On 27 October the co-founders each sold exactly 263,900 Class A shares — Biswas at a weighted $42.93 for $11,329,895, Bicket at $42.92 for $11,326,802 — both indirectly, through trusts. Identical counts, same day, two founders. The ordinary reading — routine Rule 10b5-1 execution carrying no information — is correct.

This was the most promising-looking shape on the tape this week, which is why it was tested rather than assumed. Four things could have turned it into a finding, and each one failed against a document:

• Matched tranche sizes implying coordination. Both plans were adopted on the same date, 29 September 2025, more than a year before the trade, and are footnoted as such on every tranche. Two co-founders papering their liquidity together at the same moment explains identical mechanics without implying a shared view. Fails.
• Two tranches inside one month implying acceleration. Bicket sold 263,900 across 13–15 October and 263,900 again on 27 October. But the same 263,900 appears on 17 February 2026 as well. The quantity is the plan’s unit, not an escalating amount. Fails.
• The institutional trustee as a tell. The non-revocable vehicles — Bicket-Dobson Trust I (u/a/d 11/10/2021) and Trust II (u/a/d 10/8/2021) — are administered by Jordan Park Trust Company LLC, with the same structure mirrored on the Biswas side. That is a 2021 arrangement executing a 2025 plan. Nothing recent happened. Fails.
• Selling into strength implying a top call. This is where it dies decisively, below.

The fact that settles it. The tranche is fixed in shares, not dollars. Bicket’s 263,900 on 17 February executed at a weighted average near $25.83 — about $6.82m. The same 263,900 on 27 October executed at $42.92 — $11.33m. The headline dollar figure rose 66% while the decision variable did not move by a single share. Every write-up calling this “an $11m sale” is reporting a price change as though it were a decision. There is no view embedded in a constant.

Recorded because the opposite discipline — reading every founder disposal near a high as a signal — is the single most likely way this report gets a year wrong, and today it would have cost two false findings.