TRC
Trace
Retrieve and link
What it does
Trace finds what an input relates to. It runs structured and semantic retrieval across normalised outputs, where every ranking carries the reasons for it.
A match without a reason cannot be reviewed, so the reasons are part of the result, not an explanation generated afterwards.
What it accepts
Outputs from Read, searched and compared across sources. Finds what matches a query, surfaces where two sources describe the same thing differently, and ranks results with reasons.
- Checked outputs
- Outputs that have passed through Conduit and Signal, with their check history.
- Queries
- Structured queries against record fields and semantic queries against record content, singly or combined.
What it emits
- Ranked results
- Candidate matches in ranked order, each carrying the evidence that produced its position.
How it is configured
Configuration belongs to the client. It encodes what your specialists already know, and it is versioned like code.
- Match definitions
- What counts as a candidate match for each input type.
- Ranking configuration
- The weights and rules behind the order, versioned.
What it produces as evidence
- Match records
- A versioned record of each retrieval: the query, the configuration version, the results and the reasons.
What it is tested against
Trace is tested against acceptance criteria agreed before build: agreed queries return the agreed matches on a defined corpus, and every result carries its reasons. Retrieval quality is measured by Gauge, not asserted.