frontier exhausted
r11 closed july 3. seventeen of thirty cycles, operator goal-clear. sixty seven plus cycles across r10 and r11, and exactly one survivor emerged: eur_postfix. short eurusd in the sixteen oh five to sixteen thirty london window, non-quarter-end month-ends only, riding the post-wmr-fix usd bid.
net annual sharpe zero point eight four. roughly eight events per year, plus twenty one basis points annually. seventy thirty real. sibling of the live gbp_fix strategy, same mechanism family.
every other mechanism tested was real. t-statistics to plus seventeen point seven. they all died on capturability. four arbed post twenty twenty. five sub-spread. two premium-free. one executable-negative, a sixteen year era-stable gold reopen pocket that was ask bid negative at boundaries.
mid-mid is not a cost model at boundaries. one transmission-cancelled. one burn-in artifact where the z-score blind through covid faked era-stability.
the free and owned data frontier is exhausted at current cost structure. intraday, options, fx, eu and asia sessions, gold, execution, construction. all closed at decisive power. do not re-mine.
gold mirofish collapse
three mechanical strategies from the mirofish synthesis research, tested on real dukascopy fifteen minute xau. train twenty eighteen to twenty two, validation twenty twenty three, holdout twenty twenty four to twenty five vault plus twenty twenty six year to date.
s2 asia fade: pf zero point seven seven on train, one point eleven on val, zero point ninety on hold. no edge.
s3 london momentum: pf one point sixteen train, one point ninety six val, one point twenty nine hold. the top five configs from a six hundred config sweep, train plus val pf one point five to two point two, collapsed to zero point ninety three to one point fifteen on hold.
zero of nine years pass bootstrap ci above one point zero. a daily ema regime gate does not rescue.
s4 bb squeeze: pf zero point seventy one train, zero point seventy four val, one point zero eight hold. losing on train and val, hold is noise.
a look-ahead bug was fixed in s2. ts_end was returning the first bar of the trading day group, producing exits nine hours before entries. but the lyra and vega bar requires bootstrap pf ci lower bound above one point zero on holdout. none of the three meet this.
mirofish claimed s3 frequency of seventy five per year. real frequency is ten to fifteen. the strategy is regime-fragile, losing in twenty eighteen, twenty nineteen, and twenty twenty four, winning only in trending years.
do not build an execution layer for these. aquila bot already covers xau exec-alpha at a similar pf range. adding a marginal-edge sibling is capital dilution.
vix-gated gap fade
fn book alpha hunt r7, thirty cycles plus a third deep-research pass. one new validated alpha emerged: vix-gated intraday gap-fade. fade large overnight gaps, absolute gap above one standard deviation, into the cash session. enter after the first rth hour, exit at the close.
cfd-friendly, no overnight financing. only when vix z-score is above zero point five.
sharpe roughly one point one to one point fifteen. random date p equals zero point zero zero five. generalizes across five of five instruments: spy, qqq, iwm, dia, uso. parameter-plateau across the gap threshold by vix-z grid, one point zero to one point twenty three, not knife-edge.
cost-solid at zero point ninety six with four basis points. orthogonal to the deployed daily book, correlation negative zero point zero three two.
the economic story: overnight illiquid overreaction reverts in the liquid cash session, concentrated in high-vol regimes. low vix is negative, the gate is essential. strongest single new signal of all seven fn-hunt rounds. not a one point five standalone.
build number four zero eight, needs an intraday executor path, same blocker as the gap-fade from kb five twenty six.
profit targets hurt. reversion runs to the close, hold-to-close is optimal. execution management adds nothing. edge concentrates in up-gap fades, selling faded overnight rallies. the mechanism needs a real overnight market closure, so equity and oil only. dead on twenty four hour fx.
qts audit: zero edge
edge audit on the quantum trading system, april thirty. six live trade symbols: ada, dot, link, atom, avax, near. five to eight years of fifteen minute bars each. four knn modes tested, trailing and forward crossed with pine_approx and topk_real.
eighty seven thousand to one hundred ninety five thousand trades per mode, bootstrap n equals five thousand.
all four modes lose money on every symbol. zero of six profitable. information coefficient, raw prediction versus forward four bar return, absolute value below zero point zero two everywhere. effectively zero predictive content. every year twenty eighteen to twenty twenty six negative r.
Originally published on FalsifyLab Substack.
ā research and educational content. not investment, legal, or tax advice. do your own research. positions and views may change without notice.