Quant integrity review · internal · published in full

The flagship number, put on trial.

A credential proves a strategy's numbers are real. So the numbers had better be. We turned MIZAN's own retraction ledger, every failure mode that has ever killed one of our engines, on our own flagship credential, and published the verdict. This is not a marketing tear-sheet; it is an adversarial review that assumed the number was fake until the code proved otherwise.

Supersession notice · era lawThis review examined BTC-TREND-BOOK-V1, era v10, on 2026-07-18. A later flagship, BTC-TREND-4H, was minted under era v11 on 2026-07-28; the current engine era is v12 (659bdb67…, live since 2026-10-14). This page is kept unrewritten under era law: a review binds to the artifact and era it examined, permanently. Superseded is never revoked, and never silently rewritten. A note on “flagship,” since the word carries two verdicts across the estate: what passed here is the hard-limit gate (Sharpe, CAGR, drawdown) on this strategy, verdict HONEST. The line elsewhere that “the gate refused its own maker” refers to a different seal on the maker’s flagship: its deflated-Sharpe credential returned 0.68, not significant. One strategy carries several independent seals, and they can disagree; that is the point of publishing each one.

Subject · BTC-TREND-BOOK-V1 (flagship credential at the time of review)Method · adversarial quant lens vs. the retraction ledgerDate · 2026-07-18

1.63Sharpe (after costs) · vs buy-and-hold 0.78

0.20×Avg net exposure · below buy-and-hold's 1.0×

0Lookahead · phantom fills · beta illusion

HONESTVerdict on the proof itself

§01Verdict

The credential does not misrepresent the strategy. The proof is honest.

The flagship credential attests that a hidden trend strategy cleared MIZAN's institutional gate with Sharpe 1.63, CAGR 26.6%, net +301.6%, max drawdown 20.76% over 12,917 four-hour BTC bars (2020–2026), after 7 bps fee + 3 bps slippage, verified zero-trust. Tested against every historical failure mode in the retraction ledger, the number holds: no lookahead, no phantom fills, correct annualization, no survivorship surface. Two residuals are disclosed below; neither touches the integrity of the proof.

Verdict: honest, conservative, and not disguised leverage. The strategy's entire edge is risk-adjusted. Over the identical window, simply holding BTC returned more (+432.7% vs +301.6%): the strategy trailed on raw return. It wins by running at a fraction of the risk: it averages 0.20× net exposure (below buy-and-hold's constant 1.0×), so its Sharpe of 1.63 against buy-and-hold's 0.78, at roughly a quarter of the drawdown, is real trend-following, not a leveraged bet on a bull market dressed up as alpha.

§02The load-bearing evidence

Return trails buy-and-hold. Risk-adjusted return crushes it.

The single most revealing test: benchmark the strategy against simply holding the asset over the exact same window. A strategy that is secretly just leverage would beat buy-and-hold on return with similar or worse risk. This one does the opposite: the signature of genuine risk management, not a beta illusion.

Buy & hold BTCFlagship credential

Each metric scaled to its own maximum · identical 2020–2026 window

Metric	Buy & hold BTC	Flagship credential
Total return	+432.7%	+301.6%
Sharpe (after costs)	0.78	1.63
Max drawdown	77.0%	20.8%
Avg net exposure	1.00×	0.20×

It gives up 131 points of return to buy 2.1× the Sharpe at roughly one-quarter the drawdown, and it does so while holding, on average, one-fifth of buy-and-hold's exposure. The nominal leverage ceiling in the spec is never approached; vol-targeting holds the book far below it. This is textbook trend-following: step aside in chop and bear, size into confirmed trends. The levered-beta hypothesis is refuted by the strategy's own exposure card, which the verifier recomputes from the same position series the P&L compounds.

§03What the review surfaced

Two residuals. Both disclosed, neither fatal.

The audit distinguishes "the credential misrepresents the strategy" (fatal: the proof itself would be dishonest; none found) from "the strategy is real but a sharp reader should know X" (expectations-setting). Only the second kind surfaced.

