Round 2 verification audit for C014

Model: gpt-5

Claim: After 5 us of repump (B) at 854 nm, the measured probability of leaving D5/2 is > 0.9999, with no detectable loss outside the {S1/2,D5/2} subspace.

Source alignment: source/main.tex:120-120 (Experiment)

Prior official verdict: not_verified with failure_reason data_unavailable.

Executable evidence: none. Sandbox rerun logs: none.

Independent audit: I checked the paper inventory and source context. This claim depends on measured/raw experimental data or device-specific calibration values, and inventory.json records no paper-provided dataset/code artifact. The correct action is not to synthesize data; the data_unavailable verdict is preserved.

Decision: Round 2 verdict is not_verified with failure_reason data_unavailable and limitations ['data_unavailable']. Notes: Measured value: P(leave D5/2) > 0.9999 after 5 us of 854 nm repump, no detectable loss outside {S1/2,D5/2}. No paper-provided raw data exists (inventory.json: no code/data/Zenodo/Figshare/OSF). Cannot reproduce a measurement; never synthesize data. Sanity check: 854 nm drives D5/2->P3/2 (P3/2 lifetime ~6.9 ns), so over 5 us the D5/2 is depleted over many time constants and a residual <1e-4 is physically plausible, but this does not constitute verification. Verdict: data_unavailable.