Changelog¶
Notable platform changes, newest first. The dataset catalog grows continuously — the
live count is always at GET /stats and in the dataset reference.
July 2026¶
- User accounts — self-serve signup at
/signupwith an account dashboard (/account): mint and revoke up to 5 free-plan API keys, watch usage live (per day and against the monthly quota), verify your email, reset or change your password, and delete the account (GDPR erasure — keys and usage records go with it). Sessions are cookie-based and server-side; the/v1data API is unchanged and remains Bearer-key only. Usage is now also metered per day to feed the dashboard chart. - API response polish — JSON 404s for every unknown
/v1path (no more HTML fallbacks);endpointon dataset-list rows and search hits;?domain=filtering (and strict parameter validation) on/v1/datasets;linkson every list endpoint;X-Request-IDcorrelation on all responses and in error bodies; CORS for browser consumers;HEADsupport; one RFC 3339Ztimestamp format everywhere;Last-Modifiedon exports plusformat=jsonas an ndjson alias;retry_after_secondsin 429 bodies. - Catalog past 198 datasets across 10 domains (~44M rows), onboarded through the manual-review flow in tranches.
- Multi-sheet workbooks and wide pivots — sheet unions with per-sheet header re-alignment, and unpivoting of wide year/party/age columns into long rows.
- v1 response envelope — every list endpoint now returns
{data, pagination, meta, links}, with source attribution andlast_updatedinmeta. - Codes and names — official identifiers stay verbatim; human-readable names ride alongside (CN8 descriptions, CAEN nomenclature, election column legends).
- ONRC company registry — registered firms and authorized CAEN codes, with decode tables served as datasets.
- Mislabeled-Excel sniffing: files served with the wrong extension are detected by container signature and parsed correctly.
Earlier in July 2026¶
- Universal Silver standardization — one rule set for every dataset: text hygiene, null tokens, thousands separators (both conventions), Romanian date formats. No per-dataset cleaning code.
- Self-serve API keys and platform stats endpoint; landing site and this documentation served from the API host.
- Bulk onboarding tooling — CKAN discovery, config scaffolding, and batch ingestion took the catalog from 12 to 100+ datasets.
- Performance work — Redis response cache with
ETag/304and gzip,COPY-based bulk loading, keyset cursors for deep pagination, Bronze object storage. - Hardened public surface — rate-limited key minting, proxy-aware client IPs, security headers.