Machine-readable registry
Risk archetypes
Ten recurring ways a crossing's warrant can be weaker than it looks. Each names the check that detects it, the quantity it measures, what an underwriter might require, and what it does not claim.
Download archetypes.v1.json · schema. proposed Every archetype is proposed until a fixture measures its quantity in a real domain. Physics status: FORMAL = a defined operator on declared inputs; FORMAL_MEASURE_ANALOGY_NAME = the quantity is exact, the physical name is an analogy; ANALOGY = explanatory only; NONE = no physical counterpart. No threshold has a default: a class policy declares it, or the reading is reported without one.
Physics labels
| FORMAL | A defined operator on declared inputs. |
| FORMAL_MEASURE_ANALOGY_NAME | The quantity is exact; its physical name is an analogy. |
| ANALOGY | Explanatory only. |
| NONE | No physical counterpart. Forcing one would be an error. |
hazard
Premature closure proposed FORMAL_MEASURE_ANALOGY_NAME
Does the system commit before review can catch up?
- Quantity
- review_ratio = review_seconds / commit_seconds
- Analogue
- Deborah number: relaxation time over observation time
- Threshold
- class policy max_review_ratio
- Requires
- declared review duration; declared time to commitment
- Detected by
- policy.review_ratio
- HOLD reasons
- POLICY_DENIEDUNAVAILABLE
- Underwriting condition
- Review completes before commitment for this crossing class.
- Exclusion candidate
- Losses from commitments made while review_ratio exceeded the declared maximum.
- Accumulation
- not relevant
Counterexample kept on file
- A review event recorded after commitment with a backdated valid time.
Non-claims
- Does not predict loss frequency.
- The ratio says nothing about review quality.
severity
Irreversible commitment proposed FORMAL
Can the crossing be undone within its declared scope?
- Quantity
- inverse_action exists (boolean)
- Analogue
- Absorbing state in a declared state graph
- Threshold
- definition: no inverse action means absorbing
- Requires
- declared recovery.inverse_action (null when none)
- Detected by
- appraisal only
- HOLD reasons
- none (appraisal only)
- Underwriting condition
- An inverse action is documented and tested for this crossing class.
- Exclusion candidate
- Losses whose remedy required reversing a crossing declared irreversible.
- Accumulation
- not relevant
Counterexample kept on file
- An 'undo' that restores the record but not the external effect, such as a notice already sent.
Non-claims
- Irreversibility is appraised, not forbidden; a class may warrant irreversible crossings.
severity
Branch collapse proposed FORMAL
How much of the reachable future does the crossing close off?
- Quantity
- retention = reachable_after / reachable_before
- Analogue
- Contraction of the reachable set within a horizon
- Threshold
- class policy reachable_retention_min
- Requires
- declared state model; declared horizon; counts before and after
- Detected by
- policy.reachability
- HOLD reasons
- POLICY_DENIEDUNAVAILABLE
- Underwriting condition
- Retention of reachable states stays above the declared minimum.
- Exclusion candidate
- Losses arising from options closed by a crossing below the declared retention.
- Accumulation
- not relevant
Counterexample kept on file
- Counting states in a model that omits the options the affected party actually needs.
Non-claims
- Retention depends entirely on the declared state model.
hazard
Hidden coupling proposed FORMAL
Does the action touch systems its proposal did not declare?
- Quantity
- count of effect targets absent from dependencies.declared
- Analogue
- Off-diagonal terms in a dependency matrix
- Threshold
- definition: any undeclared target is coupling
- Requires
- declared effects; declared dependencies
- Detected by
- appraisal only
- HOLD reasons
- none (appraisal only)
- Underwriting condition
- Every effect target is a declared dependency.
- Exclusion candidate
- Losses propagating through dependencies the insured did not declare.
- Accumulation
- relevant
Counterexample kept on file
- A declared dependency list that is copied forward and no longer matches the effects.
Non-claims
- Detects only coupling visible in the declared effects.
accumulation
Feedback amplification proposed FORMAL
Does repeating the crossing amplify its own error?
