Evidence-first, not trade calls

Research Lab

This platform does not optimize for finding trades. It optimizes for discovering, measuring, validating, and continuously improving statistically defensible trading edges.

What this page is

This Experiment Dashboard is a live view into five parts working together: a Research Registry of every hypothesis registered before it's evaluated (never after — no fitting a question to a result already known), an Evidence Ledger of every resolved signal, and a Confidence Engine that labels every conclusion by its real evidence strength rather than a bare number. The reasoning behind each — this project's Research Notebook — is maintained internally alongside the code that produces it. Observational only — nothing on this page gates a trade or changes scoring_engine.py's weights.

Confidence Engine labels, by real sample size

INSUFFICIENT: n<20 — too small to conclude anything
WEAK: n≥20 but effect size/significance still thin
MODERATE: n≥50 with a real effect
STRONG: n≥100 with a real effect
LOW VARIANCE: a separate check (not sample-size-based) — the factor barely varies across signals, so a low-vs-high comparison isn't meaningful regardless of n.

Hypotheses registered: 3Registered before the result is known
Resolved stock signals: 1373With a real scoring breakdown on file
Auto-trade decisions: 325091783 accepted · 30726 rejected
Executed trades on file: 29452945 with a regime snapshot

Data Quality

stock missing exit price

58/2817 resolved stock signals are quarantined (incomplete_exit, legacy, excluded from all stats)

WARN

crypto missing exit price

0/4697 resolved crypto signals have an outcome but no exit_price/pct_gain

outcome pct gain mismatch

0/2759 resolved stock signals have outcome/pct_gain sign mismatches (e.g. outcome=LOSS with a positive pct_gain)

stopped out without verdict

0/1014 stopped-out signals have no verdict field - expected for anything logged before fix #193, not a bug

shadow score neutral dilution

433/1437 shadow-scored signals have >50% of factors on neutral-credit fallback

WARN

profile cache staleness

Last full refresh 64 days ago (21769 symbols; expected roughly monthly)

FAIL

coingecko cache health

1810 cached entries, oldest 1397.2h old

regime snapshot staleness

Most recent regime snapshot 16.3h old (expected within 48h)

order flow archive readable

1 of 69 sealed order-flow archives cannot be decompressed and are silently excluded from every order-flow statistic: order_flow_journal-2026-09-16.jsonl.gz (245339 bytes) - median archive is 4,993,218 bytes

FAIL

Regime-Conditioned Performance

RANGE BOUND LOW VOL

650 trades · win rate 68.9% · Sharpe 0.35

TRENDING HIGH VOL

3 trades · small sample

TRENDING LOW VOL

2292 trades · win rate 65.7% · Sharpe 0.28

Auto-Trade Rejection Report

Exactly why signals reaching AutoTrader.on_signal() are accepted or rejected, by the specific gate that decided it — real, persisted data (fix #349), not reconstructed from logs that don't survive this project's own log retention. Covering 2026-07-30 to 2026-10-28.

32509 decisions logged — 1783 accepted · 30726 rejected

MODERATE

circuit breaker

5416 (17.6% of rejections)

min smc grade

4923 (16.0% of rejections) · 1597/4923 within 10% of clearing

max positions

4629 (15.1% of rejections)

duplicate position

4261 (13.9% of rejections)

min rr

4111 (13.4% of rejections) · 144/4111 within 10% of clearing

smc bearish bias

2836 (9.2% of rejections)

not halal compliant

1859 (6.1% of rejections)

min confluence

853 (2.8% of rejections) · 708/853 within 10% of clearing

portfolio exposure

666 (2.2% of rejections)

min rel vol

332 (1.1% of rejections) · 28/332 within 10% of clearing

compliance unresolved

181 (0.6% of rejections)

stock signal barred

147 (0.5% of rejections)

news blackout

133 (0.4% of rejections)

unpriceable symbol

101 (0.3% of rejections)

stale price invalidates bracket

84 (0.3% of rejections)

market closed

66 (0.2% of rejections)

min swing confluence

49 (0.2% of rejections) · 29/49 within 10% of clearing

crypto stop too wide

45 (0.1% of rejections)

min rr at execution

16 (0.1% of rejections)

broker rejected

11 (0.0% of rejections)

crypto signal barred

6 (0.0% of rejections)

sizing constraint

1 (0.0% of rejections)

Scalp Stop-Width Filtering

2026-07-23: real closed stock trades had stops as tight as 0.15%-2.2% of entry — tighter than ordinary intraday noise, contributing to a 16.7% stock win rate vs crypto's 57.9% over the same window. MIN_SCALP_STOP_PCT now rejects a stop that's too tight, the same way MAX_SCALP_STOP_PCT already rejected one too wide — a distinct, earlier stage than the rejection report above (a signal filtered here never reaches on_signal() at all). Covering 2026-09-01 to 2026-10-27.

too tight (avg 0.51%): 5517
too wide (avg 6.54%): 1248

BREAKOUT

184 too tight · 0 too wide

BULL_FLAG

2 too tight · 49 too wide

CANDLE

2031 too tight · 27 too wide

EMA_PULL

791 too tight · 3 too wide

FIB_BOUNCE

206 too tight · 0 too wide

GAP_GO

15 too tight · 629 too wide

HOD_BREAK

326 too tight · 31 too wide

MOMENTUM

834 too tight · 15 too wide · 374 other

ORB

40 too tight · 141 too wide

PM_HIGH

22 too tight · 103 too wide

SETUP

285 too tight · 0 too wide

VWAP

781 too tight · 250 too wide

GAP_GO Entry Slippage

Real case, 2026-07-23: a GAP_GO market order intended to enter CITR at $5.04 filled at $5.96 (18.25% slippage), while position sizing had already used the tiny intended risk-per-share — the stop stayed fixed near the pre-gap level, turning a controlled 1%-risk trade into a $1,267 loss. GAP_GO entries are now a marketable limit order bounded at entry × (1 + MAX_GAP_GO_ENTRY_SLIPPAGE) instead of an unrestricted market order.

