Track record
The consensus mirror's backtest shows +1.8 pts/quarter net vs the S&P 500 over 47 complete quarters (2014–2026), directional, NOT statistically significant (t≈1.4; 95% CI includes zero).
The consensus mirror's backtest shows +1.8 pts/quarter net vs the S&P 500 over 47 complete quarters (2014–2026), directional, NOT statistically significant (t≈1.4; 95% CI includes zero).
Risk-adjusted performance
The edge above is a raw return number. Professional allocators judge a strategy on a different question: how much risk did you take to earn it? A book that returns a lot by swinging violently is worse than one that returns a little, smoothly. So here is the whole risk panel over 47 quarters, net of cost, next to the S&P 500, with nothing hidden.
| Metric | Prior Moves | S&P 500 | What it means |
|---|---|---|---|
| Sharpe ratio | 0.69 | 0.67 | return per unit of total risk. Higher is better. This is the number allocators ask for first. |
| Sortino ratio | 0.72 | 0.55 | return per unit of downside risk only. Ignores upside volatility, which you do not mind. |
| Information ratio | 0.42 | — | active return over the S&P divided by how much we deviate from it. The direct 'beat the index at controlled risk' score. |
| Annualized return | +21.6% | +14.6% | the raw number. Higher here, but read it next to volatility below. |
| Annualized volatility | +25.3% | +15.4% | how much the return bounces around. Lower is calmer. Ours is higher — the extra return comes partly from taking more risk. |
| Beta to S&P | 1.26 | 1.00 | how much we move with the market. 1.26 means we are a slightly amplified version of the index, not market-neutral. |
| Max drawdown | -43.4% | -23.9% | the worst peak-to-trough fall. Ours is deeper than the market's — the honest cost of the higher return. |
| Calmar ratio | 0.50 | 0.61 | annual return divided by max drawdown. Punishes deep crashes. |
| Ulcer index | 15.0 | 6.9 | how deep and how long you sit underwater. Lower is less painful to hold. |
| Gain-to-pain | 2.40 | 2.18 | total gains divided by total losses. Above 1 means gains outweigh pain. |
| Return skew | 0.84 | -0.97 | the shape of the tail. Positive (ours) means a right tail of big wins; the market's is negative, a left tail of crashes. |
| Quarters beating S&P | +55% | — | how often the basket outran the index. Just over half — consistent with a small, uncertain edge. |
Prediction receipts — called it
The board makes a specific, dated call: which fund adds which name next. Below are the model’s own out-of-sample calls (predicted buy probability at or above 60%), each scored against the actual next filing. Hits and misses both shown. Because JP and UK positions re-file within days, these calls are checkable almost immediately.
JP · 大量保有 (~5 business days)
84 confident calls · hit-rate 79% vs a 31% base rate · out-of-sample AUC 0.81
| Call date | Investor | Company | P(add) | Next filing |
|---|---|---|---|---|
| 2026-06-29 | 3d_jp | Ｊ. フロント リテイリング株式会社 | 66% | confirmed ✓ |
| 2026-06-15 | sparx | 前田工繊株式会社 | 65% | confirmed ✓ |
| 2026-05-29 | simplex | 日本航空電子工業株式会社 | 64% | confirmed ✓ |
| 2026-05-12 | oasis_jp | イオンフィナンシャルサービス株式会社 | 64% | confirmed ✓ |
| 2026-03-31 | avi_jp | 株式会社アイネス | 65% | confirmed ✓ |
| 2026-03-31 | sparx | 株式会社ＬＩＴＡＬＩＣＯ | 70% | confirmed ✓ |
| 2026-03-31 | avi_jp | フォスター電機株式会社 | 69% | confirmed ✓ |
| 2026-03-31 | 3d_jp | 株式会社西武ホールディングス | 80% | confirmed ✓ |
| 2026-03-16 | strategic_cap | 株式会社Ａ＆Ｄホロンホールディングス | 66% | confirmed ✓ |
| 2026-03-05 | strategic_cap | 株式会社Ａ＆Ｄホロンホールディングス | 68% | confirmed ✓ |
| 2026-02-27 | simplex | 三菱ロジスネクスト株式会社 | 69% | confirmed ✓ |
| 2026-02-10 | strategic_cap | 株式会社Ａ＆Ｄホロンホールディングス | 68% | confirmed ✓ |
UK · FCA TR-1 (~2 trading days)
1 confident calls · hit-rate 0% vs a 15% base rate · out-of-sample AUC 0.78 (thin sample — still accumulating)
| Call date | Investor | Company | P(add) | Next filing |
|---|---|---|---|---|
| 2026-03-19 | cevian | SMITH & NEPHEW PLC | 80% | missed ✗ |
Live calls made from today forward are logged with their filing-due date and confirmed as each filing lands (332 calls currently awaiting their next filing). Receipts are the model's own out-of-sample calls (P(add)>=0.60), dated and scored against the actual next filing. Hits and misses both shown. JP/UK confirm in days, so a call is checkable almost immediately. Research, not advice.
Cryptographically timestamped
Anyone can backdate a self-reported track record. We don’t. Each day’s confident calls are frozen into one file, hashed, and timestamped on the Bitcoin blockchain via OpenTimestamps before the filings land. On 2026-07-18 that was 332 per-investor calls plus the US Q2026-06-30 board.
Verify: download commit_2026-07-18.json + commit_2026-07-18.json.ots from the public repo and run ots verify commit_2026-07-18.json.ots. The timestamp proves the calls existed on 2026-07-18 and were not edited after the filings. Self-reported receipts can be backdated; this one cannot.
Corrections
What we got wrong and fixed. We publish our mistakes because a research service you can trust is one that shows its corrections, not one that never admits them.
2026-07-03 · Demoted the Europe predicted board
We briefly promoted a Europe (AFM) predicted board at a claimed walk-forward AUC of 0.77.
An adversarial code review flagged that the fund-curation used full-sample filing counts, so information from the future leaked into earlier test folds. A point-in-time re-test (eligibility decided using only filings dated before each window) collapsed the honest AUC to 0.65, below our gate, on just eight live positions.
Europe is back to a disclosed-only board. It will auto-promote only if it clears the honest point-in-time gate on richer data.
2026-07-03 · Corrected the UK board's headline number
The UK predicted board first claimed an out-of-sample AUC of about 0.85.
Same full-sample curation issue. Unlike Europe, the UK lift survived the point-in-time re-test — but at a slightly lower, honest figure.
The UK board now states the point-in-time-validated AUC of about 0.83, and the curation caveat (the exact filing-count band was tuned) is stated on the page.