Scale-burst candidates on Exness ticks — no edge¶
2026-09-23. Four candidates in qkt-quant-live/examples/strategies/
({gold,btc}_{ema,rsi}_scale_burst.qkt): a 0.01-lot seed with no stop or target on a fast
EMA(5/13) or RSI(7)-50 cross, ten 0.05-lot STACK_AT legs when the seed's excursion reaches
$0.30 (gold) / $30 (BTC), each leg TAKE PROFIT BY $1.00 / $100 with a far dummy stop, and
EXIT AFTER 4m on the seed and every leg.
Method¶
- Data: Exness quotes pulled from the local demo gateway (
/copy_ticks_range), 2026-07-27 to 2026-09-22 — 12.9M XAUUSD ticks, 6.9M BTCUSD ticks (store/var/tmp/qkt-validation/exness-store, fetcherscale-burst-001/trace/fetch_exness_ticks.py). Venue specs fromqkt instruments pull; commission 0. - Engine: qkt build with the stack-anchoring,
EXIT AFTERand starting-balance fixes (PRs #1253, #1252, #1254),--broker mt5-sim,--starting-balance 100000, risk halts opened wide so no breaker truncates a run. Window 2026-07-28 to 2026-09-23, no parameter search. - Summary script:
/var/tmp/qkt-validation/research-scale-burst/analyze.py.
Results¶
| Candidate | Total | PF | Win % | Bursts | Seed P&L | Leg P&L | Weeks + |
|---|---|---|---|---|---|---|---|
| gold EMA | -13,009 | 0.64 | 58.9 | 715 | -373 | -12,636 | 0/9 |
| gold RSI | -23,740 | 0.69 | 60.6 | 1,549 | -625 | -23,115 | 1/9 |
| BTC EMA | -11,854 | 0.68 | 40.9 | 2,012 | -528 | -11,325 | 0/9 |
| BTC RSI | -10,799 | 0.78 | 42.6 | 2,914 | -631 | -10,168 | 2/9 |
Reading¶
- Neither entry has directional edge: seeds alone lose about the spread (-$0.35 to -$0.46 per gold seed). RSI is less bad per burst on both symbols but still negative.
- The loss is structural in the burst. Gold: a leg that hits target makes +$5.04; a leg that times out loses $11.24 on average. Break-even needs a ~69% target hit rate; realized is ~59%. The burst buys after a small up-move, and gold gives it back inside four minutes (consistent with the 2026-09-13 gold MFT noise work).
- Optimistic: mt5-sim fills all ten legs at one price; live fills them serially over 1-3 s at dispersed prices (parity row A20), so live results would be worse.
Verdict¶
Do not deploy any of the four. A burst design would need an entry with real short-horizon continuation, or a payoff where a leg's timed exit is not larger than its capped target.