Research Brief · Strategy Memo

Leverage & break-even viability

Does tiny-stake, high-leverage, move-to-break-even management actually survive real fees, slippage and liquidation — or does it just feel like it should? We tested it against 497 real trade paths and pre-registered the answer.

Paper research · not live trading 19 Jun 2026· n = 497 resolved trades· Reproducible · read-only
Executive summary

The tempting version is a ruin machine. A marginal edge survives in exactly one corner.

The honest one-paragraph readout — then the numbers that back each clause.

The “price goes briefly green before it resolves” observation is real (95% of trades touch green) and moving the stop to break-even genuinely improves the raw R. But realistic crypto taker fees eat the entire edge, and high leverage makes it worse, not better — at 50x, normal price wiggle liquidates the position before it can recover. The only configuration that is even marginally positive is a tiny lock-in profit the moment a trade ticks green, at =10x leverage, on a near-zero-fee (maker) venue. That edge is small, unproven out-of-sample, and hinges entirely on one untested assumption: that we can actually get maker fills. We are not claiming a live edge — we are claiming a disciplined map of where one could exist, and a gated plan to prove or kill it.

Trades analysed
497
Resolved bracket trades with full post-fill bar paths (of 498).
Ever go green
95%
Touch a profit at some point — but most of it lives inside the fee band.
Best rule @ maker fee
+0.038R
Per trade, near-zero fee. At realistic taker fee: −0.219R.
50x liquidation rate
55%
Risk-of-ruin 100% by 500 trades. 10x liquidates ~1%.
Finding 1 · Leverage is a liability, not a multiplier

Why 50x can’t work: the noise is bigger than the liquidation band

Leverage doesn’t change whether a trade wins — it changes how far price can move against you before you’re force-closed (≈ 1/leverage). The trades’ ordinary adverse wiggle (median 1.62%, 90th pct 3.87%) routinely exceeds the high-leverage bands.

LeverageLiquidation band% of trades liquidatedRisk of ruin (1000 trades)Verdict
10x≈ 9.5%~1%0%survives
25x≈ 3.5%~12%92%fragile
50x≈ 1.5%~55%100% (by 500)ruinous
10x
~1%
25x
~12%
50x
~55%

Share of trades whose worst adverse move breached the liquidation band — the stake is wiped regardless of the eventual outcome. High leverage converts survivable noise into guaranteed losses.

Finding 2 · Fees decide everything

The same rule is positive at maker fees and deeply negative at taker fees

Best management variant — lock +0.1R the moment it ticks green — expectancy per trade, gross then net of each fee model. Risk is measured in R (= stop distance), so a small price-fee becomes a large R-cost on tight stops.

Grossbefore fees
+0.077R
Maker / zero-feethe only positive lane
+0.038R
Realistic takercrypto, default
−0.219R
Harshtaker both sides + funding
−0.518R

Centre line = break-even (0R). The maker bar is a sliver above zero on purpose — that is the whole edge, and it is thin. Break-even management adds ≈ +0.36R of gross improvement; realistic friction is ≈ −0.30R and eats all of it.

Finding 3 · Out-of-sample shadow test (Tier 1)

Pre-registered, paper-only, zero risk — and it refuses to over-claim

We re-ran the rule out-of-sample on two venue lanes, with success and kill thresholds fixed in advance. The genuinely zero-fee book is still too small to call; the lane that looks great is the one that assumes a fill we haven’t proven.

Primary · genuinely zero-fee book
INCONCLUSIVE
n = 25 (< 150 pre-registered gate). Rule net −0.226R so far — too small to trust either way. Keep collecting.
Secondary · maker-assumed crypto
PASS*
n = 314. Rule +0.106R net (90% CI excludes 0); beats baseline by +0.427R. * assumes maker fills we can’t yet verify.

⚑ So what?

The edge is real conditional on getting maker fills at ~0 fee. That single assumption — can we actually be filled as a maker on the entry and the break-even exit — is now the make-or-break variable. Everything else is mapped. A Tier-1 “pass” is necessary but not sufficient; it cannot test fill feasibility. That is what Tier 2 exists to settle.

Finding 4 · Tier-2 maker-fill feasibility (the make-or-break)

The one untested assumption, now tested: maker entries actually fill ~87% of the time

The entire positive case depends on getting a maker fill at ~0 fee. The live paper engine already enters with a maker-retrace limit and records filled-vs-cancelled, so we can measure it directly — read-only, no live orders. Of decided entries, 86.6% filled (557 / 643), median time-to-fill 0.22h. Pre-registered kill was < 60%.

Overall
86.6%
Crypto
87.6%
Zero-fee venue
76.4%
Kill line
60%

⚑ So what?

The make-or-break leg clears its bar with room to spare — maker entries are achievable, fast, and consistent across venues and timeframes. That removes the single biggest unknown from Finding 3.

Honest scope: this measures the entry fill. The break-even exit is typically a taker on crypto (only the zero-fee TradFi venue is fee-free on both sides), and historical paper fills aren't a guarantee of real-exchange fills at micro size. So this is strong evidence, not final proof — the remaining work is the exit-side fee and a live micro-stake confirmation.

The validation plan · how this becomes real (or gets killed)

A gated, two-tier paper-forward test — cheapest and safest first

Nothing touches live money or the validated trading engine until each gate is cleared. Pre-registered success and kill criteria are committed before data is collected, so the result can’t be fitted after the fact.

Built & running · zero risk

Tier 1 — shadow overlay

A read-only report that replays the rule over real trades as they resolve. No engine change, writes nothing to the live journal. Status: live; primary lane gathering toward the n=150 gate.

2
Entry leg measured · PASS

Tier 2 — maker fill feasibility

Measure whether the maker entry actually fills. Done, read-only: ~87% fill rate, median 0.22h — clears the <60% kill. Remaining: the break-even exit fee on crypto, and a live micro-stake confirmation that paper fills hold on a real exchange.

3
Only if both clear

Graduation — micro stake

Net expectancy > 0 with statistical significance at n=150, plus liquidations ~0 at =10x, plus verified fills. Even then: micro stake only, never full-account risk.

Pre-registered gates (fixed in advance):

  • Success: net-of-fee expectancy > 0 and a statistically significant improvement over current management, at n ≥ 150 maker-venue trades.
  • Kill: maker fill rate < 60%; or net < −0.10R over any rolling 100 trades; or liquidation rate above ~2% at =10x.
  • Horizon: ≥ 2 weeks of the hourly paper engine, or n ≥ 150, whichever is later.
Risk & honest assumptions

What this brief is — and what it is not

This is not a track record or a return claim

No live trading has occurred. These are simulations over historical bar paths with explicit, labelled friction assumptions. “95% go green” counts touches, not guaranteed fills. The sample pools a known market-regime shift and is descriptive, not causal.

  • Intrabar conflicts are resolved adverse-first, so break-even and trailing logic is never over-credited.
  • Maintenance margin is fixed at 0.5%; real exchange ladders are size/tier-dependent and worse for altcoins.
  • Risk-of-ruin assumes independent trades; real losing streaks cluster, so treat ruin numbers as an optimistic floor.
  • The zero-fee/maker venue sample is small (tradfi n=41) — directional, not conclusive.
  • The entire positive case rests on unverified maker fills. Until Tier 2 measures that, treat the edge as a hypothesis.

Reproduce the full analysis (read-only over the trade journal):

PYTHONPATH=. python scripts/leverage_be_study.py
PYTHONPATH=. python scripts/leverage_be_shadow.py