Institutional memory · Investigation
A decision raises the next question.
Recover the context of a security exception or AI-assisted risk decision. Then ask where else the same policy, situation or evidence gap appears.
Roadmap · Four investigation questions
Ask what changed.
Ask what is missing.
These are proposed investigation tasks, not queries executed against a tenant. No results, rankings or counts are implied.
Find every decision like this.
Start with a decision, define the population, then look for related records. “Every” is the question to investigate, not a completeness claim.
Which similar situations had different outcomes?
Compare the recorded context, policy references and outcomes. A similarity score would guide review; it would not establish that two situations were equivalent.
Which decisions relied on this policy version?
Filter by a recorded policy reference and inspect the source record. A reference shows what the sender recorded, not that the policy was followed.
Which exceptions lack required evidence?
Define the evidence required for the review, then expose gaps in the collected records. Missing from the collection does not mean an event never happened.
Roadmap · Proposed result disclosure
Keep the boundary beside the answer.
A useful result needs a visible scope. Semantic search cannot establish completeness. Exact filters can define a population, but cannot establish that every real-world decision reached Recourse.
Disclosure for the questions above
Illustrative specification only. No search has run.
- Coverage
- No indexed population in this preview. Proposed searches would state the included sources, population filters and known collection gaps.
- Query
- The four example questions above are proposed tasks. No query was executed; an actual result would preserve its query and exact filters.
- Index version
- Not available in this preview. There is no search index.
- Collection cutoff
- Not set. No collection was selected for these illustrations.
- Ingestion watermark
- Not available for these illustrations. No ingestion population or time boundary was measured.
- Searchable-through
- Not available in this preview. The planned index coverage watermark is separate from receipt and durability states.
A recent receipt does not establish that an index includes the record. An index watermark does not establish that the sender captured every decision.
Start with the record you have.
The demo lets you inspect synthetic ledger entries and look up an exact record and its receipt. Use those bytes to inspect what was recorded and run the offline verifier.
Available in the preview
Roadmap Investigation workspace
Exact filters for model, policy version, date, subject and outcome; semantic retrieval; and review of similar situations with different outcomes.
Results will need source records and the coverage disclosure above before a reviewer can assess them.
Development signer means integrity under a published dev key and no independently trusted time.
Roadmap · Precedent library
Remember the conditions, too.
A past approval is useful only with its context. The planned precedent library keeps approved decisions with the conditions that bounded them and the dates they need review or expire.
People and agents could retrieve that context before making another decision. A retrieved precedent would remain evidence for review, not permission to repeat the outcome.
Approval and source
Who recorded the approval, what was approved and the underlying decision.
Conditions and policy
The applicable policy version, constraints and evidence needed to reuse the context.
Review and expiry
When the precedent needs another review and when its permission ends.
This is a proposed library structure. No approved precedents or review dates are asserted by this page.