Vol-target rebalancing turnover is not fully costedCost is charged at entry/exit (scaled by leverage), but the vol-target's bar-to-bar leverage drift rebalances notional without a modeled cost. Bounded impact: roughly 1–2%/yr of CAGR (worst case ~4%) against a 26.6% CAGR with ~6 points of headroom over the 20% gate floor; real but immaterial to the verdict. The strategy is slightly flatter than it looks; the proof does not lie. Fix (cost |Δ(pos·lev)|) queued for the next engine re-anchor; disclosed in the interim.

Disclosed

One config, one asset, soft out-of-sample gateA point-in-time credential proves "this strategy cleared the gate on this bound data," not "a durable multi-regime edge." The gate's walk-forward term is a last-30%-of-window tail split at Sharpe ≥ 0.50, a consistency filter rather than a pre-registered holdout. Honestly scoped by the credential's own "a hidden strategy cleared the gate" language; consider disclosing the committed out-of-sample Sharpe.

Scope

Framing rule: never show "+301.6%" alonePresented as a bare return, it reads as outperformance, but it trails buy-and-hold. The honest headline is the risk-adjusted one: Sharpe 1.63 vs 0.78 at a quarter of the drawdown. A hostile reader spots the raw-return gap before you do; always show the benchmark beside it.

Framing

§04Verified clean

Checked against every failure mode that ever killed a MIZAN engine.

MIZAN's retraction ledger is a public list of its own past mistakes: lookahead, phantom fills, survivorship, monthly sampling hiding intramonth drawdowns, carry that was gross-of-borrow. The flagship was checked against each. All clean.

◆ Verified honest

No lookahead. The signal is explicitly lagged: the position over each bar uses only data known before it. Traced through the trend-signal construction, not assumed.
No phantom fills. The trend book has no intrabar stop anywhere, so the phantom-fill class that retracted the squeeze family is structurally impossible.
Drawdown marked on wicks. The 20.8% figure is measured against intrabar extremes at the strategy's real leverage, not close-only; near-liquidation wicks are visible.
Annualization bound to data. The proof enforces that you cannot claim more periods per year than the bar spacing allows; Sharpe cannot be inflated by a false basis.
No survivorship surface. Single canonical asset; the failure mode that killed the equity families does not apply.
Metrics bound by the receipt. Image-id, real STARK, journal-match and data-root all verified; the issuer cannot fabricate or swap the numbers.
Inside the published-honest range. Sharpe 1.63 / CAGR 26.6% sits squarely within, on the conservative side of, MIZAN's own previously published trend-book research.

○ Disclosed limits

Vol-target rebalance turnover not fully costed (~1–2%/yr optimism, bounded; fix queued).
Single config, single asset, single window: a point-in-time credential, not a durable-edge claim.
Walk-forward gate is an in-window tail split, not a pre-registered holdout.
Raw return trails buy-and-hold; the edge is entirely risk-adjusted and must be framed that way.

Standing posture, same as the ZK review. Credential-grade, not capital-grade. This audit establishes that the flagship credential is an honest representation of a real, conservative, risk-managed strategy, not that any strategy is a durable future edge. A credential is a seal on a historical or forward record, not a forecast. Live capital remains gated behind an external paid audit. No custody, orders, or funds are touched anywhere in the system.

Reproducible, not asserted. Every claim here is re-derivable: the buy-and-hold benchmark is computed from the same public CSV the credential is bound to; the exposure figures are recomputed by the free public verifier from the credential itself, a 225 KB bundle, verifiable offline in ~81 ms on the current era; the full finding register with file:line lives in the engine repository at docs/AUDIT_2026_07_18_v5_findings.md. Read the companion ZK Engine Security Review →

Reviewed · 2026-07-18

Internal, AI-driven review: not a third-party audit, investment advice, or an offer of any security. Past or backtested performance is not indicative of future results. A credential attests that a strategy cleared a fixed gate on bound historical data with the integrity properties described; it does not predict future performance.

ZK engine security review The retraction ledger Verify it yourself

ENGINEv12 · 659bdb67… ● LIVE REGISTRY102 verdicts · append-only VERIFY~81 ms · offline · trusting no one ANCHORBitcoin #962,013 SPECVTR-1 · frozen · CC BY ROOT7b655dac…1fbf6d CODEGITHUB ↗