The CFTC publishes what share of a market the four largest traders hold — long side and short side, separately. Almost no retail COT tool shows it. COTVault reads it, and the central finding contradicts what everyone teaches about it.
The usual interpretation of concentration — the one the CFTC's own framing invites — is fragility: "concentration rising while open interest falls means few holders are left and the crowd is exiting, so a violent unwind is coming."
Tested across 12 markets and 26 years: range expanded within 2 reports 21.9% of the time against a 21.2% rate it would have managed anyway. A +0.7 point edge on 7 of 12 markets. The related air-pocket, squeeze-fuel, distributing and covering readings all landed inside noise as well. None of them appear anywhere in COTVault.
Compare the two regimes that both have open interest falling — so the only difference between them is the direction of concentration, and each is a clean control for the other:
| State | Participation held, 8 reports on | Markets |
|---|---|---|
| Concentration rising | 57.1% | — |
| Concentration falling | 51.8% | — |
| Lift | +5.4 | 12/12 |
Rising concentration is when the market holds together, not when it breaks. A market gathering into fewer, larger hands is the stable state. Every one of the twelve markets agrees.
The four-regime grid was replaced by two honest states: CONSOLIDATING and DRAINING. The other two regimes measured nothing and were removed rather than kept for symmetry.
| Still concentrated (≥80th pct) after | Accuracy | Free rate | Lift | Markets |
|---|---|---|---|---|
| 2 reports | 65.3% | 34.6% | +30.7 | 12/12 |
| 4 reports | 51.8% | 36.7% | +15.1 | 12/12 |
| 8 reports | 44.0% | 33.9% | +10.2 | 11/12 |
n=783 episodes. The free rate comes from the adjacent 76–80 band, so "high things stay high" is already inside the baseline and is not being claimed as insight. Median episode 7 reports.
Note the shape of the accuracy column: 51.8% at four reports looks unimpressive on its own. Read it against the 36.7% free rate and it is a +15.1 point edge with unanimous market agreement. This is exactly why every figure in this suite is published next to its baseline — a bare accuracy number is marketing.
Teal shading on the price chart marks where two independent things agree:
P3A is COTVault internal math — a three-period average of the pivot, computed on weekly bars. Its lower band sits roughly where the last three weeks of trade found support, so a close beneath it means price has left its recent value area to the downside. All three support tools are computed on weekly data whatever chart period you are on.
Measured as a barrier race: from the zone bar, does price reach +1 ATR before −1 ATR within 13 reports?
| Accuracy | Free rate | Lift | Markets | n | |
|---|---|---|---|---|---|
| Watch zone, ±1 ATR | 59.4% | 52.8% | +6.7 | 10/12 | 843 |
| In-sample (pre-2010) | — | — | +7.2 | — | — |
| Out-of-sample (2010–26) | — | — | +6.3 | — | — |
+7.2 in-sample versus +6.3 out-of-sample. Almost no decay across a sixteen-year holdout. Compare COT Classic's zone, which falls from +14.0 to +4.6 — real, but far more era-dependent. If you trade one zone from this pair, this is the more dependable one, and it has better breadth too: 10 of 12 markets against Classic's 8.
The individual components were measured separately before being combined: compressed range +8.9 (out-of-sample +8.1), EMA14 pullback +8.9, P3A band +6.2 across 9 of 12 markets. The combined OR-zone scores lower than its best component — that is expected and correct. Widening a condition dilutes it. The combination was measured on its own rather than assumed from the parts.
The zone appears about 1.8 times per market per year — roughly twice as often as Classic's.
Long-side whales at resistance, going short: 48.7% against a 47.2% base rate. A +1.4 lift on 7 of 12 markets, with a bootstrap CI spanning zero. It ships as a toggle that is off by default.
The mechanism is the same one that appears in COT Classic and COT Divergence: a crowded short is forced — margin and unlimited loss compel covering. A crowded long can sit indefinitely. Only one side of the book has something that makes the crowd move. Three separate indicators, built on three different series, found the same asymmetry independently.
There is no CFTC concentration series for a synthetic cross, so on GBPNZD or CADJPY COTVault reads the base currency's leg only — the pound, the Canadian dollar — and knows nothing about the other side of the pair. The panel label says so on the chart.
This is not theoretical. Measured on CADJPY, the watch zone resolved 39% of 33 occurrences against the same 52.8% base rate that gives 59.4% on the tested twelve. That is well below baseline — worse than no filter at all. Treat Crowding as a futures and USD-pair tool. On a cross, read the concentration for context and ignore the zone.
59.4% is a barrier-race rate, not a profit figure. It says nothing about your entry, stop, size or costs. And it is a 6.7-point edge, not a 59-point one — the baseline does most of that work.
12 markets, CFTC Legacy futures-only, joined to weekly OHLC. The CFTC series reaches back to 1986, but the
measured sample is bounded by weekly price history: 2000–2026 for ten markets, 1992 for the US Dollar Index,
2003 for NZD — 16,284 market-weeks in total. Concentration is the
Concentration Net LT 4 TDR series, long and short, percentile-ranked over 156 weekly reports —
the same construction COTVault uses live, so the backtest and the dashboard cannot drift apart.
Every figure is published as accuracy against a base rate. Persistence uses a matched adjacent-band control. The regime comparison holds open-interest direction constant so that concentration is the only variable — which is what exposed the sign error in the original FRAGILE claim: measured against all bars it appeared to have a +8.5 lift, but that trigger is defined by falling open interest, so the trigger was leaking into the resolution. Measured forward from the trigger bar, it inverts.
Zone guards: per-market breadth, market-resampled bootstrap CI, replication at 1.5 and 2 ATR, and a walk-forward split at 2010.