Trust, in the open
The trust model is not a claim on this page. It is Apache-2.0 code with a public eval suite that locks its behavior. Monetizing operations never means hiding correctness.
The one sentence that matters
Confirmation means an actor attested; it does not prove objective truth.
When a user says “yes, Thursday works,” ContextDB records that a specific actor attested a specific fact at a specific time. ContextDB can show who said yes to what, when. That is an audit trail. It is not a truth oracle, and we do not sell it as one.
When a fact is trusted
From the SDK's public trust policy: a fact is trusted only when it is action-relevant and one of:
| Path | Rule |
|---|---|
| Confirmed | An actor attested it (confirmed). |
| Corroborated | Independent corroboration meets the class threshold. |
| First-party shortcut | user_stated at or above the policy
confidence threshold, and the slot class is not excluded (health/legal/identity
are stricter). |
Injection-suspect and contested facts are never trusted. Unknown slots cannot use the first-party shortcut by default.
What is locked by evals
The SDK's
trust-model evals
are the executable spec: epistemic typing, corroboration, contest,
VerifyBeforeAct, PII-before-embed, injection rendering, and
verifiable forgetting. Cloud never patches that correctness privately. A
change lands in the Apache SDK with an eval, or not at all.
What we do not claim
- No compliance certifications are held or claimed today.
- No uptime percentages, SLAs, or latency numbers are published.
- No high-availability or multi-region topology is claimed.
- ContextDB does not replace your database, Supabase, or Hasura.
- The hosted cloud is internal alpha; the eval-locked trust model in the SDK is what is real today.
Read the boundary: OPEN_CORE.md and COMMERCIAL.md.
Inspect the rules, not a trust badge.
The policy matrix and executable evals are public.