Known Limitations

Updated: 2026-07-29

1. Real history is shallow. Snapshot-based change detection grows only from collection onward. Fixture snapshots prove the mechanism, not historical real-world changes.
2. Posting-level jobs are synthetic. There is no official, freely reusable BA postings API in the selected design. Skill extraction demonstrates the method on visibly synthetic postings; it does not describe current demand.
3. Company monitoring is synthetic and incomplete. Entity resolution, review queues, historization, and change detection work, but the public demo has no evidence-backed real company universe.
4. TED is the only implemented live source. The collector uses the official anonymous Search API and was verified live on 2026-07-29. GENESIS, CORDIS, jobs, and company inputs still use fixtures or synthetic records.
5. Procurement coverage is biased toward larger notices. TED covers above-threshold EU procurement. Sub-threshold German purchasing depends on a future verified integration with Datenservice Öffentlicher Einkauf.
6. The live TED request is intentionally bounded. It selects configured Region-10 NUTS codes, active notices with a future tender deadline, sorts by publication date, and requests at most 100 records per run. This is a current radar, not a complete historical export.
7. National research funding is underrepresented. CORDIS-shaped fixtures demonstrate EU research data. Förderkatalog and GEPRIS remain unsuitable for automated reuse until access and terms are verified.
8. The warehouse backend is SQLite. DuckDB and PostgreSQL remain architectural targets, not currently proven implementations.
9. Skill extraction is dictionary-based. It is negation- and inflection-limited. Each extracted or inferred value retains its method, and a coverage warning surfaces likely under-extraction.
10. The dashboard is a static snapshot. It has text/NUTS filters, company profiles, tender/source details, and quality status, but it does not refresh without regenerating the file and has no API-backed multi-user state.
11. No causal claims. Co-movement between skills, companies, tenders, and research does not prove causation.
12. Automation is prepared but not externally proven. CI and GitHub Pages workflows exist locally. Until a GitHub remote is configured and the first run is green, the project must not claim a public deployment.