- Quantity
- loop_gain
- Analogue
- Loop gain on a declared cycle (control theory)
- Threshold
- definition: gain above 1 amplifies; class policy max_loop_gain gates
- Requires
- declared feedback cycle; estimated gain with method
- Detected by
- policy.loop_gain
- HOLD reasons
- POLICY_DENIEDUNAVAILABLE
- Underwriting condition
- Declared feedback cycles are damped below the policy gain.
- Exclusion candidate
- Losses compounded by a feedback cycle operating above the declared gain.
- Accumulation
- relevant
Counterexample kept on file
- Denials feeding a model's training data so that the next model denies more.
Non-claims
- The gain is supplied by the proposer; this site does not estimate it.
accumulation
Accumulation proposed FORMAL_MEASURE_ANALOGY_NAME
How many crossings would fail together through a shared dependency?
- Quantity
- concentration = crossings sharing a node / crossings in portfolio
- Analogue
- Percolation or contagion on a dependency network
- Threshold
- none declared; reported as measured
- Requires
- a portfolio of proposals with declared model, vendor, policy source and data source
- Detected by
- appraisal only
- HOLD reasons
- none (appraisal only)
- Underwriting condition
- Concentration on any single model, vendor or data source is disclosed.
- Exclusion candidate
- Aggregated losses from a common dependency beyond a declared concentration.
- Accumulation
- relevant
Counterexample kept on file
- Two vendors that resell the same underlying model under different names.
Non-claims
- Measured on a portfolio, not a single crossing.
- Concentration is not a cascade probability.
hazard
Expired evidence proposed FORMAL
Was the evidence still valid when the crossing took effect?
- Quantity
- valid_until and age at valid time
- Analogue
- Decay against a declared lifetime
- Threshold
- evidence valid_until; class policy evidence_max_age_days
- Requires
- captured_at; valid_until or a policy maximum age
- Detected by
- time.*
- HOLD reasons
- EXPIREDINVALID
- Underwriting condition
- Evidence is current at the crossing's valid time.
- Exclusion candidate
- Losses from crossings warranted by lapsed evidence.
- Accumulation
- not relevant
Counterexample kept on file
- Evidence recaptured after the crossing and backdated.
Non-claims
- Currency is not accuracy.
hazard
Authority drift proposed ANALOGY
Has authority been handed on further than the class allows?
- Quantity
- delegation chain length
- Analogue
- Biased random walk away from the authorising root
- Threshold
- class policy max_delegation_depth
- Requires
- declared delegation chain
- Detected by
- policy.delegation_depth
- HOLD reasons
- POLICY_DENIEDUNAVAILABLE
- Underwriting condition
- Delegation depth stays within the declared maximum.
- Exclusion candidate
- Losses from crossings authorised beyond the declared delegation depth.
- Accumulation
- not relevant
Counterexample kept on file
- Delegation hidden inside a tool call and absent from the declared chain.
Non-claims
- Chain length is measured; 'drift' is a description.
hazard
Self-certification proposed NONE
Does the proposer authorise or approve its own crossing?
- Quantity
- trust root independence and preparer/approver identity
- Threshold
- definition
- Requires
- declared trust root; named preparer and approver where required
- Detected by
- authority.trust_root, authority.segregation
- HOLD reasons
- UNAUTHORIZEDUNAVAILABLE
- Underwriting condition
- Authority traces to a root independent of the proposer; preparation and approval are separated.
- Exclusion candidate
- Losses from crossings the proposer authorised or approved itself.
- Accumulation
- not relevant
Counterexample kept on file
- Two agents from one deployment approving each other's crossings.
Non-claims
- The site checks the declaration of independence, not the root's real-world standing.
reserve
Unwarranted execution (epistemic IBNR) proposed NONE
Did the crossing take effect without a warrant?
- Quantity
- executed while the determination is HOLD, or executed_at earlier than the determination's record time
- Threshold
- definition
- Requires
- declared execution state; executed_at, when executed
- Detected by
- appraisal only
- HOLD reasons
- none (appraisal only)
- Underwriting condition
- No crossing in the class executes without an EXECUTE determination and receipt.
- Exclusion candidate
- Losses from crossings executed while held.
- Accumulation
- relevant
Counterexample kept on file
- A crossing executed before any determination was requested, then submitted afterwards and found EXECUTE: warranted after the fact is not warranted at the time.
Non-claims
- A count of unwarranted executions is a leading indicator, not a loss estimate.