Recipe 42: Outbound disclosure screening #42

Recipe 42 of 60 · Level 4 (Advanced) · Trust & security · Noul

Use case

Screen proposed outbound messages for disclosure concerns against supplied sharing rules while deterministic controls hold flagged or uncertain samples for review.

What to build

The folder recipes/42-outbound-disclosure-screening/ with notebook.ipynb, README.md, and fixtures/, meeting the recipe contract. Start from python tools/new_recipe.py 42 and the template in recipes/_template/.

Build notes

• One independent Noul per supplied sharing rule for each proposed outbound message.
• Deterministic controls hold flagged or uncertain samples for review. Nothing is sent.
• Synthetic messages only.

Decision design

• Noul. Write each proposition as a statement that is true or false. One independent question per label or concern, sent together in one request. Noul has no confidence field, so each threshold is chosen from examples and evaluated.

Level 4 scope

Stateful workflows, provenance, adversarial cases, model calibration, or several integrated decision stages.

Acceptance criteria

• recipes/42-outbound-disclosure-screening/notebook.ipynb runs top to bottom offline, with no key and no network, and its outputs are committed
• Live mode is opt-in (JEV_COOKBOOK_LIVE=1 with TYPESAFE_API_KEY) and uses the same question definitions as the offline run
• Fixtures are synthetic, include the hard cases named in the build notes, carry gold labels, and pass the fixture validator
• Every response fixture records its provenance, and the notebook states which mode it ran in
• Evaluation: Missed-disclosure rate, false-hold rate, and overall hold rate.
• State, provenance, or policy rules are enforced in helpers.py and covered by tests that hold for any model answer
• Fixtures include adversarial, conflicting, or stateful cases, and the notebook walks through at least one
• No statement about Jev's quality, latency, or cost appears without recorded live inference; in a synthetic run, metrics are labelled as pipeline checks
• recipes/42-outbound-disclosure-screening/README.md says what the recipe teaches, how to run it, and what was and was not measured live
• The pull request changes only recipes/42-outbound-disclosure-screening/ and CI is green

Dependencies

• Blocked by Foundation: CI executes every notebook offline and validates fixtures #69 (CI executes every notebook offline and validates fixtures)
• Builds on Recipe 15: Sensitive text triage #15 (Sensitive text triage). Merge those first and reuse their patterns; do not copy their fixtures.
• Tracked in Epic: Level 4 · Advanced recipes #78

Sources

Read these before writing the questions.

• S02 · TypeSafe AI: Primitives (Questions): Choice, Noul, Score, independent questions, and dependent follow-up calls.
• S03 · TypeSafe AI: Confidence: Distribution-derived confidence and task-specific thresholds informed by observed outcomes.
• S04 · TypeSafe AI: How to build with TypeSafe: Narrow semantic judgments integrated with workflows, deterministic logic, and side effects controlled by code.
• S07 · TypeSafe AI: Jev 1.13 jaggedness: Version-specific limitations affecting arithmetic, context selection, adversarial content, and generation.

Out of scope

• Changes to shared code, the catalog, or the root README. If the shared package is missing something, open a foundation issue and link it here.
• Real side effects of any kind. Actions are simulated.