Source: assist/REVIEW_REQUEST.md
# Independent annotator dispatch is pending Astra3 has generated 200 proposals for 40 fresh cases. These are **not accepted annotations**. The shared repository copy is at `/home/mma/repos/neochart/neov2/assist/proposals/`; read `ANNOTATION_PROTOCOL.md`. Lead was requested through `lmctl prompt ... Lead --prompt-file ...` three times. All attempts returned `receiver_busy`: Lead is held by an active terminal. No queued delivery or assigned annotator is claimed. A direct asynchronous request was also submitted to the user/Lead in this session. Required next step: Lead dispatches an independent seat to inventory all 40 case sources blind to proposals, then adjudicate each proposal's element and exact variant separately. Write `annotations/<sample_id>.json` using the protocol. Astra3 must not supply these labels. Partial reading must be marked partial. After independent files arrive: ```bash python3 neov2/tools/score_assist.py ``` Until then `metrics.json` reports null precision/recall, zero reviewed cases and no established effort savings. Accept/reject alone supports precision but not an exhaustive recall denominator. Sequential review times do not establish causal savings; a randomized unassisted control is still required for that claim.