# CL-WP-005 v0.9 Release Review

## Decision
**PASS as a methods + open-measurement synthesis working release.**
**NOT PASS as a completed Critical Load primary empirical benchmark or model-ranking paper.**

## Review criteria

| Criterion | Result | Reason |
|---|---|---|
| Research question | PASS | Clear, decision-relevant and falsifiable through measured task outcomes. |
| Evidence integrity | PASS | External measurements are identified as external; no fabricated Critical Load telemetry. |
| Method originality/usefulness | PASS | Extends reporting from invocation units to acceptance-gated task accounting with explicit retry and boundary rules. |
| Reproducibility assets | PASS | Run schema, task registry, metric code, NVML wrapper and validation script are included. |
| Practical usability | PASS | Can guide benchmark design, model routing, engineering regression and vendor measurement requirements now. |
| Primary empirical depth | FAIL FOR v1.0 | No Critical Load-owned repeated GPU/node campaign has been executed in this environment. |
| Statistical validation | FAIL FOR v1.0 | Requires measured repetitions and uncertainty estimates from the primary campaign. |
| External review | FAIL FOR v1.0 | Independent systems/energy and task-domain review not yet completed. |
| Universal claims | PASS | Package explicitly avoids universal model/agent efficiency claims. |

## Publication class approved
Publish as **Working Methods Paper / Release Candidate v0.9**, not as a completed benchmark leaderboard.

## What would make v1.0 publishable as the flagship empirical result
1. Execute the frozen benchmark matrix on a declared NVIDIA GPU/node setup.
2. Collect raw cumulative energy deltas for every attempt, including failures.
3. Run >=3 independent repetitions per core comparison cell.
4. Audit scoring on complex tasks.
5. Reproduce every figure/table from released raw logs.
6. Complete independent measurement and domain review.

## Website treatment
May be listed in Publications as **CL-WP-005 / Working Methods Paper / v0.9** with a visible badge: `Primary measurement campaign pending`.
It should not yet be promoted as a Critical Load model-efficiency ranking or completed UAW benchmark.