2026-02-19: Timing Resolution and Settling

Solves the marginal dominance problem via pattern chain length within a single f. Empirical validation: sharpest-LPP strategy drops from 4.9 to 2.0 bpc. The 2.8 bpc gap between max-min and external interpolation IS the settling problem.

Papers

The Timing Resolution: Solving Marginal Dominance via Pattern Chain Length
4 pages. The “the” problem: marginal “h”→“e” always dominates bigram “th”→“e” under max-min. Resolution: longer pattern chains complete later within a single f, and the output ES settles on the sharpest arrival. Introduces conjunction patterns (multi-source min). Establishes: experiments construct models, the runner is generic, capabilities must live in the model.
 Settling: Sharpness Preference and ES Epistemics
5 pages. Empirical validation of the timing resolution. Sharpest-LPP strategy: 4.883 → 1.955 bpc (4K), 4.985 → 3.304 bpc (64K). Bigram LPP selected as sharpest 78–94% of the time. Ring pattern as sharpness detector. Settling conjecture: find total thought maximizing per-ES sharpness subject to pattern consistency. Bidirectional LPP propagation makes multi-directional settling coherent.
 Twenty Days of the Universal Model: A Research Summary and Critical Review
12 pages. Reviews the entire research arc: Phase I (sat-rnn, doubled-E, pattern chains, factor map), Phase II (total interpretation, weight construction, KN scaling), Phase III (lexical injection, extended ES), Phase IV (Hutter Prize compression, context events). Catalogs 11 dead ends, 12 dropped threads, 6 instances of double work. Identifies marginal dominance as the critical blocker. Best result: 1.588 bpc = 189.3 MB (1.79× record).

Interactive Pages

Research Timeline
Linear timeline with color-coded entries for breakthroughs, dead ends, dropped threads, and double work. Covers all 21 archive dates. Links to papers and results.

Key Findings

← Previous: Feb 18 Archive Index Next: Feb 20 →