v1.0 — DRAM public release

First public release of the DRAM dataset and reproduction code accompanying the paper
"Set but Unseen: Baseline-deviation and redaction measurement of risk-allocation terms
in public procurement contracts."

What's included

• Dataset (dataset/) — 8,534 UK framework call-off contracts, one derived record each:
  framework, frozen baseline, and per-term coverage state / value / provenance for the four
  liability terms (cap, exclusions, insurance, service credits), plus per-call-off metadata
  and the corpus identifiers.
• Validation (validation/) — the 292-case reference set (labels only).
• Baselines (baselines/) — the frozen per-framework defaults.
• Reproduction code (code/) — reproduce.py recomputes every headline figure and
  interval offline from the released inputs; score_reference.py reproduces the validation
  table.
• Expected outputs, robustness, docs — frozen targets, representativeness tables, data
  dictionary, and masking policy.

Reproduce

pip install -r requirements.txt
cd code && python reproduce.py        # -> "12/12 checks passed."

Data handling

Derived, structured data only. 67 value cells whose source documents carry a recoverable
visual redaction are withheld; short (<=160-char) provenance snippets are retained and removed
on those cells. No full clause text and no personal data are released. See MASKING_POLICY.md.

Integrity

All files are checksummed in MANIFEST.sha256 (54 files).

Licence

Code: MIT. Data: CC BY 4.0, derived from Crown-copyright records on Contracts Finder (Open
Government Licence). See LICENSE.