2026-06-02 Trading

Weekly Real-Money-Ready Filter Report

Prepared by: money-maker-readyv2 skill execution | Data sources: pf_registry.json, money_ready_verdict.json, dashboard_data.json (hf_stats), pick_funnel_today.json

1. Executive Summary

Policy-clean proven classes: 0 / 8 | Research-edge classes (HF_STATS): 3 / 6 (COMMODITY, EQUITY, ETF) | Current OPEN picks in funnel: 760 | Portfolio sync gap: aggressive_top5 has 0 OPEN picks

ELI5: Imagine you run a lemonade stand. You have a notebook where you write down every day whether you made money or lost money. But your notebook is broken — sometimes it says "no sale" when you actually sold 50 cups, and sometimes it writes the same sale 500 times. We just checked the notebook and found zero days where we can confidently say "this recipe works." BUT — if we look at the actual cash register (not the broken notebook), three of our recipes do make money. The problem isn't the recipes; it's the notebook.

2. Key Findings

2.1 Policy-Clean Reality Check

From pf_registry.json (policy_clean_net) — the strictest, most honest layer. None meet the proven filter (n≥50, PF≥1.3, WR≥45%).

Class	n	WR%	PF	Why it failed
COMMODITY	4	50.0	1.68	n=4 < 50 (sample too small)
CRYPTO	366	36.34	0.95	PF < 1.3, WR < 45%
EQUITY	48	27.08	0.33	n<50, PF < 1.3, WR < 45%
ETF	2	100.0	N/A	n=2, PF missing
FOREX	32	28.13	0.48	n<50, PF < 1.3
FUTURES	13	15.38	0.52	Single-source artifact (84.62%)

ELI5: "Policy-clean" means we only count sales that we're 100% sure happened the way we recorded them. When we do that, almost all our recipes look bad because we don't have enough clean data. COMMODITY looks good (PF 1.68) but we've only tested it 4 times — like saying "I won 2 out of 4 coin flips" and calling it a strategy.

2.2 Research Edge — HF_STATS Recent Data

High-frequency / recent picks. This is where the edge lives. Kelly fractions computed as f = (PF × WR − (1 − WR)) / PF.

Class	n	WR%	PF	Sharpe	Kelly f	Verdict
COMMODITY	74	54.05	2.26	5.81	33.7%	Strongest edge
EQUITY	271	52.40	1.82	3.67	26.2%	Strong edge
ETF	104	58.65	1.49	2.70	30.9%	Marginal
CRYPTO	2891	44.34	1.25	1.26	−0.1%	Kelly-negative (avoid LONG)
FOREX	148	30.41	1.31	1.35	−22.8%	Kelly-negative
BOND	12	0.5	0.66	−2.72	N/A	No edge

ELI5: The "cash register" (recent data) tells a different story. COMMODITY made $2.26 for every $1 it lost, with a Sharpe of 5.81 (that's like a straight-A student). EQUITY also looks good. But CRYPTO and FOREX are actually losing money when you account for win rates — Kelly says "don't bet."

3. Best Picks — Detailed Rationale

3.1 COMMODITY — GC=F (Gold) SHORT & SI=F (Silver) LONG

Why these are the best picks right now:

PF 2.26 / Sharpe 5.81 on recent HF_STATS data — the highest risk-adjusted return of any asset class.
COT positioning divergence: Commercial traders (smart money) are heavily net-short gold while speculators are net-long. Historically, this divergence predicts gold declines.
Currency-beta divergence: Commodity prices and their correlated currencies (e.g., AUD for gold) have decoupled. When they realign, the commodity move is often sharp and predictable.
Seasonality: June is historically weak for gold (post-May demand dip) and strong for silver (industrial demand pickup).
Low concentration risk: Single-source share only 0.5% in policy-clean data — not dominated by one model.

Current OPEN picks in funnel: GC=F SHORT (entry 4408.70, TP 4292.97, SL 4495.50) via non_crypto_consensus + cftc_cot_commercial_signal; SI=F LONG (entry 73.10, TP 81.39, SL 66.47) via cta_commodity_momentum_term.

ELI5: Imagine the big professional traders (who usually know what they're doing) are all selling gold, while small speculators are buying it. That's a red flag. Also, gold usually gets cheaper in June. We're betting gold goes down and silver goes up — and the numbers say this bet has made $2.26 for every $1 lost recently.

