Trust
Data and accuracy
What TenderReader knows today, how extraction is tested, and what is still deliberately unpublished.
Source
| Source | Count | Freshness |
|---|---|---|
| Find a Tender notices | 626 | 2026-06-12 |
| BOAMP notices | 1195 | 2026-06-14 |
| Contracts Finder awards | 847 | 2026-06-12 |
Dated evaluation results
- Corpus
- v2
- Engine
- deterministic-stub-v0
- Cases
- 15
- Fields
- 150/150
- Overall accuracy
- 100.0%
| Field class | Correct | Accuracy | Bar | Status |
|---|---|---|---|---|
| Deadlines | 30/30 | 100.0% | 99% | Clears generated-corpus bar |
| Values | 15/15 | 100.0% | 97% | Clears generated-corpus bar |
| Eligibility | 26/26 | 100.0% | 97% | Clears generated-corpus bar |
| Criteria | 26/26 | 100.0% | 97% | Clears generated-corpus bar |
| Scope | 15/15 | 100.0% | 97% | Clears generated-corpus bar |
| Deliverables | 15/15 | 100.0% | 97% | Clears generated-corpus bar |
| Quantities | 15/15 | 100.0% | 97% | Clears generated-corpus bar |
| Lots | 8/8 | 100.0% | 97% | Clears generated-corpus bar |
Language split: v2 is primarily generated English fixtures; real French evidence exists for one France Travail Services RC/DCE pack, but there is not yet an automated French corpus or per-language score.
Status: the generated corpus clears the bar, but the public 97% critical-field / 99% deadline bar is not yet proven on real packs with a production model.
Provenance
Notices and awards come from public feeds: Find a Tender, Contracts Finder, and BOAMP. Product surfaces keep the source name and the OGL/Etalab attribution with the data.
Evaluation method
The current CI gate uses the v2 corpus: 15 generated tender-pack fixtures, 150 labelled fields, multiple formats and sectors, and per-class scoring.
That corpus protects against regressions. It is not yet a public accuracy claim for real customer packs.
Accuracy numbers
The numbers below come from the v2 evaluation corpus and render from the evaluator data, not hand-written copy.
They protect regressions but are not yet a claim of real customer-pack or production-LLM accuracy.