Home / Trust

Scope. Authority. Proof.

Every change.
Accounted for.

Refinery is designed to fail closed. Uncertain data is limited or blocked. A gateway read cannot execute a repair. No completed repair claim exists without an exact source readback and a durable receipt. These safeguards keep the limits visible before a decision is made.

Control before change. Evidence after.ILLUSTRATION
01

Approve the purpose and the fields.

The example consumer may read account status. An unrelated credit-limit field stays outside its scope.

statusApproved for this purpose
credit_limitOutside the approved scope
Scope before reads. Authority before writes. Proof before a fixed claim.Sample illustration · no live connection or changes.
Two separate paths · planned availability

Data for this task. Proof for this change.

Your AI needs more than database access. It needs to know whether the data is usable for this task, right now.

  1. AI or applicationDeclares its task and requirements
  2. Read-only gatewayChecks the approved request
  3. Approved source dataSource database stays authoritative
  4. Result with evidenceEligible, limited or blocked
Read evidence

Why this data could be used.

Explains why data was released for a specific purpose and time, checked against approved freshness, quality, scope and policy requirements. A limited result carries explicit restrictions; a blocked result withholds the requested data.

A read check does not create a repair receipt.
Repair receipt

What changed at the source.

Proves an authorized source correction only after exact fresh readback and durable receipt evidence match the operation. A proposal, a routine check or an AI request is not proof of a repair.

The gateway only reads. Repair uses a separate governed path.

Existing applications may keep direct database connections. Only consumers using the optional gateway receive its eligibility check; the source database remains authoritative. An AI can request a supported repair, but cannot authorize or execute one merely by asking.

The browser sample is available. Live customer connections, API/MCP access and repairs are not publicly released.

Freshness is currently table-level. Row-level freshness is not release-proven.

Three structural rules

The same safeguards in every tier.

Access

Purpose-bound and read-only

An approved AI use case receives only its allowed fields. A gateway request cannot call the repair executor — reading and writing are separate authorities, not separate permissions on the same one.

Uncertainty

Unknown means limited or blocked

When the freshness signal is stale or unknown, or data is conflicted, disallowed, wrong-identity, or insufficiently proven, it is never quietly returned as trusted.

Proof

Readback before fixed

After an allowed repair, Refinery re-reads the authoritative source and keeps a durable receipt of the exact result. A successful write is not a verified outcome.

The standard

Never claimed without proof.

  • No trusted-data claim without a current purpose-bound check and the applicable table-level freshness signal
  • No row-level freshness claim until that gateway is release-proven
  • No repair claim without governed execution
  • No fixed claim without an exact source readback
  • No proof claim without a durable receipt chain
  • No silent resolution of identity conflicts or ambiguity
The gaps

Not claimed today.

  • Universal or fully certified connector coverage
  • Row-level freshness certification
  • General availability or a production uptime SLA
  • Formal security or compliance certifications
  • Customer counts, logos, ratings, or case-study outcomes
  • Autonomous identity merges, destructive changes, or ambiguous finance repair
Clear boundaries

Know what the evidence supports.

Every tier follows the same rules. Scope, authority and proof determine what Refinery may do.

BoundaryWhat it means for your data
Approved readsOnly the fields and purpose you approve may pass. The gateway is read-only and cannot execute a repair.
FreshnessTable-level freshness is evaluated against the agreed window. Row-level freshness is not release-proven. A table signal cannot establish the age or correctness of every record.
Source changesSupported repairs require current policy, exact authorization, connector and repair certification, an exact expected-before check, fresh source readback and durable receipts.
Uncertain dataUnknown values, identity conflicts and ambiguous financial facts are blocked or escalated. A model cannot invent the missing evidence.
Access todayThe sample demo is available. Product access and purchasing are not open yet. Join early access to register interest.
Before any engagement

Five things must be explicit.

No customer-data engagement begins on trust alone. Purpose, scope, access, decision policy, and evidence plan are reviewed and documented first.

01

Use case and source

The exact AI or workflow decision, the PostgreSQL scope, the data class and volume, and the accountable business owner.

02

Access model

The minimum credentials, permissions, network path, storage, and retention needed for the agreed mode — and nothing beyond it.

03

Connector posture

Whether the path is read-only, setup-ready, certification-required, or proven for bounded repair. These are not interchangeable.

04

Decision policy

Which cases may be decided deterministically, which need a named human, and which are refused outright.

05

Evidence plan

What will be recorded, how receipts are retained, and how the engagement is reconciled and audited afterwards.

And before that

Nothing at all

The fit check and the waitlist need no credentials, no exports, and no production records. If someone asks you for those before the release gate, it is not us.

Boundaries you can check yourself.

The demo runs the same refusals described on this page — including the one case that is deliberately never repairable.