# Publication 8 — pending independent verification

475 distinct frames = unchanged parent 378 + actual requested saved-human 24 + authorized existing-human reuse 7 + strict new-machine 66. Goal 100 per class / 500 total; missing 25.

Prior publication-7 new machine reviews: actually saved 39/84; 45 UNSAVED, not seeds. Claims of completion are not evidence. Pending keys are in manifest.saved_review_status.

`new_proposals.jsonl` is ONLY this round’s machine drafts. `manual_proposals.jsonl` and `manual_source_records.jsonl` contain 31 actual human sources; `all_manual_sources.jsonl` freezes all 118 = inherited 87 + added 31, with exact current and actual non-automatic human event payloads/raw hashes. All 87 inherited human proposals were selected from canonical parent proposals, NOT from source-freeze records or only the last 57.

Four-point sources ['3b702879c30d94bb85fd9a56'] count as human coverage, not templates. Their actual shape is unchanged; `human_source`, not `template`, carries provenance. Rendering metadata is separate from the propagation template inventory. No generated points or machine scores.

Latest human gallery edits are validated against frozen publications 3–7 and win duplicate seeds. `quota100_continue_generation` is new; all five parent template namespaces and 378 canonical assets are byte-identical. Parent evidence is archived, not represented as new verification.

Original attempt_target and .55 initial / .84 ECC / .78 boundary NCC / 1.5 forward-backward / beam gates unchanged; select_diverse goal explicitly 100. No multiscale research, threshold relaxation, forced fill, recursive candidates, database/config/service writes, automatic confirmation, training or robot eligibility. Current sources/events were checked again at completion; concurrent saves require a new generation.
