Did the navigation-centre learning site, rather than the extra propagation step, lift the score?

E178 — the steps-4 control: is it the fan-shaped body or the fourth step?

E171b's protocol with FORAGER_SITES=oa,5ht (no FB site; PFL not added to the readout), FORAGER_STEPS=4, modulation on, tag e178, 200 × 4 × 2, into the free search slot. Predictions.

1. If the FB site carries it: E178 mean AUC_Q ≤ 32 (E170's level or below) and finds ≤ 72 — the fourth step alone buys nothing beyond heterogeneity.
2. If the fourth step carries it: E178 ≥ 36 with finds ≥ 78 — then "the central complex learns" is withdrawn as the reading of E171b, and what the fourth step changed (deeper propagation to the readout's LAL/DN inputs) is the next thing to measure.
3. Between 32 and 36: undecidable at two seeds; a third seed on both arms before any claim.

E178 result — undecidable at two seeds, as the third prediction allowed for. Steps = 4, no FB site, otherwise E171b: AUC_Q 34.77 ± 4.28, 74 ± 10 finds, −136.

arm                                   AUC_Q            finds
E170  steps 3 (shuffled gains)        31.54 +- 18.81   68 +- 24
E178  steps 4, no FB site             34.77 +-  4.28   74 +- 10
E171b steps 4, FB site                39.96 +-  7.06   82 +-  1

Neither bar was crossed (≤ 32 → FB; ≥ 36 → fourth step). The fourth step alone accounts for most of the move above E170; the FB site's share is ≤ 5 AUC_Q and ≤ 8 finds, inside the seeds' spread. "The central complex learns" is not supported at two seeds — not refuted either. Branch as written: a third seed on both arms (seed offset 2), chained after E173c. Predictions for the replications: E171b seed 2 in [33, 47] and E178 seed 2 in [30, 39] (each arm's own pair, ±1 sd widened to the pair's range); if E171b's third seed is below E178's, the FB claim is withdrawn outright; if all three E171b seeds exceed all three E178 seeds, the ≤ 5 AUC_Q effect is real and small — worth stating, not headlining.