Ataccama brings data trust to Apache Ossie — open-source converter for DQ signals in the semantic layer (press release + blog)
Tag: S-2026-09-09-ataccama-apache-ossie-converter Type: article (vendor press release via GlobeNewswire, vendor blog, plus one independent trade-press report) Author(s): Ataccama (release; quotes Jessica Smith, CPO, Ataccama, and Josh Klahr, Head of Product Management, Snowflake); Anja Duricic, Ataccama (blog); Erik van Klinken, Techzine Global (independent coverage) Date of source: 2026-09-09 Date ingested: 2026-09-10 Authority weight: medium — a dated, technically specific primary announcement from the vendor, corroborated on the facts of the announcement (not on product performance) by Techzine; low for outcome claims about agent behaviour and for the relayed Gartner/IDC statistics Raw file: S-2026-09-09-ataccama-apache-ossie-converter.md
What it claims
Ataccama announces that it will open-source a converter that reads catalogue items and data-quality scores from Ataccama ONE via the Ataccama API and emits YAML compliant with Apache Ossie (Incubating) — the semantic-interchange specification that began as Snowflake’s Open Semantic Interchange (OSI) and entered the Apache Incubator in mid-2026. The output is meant to load into Snowflake Semantic Views, Databricks metric views, the dbt Semantic Layer, or any other Ossie consumer.
The converter carries four things into each Ossie definition: steward-curated business terms mapped into Ossie’s ai_context field; a structured data-quality block (DQ state, pass-rate percentage, the threshold measured against, a below-threshold flag, and the count of active findings) added as an Ataccama extension; optional plain-language AI warnings appended to the AI instructions for datasets flagged with quality issues; and an automated refresh design in which the converter runs as a scheduled job (Airflow DAG, GitHub Action or dbt job step) so exports refresh at the cadence of DQ runs. Detailed evidence — which checks failed and which records are affected — is deliberately kept out of the file and served live from Ataccama’s MCP Server when an agent needs to explain or act on a flag.
The argument is that semantic layers settle what a metric means and how it is calculated but cannot say whether the underlying data is complete, current or in range; as agents move from interpreting to acting “with less human review”, quality must become part of the context they reason over rather than a separate dashboard. Because Ataccama evaluates quality in source systems as well as cloud platforms, it claims the signals travel across heterogeneous estates without consolidating data onto one platform. The converter will “ship open source, with a documented field mapping, a README, and tests”, joining Ossie contributions from Snowflake, Databricks and dbt. A webinar with Snowflake is scheduled for 15 September 2026.
Techzine independently reports the announcement, characterises Ossie as “a specification, so not a product”, notes the Ossie ecosystem listing (Snowflake, Databricks, Kyvos, Honeydew, Strategy, Veezoo; 35 GitHub contributors), and observes that the Snowflake connection is partly financial because Snowflake Ventures invested in Ataccama.
Notable quotes
“An agent may know what revenue means, but not whether the revenue data its reading is complete or current, and it will still produce a convincing answer.” — Jessica Smith, CPO, Ataccama (press release, para. 6)
“Trust signals that are current only on launch day aren’t trust signals; this keeps them live.” (blog, “Automated refresh” bullet)
“None of this is about replacing the layers the industry is building. It’s about making sure they can be believed.” (blog, closing of launch section)
“It is a specification, so not a product, in the form of a declarative YAML format for metrics, dimensions, joins and relationships that different platforms can read and write.” — Techzine, “What Ossie is, and isn’t”
What’s speculative vs. asserted
- Asserted (vendor): the converter will be open-sourced; the field mapping described (ai_context enrichment, DQ extension block, AI warnings, scheduled refresh); the MCP Server as the route to detailed evidence; source-system DQ evaluation; Snowflake/Databricks/dbt as fellow Ossie contributors.
- Asserted (independent, Techzine): Ossie’s Apache Incubator status; the ecosystem/contributor listing; Snowflake Ventures’ investment in Ataccama.
- Speculative / unverified: that agents will in practice “qualify or investigate questionable data before it informs an answer or action” — this depends on the consuming agent honouring the warning and is not demonstrated; the release date, licence and repository for the converter (none given — future tense throughout); the relayed Gartner (“more than 60% of enterprises … agentic AI within the next two years”) and IDC (“15% productivity loss by 2027”) figures.
- Not addressed: any regulator, standard, evidentiary threshold, or financial-services customer.
Topics this feeds
- Ataccama — company page: product/positioning update and Tracked changes entry.
- Agentic Data Access Governance — topic: extends the “trust layer” thread (Data Trust Index + MCP Server) to an open semantic standard; records the OSI → Apache Ossie rename.
Open questions raised
- When will the converter actually ship, under which licence, and will the DQ extension block be adopted by the Ossie spec or remain an Ataccama-specific extension?
- Do the consuming platforms (Snowflake Semantic Views, Databricks metric views, dbt Semantic Layer) surface or enforce the below-threshold flag, or merely pass it through?
- For a regulated firm, is a pass-rate/threshold/flag block in a semantic YAML file — refreshed on a schedule — acceptable evidence of data-quality control for BCBS 239 or EU AI Act record-keeping, and how is its history retained? [inference — not raised by the source]