quant·loop

dry-run (paper) ● kill switch armed
Autonomous, self-verifying trading loop — propose, verify, execute only what passes.
Cycles run
1
Equity
$100,000
Drawdown / HWM
+0.00%
Last verdict
REJECT
Open positions
0
Last run
2026-06-28 20:23 UTC

The 5-stage cycle

ingestmarket data→ signalmaker (Claude)→ verifychecker (math)→ executegated on PASS→ riskkill switch

Latest candidate & verdict

lookback 60 · top_k 3 · hold 20
Verdict: REJECT ✗

GateValueOK
friction_survival 2.899 ✓
sharpe 1.202 ✗
max_drawdown -21.375 ✗
newey_west_t 1.778 ✗
oos_span_years 2.062 ✓
mean_per_trip 4.558 ✓

The maker proposes; this deterministic gauntlet decides. Nothing executes that fails here.

Lessons learned · self-improving memory

reject Candidate {'lookback': 60, 'top_k': 3, 'hold': 20} REJECTED (failed: sharpe, max_drawdown, newey_west_t). Ret-DD 2.90, Sharpe 1.20, maxDD -21.4%.
2026-06-28
self cap top_k at 3 unless drawdown stays under the limit — wide baskets diluted the per-name edge in testing.
seed
validated The validated edges are momentum with risk-managed intraday exits (Scanner 5m-exec) and crypto trend on BTC/ETH. Build near these.
seed
rejected Already verified and REJECTED — universe expansion to 1000, new-listing pickup, news/sentiment ranking, sector caps, regime filters, 5% take-profit, gap-fade (all thresholds). Don't re-propose without new evidence.
seed
benchmark Friction-robust ≠ beats the benchmark. Long-only factor tilts survived friction but still lost to SPY in 2016-26. Clearing the cost floor is necessary, not sufficient.
seed
survivorship Never rank/select on today's-liquid universe. It printed a fake +1400% momentum result; point-in-time corrected it to ~13% that loses to SPY. Always point-in-time.
seed
edge-per-trip Friction survival = edge-per-trip ÷ cost-per-trip, NOT how many legs pay. A 3,300-trip, ~0.17%/trip coin-flip can't absorb ~0.19%/trip exit cost. Avoid thin-edge high-frequency mean-reversion.
seed
friction A backtest at default 4 bp friction is meaningless for any strategy whose exits are market/touch-stops. Always verify at realistic friction (~38 bp RT intraday, ~12 bp RT daily-close). EOD and GapFade both inverted from big positive to negative under this and lost real money.
seed

Recent cycles

2026-06-28 20:23 UTC · verdict=REJECT equity=$100,000 dd=+0.0%