Chain of custody and audit access

19-00 named OVERT’s three components. Schema is the easy one — a format. This page covers the two that make the format evidence: how a record’s integrity is preserved from signing to audit (chain of custody), and who may check it, how, without anyone’s permission (audit access).

Verification without Glacis

The design goal, from the brochure: any “syndicate, reinsurer, actuary, or capital-markets modelling firm” can verify an OVERT-compliant record “without relying on Glacis’s interpretation.” GROUNDED · Brochure Mechanically, that means a third party holding the published signing keys, the receipt chain, and open tooling can answer every question below offline — no API call to Glacis, no cooperation from the vendor whose behavior is in dispute: GROUNDED · Labs

Question a stranger can answer	What proves it	Custody link
Is this attestation authentic?	Signature over canonical content	Arbiter signing keys
Was it witnessed when claimed?	The Notary’s receipt over its hash	Independent Notary keys
Is it really in the history?	Merkle inclusion proof	Append-only receipt chain
Was history rewritten?	Chain consistency check	Any tampering breaks the chain visibly
What policy governed it?	The policy’s own hash in the record	Policy-as-artifact (below)

Two properties carry the whole table. Two independent signers — the Arbiter attests, the Notary witnesses, on separate infrastructure with separate keys (mechanics in receipts and the Notary) — so forging a record requires compromising two parties who never share custody. And append-only chaining — a signed receipt is immutable, so custody is continuous by construction rather than by procedure. GROUNDED · Labs

Audit access is the protocol layer on top: OVERT specifies who may verify and under what access terms, so that an auditor or an arbitrating party in a trigger dispute can check the evidence even when the two sides have stopped cooperating. A parametric clause pays real money on these records (section 20-00); the access protocol is what keeps the fight about facts. GROUNDED · Brochure

The scope claim, verbatim

Chain of custody guards the records that exist. It says nothing about traffic that never generated one. The canonical scope claim — memorize it word for word, it is the most-used sentence in this curriculum: GROUNDED · Labs

“For governed paths, Glacis produces independently verifiable evidence of policy execution. Receipts attest what was governed and are silent about un-mediated paths.”

OVERT builds this honesty into the schema itself: §3.16 of the standard defines mediation scope, a coverage ratio (attested actions over total actions), and a denominator-source classification — the record declares who counted the denominator rather than letting a strong number pass as a stronger one. GROUNDED · Labs The full trust ladder and its sales use is insurance signals; the framing discipline is the claim taxonomy.

OVERT-as-Code, honestly

One paragraph, because it earns exactly one. The policy side of the standard has an open (Apache-2.0) implementation: governance rules written as declarative TOML, evaluated deterministically on the hot path (the labs benchmark reports sub-millisecond P50), with staged enforcement modes — shadow → warn → enforce → strict — so a customer can watch what a policy would do before letting it block anything. The policy file is hashed, and that hash rides in the attestation, which is what makes the governing policy itself part of the chain of custody. What it is not: an underwriting-grade profile of what an OVERT record must contain for a parametric policy form — the draft is candid that this “does not exist anywhere in the market today” and is proposed as cohort work. GROUNDED · Labs The shadow-to-strict ladder returns as the customer-success motion in 22-04.

The AAL honesty ladder

Evidence strength has levels, and claiming the wrong one is the fastest way to lose a security reviewer. The internal doctrine grades attestation assurance levels: GROUNDED · Labs

Level	Who vouches	Example
AAL-1	The operator, self-attested	A vendor’s own compliance PDF
AAL-2	Software-signed in the operator’s environment	Arbiter attestations, software-keyed
AAL-3	Software-signed + independently witnessed	Attestation hash receipted by the Notary
AAL-4	Independent cryptographic assurance end to end	Hardware-rooted or independent signer — not claimed today

The rule is absolute: a software-keyed deployment is AAL-2/3. Never claim AAL-4 for it. The schemas are deliberately designed to be AAL-4-capable so nothing breaks when stronger key custody arrives — but capability is not a claim. The same discipline covers attribution (“bound, declared-not-verified” unless stronger binding exists) and streaming output (observe and truncate, never “prevention”). GROUNDED · Labs This candor is not a legal disclaimer bolted onto the pitch; with a security-reviewer persona it is the pitch — the vendor that grades its own assurance level honestly is the one you believe about everything else. STABLE

With the standard’s independence and verification machinery in place, the curriculum turns to what underwriters build on top of it: triggers, verified loss ratios, and — eventually — capital markets.

→ 20-00 Parametric triggers

Drill this page →18 bank questions stand behind what you just read. Check it while it’s warm.