Bot Titles
PLS-TRI-ARB-V1Performance So Far
PLS-TRI-ARB-V1Recent Fire Attempts
PLS-TRI-ARB-V1Token Reverters (24h)
PLS-TRI-ARB-V1Biggest Block
PLS-TRI-ARB-V1Status 2026-05-25 (verified via tmux): PAUSED — tri-backrun not running. Snapshot below from 2026-05-21, when it was last live (paper mode).
Fleet status 2026-05-21: RUNNING · MOSTLY PAPER MODE. Scanning 258 scans/60s but only 11 live wins ever (vs 1,394 PAPER wins + 51 live errors). The Performance So Far counter says "1405 wins / 96.5% WR" — that's the wins-canonical classifier counting paper-mode entries as wins (per trade-log-reader.js:131 the "paper" status is intentionally treated as a paper-win marker). HONEST live record from logs/pulse-tri-backrun-trades.ndjson: 11 live_win / 51 live_error → 17.7% live WR. 168 state_override_caught entries are pre-broadcast catches that never reached the chain.
The bot's default is paper mode (per the pulse-tri-backrun.js header: "Default: paper mode. Live requires --live flag."). To get to honest live numbers the daemon needs --live AND the V3-leg sim reverts need to clear.
The dominant block: edge erosion on PulseChain triangle paths. Three-token cycles like WPLS→TOKEN_A→TOKEN_B→WPLS depend on either (a) one leg in a thin pool that the victim trade shifts, or (b) cross-DEX price drift surviving long enough for our backrun to fire before the next arb bot. Both have decayed over 2026-04 → 2026-05 as competitor density on the major triangles rose. The bot picks up 226 historical "wins" but a chunk of those are state_override_caught (pre-broadcast sim catches, NOT real fires) — L3 2026-05-19 flagged the mis-labeling as a composite-score-elevation bug. Cleanup pending.
Plus: 9x sim-to-actual drift (-29.24% avg, -22.2% median per L4) compounds across all three legs of a triangle, so a 9x leg in any position inflates projected gross vs realized. Today's PLS-AGG-ARB retune (route-discount tighten on 9x routes) does NOT propagate to this bot — would need a parallel route-discount table here.
Two practical levers:
• Fix the state_override_caught label (cheap, ~30min, fixes composite scoring across the fleet)
• Apply the PLS-AGG-ARB route-discount table here (ROUTE_SIM_DISCOUNT mirror for triangles touching 9x). Likely tighter still given 3-leg compounding.
Without one of these, the bot stays alive as a backrun observer but won't fire meaningful trades. Worth keeping in the 8-keeper slate for cheap diversification on the revenue chain.
Fantasy
PLS-TRI-ARB-V1If we had some way to instantly remove this bot's biggest constraint — capital, latency, signal coverage, or counterparty risk — what would the world look like? (Pending creative assessment.)
Suggestions for Improvement
PLS-TRI-ARB-V1- 01CRITICALFix state_override_caught mis-labeling (L3 2026-05-19). 226 historical rows tagged as liveWins are actually pre-broadcast catches that never executed — they inflate composite score above the real champion. ~30min: relabel as sim_caught in the trade-log schema + backfill. Frees champion scoring across the fleet.
- 02HIGHApply PLS-AGG-ARB ROUTE_SIM_DISCOUNT mirror here, but tighter — 3-leg triangles compound the 9x drift across all three legs, so a triangle touching 9x even once should be discounted more aggressively than a 2-leg cross-agg trade. Suggested: 9x in any position → multiply triangle discount by 0.42 (already-tightened single-leg rate); 9x in two positions → 0.42² = 0.18.
- 03HIGHAudit pulse-sentinel mempool feed for victim-tx capture lag. If our backrun lands consistently 1-2 blocks AFTER the victim, our triangle is fighting whatever liquidity the next-arb bot left behind. Verify by stamping signal-receive-ts vs victim-confirm-ts on each fire.
Conclusion
PLS-TRI-ARB-V1Status 2026-05-25 (verified): PAUSED — tri-backrun tmux not running.
Verdict: keeper slot for diversification on the revenue chain; not currently producing. Triangle backrun on PulseChain. At its last live snapshot it was heartbeating + scanning (258 scans/60s) but landing 0 fires in 24h, with 13 sim-failures over the window. Historical "226 wins" include a chunk of state_override_caught pre-broadcast catches (NOT real fires) — pending label-cleanup per L3 2026-05-19.
Fixable? Yes, two cheap interventions: (1) fix the state_override_caught label so composite scoring stops over-promoting this bot above the real champion; (2) apply 9x route-discount table (mirroring today's PLS-AGG-ARB tighten), with extra discount for triangles touching 9x in multiple legs (compounding). Both ~1h work. After: expected 1-3 fires/day at ~5-30 WPLS each.
Place in fleet: keeper #6 (PulseChain diversification — 2-leg PLS-MEM-BACKRUN is #7, atomic-scan PLS-AGG-ARB is #1). Three different signal-source surfaces on the only chain that has produced revenue. Stays unless it produces 0 fires for 7 consecutive days post-cleanup.