Sleeve quality decomposition
Sleeve quality decomposition is one of 21 measurements this engine publishes in full. It is regenerated from a real run rather than transcribed, and it is shown here with its own claim boundary so the number and its limits arrive together.
The measurement
- Largest cost burden
- AlphaForge
- S bar measured
- 0.636823
- Schema
- canli.alphac-sleeve-quality-decomposition.v1
- Weakest sleeve
- AlphaTrend
Largest recoverable component
- Sharpe points
- 0.274529
- Sleeve
- AlphaForge
- Turnover ann
- 33.221809
What recovering half would be worth
- 0.137265
- Sharpe points on that sleeve, which is roughly a third of that on s_bar across three sleeves.
Why. Transaction cost is the only component here that better execution could give back. Signal is construction and funding is a market fact; neither is recovered by executing better. This sleeve carries the largest cost burden by a wide margin, and it is the turnover that drives it.
S bar basis note. s_bar_measured here is the mean of these three FULL-HISTORY backtest Sharpes, each on its own window. It is NOT the 0.469 measured across the four live sleeves on their 1,061-day common window in artifacts/analysis/book_without_alphavintage, and the two must not be compared. Different sleeves, different windows.
| Asset class | Commission bps realised | Commission dollars MEASURED | Commission sharpe points MEASURED | Funding dollars MEASURED | Funding sharpe points MEASURED | Half spread bps used | Market impact | Net sharpe published | Notional traded total | Realised vol ann | Residual signal sharpe | Sleeve | Spread dollars MODELLED | Spread sharpe points MODELLED | Turnover ann | Walkforward | Years |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| equity | 1.000000 | 86.585143 | 0.003319 | 0 | 0 | 5 | NOT SEPARABLE — see claim boundary | 0.907084 | 865851.427556 | 0.130716 | 0.926997 | AlphaMax | 432.925714 | 0.016594 | 4.111369 | k30_dn_63 | 1.995893 |
| crypto | 5.000000 | 7372.717709 | 0.144489 | 19500.019816 | 0.382158 | 4.500000 | NOT SEPARABLE — see claim boundary | 0.676610 | 14745435.418976 | 0.118332 | 0.568982 | AlphaForge | 6635.445939 | 0.130040 | 33.221809 | crypto_carry_wk | 4.312115 |
| equity | 1.000000 | 1566.954815 | 0.020755 | 0 | 0 | 5 | NOT SEPARABLE — see claim boundary | 0.326774 | 15669548.154843 | 0.053132 | 0.451304 | AlphaTrend | 7834.774077 | 0.103775 | 9.693324 | mf_live_fwd | 14.209446 |
The finding. AlphaTrend is the weakest at 0.327 net Sharpe. Its total modelled-plus-measured cost burden is 0.125 Sharpe points, so recovering ALL of its transaction cost would still leave it at 0.451 before spread and impact are re-charged. The drag is construction, not execution.
What is measured and what is not
MEASURED
- commission — summed from the `fee` column of every fill
- funding — the walk-forward's own funding_net
MODELLED
- half-spread and latency — turnover x notional x the DECLARED bps, because the spread is applied to the FILL PRICE and never recorded as a line item
NOT SEPARABLE
- market impact — same reason as spread, and it depends on ADV per name per day, which the published fills do not carry. It is inside the residual.
The trap. `fees_paid` in every summary.txt is COMMISSION ONLY. Reading it as total transaction cost understates the cost component badly: measured commission is about 1bp against a declared 3bp equity half-spread plus a 2bp latency add-on plus impact. This decomposition reports a RESIDUAL rather than pretending to a full attribution.
What this does not settle. Whether the modelled spread is the spread actually paid. That is B5's question and it needs the live fills, which the forward record is currently too short to answer.
Check it yourself
Every figure above is read from /glassbox/sleeve_quality_decomposition.json, the
artifact the engine wrote. Nothing on this page is typed by hand: the page is generated from that
file, so a figure that moves in the artifact moves here and a figure that is not in the artifact
cannot appear here at all.