## 3. Its independent replication of the root cause — stronger than ours

Mechanically sampled, no anomaly screen: **103 control wallets**.

- **7** have a genuine interior 4-hour hole at 2026-06-22 22:00 → 2026-06-23 01:00.
- **6 of those 7 publish DAY = cumulative** (residuals $0.0000–$0.0145). **0 of 96** without the hole
  do. Fisher one-sided **p = 4.9 × 10⁻⁹**.
- Five are instances this package never saw: `obamaobamaobama` (rank 359, $6,569.64),
  `0xB5269581…` (784, $2,287.88), `0xcbedc34315` (903, $1,873.20), `PurpleTurtleM8` (971, $1,733.29),
  `0xC5111AB9…` (SPORTS 573, $2,428.41).
- **Two of the six traded that day** ($3,895 and $10,060 of volume) — which kills any
  "dormant-account artifact" reading.

**And a natural experiment this package missed, which isolates the cause precisely:** six wallets
carry the same 19:00–21:00 corrupted block but **retain the 01:00 baseline point — and all six
published correct DAY figures.** `lev1394`, whose series simply begins at 01:00, also published a
correct figure. It is specifically the missing baseline point, not the wider incident, that produces
the fault. No forward-fill is occurring: for `Inaccuratestake`, `cum(05:00) − cum(21:00)` =
$1,779,051.80 ≠ the displayed figure. The baseline is read as **zero**.

The gap is in stored data, not 1-hour serving: it persists at `fidelity=3h` and `12h`.

**One further correction it forced.** `16` §2 claimed missing points "would appear on every date, and
they appear on exactly one". False as a statement about the endpoint: dormant `RWCS` has **103
interior missing hours across ~80 distinct dates**, including runs of 7 and 8 consecutive hours, in
Feb–Apr 2026. The true, narrower statement — which it verified — is that **within 2026-06-01 →
2026-09-05 there are no other interior gaps**, so gap-freeness is a property of this window, not of
the endpoint.

---