GAP_GO trades: 7
with bounded entry: 2
avg slippage: 3.628%
max slippage: 18.254%

Halal ETF Constituents Not Yet Tracked

Real holdings of SPUS/HLAL/MNZL (from their own SEC Form N-PORT filings) that don't match any symbol already on the watchlist by company name. A review list only — N-PORT filings never include a ticker, so matching is by name and inherently fuzzy. Nothing here is added automatically; each candidate needs its own live ticker/listing check first, same discipline that caught "AMAL" resolving to an unrelated bank.

393 watchlist symbols confirmed held by a real fund · 94 real holdings not matched to the watchlist

as of 2026-10-24

AECOM (00766T100)

MNZL · 0.03%

AGCO Corp (001084102)

MNZL · 0.02%

Acuity Inc (00508Y102)

MNZL · 0.02%

Antero Resources Corp (03674X106)

MNZL · 0.04%

Apellis Pharmaceuticals Inc (03753U106)

MNZL · 0.01%

Ashland Inc (044186104)

MNZL · 0.01%

AvalonBay Communities Inc (053484101)

MNZL · 0.10%

Avantor Inc (05352A100)

MNZL · 0.02%

Avnet Inc (053807103)

MNZL · 0.03%

BellRing Brands Inc (07831C103)

MNZL · 0.01%

BorgWarner Inc (099724106)

MNZL · 0.04%

Bright Horizons Family Solutions Inc (109194100)

MNZL · 0.01%

Brookfield Renewable Corp (11285B108)

MNZL · 0.02%

Builders FirstSource Inc (12008R107)

MNZL · 0.03%

CDW Corp/DE (12514G108)

MNZL · 0.09%

CH Robinson Worldwide Inc (12541W209)

MNZL · 0.07%

CLEAN HARBORS INC COM (184496107)

MNZL · 0.06%

Caris Life Sciences Inc (142152107)

MNZL · 0.00%

Cava Group Inc (148929102)

MNZL · 0.03%

Choice Hotels International Inc (169905106)

MNZL · 0.00%

ConocoPhillips (20825C104)

HLAL, MNZL · 0.51%

Crocs Inc (227046109)

MNZL · 0.02%

DR Horton Inc (23331A109)

HLAL, MNZL · 0.15%

Dick's Sporting Goods Inc (253393102)

MNZL · 0.05%

Dillard's Inc (254067101)

MNZL · 0.00%

Eagle Materials Inc (26969P108)

MNZL · 0.02%

Elastic NV (N/A)

MNZL · 0.02%

Electronic Arts Inc (285512109)

HLAL · 0.18%

Eli Lilly & Co (532457108)

HLAL, MNZL · 2.68%

Envista Holdings Corp (29415F104)

MNZL · 0.01%

Equity Residential (29476L107)

MNZL · 0.09%

Esab Corp (29605J106)

MNZL · 0.02%

Everpure Inc (74624M102)

MNZL · 0.08%

Expeditors International of Wa (302130109)

HLAL · 0.07%

Ferguson Enterprises Inc (31488V107)

HLAL, MNZL · 0.18%

Floor & Decor Holdings Inc (339750101)

MNZL · 0.02%

Fortune Brands Innovations Inc (34964C106)

MNZL · 0.01%

GXO Logistics Inc (36262G101)

MNZL · 0.02%

Gap Inc/The (364760108)

MNZL · 0.02%

Gates Industrial Corp PLC (N/A)

MNZL · 0.02%

Globant SA (N/A)

MNZL · 0.01%

Hayward Holdings Inc (421298100)

MNZL · 0.01%

Hologic Inc (436440101)

HLAL · 0.06%

Huntsman Corp (447011107)

MNZL · 0.01%

IAC Inc (44891N208)

MNZL · 0.01%

IFF (459506101)

HLAL · 0.08%

Insmed Inc (457669307)

MNZL · 0.10%

JB Hunt Transport Services Inc (445658107)

HLAL, MNZL · 0.06%

Kyndryl Holdings Inc (50155Q100)

MNZL · 0.01%

Lear Corp (521865204)

MNZL · 0.02%

+ 44 more, not shown

Signal-Level Factor Research

Based on 1373 resolved stock signals with a real scoring breakdown on file — low vs high tercile win rate per factor.

liquidity structure

stuck at 0.0 for most signals — not enough spread to compare yet

LOW VARIANCE

vwap

stuck at 7.2 for most signals — not enough spread to compare yet

LOW VARIANCE

momentum

low `10.7%` (n=457) vs high `14.2%` (n=459)

MODERATE

volume

low `12.0%` (n=457) vs high `11.8%` (n=459)

MODERATE

smc

stuck at 0.0 for most signals — not enough spread to compare yet

LOW VARIANCE

trend

low `8.5%` (n=82) vs high `12.0%` (n=83)

WEAK

sentiment

low `20.8%` (n=457) vs high `8.3%` (n=459)

MODERATE

fundamentals

low `12.7%` (n=457) vs high `9.8%` (n=459)

MODERATE

risk liquidity

low `10.3%` (n=457) vs high `16.6%` (n=459)

MODERATE

Feature Redundancy

liquidity_structure and smc — insufficient variance to assess

r=0.922 (artifact)

Research Registry

Every hypothesis this platform tests, registered here before its result is known — a question can't be quietly reworded after the fact to match whatever the data happened to show.

STRONG n=664 vs 1326