Each name’s share of portfolio variance at the three declared windows, sized by the actual positions — the number a vendor without position sizes cannot compute. An option counts as its delta times spot, never its notional, which is why WDC nets short here: the puts outweigh the shares. Cash is at par and outside the table; the pairs module says which names move together.
core.broker_lots (imported 2026-09-15 17:06 ET, lots derived 2026-09-15 17:07 ET, last trade 2026-09-14) · close_adj from core.metric_daily to 2026-09-22 · deltas from core.option_strikes · positions opened after 2026-09-14 are invisible here
Every pair at 21, 63 and 252 sessions, each with its noise floor of about 1/√(n−3): a correlation from 21 observations carries ±0.24, so two names that truly move independently routinely show |rho| of 0.3 over three weeks. A reading within two floors of zero is marked. At 63d, 3 of 6 pairs are inside it. The regimes module says whether the average held when it mattered.
| Pair | rho @21d ± SE | rho @63d ± SE | rho @252d ± SE |
|---|---|---|---|
| CRSR-MU | +0.28 ±0.24inside the noise floor | +0.21 ±0.13inside the noise floor | +0.21 ±0.06 |
| CRSR-SNDK | -0.07 ±0.24inside the noise floor | +0.13 ±0.13inside the noise floor | +0.16 ±0.06 |
| CRSR-WDC | -0.08 ±0.24inside the noise floor | +0.19 ±0.13inside the noise floor | +0.13 ±0.06 |
The same pairs over their whole shared history, split by the stored regime label (Regime Monitor’s code: trend × volatility). Correlation in a drawdown is the only correlation that matters for a concentration question, and it is not the average one. A cell below 20 sessions is withheld, and a change from calm to stressed within two combined standard errors (each cell’s 1/√(n−3)) is marked as noise. Descriptive: nothing here is graded against what followed.
| Pair | up-calm | up-stressed | down-calm | down-stressed | calm → stressed |
|---|---|---|---|---|---|
| CRSR-MU | +0.28 n=1,122 | +0.14 n=109 | +0.57 n=160 | +0.71 n=114 | +0.43 |
| CRSR-SNDK | +0.18 n=294 | +0.13 n=48 | insufficient data (n=12) | +0.73 n=42 | +0.55 |
| CRSR-WDC | +0.24 n=1,122 | +0.12 n=109 | +0.55 |
Each pair’s rolling 63-session rho ranked among every earlier window of the same length — the pair against itself, never against another pair. Windows overlap 62/63, so the floor is 756 shared sessions (twelve independent windows): with fewer, a percentile pins at the ends and reports the sample, not the pair. Never reported at 252d, where even the four-year pairs have four independent windows.
| Pair | Today @63d | vs its own history | Median | p10 – p90 | Windows |
|---|---|---|---|---|---|
| CRSR-MU | +0.21 | 20.1% | +0.36 | +0.12 – +0.66 | 1,443 from 2020-12-22 (~22 independent) |
| CRSR-SNDK | +0.13 | CRSR and SNDK share 396 sessions since 2025-02-25; 756 (12 independent 63-session windows) are needed before a percentile means anything | — | — | 334 from 2025-05-23 (~5 independent) |
| CRSR-WDC | +0.19 |
The book before and after a hypothetical position, at the same 63d window on both sides — comparing a 63-day before with a 21-day after would attribute the window to the trade. Any of the 39 names with a close_adj series; the amount is delta-adjusted dollars, so a share position is its market value. Adds to a name already held.
?candidate=STX&dollars=20000 to the URL, or use the form above. Without an amount, $10,000 is assumed and said so.?candidate=STX&dollars=20000 · the same exposure rule as the book module
The per-holding driver map DESIGN asks for (XOM to WTI, TSM to USD/TWD) cannot be drawn: no commodity, FX or memory-spot series exists in the warehouse. Measured 2026-09-23, core.macro_observations holds 5 series, all CBOE volatility indices, and core.fred_observations holds 20 - rates, spreads, the Fed balance sheet and the macro prints - with no DCOILWTICO and no DEXTAUS; 0 registered metrics match oil, WTI, TWD, FX or DRAM. XOM is in the universe but not held, and TSM is not in the 39-name universe at all. The strip below is a factor-proxy view inside the stored universe, labelled as such, not a substitute for the map.
