Plan, apply and roll back changes to a production ontology, with a blast radius report and a proof an auditor can re-check.
Rust ai-nativedescription-logicsformal-verificationinfrastructure-as-codeisabelle-languageknowledge-graphleanlean4mcpmcp-serverontologyowlowl2reasoningrustshaclsparqltheorem-proving
Merge pull request #282 from rudycelekli/fix/ingest-tabular-header-order-signed-20261006
FABIOTESS8120ab343182
diff --cc CHANGELOG.mdindex 77540103,852b5b62..fad5b3d1--- a/CHANGELOG.md+++ b/CHANGELOG.md@@@ -55,26 -46,8 +55,28 @@@ All notable changes to Open Ontologies### Fixed+- XML ingest preserves CDATA and text split by comments or CDATA boundaries,+ and rejects undecodable entities rather than loading a silently empty value.+ Whitespace normalization applies once to the completed field.+- Explicit ingest formats are honored by the local CLI and the MCP ingest/map+ tools rather than silently falling back to filename detection. Omitted formats+ retain extension detection, and unsupported explicit formats return an error.+- Read extraction-scaffold metadata and mapping-generator declarations from+ every loaded graph, so TriG and N-Quads retain the same discovered classes,+ properties, typed extraction checks, and prompt fields as Turtle schemas.+- Materialize documented row-number identifiers in batch and MCP induction+ when no input column identifies the rows. Choose an unused internal field so+ existing `__row` data remains a property instead of merging different rows.++- Render XLSX numeric calendar cells as ISO dateTime values instead of Excel+ serial numbers, respecting the workbook's 1900 or 1904 calendar. Duration+ cells retain their existing representation; calendar cells do not distinguish+ date-only values from midnight dateTimes.+- Parquet ingest formats each non-primitive cell with Arrow's value formatter,+ preserving date and decimal lexical values instead of embedding a debug dump+ of the complete column in every row. Formatting errors fail the import.+ - Preserve XLSX header and Parquet schema order during induction so the first+ filled, unique column remains the identifier candidate, as it is for CSV.- **The embedding model is loaded once per process rather than once per HTTPsession** (#262). The HTTP arm hands the same server constructor to