QMCE Sesja 14: An Unresolved Elo Effect for a Chess Meta-Engine Overlay under Author-Declared Thresholds, with an Analytical Bound Against the Queueing-Overload Explanation and the Withdrawal of a Depth-per-Clock-Second Deficit

QMCE (Quantum Meta-Chess Engine) was integrated as a Path A overlay on Amelia Chess via PyO3 0.20 in-process embedding (sessions 10–14). Nine match estimates across eight configurations give centers in -12.2 to +9.9 Elo, every 95% interval bracketing zero as reported (eight pre-cloud rows are legacy game-level approximations); three same-binary long-TC probes pool to -0.0 ± ~19 Elo (N=634). A queueing-overload explanation is bounded below its own overload band for the long-TC configuration the flip concerns: long-TC offered load cannot exceed 0.412 under any mixture of the two measured service times (0.012 warm), against an H0 boundary of 0.5 and overload at 0.9. That is a bound, not a test: the model's arithmetic makes the declared H1 bands unreachable (§3.2). v0.5 is an erratum revision of v0.4 (10.5281/zenodo.22701150) with three corrections. (1) The depth-per-clock-second deficit reported in v0.1–v0.4 (mean ratio 0.874) is withdrawn: it averages per-move quotients whose denominator reaches 0.001 s, moves to 0.972 after trimming 0.5% of moves, and its pair-resampled CI95 [0.760, 1.009] contains parity; the aggregate rate ratio sum(depth)/sum(time) is 0.9996, CI95 [0.9955, 1.0036]. A residual paired depth difference (-0.048 plies, CI95 [-0.104, +0.007]) is unresolved. (2) The K2 degraded-cache case averaged service rates instead of service times: short-TC production carries an offered load of 1.116, not 0.286, so the claim that every production scenario stays below 0.5 is withdrawn. (3) The S5 cold-start amortization is 0.184, not 0.304 (production rows unaffected). A 4.02× microbenchmark speedup on the supposed hot path produced no detectable Elo change (-12.2 ± 32.9). The 1500-game cloud follow-up (putative replication; binaries never identified) gives +3.0 ±12.3 Elo over 749 colour-swapped pairs, containing both zero and the +5 Elo H1 boundary: H1 is unresolved rather than rejected.