{"kind":"reversal-self-audit","the_test":"would the read come out the same if the operator's incentives were reversed? (gnon / rectification-of-names — the reversal test; the who-watches-the-watcher answer)","the_answer":["the read is a PURE FUNCTION of the trace — turn_record.channel_profile over (Y, M, D) with fixed, published detectors (violated / flagged_reasoning / the context key). Payment, caller, and operator preference are not inputs; there is no lever.","the read is BOUND TO THE TRACE BY SIGNATURE — trace_sha256 + context_key + the profile sit under one Ed25519 signature, so the trace and the conditioning axis cannot be swapped silently.","the math is REUSED, NOT FORKED — server, the MCP sidecar, and the pip client call the same channel_profile, so any third party recomputes the read from the trace offline and catches a tilt (the-last-outside: independence, not number — a self-signing channel is false by construction)."],"live_check":{"incentive_invariant":true,"incentive_invariant_means":"the reference read is byte-identical with and without operator-incentive signals (fee/caller/preference) stamped on every turn","trace_sensitive":true,"trace_sensitive_means":"the same read DOES move when the trace's own content changes (one violated flag) — proof it is a measurement, not a constant","read_plain":{"n":8,"n_unmeterable (no reasoning channel)":0,"I(C;D) bits":0.0,"I(C;M) bits":1.0,"I(C;M | D-clean) bits  [switch signature]":1.0,"I(C;D,M) bits  [joint]":1.0},"read_with_operator_incentive":{"n":8,"n_unmeterable (no reasoning channel)":0,"I(C;D) bits":0.0,"I(C;M) bits":1.0,"I(C;M | D-clean) bits  [switch signature]":1.0,"I(C;D,M) bits  [joint]":1.0},"recompute_it_yourself":"pip install scry-client; the read is turn_record.channel_profile — run it on the trace and compare to the signed profile"},"honest_limit":"this proves the operator cannot bias a read at compute/sign time. It does NOT solve the burned probe (metrology-shelf: the subject can game the trace over time) or the scarab (weighing-of-the-heart: fixed, published detectors are exactly what gets learned and addressed). Fixed detectors are the anti-tilt property AND the scarab exposure — both true, both stated.","not_a_measurement":"this is a self-audit of the instrument, not a read of an agent; nothing here is score-blind-relevant and no fee touches it. The recomputation is the record — re-run it against any trace you bring."}