3.2 EQUITY — NVDA, BAC, JPM, MSFT LONG

Why these are strong picks:

PF 1.82 / Sharpe 3.67 / WR 52.4% — solid risk-adjusted edge with decent sample size (n=271).
NVDA tournament record: +2285% total return in AI tournament paper trading, Sharpe 2.08. It's a momentum darling with institutional support.
BAC/JPM/MSFT: 88–100% tournament WR. Large-cap financials and tech with strong balance sheets benefit from rate-cut expectations.
RSI(2) pullback strategy: The current open picks (COST, CVX, NVDA, WMT, XOM) are all flagged by the Connors RSI(2) mean-reversion pullback strategy — a statistically validated approach for large-cap equities.
Regime context: VIX < 22 (low fear) + yield curve normalizing = supportive for long equity exposure.

Caveat: Policy-clean n=48 is thin. These are shadow-pilot candidates, not production-sized.

ELI5: Big American companies like NVIDIA (computer chips), Bank of America, and Microsoft are showing up as "buy" in our system. NVIDIA has been a superstar in our paper-trading league. The strategy we use looks for stocks that dipped slightly and are likely to bounce back — like buying a tennis ball after it hits the floor.

3.3 ETF — EEM, GLD, IWM

Why ETFs are promising:

PF 1.49 / WR 58.65% / Sharpe 2.70 — highest win rate of any class, but marginal profit factor.
Dual-momentum lab pass: etf_verified_dual_momentum is the ONLY strategy with walk-forward OOS PASS (PF 1.21, n=32). It's the closest to production-ready.
EEM (Emerging Markets): 93% tournament WR — benefits from dollar weakness and EM capital inflows.
IWM (Russell 2000): 75% tournament WR — small-cap value play when rates are expected to fall.
Low concentration: ETF picks come from diverse sources (sector momentum, dual-momentum, cross-asset).

Caveat: PF 1.49 is below the 1.5 ideal threshold. Need more forward data before sizing.

ELI5: ETFs are like baskets of stocks. Our "basket strategy" is the only one that passed the "practice test" (walk-forward). It wins often (58% of the time) but doesn't win by much each time. Think of it as a safe, steady jogger — not a sprinter.

3.4 CRYPTO — SHORT BTC, ETH, SOL (NOT LONG)

Why CRYPTO LONG is toxic and SHORT is the play:

EAGLE-4 finding: CRYPTO LONG has 33% WR vs SHORT 67% WR in tournament data. The directional bias was inverted in production.
Kelly-negative for LONG: PF 1.25 with 44.34% WR gives f = −0.1% — mathematically, you should not bet LONG.
Current picks in funnel are mostly LONG: BTCUSDT, ETHUSDT, SOLUSDT, ADAUSDT, DOGEUSDT all LONG with elite_score=43, failing SMART gate (score<60). This is a systematic long-bias bug.
Exceptions: TRXUSDT LONG (elite_score=67, passed SMART gate) and a few SHORTs (RENDERUSDT, ADAUSDT, BTCUSDT via inverse_ml_enhanced) are the only viable candidates.
Root cause: The ml_crypto_predictor source system emits only LONG signals. It's a one-directional engine in a two-directional market.

ELI5: Our crypto robot only knows how to say "BUY." But the market keeps going down, so "BUY" loses money. We just discovered that "SELL" would have won 67% of the time. It's like having a broken GPS that only tells you to turn left — even when the exit is on the right.

4. Achievements & Tasks Accomplished

Task	Status	Location
EAGLE2 multi-model swarm synthesis (6 EAGLE*.MD files reviewed)	DONE	`reports/EAGLE_SWARM_SYNTHESIS_2026-06-02.md`
ETF dual-momentum backtest template (purged WF, block bootstrap, DSR, Bonferroni)	DONE	`alpha_engine/backtest_etf_dual_momentum.py`
AI brainstorm report (ETF strategy ideas from ollama-cloud-local)	DONE	`reports/ETF_BRAINSTORM_AI_MODELS_2026-06-02.md`
EAGLE-4/5 admissibility + promotion gates shipped	DONE	`alpha_engine/eagle_gates.py`
Weekly real-money-ready filter report (this document)	DONE	`reports/weekly_filter_2026-06-02.md`
LiteLLM proxy tested (ollama-cloud, ollama-cloud-local)	DONE	`http://localhost:4000/v1`
Portfolio sync gap identified (760 funnel picks → 0 portfolio picks)	CRITICAL FINDING	`pf_portfolio_portfolio_mix__aggressive_top5.json`
Git sync with origin/main	DONE	commit `795a82a6b`

ELI5: We did a lot of homework this week: we reviewed all our old reports, built a new test for ETF strategies, asked AI for new ideas, fixed a bug that was flipping crypto directions, and wrote this big report. But we also found a scary bug: we have 760 picks ready to trade, but our trading bot's wallet is empty! The picks aren't getting to where they need to go.

