Underwritability catalogue

Crossing classes

A crossing class declares what a domain must supply before its crossings can be evaluated, and what an underwriter would need to see. No class is underwritten: these are statements of what cover would require, not offers of cover.

Schema: crossing-class.v0.schema.json. Proposing a new class means writing one against this schema; there is no submission service yet.

insurance claims

Automated denial of an insurance claim proposed

An automated claims workflow proposes to deny a claim after document mismatches and send closure notices. The crossing moves a claim from open to denied. Synthetic specification; no insurer has reviewed it.

Class: insurance.claim-auto-denial@0.1.0
Class ref: ntc:insurance.claim-auto-denial@0.1.0:sha256:864ad01bd45369bbb95b1301d00c292bd08ef1ddb434741318353cfe807d6dff
Policy ref: ntp:nt-policy:claims-auto-denial@0.1.0:sha256:f0eb14ee6e9d4d8384519c1debc492f6c7aabc5731a50bba0c77129a306630fc
Actions: deny_claim
Dimensions: material stake · compensable action · explicit authority · bounded freshness · E2+ evidence
Class fields from crossing-graph-spec v0.4 §1.2.
Admissibility (I2): PASS satisfies the source's class admissibility rules
Profiles: Recorder: nt-receipt-chain:v0 (this site's chain; not CER-17)
Underwriting: nt-underwritability-conditions:v0 (conditions document; not UW-12)
Failure mode: fail-closed
A denial is not committed and no notice is sent unless the gate returns EXECUTE.
Consequence: Unbounded: no ceiling
The direct effect is the value of the denied claim, but a wrongful denial can also create liability beyond the claim value, such as regulatory action or bad-faith claims, so no ceiling is declared.
Authority: Trust root independent of the proposer

Policy parameters

Neverthought (illustrative class policy; not an insurer's policy). Parameter values are illustrative. A real class takes them from the insurer's claims policy and regulator.

max_review_ratio	1
max_delegation_depth	2
evidence_max_age_days	30
reachable_retention_min	0.5
max_loop_gain	1

Invariants

Claimant keeps review, correction and appeal routes after denial (measure routes_preserved, tolerance 0)

Evidence required

mismatch_record: E2 or higher
policy_wording: E4 or higher
model_version: E2 or higher

Tiers from the source spec (§1.6): E1 self-attested, E2 single-source, E3 corroborated, E4 independently verified, E5 adversarially tested.

Consequence paths represented

claimant_routes_closed: Review, correction or appeal routes close after the denial.
wrongful_denial_settlement: The denial is overturned and the claim is settled, with handling cost.
reserve_released: The claim reserve is released in reliance on the denial.
feedback_to_model: Denials feed training labels and shift later denials.

Structure status: UNTESTED no structural test has been run against this class. A consequence that follows none of these paths challenges the model; an absence of such observations is not support.

Risk archetypes read

Premature closure Irreversible commitment Branch collapse Hidden coupling Feedback amplification Accumulation Expired evidence Authority drift Self-certification Unwarranted execution (epistemic IBNR)

What an underwriter would need

A receipt for every automated denial, issued before closure notices are sent.
HOLD rate and reasons for this class over a declared period.
Count of denials executed without an EXECUTE determination (epistemic IBNR).
Concentration of denials on a single model, vendor or data source.
Evidence that claimant review, correction and appeal routes remain open after denial.

Conditions

Human review completes before a denial is committed.
Denial authority traces to a root independent of the claims system.
Feedback from denials into model training is declared and damped.
Evidence relied on is no more than 30 days old at valid time.

Exclusions an underwriter might apply

Denials executed while held.
Denials where appeal or correction routes were not available.
Losses propagating through undeclared dependencies.

Mutation battery: 9 of 9 mutants held as expected

INS-001, a warranted crossing, broken one way at a time. Each mutant must hold for exactly the expected reason and no other.

