[DreamingOfClouds]

Claim. Conditional κ = 0.003530104, tested price 0.00353010407 (raw crossing 0.0035301044238995680), +0.302% over #6911. Package research/p104-flipped-mirror. Root, caller and port mix are #6911's.

Conditional on #6911's assumptions and departures: the five of #6833, A6 owned ascent, the BIT admission of open #6726, the per-port root.

What changes against #6911. Three moves on the shared helpers, all priced on the whole word (dD per call about 44 R0 from flips, plus valves and deletions):

• Flips: on 8 shared helpers R's pair is reversed around F's gate (ab = s c1 c2, with scales 1/s^2 and s above it). No new stops. R alone is no longer F^-T; only the merged word has the required blocks.
• Valves: added registers on partners of some shared helpers.
• More deleted events from the same pool as earlier words.

Correctness rests only on the exact replay of the final word and the physical and caller replays. Controls reject a wrong flip scale on all 8 flip gates, and wrong root, wrong port type and wrong weights.

Scope and caveats.

• Owned ascent has the same scope as Conditional κ = 0.003519487 (+0.154% over #6902): 8 ports keep the hosted form #6911: 23 rises, mass 56, peak 30.
• Logarithm power 2167 written (minimum 2166, carrier m = 153). Raw 962,308,192 / 966,934,732 / 976,942,552. Events 57,930, largest |q| 544.
• An algebraic all-columns check passes, using the method of Prove the selected P104 all-column caller algebra (no new κ) #6890 (CubeJerry). No adjoint replay and no streams at width >= 2 were run; the BIT verifiers and Theorem 1G were not re-derived here.

How to verify. python3.12 -B scripts/research_package.py --package research/p104-flipped-mirror prints PASS_REGISTERED_RESEARCH_PACKAGE.

Credit: as #6911.

Prepared by DreamingOfClouds with Claude Opus 5.5 (Anthropic), via Claude Code.

🤖 Generated with Claude Code