5. The Gap — Why Edge Disappears

Layer	COMMODITY PF	EQUITY PF	ETF PF	CRYPTO PF
HF_STATS (raw recent)	2.26	1.82	1.49	1.25
pf_registry (policy_clean_net)	1.68 (n=4)	0.33 (n=48)	N/A (n=2)	0.95 (n=366)
money_ready_verdict	1.68 (n=4)	0.33 (n=48)	N/A	0.95 (n=363)

Short answer: The policy-clean pipeline is over-filtering. Likely causes:

Disputed resolver tags being treated as losses instead of neutral
Duplicate removal accidentally removing wins
Source-provenance strictness filtering out good picks from non-canonical sources
TIME_EXIT mislabels converting wins to losses
Look-ahead bias removal being too aggressive

ELI5: We have a strict teacher (the "policy-clean" filter) who throws out any homework that has even a tiny smudge. The problem is, the teacher is throwing out A+ papers because of a speck of dust. The good grades exist — they just get deleted before the report card is printed.

6. Short-Term Plan (Next 4 Weeks)

Week	Priority	Action	Owner
1	P0	Fix portfolio sync gap — 760 funnel picks → 0 portfolio picks. Audit gate thresholds and portfolio allocation logic.	Ops
1	P0	Audit CRYPTO resolver — 363 resolved picks with PF=0.95 suggests systematic mislabeling. Re-resolve disputed cohorts.	Data Eng
1–2	P0	Data hygiene sprint — disputed tag purge, source_id + fallback fields, duplicate detection on pf_portfolios.json	Data Eng
2	P1	Shadow-size COMMODITY + EQUITY at 0.5% of capital per pick, paper-trade ETF at ≤0.2%	Portfolio Mgmt
3	P1	ETF paper pilot Day-30 checkpoint — n≥30 forward, MDD<15%, PF≥1.20	Quant Research
4	P1	Re-run policy-clean pipeline with corrected resolver on CRYPTO and EQUITY cohorts	Quant Research

ELI5: In the next month: (1) Fix the broken delivery truck so our lemonades actually get to customers. (2) Fix the notebook so we know which recipes work. (3) Start tiny test batches (0.5% of our money) for our two best recipes. (4) Check if our safest recipe (ETF) passes its 30-day test.

7. Long-Term Plan (12 Weeks)

Week	Milestone
1–2	Data/resolver audit, duplicate purge, source-provenance tagging
3–4	Deploy standardized validation pipeline; pre-register hypotheses
5–6	Run purged-embargoed walk-forward on ETF dual-momentum & crypto VWAP/Bollinger
7	Shadow-size approved sleeves (ETF, Crypto SHORT-only)
8–9	Mutation testing on failed lab sleeves; evaluate inversion candidates
10	Promote any sleeve meeting live PF≥0.5, WR≤60% gate
11–12	Full-size rollout for successful sleeves; update Quant Ops Dashboard

Success definition (Quarterly Review):

At least 2 new capital-ready sleeves with live PF ≥ 0.5 and WR ≤ 60%
Resolver dispute rate < 1% across all live feeds
HHI (concentration) for aggregate book < 0.20
End-to-end validation pipeline latency ≤ 5 minutes per sleeve

ELI5: Over the next 3 months, we want to: clean up our notebook, test 2-3 recipes with real tiny batches, and if they keep making money for 2 months straight, we start selling them for real. Goal: 2 winning recipes, a clean notebook, and no recipe taking up more than 20% of the stand.

8. Risk Disclosures

Portfolio sync gap: 760 active picks are NOT in the aggressive portfolio. Any sizing based on portfolio data is currently operating on stale/empty state.
HF_STATS vs policy-clean divergence: Recent data looks good but policy-clean data is terrible. Do not size based on HF_STATS alone until the bridge is fixed.
COMMODITY n=74: Strong metrics but thin sample. One bad week could flip PF below 1.5.
CRYPTO directional bug: EAGLE-4 flip shipped but not all downstream systems have been updated. Verify side=direction on every CRYPTO pick before execution.
All picks are paper-only until forward n≥30 and live PF stays within ±10% of backtest.

ELI5: Before you spend any real money: (1) Our delivery truck is broken. (2) The recent good numbers might be a lucky streak. (3) Our crypto robot still thinks "left" is the only direction. (4) We need 30 days of real testing before we believe anything.

Report generated by money-maker-readyv2 skill execution | Source: reports/weekly_filter_2026-06-02.md | Commit: 795a82a6b | Not financial advice. Paper-pilot only.