Mutation	Expected	Result
Remove the evidence for the class's first requirement drop-evidence	HOLD for UNAVAILABLE only	PASS HOLD UNAVAILABLE
Lower one evidence item to the tier just below the class minimum lower-tier	HOLD for INVALID only	PASS HOLD INVALID
Make one evidence item refer to a different proposal rebind-evidence	HOLD for INPUT_BINDING_FAILED only	PASS HOLD INPUT_BINDING_FAILED
Make one evidence item lapse a day before valid time expire-evidence	HOLD for EXPIRED only	PASS HOLD EXPIRED
Have the proposer declare its own trust root self-authorise	HOLD for UNAUTHORIZED only	PASS HOLD UNAUTHORIZED
Have the preparer approve their own crossing self-approve	not applicable to this class	—
Hand authority on one hop further than the policy allows lengthen-delegation	HOLD for POLICY_DENIED only	PASS HOLD POLICY_DENIED
Declare a measurement as a range straddling its limit straddle-range	HOLD for UNAVAILABLE only	PASS HOLD UNAVAILABLE
Change the proposed state so the class's first invariant fails break-invariant	HOLD for INVARIANT_FAILED only	PASS HOLD INVARIANT_FAILED
Make one declared mapping of the evidence report a different value split-mappings	not applicable to this class	—
Execute the crossing before it is determined execute-early	EXECUTE, flagged as epistemic IBNR	PASS EXECUTE

Examples

Proposal	Scenario	Expected
INS-001	Denial after completed human review, appeal routes kept	EXECUTE
INS-002	Denial committed in minutes, before review, appeal route closed	HOLD INPUT_BINDING_FAILEDEXPIREDPOLICY_DENIEDINVARIANT_FAILED
INS-003	Denial already sent, authority self-asserted	HOLD UNAUTHORIZED
INS-004	Denial like INS-001, but the feedback gain is only known as a wide range	HOLD UNAVAILABLE
INS-005	Denial warranted when made; a model withdrawal learned later defeats it	HOLD EXPIRED

Non-claims

Not an insurer's claims policy or a regulatory standard.
Parameters are illustrative.
No insurer, broker or regulator has reviewed this class.

Download class JSON · Propose a crossing in this class

finance operations

Automated ledger posting proposed

An agent proposes a journal entry that moves the ledger from one balance to the next. The invariant is double-entry conservation. Synthetic specification; no auditor has reviewed it.

Class: finance.ledger-posting@0.1.0
Class ref: ntc:finance.ledger-posting@0.1.0:sha256:0e759b311284b657922ef13ad05dd21f16e874430212b003c2dcb65b37b944b5
Policy ref: ntp:nt-policy:ledger-posting@0.1.0:sha256:9a1579350abfd50d97fb8d7b36b2fd2846f4a628dfc8bee18ceb9a5630121d4f
Actions: post_journal_entry
Dimensions: material stake · reversible action · explicit authority · bounded freshness · E2+ evidence
Class fields from crossing-graph-spec v0.4 §1.2.
Admissibility (I2): PASS satisfies the source's class admissibility rules
Profiles: Recorder: nt-receipt-chain:v0 (this site's chain; not CER-17)
Underwriting: nt-underwritability-conditions:v0 (conditions document; not UW-12)
Failure mode: fail-closed
The entry is not posted unless the gate returns EXECUTE.
Consequence: Bounded: The absolute value of the entry's postings (direct misstatement only)
A single posting misstates the ledger by at most its own amounts. Consequences in downstream reporting, tax or audit are outside this bound and are not declared here.
Authority: Trust root independent of the proposer; preparer and approver must differ

Policy parameters

Neverthought (illustrative class policy; not an accounting standard). Parameter values are illustrative.

max_delegation_depth	2
evidence_max_age_days	90

Invariants

Debits equal credits (postings sum to zero) (measure double_entry_balance, tolerance 0)

Evidence required

source_document: E2 or higher

Tiers from the source spec (§1.6): E1 self-attested, E2 single-source, E3 corroborated, E4 independently verified, E5 adversarially tested.

Consequence paths represented

misstatement: The ledger is misstated by the entry's amounts.
reversal_cost: A reversing entry and its review cost.

Structure status: UNTESTED no structural test has been run against this class. A consequence that follows none of these paths challenges the model; an absence of such observations is not support.

Risk archetypes read

Irreversible commitment Hidden coupling Accumulation Expired evidence Authority drift Self-certification Unwarranted execution (epistemic IBNR)

What an underwriter would need

A receipt for every automated posting, issued before the period closes.
Evidence of segregation between preparer and approver.
Count of postings executed without an EXECUTE determination.
Concentration of postings on a single model or vendor.

Conditions

Every posting balances exactly.
Preparer and approver are different principals.
Source documents are independently retrievable.

