Quant Swarm Audit · Reproducible Evidence Report

momentum_breakout_v2_remediated

Job: audit_b05e984c671fe229 Assessment: PASSED_LOCAL_AUDIT Input SHA-256: b287ae1fc4f67c6204ec64a4… Package: quant-swarm==0.7.0

Remediation Comparison: momentum_breakout_v1_flawed → momentum_breakout_v2_remediated

REMEDIATION_VERIFIED

Before Job: audit_906ea84d9a70a362 (PROBLEMS_OBSERVED, SHA-256 158d402e172551c6…) → After Job: audit_b05e984c671fe229 (PASSED_LOCAL_AUDIT, SHA-256 b287ae1fc4f67c62…) · Resolved findings: 5

Metric Before (momentum_breakout_v1_flawed) After (momentum_breakout_v2_remediated) Delta
Gross Annualized Sharpe -1.642 3.089 +4.731
Net Annualized Sharpe (1.0x House Costs) -2.065 2.161 +4.226
Mean Gross Edge / Trade (bps) 7.79 39.68 +31.89
Mean Net Edge / Trade @ 1.0x Cost (bps) -7.21 24.68 +31.89
Mean Net Edge / Trade @ 2.0x Cost (bps) -22.21 9.68 +31.89
Breakeven Cost Multiplier (x) 0.52 2.65 +2.13
Max Drawdown (%) 23.74 3.13 -20.61
Top-5 Trade PnL Share (%) 93.2 15.2 -78.0
Diagnostic Check Before After Post-Remediation Evidence
Sample Depth & Statistical Power Floor OBSERVED_PROBLEM PASSED Observed 180 sessions across 37 weekly blocks (clears 60-session / 20-week floor). Gross Sharpe 3.09, 90% bootstrap CI [1.09, 5.48].
Pessimistic Execution Cost & Slippage Stress Survival OBSERVED_PROBLEM PASSED Gross edge of 39.68 bps/trade survives 1.0x (15.00 bps -> net +24.68 bps) and 2.0x pessimistic cost stress (30.00 bps -> net +9.68 bps; breakeven multiplier: 2.65x; net Sharpe: 2.16).
Drawdown Tail & Volatility-Regime Consistency OBSERVED_PROBLEM PASSED Max drawdown contained at 3.13% (worst 5-session return: -2.18%); positive mean return in both calm (+15.65 bps/day) and elevated-volatility (+7.49 bps/day) windows.
Trade Outlier Concentration & House Position Cap Compliance OBSERVED_PROBLEM PASSED Across 45 trades, top-5 trades account for 15.2% of gross gains (well below 45% threshold) and max single-position size is 16.0% (0 breaches of the 25% house floor).
Point-in-Time Feature Provenance & Lookahead Audit OBSERVED_PROBLEM PASSED All 3 declared feature(s) passed point-in-time timestamp checks (`available_at <= as_of`), zero banned actuals sources, and coverage >= 90%.
Benchmark Active Return & Information Ratio NOT_TESTED PASSED Matched 180 overlapping sessions vs benchmark; annualized active return is 2754.4 bps with Information Ratio 1.72.
Underlying Strategy Code & Multiple-Testing Trial Ledger Boundary NOT_TESTED NOT_TESTED Tabular artifact audit only — strategy source code and discarded trial count were not evaluated locally.
Observations / Weeks180 / 37w
Gross / Net Sharpe3.09 / 2.16
Net Edge (1.0x / 2.0x Cost)24.7 / 9.7 bps
Breakeven Cost Multiplier2.65x
Max Drawdown3.13%
Top-5 Trade Share15.2%

Structured Diagnostic Findings

1. Sample Depth & Statistical Power Floor

PASSED
Evidence
Observed 180 sessions across 37 weekly blocks (clears 60-session / 20-week floor). Gross Sharpe 3.09, 90% bootstrap CI [1.09, 5.48].
Why it matters
Provides sufficient time-series degrees of freedom to evaluate regime stability and cost drag.
Next experiment
Proceed to multi-seed paired comparison against the live champion (`tournament.submit`).
Limitation
Single-path historical depth does not account for unrecorded parameter trials (Deflated Sharpe).

