From 6d376018ec5ca998e27db131e03214cd1c583036 Mon Sep 17 00:00:00 2001 From: leetcrypt Date: Sun, 19 Jul 2026 23:08:16 -0700 Subject: [PATCH] =?UTF-8?q?docs(sor):=20NEEDS-OPERATOR=20=E2=80=94=20RQ2?= =?UTF-8?q?=20per-circuit=20posterior=20underspecified=20vs=20instrument?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Frozen §3/§4/§6 pin RQ2 DV as a per-circuit adversary sender-posterior entropy (Miller-Madow, bootstrap over circuits; RQ2-P3 needs per-circuit H). The committed executor measures a per-run realized-frequency entropy instead — no per-circuit posterior, no per-circuit H. No formula for the posterior construction exists in the frozen prereg or the instrument. Flagged rather than invent a posterior model post-freeze (HARKing/ fabrication on a confirmatory DV). Battery left running (raw records recoverable → RQ2 recomputable offline under any ruling). RQ1 proceeds. Co-Authored-By: Claude Opus 4.6 --- OVERSEER-STATUS.md | 3 +++ 1 file changed, 3 insertions(+) diff --git a/OVERSEER-STATUS.md b/OVERSEER-STATUS.md index 5ceec4a..513203e 100644 --- a/OVERSEER-STATUS.md +++ b/OVERSEER-STATUS.md @@ -1,5 +1,8 @@ # OVERSEER-STATUS — sor-consent build agent +NEEDS-OPERATOR: How is the RQ2 per-circuit adversary sender-posterior constructed from a measured confirmatory circuit? The frozen prereg pins RQ2's DV as "Shannon entropy of the adversary's posterior over candidate senders PER CIRCUIT" (§3 L84-88), unit = "the circuit (entropy of the per-circuit sender posterior)" (§4 L108-110), Miller–Madow, BOOTSTRAP OVER CIRCUITS, and RQ2-P3 (§6 L205-207) needs a PER-CIRCUIT H for its Spearman ρ. But (a) the frozen prereg gives no explicit formula for building that per-circuit candidate-sender posterior from an observed circuit (it cites [Serjantov2002] S=2^H and [Diaz2002] d=H/log2 N for effective-size/normalization only), and (b) the built+committed executor (ce68d41) measures RQ2 as ONE per-RUN Shannon entropy of the realized entry-node FREQUENCY across the C=50 circuits — not a per-circuit posterior, and it yields no per-circuit H, so RQ2-P3 is uncomputable as-collected. I will NOT invent a posterior/adversary model post-freeze (that is HARKing + fabrication on a confirmatory DV). Need a ruling, one of: (A) ratify an explicit per-circuit posterior-construction rule (e.g. posterior mass over candidate senders as a function of the observed exit-bridge/house) logged as a stage-05 clarification like the Holm one → RQ2-P1/P3 then computable OFFLINE from the immutable raw records; (B) confirm the per-run realized-frequency entropy IS the intended RQ2 op (still leaves RQ2-P3's per-circuit H undefined); or (C) scope/relabel RQ2-P3. NOTE: the running battery is NOT wasted under any ruling — per_circuit_seeds are persisted + assemble() is deterministic, so every circuit's path/bridge/house is recoverable and RQ2 can be recomputed offline; RQ1 is fully specified and proceeds now. Battery left RUNNING. + + - 2026-07-19 R2 DONE + committed (7ed2370): provenance writer — immutable manifest.json (seed, topology/selector/churn id, per-node persona fingerprints, git SHA, isolated engine+image, pip/cargo freeze, timestamps); events SHA reserved for R3. - Checks green: pytest R1+R2 = 15 passed (via `rtk proxy python -m pytest`); full cargo suite = 63 passed. Cross-language fingerprint parity locked (persona.rs <-> provenance.py). Deps installed into worktree venv. - 2026-07-19 R3 DONE + committed (a9308ed): append-only events.jsonl (fixed 10-event vocab, metadata-only, no plaintext) SHA-256 sealed once into manifest; deterministic replay_fixture_circuit (no engine/traffic). Provenance spine R1->R2->R3 = protect-first minimum unit, GREEN. Checks: pytest sor suite 21 passed; cargo 63 passed.