Exclusions an underwriter might apply

Postings executed while held.
Postings approved by their preparer.

Mutation battery: 9 of 9 mutants held as expected

LED-001, a warranted crossing, broken one way at a time. Each mutant must hold for exactly the expected reason and no other.

Mutation	Expected	Result
Remove the evidence for the class's first requirement drop-evidence	HOLD for UNAVAILABLE only	PASS HOLD UNAVAILABLE
Lower one evidence item to the tier just below the class minimum lower-tier	HOLD for INVALID only	PASS HOLD INVALID
Make one evidence item refer to a different proposal rebind-evidence	HOLD for INPUT_BINDING_FAILED only	PASS HOLD INPUT_BINDING_FAILED
Make one evidence item lapse a day before valid time expire-evidence	HOLD for EXPIRED only	PASS HOLD EXPIRED
Have the proposer declare its own trust root self-authorise	HOLD for UNAUTHORIZED only	PASS HOLD UNAUTHORIZED
Have the preparer approve their own crossing self-approve	HOLD for UNAUTHORIZED only	PASS HOLD UNAUTHORIZED
Hand authority on one hop further than the policy allows lengthen-delegation	HOLD for POLICY_DENIED only	PASS HOLD POLICY_DENIED
Declare a measurement as a range straddling its limit straddle-range	not applicable to this class	—
Change the proposed state so the class's first invariant fails break-invariant	HOLD for INVARIANT_FAILED only	PASS HOLD INVARIANT_FAILED
Make one declared mapping of the evidence report a different value split-mappings	not applicable to this class	—
Execute the crossing before it is determined execute-early	EXECUTE, flagged as epistemic IBNR	PASS EXECUTE

Examples

Proposal	Scenario	Expected
LED-001	Balanced accrual, prepared and approved by different principals	EXECUTE
LED-002	Entry off by one penny, approved by its preparer	HOLD UNAUTHORIZEDINVARIANT_FAILED
LED-003	Entry backed by lapsed evidence, authority handed on four times	HOLD EXPIREDPOLICY_DENIED
LED-004	Balanced, properly approved entry, posted before anyone asked the gate	EXECUTE

Non-claims

Not an accounting standard or audit opinion.
Balancing is necessary, not sufficient, for a correct entry.

Download class JSON · Propose a crossing in this class

AI operations

Promotion of a model to deployment proposed

A model moves from candidate to deployed in named contexts. Every crossing that later depends on the model inherits this one, so it is the root of accumulation. Synthetic specification; no model-risk or insurance reviewer has examined it.

Class: ai.model-promotion@0.1.0
Class ref: ntc:ai.model-promotion@0.1.0:sha256:ef2a6c30024a521be9afec45c1b3aae6cfc3103aaa55ccc9e4fbe6b5db5024a4
Policy ref: ntp:nt-policy:model-promotion@0.1.0:sha256:a49f52bdc99c6f122c105b138fb31cff72b0a4362243473fe597db3f9045c954
Actions: promote_model
Dimensions: material stake · reversible action · explicit authority · bounded freshness · E2+ evidence
Class fields from crossing-graph-spec v0.4 §1.2.
Admissibility (I2): PASS satisfies the source's class admissibility rules
Profiles: Recorder: nt-receipt-chain:v0 (this site's chain; not CER-17)
Underwriting: nt-underwritability-conditions:v0 (conditions document; not UW-12)
Failure mode: fail-safe
If the gate holds or is unreachable, traffic stays on the previously deployed model. That is safe only when a previous model exists to fall back to.
Consequence: Unbounded: no ceiling
Every crossing that later depends on the promoted model inherits this one, so its consequence grows with use and has no ceiling.
Authority: Trust root independent of the proposer; preparer and approver must differ

Policy parameters

Neverthought (illustrative class policy; not a model-risk standard). Parameter values are illustrative. A real class takes them from the organisation's model-risk policy.

max_review_ratio	1
max_delegation_depth	1
evidence_max_age_days	60
max_loop_gain	1

Invariants

Every deployment context was covered by evaluation (measure evaluation_coverage, tolerance 0)

Evidence required

evaluation_report: E4 or higher
data_provenance: E2 or higher
rollback_plan: E2 or higher

Tiers from the source spec (§1.6): E1 self-attested, E2 single-source, E3 corroborated, E4 independently verified, E5 adversarially tested.

