small set of hemostasis profiles; outside them the dimension is reported as not evaluated rather than as zero risk.
Absence of outcome evidence sets no publication floor; it is disclosed through reliability_state on every output.
Conformance fixtures prove numerical compatibility with the reference numerics. They are not certification.

Licensing

Specification. The CIRE specification, canonical fixtures, and the independent reference formula are public. Anyone may read, implement, and verify them.
Reference formula. An independent, standard-library Python implementation of the concentration formula is published with the specification.
Fixtures. Canonical example and conformance fixtures are published with their SHA-256 digests.
Reference implementation. The reference implementation (@vetios/cire-engine) is licensed under the MIT License.
Managed infrastructure. VetIOS-managed infrastructure is a proprietary commercial service: hosted outcome lineage, the evidence graph, calibration history, model governance, clinical validation, policy execution, federation, audit, deployment gates, and network evidence.
Marks. No code licence grants rights to the VetIOS or CIRE names, logos, or any certification mark. Use of marks is governed by a separate conformity and trademark policy (pending counsel).
What an implementer may say. “Our model emits CIRE-compatible telemetry.”, when: The implementation passes the published CIRE conformance fixture for the stated standard version.

Deprecated 1.0 fields

reliability_badge — The 1.0 badge was derived from CPS alone and did not measure clinical reliability. Replaced by publication.publication_state, structural.structural_state, reliability.reliability_state.
safety_state — Now a lossless projection of publication_state (NOMINAL→nominal, REVIEW→warning, CAUTION→critical, SUPPRESSED→blocked). Replaced by publication.publication_state.
phi_hat — phi_hat is differential concentration under policy cire-concentration-v2. It is a structural signal, not correctness. Replaced by differential_concentration.value.