Episode Details

Two major corrections reveal how even verified data can mislead when arithmetic gets applied carelessly, and how rigorous retesting transforms preliminary findings. The hosts break down the clip boundary percentage error from Episode 11—a live-computed mistake that looked correct because the underlying numbers were—and detail how Episode 9's 'human predictions beat the machine' finding fell apart under proper lab conditions. Key timestamps: 00:00:00 - Introduction and corrections overview, 00:05:15 - The arithmetic error and verification protocols, 00:12:30 - The 'lab coat effect' problem in presentation, 00:18:45 - Machine learning correction and retesting methodology. --- Sources & further reading: • game-senser-fable — docs/podcast/LEARNING-CONTEXT.md §11 (§11.0 alt-camera • and the narrowed gold-notes claim; §11.3 EXP-9; §11.6 EXP-13; §11.7 what stands and what • an episode must not say) • game-senser-fable — docs/touch-arc/frontier-touches/EXP-9-NULL-DECLARATION.md (the • amended gate, the below-chance control table, 20 seeds) • game-senser-fable — docs/touch-arc/frontier-touches/EXP-9-NULL-RESULT.md (243/243, the • recall table with chance columns, 0.144s MAE, the 164/40/39 error budget, per-type recall • the attack↔block analysis, thinking at 72% of spend, the scored predictions) • game-senser-fable — docs/touch-arc/frontier-touches/EXP-13-GAPSCAN-RESULT.md (168/243 • against a gate of 172, the 16-of-21 empty lists, the count-matched null, the tolerance • instability, 0.414s recovery MAE, the scored predictions, the disclosed ledger fix, the • two-arm fidelity test, clip-edge exposure) • game-senser-fable — docs/touch-arc/v91/EXP-2-RESULTS.md (alt-camera 15/15, pooled 73/74 • $2.14, the wrong pre-registered prediction) • guist-bot — data/series/ml-volleyball/ERRATA.md (ep 11's invented "almost 60%", the real • deltas, and the derived-numbers rule that came out of it) This podcast episode was fully generated by AI — research, script, voices, and production. Built with Claude, Piper TTS, and automated pipeline tooling.