iter 9 — temporal epoch transfer of the §B gate

SURVIVES 2026-07-10 · novel library extension · 2 historical epochs × 7 pairs × 62 cells · 14/14 verdicts reproduce · clean 2/3

Claim under attack (assumed true only to destroy it)

Every campaign iteration so far ran on the MOST RECENT 80k bars per cell. The claim: the hardened §B machinery's verdicts are properties of the column relationships — not artifacts of the recent market regime — so the 7 frozen worked-example verdicts must reproduce on OLDER epochs. (The declaration itself was calibrated on 2021–2025 slices; cross-epoch validity is its own implicit claim, silently assumed by the probe's recent-only fetch.)

Attack construction

The iter-1 A5 gate, shifted in time: cells fetch their most recent 80k bars strictly before a hard calendar cutoff — E1: pre-2026-01-01 · E2: pre-2025-01-01 (per-cell time spans recorded in evidence; slower cells reach further back).
Same 62-cell grid, strided substrate, FLOOR_1000, hardened xi_max_pair (identity-key tie-break), frozen §B arithmetic — nothing changed but the epoch.
Falsifier F1: any (epoch, pair) verdict mismatch vs the declaration.

Result — verdicts transfer across market epochs

Pair	Declared	E1 pre-2026 (xi_worst · breadth)	E2 pre-2025 (xi_worst · breadth)
`ofi \| turnover_imbalance`	BAN	BAN — 0.9991 · 1.000	BAN — 0.9991 · 1.000
`agg_record_count \| individual_trade_count`	BAN	BAN — 0.9993 · 1.000	BAN — 0.9993 · 1.000
`price_impact \| volume`	BAN	BAN — 0.9992 · 0.984	BAN — 0.9798 · 0.968
`price_impact \| sell_volume`	WATCH	WATCH — 0.7683 · 0.871	WATCH — 0.7907 · 0.823
`kyle_lambda_proxy \| ofi`	PASS	PASS — 0.4052 · 0	PASS — 0.4585 · 0
`bar_cecp_velocity \| ofi`	PASS	PASS — 0.0385 · 0	PASS — 0.0267 · 0
`bar_dispersion_entropy \| trade_intensity`	PASS	PASS — 0.4312 · 0	PASS — 0.2129 · 0

Honest reading: the numbers move with the regime — the PASS ceiling pair breathes (0.405 ↔ 0.459), the BAN dial dips (0.9992 → 0.9798 pre-2025) — but nothing crosses a frozen band in either epoch, and 62/62 cells grounded in both. The PASS↔redundancy moat holds across three market eras (recent, pre-2026, pre-2025).

Evidence

File	What
a9_epoch_transfer_attack.py	The attack — hard calendar cutoffs, falsifier pre-declared in-file.
a9_results.json	Per-epoch, per-pair verdicts + per-cell time spans (62 cells × 2 epochs).
a9_stderr.log	Run log under the 2 CPU / 2 GB scope.
regression_tests/test_a9_epoch_transfer.sh	Permanent regression test — re-runs both epochs, asserts SURVIVES.

Consequence

The recent-fetch substrate's silent time-transfer assumption is now evidence-backed. Clean rounds: 2 of 3 — one more clean round triggers ROBUST-ENOUGH — OPERATOR REVIEW.

iter 9 · epoch transfer · verdict SURVIVES · one atomic commit: iter 9: epoch transfer — SURVIVES · append-only.