PairsCopilotPublished record

What the full report published

Every pair the model flagged in the full report, listed 30 days after the evening it was published — the same record for every reader. Each row shows how far apart the two stocks had drifted from their usual relationship when it was published, and how long the model expected the gap to take to close.

Published	Pair	Apart vs normal (at publication)	Typical time to close (modeled)
2026-09-16	MS / BAC	9.7% further apart than normal	about 15 trading days
2026-09-14	BAC / COF	5.2% further apart than normal	about 11 trading days
2026-09-10	CSX / NSC	4.5% further apart than normal	about 10 trading days
2026-08-25	MRK / JNJ	7.7% further apart than normal	about 8 trading days
2026-08-20	TJX / ROST	4.9% further apart than normal	about 6 trading days
2026-08-19	L / AFL	5.5% further apart than normal	about 6 trading days
2026-08-10	AFL / ALL	4.0% further apart than normal	about 7 trading days

These percentages measure how far the two stocks have drifted from their usual relationship with each other. They are a distance, not a return: they are not a gain, not a yield, and not a forecast of profit. Times to close are modeled. Modeled, not achieved. Hypothetical results have inherent limitations and do not represent actual trading.

Where are the outcomes?

Not on this page. Results are published on the performance ledger only once the live record has at least 30 closed signals and 6 months of history — so no early streak, good or bad, is shown out of context.

How strict the pipeline is, on Fri, Oct 16

We publish process, not promises. Every pair has to re-qualify on every scan, and is benched the moment it stops passing.

22 pairs

Qualified on Fri, Oct 16

Of 29 active pairs, these passed the stationarity and half-life gates and were measured in the report.

Benched on Fri, Oct 16

Pairs failing a gate publish nothing that night — re-tested on every scan, never carried forward.

173

Suspended right now

Repeated stop-level stress auto-suspends a pair; reactivation is manual, never automatic. 2014 candidates wait in validation.

0 admitted

Placebo control

The same admission bar is run on shuffled placebo pairs; a batch only counts if the placebo set fails. Most recent control: 0 of 29 placebos passed.

The qualified, benched and suspended counts are live from the production database, as of the latest nightly scan; the placebo figure is the most recent recorded control run. Subscribers see how every published signal resolves — losers with the same weight as winners — inside the report.

See them the evening they are published

This page runs 30 days behind. The full report carries every opportunity the evening the model flags it.

Start a 7-day free trial →

Available to US residents only — US-listed equities.