Eval

Evaluate the system

Before running

Name the exact system under test. Record its model, instructions, skills, agents, helpers, knowledge, workflow, tools, connector configuration, and data snapshot.

Use one of these modes honestly:

• Working mode supports iteration. Its references may be visible to the evaluator, but never to the child trial before its output is locked.
• Course heldout mode sends locked outputs to the course-controlled grader. The expected results do not live in the student clone.
• A local visible answer file is development material. Do not call it a secret heldout evaluation.

Run

1. Load the question-only manifest from data/evals/public/.
2. Select the exposure, purpose, named cases, and any slice before the run starts. Exposure is working or heldout. Purpose is capability or regression. Do not treat these as one dimension.
3. Use helpers.evals.controller.EvaluationController to launch and record the trials.
4. Give each trial only its public task, permitted system files, permitted data, and permitted tools.
5. Lock every trial output before grading begins.
6. Grade deterministic criteria first. Keep model-based grades separate.
7. Preserve pass, fail, blocked, error, invalid, and unknown as different results.
8. Report every case and slice before discussing the aggregate.

The local controller is available through python3 -m helpers.evals.cli run-suite. Use --model claude-opus-4-6, --exposure, optional --purpose, and repeated --case-id arguments when selecting a subset. General code access is not required for routing or contract cases. When local data analysis requires --allow-code, state that local process isolation is not the same as course-heldout answer isolation.

For a reviewed working suite with local references, lock the trial outputs first, then grade them with python3 -m helpers.evals.cli grade-suite. Pass the run ID, public manifest, and reviewed reference file. Never copy the reference file into the trial workspace.

Run a full-analysis development case

Full-analysis cases live under evals/cases/public/. They produce a complete analysis bundle rather than one scalar answer.

1. Start a new run. For the Session 6 case:

   python3 -m helpers.evals.full_analysis start \
     --case evals/cases/public/novamart-monthly-operating-review-001/v1
   

2. Read the generated RUN-INSTRUCTIONS.md, public case, and result schema. The start command creates the trial's draft/ directory but does not begin the analysis trace.

3. Before querying data, start exactly one analysis trace with the trial draft as its output directory. Keep that analysis ID for every query, finding, receipt, and trace artifact in the trial.

4. Perform the analysis through the existing AI Analyst system. Save exactly the six requested files in that draft directory. Do not inspect any course reference repository before the first run is locked.

5. Register the reported findings, build the trace, and confirm the trace HTML path. Do not start a second analysis trace.

6. Lock the run:

   python3 -m helpers.evals.full_analysis lock --run-id <run-id>
   

   Locking verifies the Snowflake snapshot, validates the output contract, requires a complete trace, copies the six-file bundle, hashes every artifact, and snapshots the analysis record, query log, action log, provenance, receipt, and trace HTML. Never modify the locked submission.

7. After the course evaluation repository is released, run its grader against the locked run. Read the resulting student-report.md and individual grade records.

8. Diagnose failures using the saved trace. Start any changed system as a new development run:

   python3 -m helpers.evals.full_analysis start \
     --case evals/cases/public/novamart-monthly-operating-review-001/v1 \
     --baseline-run-id <baseline-run-id> \
     --intended-change "<one concrete system change>"
   

9. After grading the candidate, compare the two immutable runs:

   python3 -m helpers.evals.full_analysis compare \
     --before-run-id <baseline-run-id> \
     --after-run-id <candidate-run-id>
   

The shared course answers make these development cases. A genuinely held-out set must remain outside the system and be graded through a course-controlled boundary.

Design before running

When a user is creating a new case, interview them for the intended user, decision, consequence if wrong, observable criteria, independent reference plan, grader per criterion, human-review boundary, task and risk slices, and lifecycle status. Preserve the user's decisions rather than silently choosing for them.

Keep a new case proposed until its reference and graders receive independent review. Validate its structure with python3 -m helpers.evals.cli validate-case. Structural validation does not verify the reference or promote the case.

When a user is assembling a proposed set from a candidate pool, require an explicit selection, at least one rejected candidate with a reason, and named missing coverage. Validate it with python3 -m helpers.evals.cli validate-suite. Do not replace the user's proposed set with a canonical set during comparison.

Compare a change

Hold the suite, data snapshot, model, evaluator, tools, and trial count fixed. Name one intended system change. If more than one material input changed, label the comparison confounded rather than attributing the score movement.

Use --intended-change context for a candidate context run. Do not expose expected values, reference queries, private grader prompts, or a heldout answer key in the report.