Consequence paths represented

unevaluated_context_error: Errors in a deployment context the evaluation did not cover.
downstream_inheritance: Every crossing depending on the model inherits its errors.
rollback_cost: Rolling back the model and its downstream consumers.

Structure status: UNTESTED no structural test has been run against this class. A consequence that follows none of these paths challenges the model; an absence of such observations is not support.

Risk archetypes read

Premature closure Irreversible commitment Hidden coupling Feedback amplification Accumulation Expired evidence Authority drift Self-certification Unwarranted execution (epistemic IBNR)

What an underwriter would need

A receipt for every promotion, issued before the model serves traffic.
Evaluation coverage for each deployment context, from a team independent of the builders.
The list of downstream crossing classes that depend on the model.
A tested rollback, including downstream consumers.
Whether deployed outputs feed back into training, and the declared gain.

Conditions

No model serves a context its evaluation did not cover.
Builder and approver are different principals.
Evaluation evidence is no more than 60 days old at valid time.
Feedback from deployed outputs into training is declared and damped.

Exclusions an underwriter might apply

Losses in contexts the promoted model was not evaluated for.
Losses from models promoted while held.
Aggregated losses across crossings sharing a model beyond a declared concentration.

Mutation battery: 10 of 10 mutants held as expected

MOD-001, a warranted crossing, broken one way at a time. Each mutant must hold for exactly the expected reason and no other.

Mutation	Expected	Result
Remove the evidence for the class's first requirement drop-evidence	HOLD for UNAVAILABLE only	PASS HOLD UNAVAILABLE
Lower one evidence item to the tier just below the class minimum lower-tier	HOLD for INVALID only	PASS HOLD INVALID
Make one evidence item refer to a different proposal rebind-evidence	HOLD for INPUT_BINDING_FAILED only	PASS HOLD INPUT_BINDING_FAILED
Make one evidence item lapse a day before valid time expire-evidence	HOLD for EXPIRED only	PASS HOLD EXPIRED
Have the proposer declare its own trust root self-authorise	HOLD for UNAUTHORIZED only	PASS HOLD UNAUTHORIZED
Have the preparer approve their own crossing self-approve	HOLD for UNAUTHORIZED only	PASS HOLD UNAUTHORIZED
Hand authority on one hop further than the policy allows lengthen-delegation	HOLD for POLICY_DENIED only	PASS HOLD POLICY_DENIED
Declare a measurement as a range straddling its limit straddle-range	HOLD for UNAVAILABLE only	PASS HOLD UNAVAILABLE
Change the proposed state so the class's first invariant fails break-invariant	HOLD for INVARIANT_FAILED only	PASS HOLD INVARIANT_FAILED
Make one declared mapping of the evidence report a different value split-mappings	not applicable to this class	—
Execute the crossing before it is determined execute-early	EXECUTE, flagged as epistemic IBNR	PASS EXECUTE

Examples

Proposal	Scenario	Expected
MOD-001	Model promoted only to contexts it was evaluated for	EXECUTE
MOD-002	Retrained model extended to an unevaluated context, approved by its builder	HOLD UNAUTHORIZEDPOLICY_DENIEDINVARIANT_FAILED
MOD-003	Model already serving, stale evaluation, no declared trust root	HOLD EXPIREDPOLICY_DENIEDUNAVAILABLE

Non-claims

Not a model-risk management standard or regulatory requirement.
Evaluation coverage says which contexts were evaluated, not how well the model performs in them.
No model-risk, insurance or regulatory reviewer has reviewed this class.

Download class JSON · Propose a crossing in this class

empirical research

Promotion of an experimental result to a claim proposed

A result moves from 'reported' to 'claimed with standing' so others may reuse it. The path is instrument → mapping → analysis → interpretation → claim → reuse. A missing or disputed measurement is value uncertainty; an analysis path that was never executed from authoritative input challenges the structure; competing mappings of the same evidence end UNRESOLVED. Not an insurance class.

