Methodology / Backtest

Compass Backtest — v1.0 Test Period

Compass v1.0 · window 2022-10-31 → 2026-09-18 (48 months) · universe size 50 stocks · forward horizon 21 days · 2400 scored month-ticker observations.

Survivorship

This backtest uses today’s companies only. Companies that were delisted, acquired or went bankrupt are missing, which flatters results. We judge the Compass on how it ranked stocks against each other, not on how much they returned.

Avg rank correlation

0.045

Spearman, Compass score vs fwd return · 30/48 months positive

Decile spread (top−bottom)

+1.85%/mo

Top 2.09% · bottom 0.24%

Monotonicity

56%

Adjacent deciles rising with the score

Skill gate

NOT YET

Threshold rank corr ≥ 0.02 & monotone stance bands

Hit rate by stance (beat the monthly cross-sectional median)

Constructive

48.7%

n = 1039

Neutral

47.4%

n = 1355

Cautious

66.7%

n = 6

Stance bands are scored on whether each stock beat the median return of its month (a rank-based, sector-neutral measure, not an absolute return). Monotone bands (Constructive > Neutral > Cautious) are required to pass the skill gate; an under-populated Cautious band fails the gate until more observations accrue.

Average forward return by Compass decile

0.2
1
0.1
2
0.8
3
1.3
4
1.4
5
0.8
6
1.8
7
0.8
8
0.1
9
2.1
10

Decile 1 = lowest Compass score · Decile 10 = highest

How this was built

Fundamentals are point-in-time annual SEC EDGAR filings (filed-date filtered); the Macro layer is omitted (historical earnings-catalyst dates unavailable) and its weight redistributes pro-rata.

Rank correlation is the average monthly Spearman correlation between the Compass score and the following month's return across the universe. The decile spread is the difference in average forward return between the top and bottom score decile. Hit rate is the fraction of each stance beating the monthly median.

The skill gate has not passed: the mapping is not yet demonstrably better than chance (monotone stance bands not established). The Aether Outlook remains in calibration. No probability is published until the gate clears.

Last run 10/14/2026, 8:56:38 AM. Point-in-time fundamentals from SEC EDGAR companyfacts, filed-date filtered. This is research methodology, not investment advice.