Working thesis · Published · 6 min read
Trust should be a shared fact
Security questionnaires exist because trust facts have no durable home. Answering faster is useful. Making repeated re-derivation unnecessary is the actual goal.
Somewhere today, a security engineer is answering whether their company encrypts data at rest again. The answer may already exist in an audit report, a policy, and dozens of prior questionnaires. A new buyer asks in a new dialect, so the company reconstructs the representation one more time.
01 · Facts with no home
The questionnaire is a protocol workaround
A completed questionnaire carries two legitimate things: a point-in-time attestation and accountability for what the vendor represented. Any replacement has to preserve both. The waste is not attestation itself. It is bundling attestation with fresh reconstruction of facts the parties have already established elsewhere.
Prices have catalogs. Identity has certificates. Ownership has registries. Trust facts like encryption, access review, background screening, and subprocessors mostly live in prose, private spreadsheets, and buyer portals. Nothing durable accumulates between the parties.
02 · Smaller state
Reality is smaller than the questionnaire space
There are many ways to ask about encryption, access, incident response, or business continuity. The underlying facts are fewer. A complex buyer question may compose several facts, but it should not require a new hand-maintained truth simply because the wording changed.
In database terms, the questionnaire is a query and the vendor's posture is the state. Current practice materializes every query result as a separate artifact and lets each artifact age independently. A better system stores governed state and computes the buyer-specific view.
03 · AI's actual role
AI makes the shared layer bootstrappable
Models are useful at translating arbitrary buyer language into and out of a canonical representation. They can map phrasings, retrieve evidence, and draft a projection for review. That changes the economics of semantic normalization.
The model is migration technology, not the authority. A terminal world in which one model writes questionnaires for another model to grade would move more prose without creating more knowledge. The durable value is the maintained state, provenance, and accountability beneath the exchange.
04 · The representation
A shared fact is a governed claim, not a decree
A useful trust fact needs four properties:
- Canonical
- One addressable representation beneath many buyer phrasings.
- Provenanced
- The claim carries evidence, scope, owner, observation date, and assurance type.
- Governed
- Approval history and composition rules remain inspectable and versioned.
- Living without forgetting
- Current state can change while transaction-time snapshots remain durable.
Provenance does not make a claim true. It makes the trust calculus legible. An auditor's evidence, a platform capability, and a company assertion should not all collapse into an indistinguishable “yes.”
05 · Adoption path
The questionnaire can distribute the thing that replaces it
Vendors first adopt a canonical profile for a selfish reason: it helps answer the questionnaires buyers already send. Buyers change nothing. Over time, answers can arrive with durable references to their facts, evidence, owners, and approved snapshots.
If those references make reviews easier, some buyers may eventually accept the snapshot instead of requesting a fresh reconstruction. That is a future adoption decision, not a capability or network Tyrvar claims today. The path matters because it proves value before asking either side to adopt a new ontology.
Boundary
Shared facts do not standardize risk appetite
Two buyers can inspect the same facts and reach different risk decisions. That is legitimate. Questionnaires conflate “what is true about this vendor?” with “is that acceptable under our policy?” Tyrvar's thesis applies to the first. Buyer judgment remains buyer-specific.