Recursive Viability · accumulation
Portfolio
One crossing can be warranted while many crossings together are fragile. Run a set of proposals to see which models, vendors, policy sources and data sources they share, how often each class holds, and how many crossings executed without warrant.
Epistemic IBNR borrows the actuarial idea of losses incurred but not reported. Here it counts crossings that took effect while held, or before they were determined: a later EXECUTE does not warrant a crossing retroactively. No loss may have surfaced yet, which is why the count is a leading indicator and not a loss estimate.
Portfolio input (12 synthetic examples preloaded)
Crossings evaluated
12
Epistemic IBNR
4
executed while held, before being determined, or defeated afterwards: INS-003, INS-005, LED-004, MOD-003
Shared dependencies
8
nodes used by more than one crossing
Concentration on shared dependencies
Concentration is the share of evaluated crossings that depend on a node. It is a formal measure; calling it percolation is an analogy. No threshold is applied, and it is not a cascade probability.
| Dependency | Concentration | Crossings | Held |
|---|---|---|---|
| vendor vendor-alpha | 92% | INS-001, INS-002, INS-003, INS-004, INS-005, LED-001, LED-002, LED-004, MOD-001, MOD-002, MOD-003 | 7 of 11 |
| data_source claims-intake | 58% | INS-001, INS-002, INS-003, INS-004, INS-005, MOD-001, MOD-002 | 5 of 7 |
| model model-m3 | 50% | INS-001, INS-002, INS-003, INS-004, INS-005, MOD-002 | 5 of 6 |
| policy_source policy-wording-2026 | 42% | INS-001, INS-002, INS-003, INS-004, INS-005 | 4 of 5 |
| model model-k1 | 33% | LED-001, LED-002, LED-003, LED-004 | 2 of 4 |
| policy_source close-manual-2026 | 33% | LED-001, LED-002, LED-003, LED-004 | 2 of 4 |
| data_source ap-inbox | 25% | LED-001, LED-004, MOD-003 | 1 of 3 |
| policy_source model-risk-policy-2026 | 25% | MOD-001, MOD-002, MOD-003 | 2 of 3 |
HOLD rate by crossing class
The source spec treats denial rate as “a primary underwriting signal, not an error stream” (crossing-graph-spec v0.4 §11.2).
| Class | Crossings | Held |
|---|---|---|
| insurance.claim-auto-denial | 5 | 4 |
| finance.ledger-posting | 4 | 2 |
| ai.model-promotion | 3 | 2 |
Determinations
| Proposal | Outcome | Reasons | Executed |
|---|---|---|---|
| INS-001 | EXECUTE | no | |
| INS-002 | HOLD | INPUT_BINDING_FAILEDEXPIREDPOLICY_DENIEDINVARIANT_FAILED | no |
| INS-003 | HOLD | UNAUTHORIZED | yes FIRED |
| INS-004 | HOLD | UNAVAILABLE | no |
| INS-005 | HOLD | EXPIRED | yes FIRED |
| LED-001 | EXECUTE | no | |
| LED-002 | HOLD | UNAUTHORIZEDINVARIANT_FAILED | no |
| LED-003 | HOLD | EXPIREDPOLICY_DENIED | no |
| LED-004 | EXECUTE | yes FIRED | |
| MOD-001 | EXECUTE | no | |
| MOD-002 | HOLD | UNAUTHORIZEDPOLICY_DENIEDINVARIANT_FAILED | no |
| MOD-003 | HOLD | EXPIREDPOLICY_DENIEDUNAVAILABLE | yes FIRED |
The joined record
Configuration → controls → execution → failure → reliance → consequence → loss, joined per crossing. The actuarial value lies in the later joins, which only real deployments can supply. This table shows how many crossings have each stage represented. An unrepresented stage means nothing was recorded, not that nothing happened: no observed effect is not no effect.
| configuration | controls | execution | failure | reliance | consequence | loss |
|---|---|---|---|---|---|---|
| 12 of 12 represented | 12 of 12 represented | 12 of 12 represented | 12 of 12 represented | 1 of 12 represented | 1 of 12 represented | 1 of 12 represented |
Insurer export
Redacted views and per-class summaries, with timestamps coarsened to the hour. Any class with fewer than k crossings is suppressed entirely, because a rare combination can identify a policyholder even without names (crossing-graph-spec v0.4 §11.6; source default k = 5).
| Class | Released | Crossings | Hold rate | Executed without warrant |
|---|---|---|---|---|
| insurance.claim-auto-denial | PASS | 5 | 80% | 2 |
| finance.ledger-posting | UNAVAILABLE | suppressed | — | — |
| ai.model-promotion | UNAVAILABLE | suppressed | — | — |