| Held | SPY | QQQ | XLK | SOXX | SMH | STX |
|---|---|---|---|---|---|---|
| CRSR | +0.46 ±0.13 | +0.45 ±0.13 | +0.42 ±0.13 | +0.37 ±0.13 | +0.38 ±0.13 | +0.09 ±0.13inside the noise floor |
| MU | +0.42 ±0.13 | +0.67 ±0.13 | +0.76 ±0.13 | +0.85 |
This page describes: a correlation, a risk share and a regime-conditioned rise are statements about past sessions, and none of them has been graded against what followed. When the Signal Board (DESIGN.md page 20) can grade them, its result belongs here. Personal account rows and greeks from a personal-use key: no export, no CSV (ADR-0014).
| 2 |
| −$5,205 |
| 8.2% |
| −4.4% |
| −6.8% |
| −5.5% |
Effective bets is the diversification ratio squared — the weighted-average volatility over the book’s own, squared — capped at the number of names. It is not DESIGN’s literal “1/HHI on risk contributions”, which cannot see correlation (the engine measured 2.0 bets for a pair at rho 0.99) and is kept beside it as the by-name spread. A window is refused below max(20, 3 × names) observations, because a rank-deficient covariance reports noise as diversification: this 4-name book needs 20, a 39-name one would need 117 and compute only at 252d.
Coverage: 6 of 6 security lots priced (100.0%; floor 90%). Cash: 12 sweep lots totalling $2.58, at par, not in the table.
| MU-SNDK | +0.85 ±0.24 | +0.90 ±0.13 | +0.77 ±0.06 |
| MU-WDC | +0.75 ±0.24 | +0.75 ±0.13 | +0.71 ±0.06 |
| SNDK-WDC | +0.74 ±0.24 | +0.78 ±0.13 | +0.73 ±0.06 |
Not a dendrogram: the engine ships connected components at a stated |rho| threshold (single linkage), because the question is which names are effectively one bet, and a threshold is a number a reader can argue with where a tree's cut height is not.
close_adj simple returns aligned on the sessions all 4 names share, to 2026-09-22 · n=63 at 63d
| +0.66 n=114 |
| +0.42 |
| MU-SNDK | +0.71 n=294 | +0.86 n=48 | insufficient data (n=12) | +0.83 n=42 | +0.11inside the noise floor |
| MU-WDC | +0.65 n=1,834 | +0.75 n=291 | +0.79 n=168 | +0.86 n=233 | +0.21 |
| SNDK-WDC | +0.67 n=294 | +0.74 n=48 | insufficient data (n=12) | +0.83 n=42 | +0.15inside the noise floor |
regime_code (entity US, app/core/regime.py) joined to close_adj returns on every shared session · S3.2 D
| +0.33 |
| +0.16 – +0.62 |
| 1,443 from 2020-12-22 (~22 independent) |
| MU-SNDK | +0.90 | MU and SNDK share 396 sessions since 2025-02-25; 756 (12 independent 63-session windows) are needed before a percentile means anything | — | — | 334 from 2025-05-23 (~5 independent) |
| MU-WDC | +0.75 | 64.9% | +0.71 | +0.51 – +0.85 | 2,464 from 2016-12-01 (~39 independent) |
| SNDK-WDC | +0.78 | SNDK and WDC share 396 sessions since 2025-02-25; 756 (12 independent 63-session windows) are needed before a percentile means anything | — | — | 334 from 2025-05-23 (~5 independent) |
rolling corr() over close_adj returns on each pair's shared sessions to 2026-09-22; midrank like core.metric_context()
| +0.82 ±0.13 |
| +0.71 ±0.13 |
| SNDK | +0.39 ±0.13 | +0.66 ±0.13 | +0.73 ±0.13 | +0.81 ±0.13 | +0.78 ±0.13 | +0.75 ±0.13 |
| WDC | +0.37 ±0.13 | +0.63 ±0.13 | +0.68 ±0.13 | +0.75 ±0.13 | +0.71 ±0.13 | +0.88 ±0.13 |
close_adj returns aligned on the sessions all 10 names share · 63d · n=63 · SE ±0.13