2. Pessimistic Execution Cost & Slippage Stress Survival

PASSED
Evidence
Gross edge of 39.68 bps/trade survives 1.0x (15.00 bps -> net +24.68 bps) and 2.0x pessimistic cost stress (30.00 bps -> net +9.68 bps; breakeven multiplier: 2.65x; net Sharpe: 2.16).
Why it matters
Confirms a structural buffer above realistic bid-ask spread, slippage, and adverse selection.
Next experiment
Verify liquidity capacity on target symbols using `market.microstructure` before scaling notional.
Limitation
Based on declared/house bps parameters; borrow fees for short legs are not included.

3. Drawdown Tail & Volatility-Regime Consistency

PASSED
Evidence
Max drawdown contained at 3.13% (worst 5-session return: -2.18%); positive mean return in both calm (+15.65 bps/day) and elevated-volatility (+7.49 bps/day) windows.
Why it matters
Demonstrates controlled tail excursion and positive expectancy across volatility environments.
Next experiment
Verify worst-regime margin across all 5 canonical regimes via `gym.probe_fragility`.
Limitation
Close-to-close series does not capture intraday stop-out slippage.

4. Trade Outlier Concentration & House Position Cap Compliance

PASSED
Evidence
Across 45 trades, top-5 trades account for 15.2% of gross gains (well below 45% threshold) and max single-position size is 16.0% (0 breaches of the 25% house floor).
Why it matters
Confirms broad-based trade breadth and full compliance with live-capital position sizing invariants.
Next experiment
Validate multi-asset rebalance trade lists using `warden.plan_rebalance`.
Limitation
Assumes trade timestamps reflect actual execution fill times.

5. Point-in-Time Feature Provenance & Lookahead Audit

PASSED
Evidence
All 3 declared feature(s) passed point-in-time timestamp checks (`available_at <= as_of`), zero banned actuals sources, and coverage >= 90%.
Why it matters
Verifies declared feature inputs are causally knowable at decision time.
Next experiment
Cross-check cross-sectional IC significance on alternative-data features with `pit.analysis`.
Limitation
Validates the feature manifest contract; does not verify internal code implementation.

6. Benchmark Active Return & Information Ratio

PASSED
Evidence
Matched 180 overlapping sessions vs benchmark; annualized active return is 2754.4 bps with Information Ratio 1.72.
Why it matters
Separates genuine active edge from passive market beta exposure.
Next experiment
Verify active return stability across drawdown and chop regimes (`gym.paired_preview`).
Limitation
Linear benchmark subtraction does not capture nonlinear tail beta.

7. Underlying Strategy Code & Multiple-Testing Trial Ledger Boundary

NOT_TESTED
Evidence
Tabular artifact audit only — strategy source code and discarded trial count were not evaluated locally.
Why it matters
Even a clean return series and manifest cannot prove that 500 parameter combinations were not swept behind the scenes to find one lucky backtest.
Next experiment
Submit the candidate genome to the hosted Shadow Tournament (`tournament.submit`) for paired replay against the live champion and Deflated Sharpe ratio accounting.
Limitation
Local tabular checks test the artifact you brought, not the research process that selected it.

Reproducibility Bundle

{
  "job_id": "audit_b05e984c671fe229",
  "preflight_id": "pf_b287ae1fc4f67c62",
  "strategy_name": "momentum_breakout_v2_remediated",
  "input_sha256": "b287ae1fc4f67c6204ec64a4f73f08a1d7235ab898028034c73677bc648617f5",
  "artifact_hashes": {
    "returns": "de053179e207ebeadd12f5d5dade4d65b1f57027d610f497e685afe65bf2541b",
    "trades": "c9cbb91116395c9bda1bc52f13a66f7ef37cbabfbdd5055f308ba2829ebbbf4d",
    "benchmark": "606aa085d26bdec4f36484ec6335fe10628018d441dd9d1d7a6a65a4829fa416",
    "cost_assumptions": "28bf32f0f5607e7aee1aa6cd2cbb41d7255a521be6922470e05b1bd35a23042d",
    "features_manifest": "1e83d3c64422303b8c47ae4166ac8fc64b6dbc8e9c526d19c34a95d0a07872fb"
  },
  "cost_assumptions": {
    "spread_bps": 2.0,
    "slippage_bps": 3.0,
    "adverse_selection_bps": 2.0,
    "commission_bps": 0.5
  },
  "mode": "local",
  "package_version": "0.7.0",
  "random_seeds": [
    0,
    1,
    2,
    3,
    4
  ],
  "approval_token": "aq1_3ae700c96b0aad931c439f2cc5f9f88a",
  "idempotency_key": "demo-clean-v2",
  "generated_at": "2026-10-31T00:24:31.745476+00:00"
}