| name | loop1-exit-check |
| description | Exit rubric for Loop 1 (ideation refinement). Load when the Main Agent needs to determine whether the current IDEA.md is sufficiently developed to proceed to Loop 2 stress testing. Returns PASS or FAIL with specific failing criteria listed. Evaluates development completeness only — not idea quality, novelty, or validity. Those are Loop 2's job. |
Loop 1 Exit Check
Read phase_0_ideation/ideas/IDEA.md. Apply the four criteria below in order.
Each criterion is binary: pass or fail. Do not score, weight, or partially credit them.
A single failing criterion means the overall result is FAIL.
Criterion 1 — Core claim is specific
Pass if IDEA.md contains at least one proposition stated in specific, falsifiable
form — what is claimed to be true and why, not just what the work aims to accomplish.
✓ Pass: "We claim that explicitly encoding [X] into [mechanism] improves [outcome] because [reason]."
✗ Fail: "We aim to improve [outcome] using [vague approach]."
Criterion 2 — Mechanism is described
Pass if IDEA.md describes the proposed approach at a level where one could ask
"does this mechanism actually work?" There must be a specific account of what it does
differently from existing approaches — not just an assertion that it is different.
✓ Pass: "Instead of [existing operation], our method [specific operation], which achieves [effect] by [reason]."
✗ Fail: "Our method improves on existing approaches through a better design."
Criterion 3 — Differentiation from closest prior work is stated
Pass if IDEA.md names at least one closely related prior approach, states a specific
difference from it, and that prior work is grounded — it must either appear in a Scout
report under phase_0_ideation/literature/ or be listed in the "Known related work" section
of IDEA.md before this revision. The difference does not need to be airtight — the Tester
will probe it — but it must be stated explicitly, not implied, and the cited work must be real.
To verify grounding: check phase_0_ideation/literature/ for any scout_report_*.md files
and the "Known related work" section of IDEA.md. If the named prior work appears in neither
source, the criterion fails regardless of how the difference is phrased.
✓ Pass: "[Paper X from scout report] does [Y]. Our approach differs in that we [Z] because [reason]."
✗ Fail: Prior work named but not found in Scout reports or Known related work — likely hallucinated.
✗ Fail: Prior work is mentioned but no specific difference is stated.
✗ Fail: No prior work is named at all.
Criterion 4 — Problem scope is defined
Note on C2/C4 independence: Criterion 2 (mechanism) and Criterion 4 (scope) are correlated
but distinct. A well-described mechanism often implies a task setting, but it does not automatically
define what a positive result looks like — that is C4's specific requirement. An IDEA.md can
pass C2 and still fail C4 if it never states the evaluation setting or success criterion. Evaluate
them independently; do not infer C4 from C2.
Pass if IDEA.md makes clear enough what problem is being solved that one could
identify what kind of evidence would count as validation — what task, what setting,
what a positive result would look like.
✓ Pass: "We target [specific task] in [setting]. Success would be demonstrated by [measurable outcome]."
✗ Fail: The scope is broad or vague enough that any result could be claimed as supporting the idea.
Output
After evaluating all four criteria, output exactly this block:
## Loop 1 Exit Check
**Result: PASS / FAIL**
| Criterion | Result | Note |
|-----------|--------|------|
| 1. Core claim specific | PASS / FAIL | [what was found, or what is missing] |
| 2. Mechanism described | PASS / FAIL | [what was found, or what is missing] |
| 3. Differentiation stated | PASS / FAIL | [what was found, or what is missing] |
| 4. Problem scope defined | PASS / FAIL | [what was found, or what is missing] |
**Failing criteria:** [list by number, or "none — all criteria passed"]
Do not suggest how to fix failing criteria. That is the Refiner's responsibility.
