Biggest block (2026-06-16): a structural latency disadvantage on the V1↔V2 niche — not a broken bot. The opportunity it hunts (same token cheaper on PulseX V1 than V2, atomic buy-low/sell-high in one tx) is real and shows up in the scan, but it is a crowded lane: a faster competitor ("The Machine", 0x82cd9533…) sits on the identical V1↔V2 routes with a pre-warmed custom contract and almost certainly private-mempool inclusion, so by the time our public broadcast lands the spread is gone or the pool has moved (K-invariant revert). That is why a healthy-scanning bot can sit at ~0 broadcasts for long stretches — the gate is correctly refusing to fire into a race it will lose at a gas-only loss.

What was actually broken (and is now fixed, 2026-06-16 hardening): the bot had been built WITHOUT the champion's safety scaffolding, so it mis-fired and burned gas. Six fixes shipped: (A) ethers-v5 tx.wait() throws CALL_EXCEPTION on a mined-but-reverted tx instead of returning status===0 — the catch now recovers the receipt and books the gas as a real loss (previously every revert silently bypassed both fail-safes); (B) sims now run against the pending block with a real gasLimit, not the head block (head-block sims false-passed ~99.6%); (C) a fat-margin gate (sim NET must clear gas×2 after a 55% sim-drift haircut) so it only fires trades fat enough to survive the latency gap; (D) a per-token revert-bench (3 reverts → 6h bench) so it stops re-firing a token that keeps losing; (E) free reverts / send-errors that cost no on-chain gas are NOT counted toward the daily-loss halt or breaker; (F) an owner==wallet boot check before it ever ARMs. Daily-loss halt 5K WPLS; notional probation cap 25K.

2026-07-02 — revived on its own stack, coupling removed: the bot now runs a DEDICATED wallet (0xfc7E301D) + DEDICATED PulseAtomicArb (0xe9e93586, funded 500K WPLS), so the shared-wallet nonce-coupling that caused the 06-21 pause (it was starving the champion) is gone — and 16 of its historical failures were exactly that shared-wallet nonce-stomp class, which is now eliminated.

The genuine remaining lever is the same as the champion's: this is WPLS profit on a tiny book, and we're racing a faster incumbent on a public mempool. It now dual-broadcasts through our local sentinel node (partial inclusion-latency help), but to fully win this niche we'd need private/bundle submission (PulseChain has no Flashbots equivalent) or a pre-warmed contract + colocated node — neither justified at current capital. It runs hardened, fires only the rare fat winners, and we measure whether even those land vs The Machine.

If our V1↔V2 fire could reach a PulseChain validator the same way the competitor "The Machine" (0x82cd9533) almost certainly does — top-of-block, before the public mempool sees it — the structural disadvantage that makes us lose nearly every race would vanish, and the fat gaps our gate already finds would actually LAND instead of reverting on a closed spread. The whole edge on this niche is inclusion speed, not trade cleverness; whoever the block builder includes first wins, full stop. The closest real-world thing accessible to us: pulse-sentinel already peers with PulseChain nodes via devp2p and can dual-broadcast through rpc.pulsechain.box — porting the champion's multi-RPC first-accept-wins racer here is the cheap first step. The true unlock is a private-builder/preconf relationship that bypasses the public gas auction entirely, but PulseChain has no Flashbots-equivalent relay market yet, so there is no clean API to buy this today. Until that exists (or we run a colocated node with a pre-warmed contract like the incumbent does), the honest ceiling on this bot is "win the occasional gap fat enough that even a slower public broadcast still lands" — which is exactly what the fat-margin gate is now tuned to catch.