Reports, sample audits, methodology, the Sentinel Industry Index, our changelog, and the API reference — all in one place.
Public research from the PartSentinel team.
Which brands' catalogues have actually leaked into model training data, and which still hold an exploitable moat. Audit of 412k references, 8 LLMs.
Methodology paper: prompt design, sampling protocol, ground-truth alignment, scoring formula and per-vertical calibration constants.
Quarterly benchmark of how each major model performs on industrial reference data. Updated each quarter.
An anonymised illustrative audit on a fictitious manufacturer (50 references). Includes the executive summary, per-reference scorecards, leak detection, AI Act dossier and raw model responses.
Open the sample report →Free utilities for evaluating your AI exposure.
Six-question self-assessment aligned with EU AI Act Article 53(1)(d). 4 minutes.
Start checklist →Drop one reference (SKU or OE number). We'll tell you what 5 LLMs say about it, instantly.
Run a probe →Open-source parser for industry catalogue formats — used inside PartSentinel, released to the community.
View on GitHub →What shipped, when.
Replaces Opus 4.6 in the default panel. Calibration constants refreshed across all 18 verticals. Existing audits get a re-run automatically.
New module that flags when confidential cross-reference tables appear inside model responses. Public reference: blog post.
Article 53(1)(d) compliance pack, exportable as a single signed PDF + JSON manifest. Includes provenance trail and per-model attestations.
For PIM, MDM and BI teams. Score any reference in your catalogue programmatically. Stable v1 endpoints, signed webhooks, OpenAPI 3.1 schema.
Read the OpenAPI spec →# Score a single reference
curl -X POST https://api.partsentinel.com/v1/audit \
-H "Authorization: Bearer $PARTSENTINEL_API_KEY" \
-d '{
"reference": "NVX-22010",
"vertical": "automotive_aftermarket",
"models": ["model-a", "model-b", "model-c"]
}'Drop one of your own references in 90 seconds, then talk to us about scoping a full catalogue audit.