Class: science.claim-promotion@0.1.0
Class ref: ntc:science.claim-promotion@0.1.0:sha256:44fc38a32a76ed6d45b3c400ed15a8591c4610c7976570783190b05e5f482980
Policy ref: ntp:nt-policy:claim-promotion@0.1.0:sha256:376aee7d66d4f340e624cd4481c877da01f642ccdfd82c1281ec541688b59658
Actions: promote_claim
Dimensions: reputational stake · compensable action · explicit authority · perpetual freshness · E2+ evidence
Class fields from crossing-graph-spec v0.4 §1.2.
Admissibility (I2): PASS satisfies the source's class admissibility rules
Profiles: Recorder: nt-receipt-chain:v0 (this site's chain; not CER-17)
Underwriting: none
Failure mode: fail-closed
A result that is held is not presented as a claim.
Consequence: Unbounded: no ceiling
A claim with standing is reused by others; errors propagate through every reuse, so no ceiling is declared.
Authority: Trust root independent of the proposer; preparer and approver must differ

Policy parameters

Neverthought (illustrative class policy; not a journal's or funder's standard). Promotion needs the raw results, an independent reproduction, and a reviewer other than the builder.

max_delegation_depth

Invariants

Every declared mapping of the evidence reports the same value (measure mappings_agree, tolerance 0)
The analysis path the claim relies on was executed from authoritative input (measure path_executable, tolerance 0)

Evidence required

raw_results: E2 or higher
independent_reproduction: E4 or higher

Tiers from the source spec (§1.6): E1 self-attested, E2 single-source, E3 corroborated, E4 independently verified, E5 adversarially tested.

Consequence paths represented

claim_reused: Others build on the claim.
claim_retracted: The claim is withdrawn or corrected.
replication_attempted: An independent party attempts to reproduce the result.

Structure status: UNTESTED no structural test has been run against this class. A consequence that follows none of these paths challenges the model; an absence of such observations is not support.

Risk archetypes read

Self-certification Hidden coupling Unwarranted execution (epistemic IBNR)

Not offered for underwriting review. See the science example.

Mutation battery: 10 of 10 mutants held as expected

SCI-001, a warranted crossing, broken one way at a time. Each mutant must hold for exactly the expected reason and no other.

Mutation	Expected	Result
Remove the evidence for the class's first requirement drop-evidence	HOLD for UNAVAILABLE only	PASS HOLD UNAVAILABLE
Lower one evidence item to the tier just below the class minimum lower-tier	HOLD for INVALID only	PASS HOLD INVALID
Make one evidence item refer to a different proposal rebind-evidence	HOLD for INPUT_BINDING_FAILED only	PASS HOLD INPUT_BINDING_FAILED
Make one evidence item lapse a day before valid time expire-evidence	HOLD for EXPIRED only	PASS HOLD EXPIRED
Have the proposer declare its own trust root self-authorise	HOLD for UNAUTHORIZED only	PASS HOLD UNAUTHORIZED
Have the preparer approve their own crossing self-approve	HOLD for UNAUTHORIZED only	PASS HOLD UNAUTHORIZED
Hand authority on one hop further than the policy allows lengthen-delegation	HOLD for POLICY_DENIED only	PASS HOLD POLICY_DENIED
Declare a measurement as a range straddling its limit straddle-range	not applicable to this class	—
Change the proposed state so the class's first invariant fails break-invariant	HOLD for INVARIANT_FAILED only	PASS HOLD INVARIANT_FAILED
Make one declared mapping of the evidence report a different value split-mappings	HOLD for UNRESOLVED only	PASS HOLD UNRESOLVED
Execute the crossing before it is determined execute-early	EXECUTE, flagged as epistemic IBNR	PASS EXECUTE

Examples

Proposal	Scenario	Expected
SCI-001	SURF-14: ablation result independently reproduced (four of five detections lost)	EXECUTE
SCI-002	SURF-10 pair 2: the pre-state was not preserved, so the comparison cannot be rebuilt from bytes	HOLD UNAVAILABLE
SCI-003	SURF-15c: the detection path compares two hard-coded dictionaries; no authoritative input is ever loaded	HOLD INVARIANT_FAILED
SCI-004	Package file count: 24 manifested payload files versus 25 files counted recursively	HOLD UNRESOLVED
SCI-005	The same count restated under an explicit, prospective counting convention	EXECUTE

Non-claims

Promotion records that a result met this class's checks; it does not certify a discovery or establish causality.
The worked examples map a review of software-governance experiments, not a natural-science study.
Methodological inference in the source review is recorded as not adjudicated, and stays so here.

Download class JSON · Propose a crossing in this class