The latest published release of every crate in this workspace receives security updates. Pre-release tags (0.y.z before 1.0.0 stabilization) are patched on a best-effort basis for the immediately-previous minor; older minors are not backported.

Do not file public issues for vulnerabilities. Public disclosure before a coordinated patch exposes every downstream user.

Instead, report privately via GitHub's Private Vulnerability Reporting:

• https://github.com/Dicklesworthstone/frankenpandas/security/advisories/new

(Enable Private Vulnerability Reporting in repo settings if you are the maintainer; until that ticks on, reports to the maintainer's email registered on crates.io are the fallback channel.)

Expected response timeline:

• Acknowledgment within 48 hours
• Initial severity assessment within 7 days
• Coordinated disclosure within 90 days (or earlier if upstream fixes land faster)

• Memory safety issues in frankenpandas-authored unsafe code (none exists today — every crate carries #![forbid(unsafe_code)] — but report anyway if you find our attribute removed).
• Logic bugs in IO parsers (fp-io: CSV, JSON, JSONL, Parquet, Excel, Feather, Arrow IPC, SQL) that allow crafted untrusted input to cause panic, OOM, unbounded memory growth, infinite loop, or integer overflow.
• Logic bugs in expression evaluation (fp-expr: eval_str, query_str, parse_expr) that allow crafted expression strings to crash or consume unbounded resources.
• Logic bugs in SQL bindings (fp-io SqlConnection trait) that allow crafted queries to panic the driver or bypass parameter binding.
• Differential conformance divergences that silently change numeric results vs pandas when the divergence affects scientific correctness (data corruption, not just dtype differences).

• Behavior differences between frankenpandas and pandas that are documented in DISCREPANCIES.md. These are intentional divergences.
• Issues in transitive dependencies (arrow, parquet, calamine, rusqlite, serde_yaml, etc.). Report upstream to the dep maintainer; we will accept a patch bump once upstream fixes and flag affected versions in our CHANGELOG.
• Denial-of-service via legitimate-but-large inputs (e.g. a 1 TB CSV file). Mitigate at the application layer by bounding input size before calling frankenpandas APIs.
• Issues requiring physical access or compromised credentials to the user's machine.

We practice coordinated disclosure:

1. Reporter files privately via GitHub Security Advisories.
2. We triage and draft a fix.
3. Reporter reviews the fix for correctness.
4. We publish a patched release with a RustSec Advisory ID.
5. Public disclosure happens simultaneously or after step 4.

We credit reporters in the CHANGELOG under a ## Security section and in the published RustSec Advisory unless the reporter prefers anonymity.

This workspace enforces (as of 2026-04 CI):

• cargo audit --deny warnings on every PR (br-frankenpandas-36qc).
• cargo deny check advisories bans licenses sources on every PR (br-frankenpandas-36qc).
• #![forbid(unsafe_code)] on every crate (verified).
• Fuzz regression corpus on every PR (br-frankenpandas-zjme).
• Differential conformance against live pandas on every PR (br-frankenpandas-d6xa).
• Dependabot weekly updates for Cargo + GitHub Actions.

These gates minimize silent introduction of known vulnerabilities. They do NOT substitute for your report when you find something novel — please report.

(Populated as valid reports arrive. See CHANGELOG.md § Security for per-release credits.)