Scribe
Makes any source speak the same language.
Scribe is the agentic conformance engine. Its single verb is render: take any data source (a scanned document, a sensor feed, a government registry, a legacy database) and turn it into records that conform to the shared schema. It owns reconciliation (resolving who and what a record refers to) and commoditisation (mapping arbitrary sources onto the standard). Every rendered record keeps its provenance and lineage, so a conformed fact never loses the trail back to where it came from.
One engine, two targets. Rendered against the cross-system schema, a source becomes a tradable commodity. Rendered against a Frame app’s configuration, it becomes data that app can ingest. The target is a parameter; the engine is one, and it runs a continuous render / evaluate / learn loop that accumulates reusable mapping knowledge per source.
Scribe is what lets DAAC reach data it did not capture (the registries and platforms that already exist but cannot talk to each other) and make it comparable to the data the ecosystem records first-hand.
Where it sits. Sibling to Spine; neither depends on the other. Consumes the shared conventions as render targets; needs no Frame app to run.
- 01RenderMap + reconcile
Map a source’s constructs to the target schema; resolve its entities to identity records, provenance intact.
- 02EvaluateScore the render
First-pass conformance and per-mapping confidence; low-confidence mappings are flagged, not accepted.
- 03CritiqueRefute, don’t rubber-stamp
An independent critic re-examines each mapping to catch false friends, grounded, yet semantically wrong.
- 04ImproveCapture the why
Corrections are captured with rationale, context and priors, from a human or a higher-trust agent.
- 05LearnNo model retrain
Corrections feed the next render; per-source templates compound in prompts, rules and corpus.
corrections compound into the corpus, the corpus is the model
the falsifiable proof it learns, must trend down
share of fields auto-validating before human touch
do confidence scores match actual correctness
false-friend mappings the critic rejects
watched for drift toward rubber-stamping
These are the vital signs Scribe is designed to report, not measured results. Until it runs, its Console panel is schema-stubbed and clearly labelled. No panel fakes data.
- Digital system development and maintenanceA described situation becomes a system people work in, and it stays in step with what it was built on instead of drifting behind it.
- Migration of existing systems for interoperabilityThe system people already depend on keeps working, on a substrate that can be maintained, and the legacy records come across with it, reconciled rather than stranded.
- Documentation mapping to schemaSource documents become schema-conforming records an operator can query, checked by an independent critic before a person reads them, and every correction the operator makes teaches the next render.
- Mapping any database to a reference schemaA database’s records render against a reference schema with that schema as a parameter, so the mapping is a run rather than a new piece of work.
- Targeted knowledge discoveryA local data supply reaches the markets that want it: discoverable, permissioned and payable at the holder’s own discretion, with nobody taking custody of it to make that true.