Every mirrored datasheet on the platform is now read, not only stored. Three layers sit on top of the typed specification layer:
- Datasheet digest — facts by section (identity, performance, electrical, interfaces and connectors, mechanical, thermal, environment and radiation, qualification and heritage, lifetime, options, compliance, commercial). Each fact is machine-read and verified against a verbatim quote from the page it comes from, and links into the PDF at that page.
- Facts across a category — one key (connector, protocol, survival temperature, TID, flight heritage, TRL, export control, lead time, price…) for every product of a category, with coverage stated honestly.
- Ask the datasheet — a question to one product’s datasheet returns the closest excerpts with page numbers; an exact-term scan across a category returns the exact list of products mentioning it.
The engineering agent uses all three, and comparisons now show a “From the datasheets” section below the typed values. Machine access: /mrd-connect/digest, /mrd-connect/facts, /mrd-connect/datasheet and /mrd-connect/gaps in REST, the same four tools in the MCP connector (v1.1.0) and in the OpenAPI document — see Connect to MRD.
Data gaps. For every product with typed specifications the platform now knows which core fields are missing and where to get them: already read from the datasheet, mentioned in the datasheet text, absent from the datasheet, or no datasheet at all. Vendors see what to publish; buyers know what to ask.