Always-In Ledger
The always-in rule — the same code behind the chart's Day Tape rail chip, drift-checked at 100% verdict agreement — replayed bar-by-bar over 186,135 five-minute sessions (11,649,937verdicts). This is a ledger of the rule's behavior: how often it commits to a side, how quickly that commitment fades, and what price did afterward. It is not a strategy and there is no P&L here.
97% of flips are off their side within the hour — the median commitment lasts 1 bar (~5 minutes) — and a flip alone carries no directional edge: 46–49% of them went the flip's way, at every horizon.
Across 801,990 flips, the follow-through distribution is centered on zero (medians ~0.00%, middle halves roughly ±0.2%). 19% see the opposite side print later the same session. In price-action terms this is the expected shape — always-in is a state read, not a signal — but the ledger says it plainly: on the broad corpus, the flip by itself predicts nothing.
Where the verdict lives
LONG 6.9%
UNCLEAR 86.7%
SHORT 6.3%
The honest headline: the rule is UNCLEAR 86.7% of the time. Always-in does not mean always-committed — most bars, by design, it declines to pick a side.
Flips per session
| 0 flips | 1 flips | 2 flips | 3+ flips |
|---|
| 3.8% | 7.7% | 12.4% | 76.2% |
After a flip
Moves are close-to-close from the flip bar, signed in the flip's direction(positive = the flip's way), shown as median and middle half (25th–75th percentile). “Faded” = the verdict left the side within 1h. A direct opposite print within the hour is structurally near-impossible (the 12-of-15 close majority needs ≥12 bars to invert), so the reversal stat is same-session.
flip to LONG
418,662 flips
faded ≤1h 97% · median 1 bars on side (p75 2) · held 1h+ 3% · on side at close 14% · opposite later that session 18%
| horizon | median | middle half | went its way |
|---|
| +30m | +0.00% | -0.15% … +0.15% | 47% |
| +1h | +0.00% | -0.20% … +0.20% | 48% |
| close | +0.00% | -0.30% … +0.31% | 49% |
flip to SHORT
383,328 flips
faded ≤1h 97% · median 1 bars on side (p75 2) · held 1h+ 3% · on side at close 13% · opposite later that session 21%
| horizon | median | middle half | went its way |
|---|
| +30m | +0.00% | -0.18% … +0.16% | 46% |
| +1h | -0.01% | -0.25% … +0.21% | 46% |
| close | -0.01% | -0.37% … +0.33% | 46% |
By asset class
| asset class | sessions | flips / session | opposite later that session |
|---|
| crypto | 28,706 | 3.96 | 20% |
| equity | 135,862 | 4.46 | 20% |
| fx | 21,567 | 3.79 | 15% |
Generated offline by
scripts/ml/gen_alwaysin_ledger.ts from the same corpus behind
Look-Alikes and
day-type base rates. The live rule runs on the
chart's Day Tape rail, where every verdict cites the conditions that fired and how long it has held.