The site describes governance evidence. This page shows its format: the fields a record carries, the states its exception field can hold, and how a counterparty reads it. The record is the governability state made legible, the evidence a counterparty prices against. Every value here is an illustrative placeholder. It is a template, not a client and not a result.
Inference is not evidence, and the gap between the two is where value is destroyed.
Format specimen · no client data · no score issued| Obligation | Evidence source | Observation | Exception state | Recorded status |
|---|---|---|---|---|
| Third-party access control | Helpdesk access logs | Continuous | Open | Signed |
| Privileged-access review | IAM change log | Event-triggered | Clear | Signed |
| Model-override governance | Override log, risk-committee minutes | Monthly review | In review | Signed |
| Incident notification | Regulator-of-record filing | On event | In progress | Filed |
| Board protocol adherence | Decision log vs protocol | Monthly | Within bounds | Signed |
{
"record": "governance-evidence-record",
"schema": "GET-v1",
"counterparty": "management-liability underwriter, renewal",
"decision_priced": "management liability renewal; automated pricing models in use",
"risk_if_misrepresented": "avoidance; model-risk exclusion bite at claim",
"obligations": [
{
"obligation": "third-party access control",
"primitive": "helpdesk access logs",
"cadence": "continuous",
"provenance": "SIEM -> engine -> signing key",
"verification": "cryptographic-signature",
"exception_state": "open"
},
{
"obligation": "model-override governance",
"primitive": "override log; risk-committee minutes",
"cadence": "monthly-review",
"provenance": "MLOps -> engine -> signing key",
"verification": "cryptographic-signature",
"exception_state": "in-review"
}
],
"maturity": { "level": "Foundational", "trajectory": "rising", "window_days": 90 },
"score": null,
"note": "illustrative placeholder values, not a client record"
}Override events are logged and signed, not asserted. As the committee tightens thresholds the rate falls. The read: the control is being exercised, not just documented.
HITL decisions are routed to the named accountable role within the protocol window, recorded against the obligation with timestamps. The read: the human decision is on the record.
Risk-committee minutes show the model risk reviewed and acted on, not noted. The read: the protocol the board authored is live, not shelfware.
Maturity rising on the record over the period. The read: an improving firm being priced, not a static snapshot.
Automated decisions have already crystallised nine and ten figure losses, reconstructed only after the event.
The trail those reconstructions assembled too late is the trail a conformant engine produces continuously, in the format Arkaya stewards.
Robodebt averaged annual income across fortnightly periods and raised welfare debts later ruled unlawful. A$1.8bn was settled in 2020, a further A$475m (about US$309m) in 2025. A royal commission found the responsible department held legal advice warning of unlawfulness in 2018 and proceeded. The decision trail surfaced years after the loss. In the schema: the unlawful-calculation control carries an open exception from the moment that advice lands, readable by a counterparty years before settlement.
Royal Commission into the Robodebt Scheme, final report 2023; Reuters, 2025.
Algorithmic home-buying priced off the Zestimate model, which overpaid as the 2021 market turned. A $304m write-down in Q3 2021, a further $240m to $265m guided for Q4, the unit closed and about 2,000 staff cut, a quarter of the workforce. The model outputs existed internally. A record a counterparty could check did not. In the schema: the model-override governance field shows the Zestimate's widening error as a populated exception in Q3, not after the write-down.
Zillow Group SEC Form 8-K, third quarter 2021.
The April 2025 cyber incident, attributed to Scattered Spider operating through the TCS-run helpdesk, sat inside a multi-year outsourcing arrangement renewed on cost. Insurance proceeds covered about three quarters of the year's booked incident cost: £100.0m against £131.3m. That is separate from the equity value the market removed. Arkaya's composed estimate puts the total equity-value impact near £2.4bn. In the schema: the third-party access-control obligation carries an open exception at the helpdesk boundary, readable by an underwriter at renewal, not reconstructed at claim.
Cyber Monitoring Centre, 2025 (Category 2, with Co-op, £270m–£440m); M&S full-year results, 52 weeks to 28 March 2026.
Contemporaneous records can support a later investigation. Arkaya reads what those records support; it does not create the underlying record.
Illustrative contrast. No premium figure is shown, and no score is issued from the schema. The counterparty reads the primitives and prices its own risk.
What this specimen is. A format, populated with example states, not a client record and not a result. The figures, obligations and exception states are placeholders chosen to show the shape a record takes; none are drawn from any engagement. No governance score is produced. A worked walkthrough, and in time a real anonymised record from a live engagement, is available to qualified counterparties under NDA. This page shows the form so a reader knows what to ask for.
Resilience Capital is built.
Not asserted.