We have been measuring the right phenomenon with the wrong unit. ESCT’s minimal dynamical equation correctly describes collective semantic attractor-well dynamics. But K2 experiments measured single-well entrance traces — averaging 18 polysemous word signals from a population of billions of heterogeneous well-cluster ensembles. The dilution was structural: like measuring a retinal image by averaging across 18 individual photoreceptors. ESCT v10.0 corrects the measurement ontology: the state variable C(t) is the collective differentiation state of a nested well-cluster ensemble, not the activation of one isolated well. This single reinterpretation unifies all K2 findings. Weak ΔD signals (+0.077): expected diluted ensemble average. Cross-model failure (R48): different training histories produce different ensemble fingerprint distributions. High-layer reversal in bark and ring (R53, R58b): ensemble-level re-negotiation overriding local well preliminary signals. Layer-12 output commitment (R59): ensemble convergence, not single-layer decision. Cue-causal output flips (R60, max |Δout| = 0.814): collective basin phase transitions triggered by removing locking cues. AGI Fly Hypothesis v0.4 derives the correct structural conclusion: the fly is not a controller behind the compound eyes. The fly IS the compound-eye ensemble’s emergent collective dynamics. The gaze is the alignment. There is no homunculus. measurement ontology; well-cluster ensemble; semantic ignition; collective alignment; cue-causal arbitration; transformer interpretability; default-prior bias; phase transition; nested attractor wells; BERT; ESCT v10.0; AGI Fly Hypothesis v0.4 well-cluster ensemble · measurement ontology · semantic ignition · collective dynamics · emergent control · cue-causal arbitration · default-prior bias · locking threshold · nested attractor wells · transformer interpretability · BERT · ESCT v10.0