Roadmap¶
Direction, not promises — ordered by what makes the tool more trustworthy, which beats making it bigger. Shaped substantially by early adopter feedback (see #1 for the kind of input that moves things here).
Next (v0.2.x)¶
- PostgreSQL & MySQL integration tests in CI (testcontainers), turning "supported by construction" into "verified on every commit". SQL Server / Oracle in a nightly/manual matrix after that.
- Run manifests: record exactly what a run touched (tables, columns, strategies, row counts, config hash) → idempotency guards ("this target was already masked by run X"), resumability, and an audit artifact.
- Detection benchmark: a public, versioned dataset with per-rule precision/recall/F1 and published false-positive/negative examples, so detection quality is a measured number, not an adjective.
- Stable public API: a small
dbmask.scan() / apply() / verify()surface with typed reports and a deprecation policy, so the library story stops leaning on internal modules.
After that (v0.3.x)¶
- Performance: bulk seed-map lookups/writes, set-based updates
(
executemany), bounded parallelism per table. - Governance export: publish classification results (column → sensitivity → rule → decision source) to OpenMetadata / DataHub, so dbmask's findings feed the catalogs organizations already run.
- Locale packs: dictionaries and patterns beyond the US-centric starter set.
- Machine-readable audit reports for scan/mask/validate in one schema.
Toward 1.0¶
The bar 1.0 has to clear, in one sentence each:
- no known way to make the CLI write when it said it wouldn't;
- no path that reports success over unverified data;
- a documented threat model (see Security model) with the claims tested;
- integration tests against at least three real engines;
- a stable API under a deprecation policy;
- independent adopters on record (ADOPTERS.md).
How to influence this¶
Real-world reports beat feature ideas. The single most useful thing: adopter feedback from an actual attempt — including the attempt that made you pick a different tool.