Status (2026-09-07): a verified implementation of an edge that failed. Every pre-registered variant of a 5m/15m pullback entry on the H1 ZTD divergence bias fell short of the criteria committed before it ran. It ships in CAT anyway, as a tracked engine whose every card carries this record, under the same rule ZMS ships under: a record that only shows what worked is not a record.
While an hourly ZTD bias is live, watch the 5-minute chart for a pullback against it to form a confirmed pivot. Place a stop-entry order one tick beyond the pullback's origin, the stop one tick beyond the pullback extreme, the target at the bias's zero-touch level. Cancel if price takes the extreme first or the order sits unfilled for 48 bars; close any position when the bias ends.
Design: entry5m/docs/2026-09-06-entry5m-design.md. Rules: pyakao/src/pyakao/entry5m/.
Indicator (chart parity verified): entry5m/pine/ztd_entry.pine v8.
Five years of tick-built bars at each instrument's measured spread, on six instruments in three groups — usd (EURUSD, GBPUSD), metal (XAUUSD), jpy (EURJPY, GBPJPY, AUDJPY). A variant passes if two of the three groups clear all five sub-criteria (sample size, profit factor above 1.15, profit factor above 1.0 with the single best trade removed, positive expectancy in R, a non-negative second half). The criteria for each variant were committed to the repository before it ran; the links below go to those commitments and to the run records.
Every variant was also run on 15-minute bars as a secondary record. None of those changed a verdict.
| # | Variant | What changed from the one before | Verdict |
|---|---|---|---|
| v1 | Pullback stop | The design as specified: stop beyond the pullback extreme, target at the zero-touch level | FAIL 0 of 3 groups |
| v2 | H1-extreme stop | Stop moved to the H1 divergence extreme | FAIL 0 of 3 |
| v3 | Fixed 3R target | H1-extreme stop, target at three times risk instead of the zero-touch level | FAIL 1 of 3 — the usd group passed on EURUSD alone (n 415, PF 1.21, ex-outlier 1.17); under pre-registration one group is not a pass |
| v4 | Arm-triggered bias, XAUUSD | Bias starts at the H1 extreme-triangle arm rather than the divergence fire; gold only, single-instrument criteria | FAIL PF 0.83, expectancy −0.09 R |
| v5 | Scale-out, XAUUSD | v4 with half off at 1R, stop to breakeven, rest at 3R | FAIL PF 0.81, expectancy −0.09 R |
| v6 | Scale-out, EURUSD + GBPUSD | v3 with half off at 1R and breakeven; both must pass | FAIL neither passed |
| v7 | Opposite-triangle take-profit | v3 plus an exit when an opposite-side H1 triangle prints | FAIL 0 of 3 |
| v8 | Arm bias + opposite-divergence exit | Arm-triggered bias on four symbols, exit on an opposite-side divergence fire | FAIL 0 of 3 |
| v9a | No take-profit, triangle exit | No target at all; exit only at the opposite-side H1 triangle | FAIL 0 of 3 |
| v9b | No take-profit, divergence exit | No target at all; exit only at the opposite-side H1 divergence | FAIL 0 of 3 |
| v10 | H4 bias | v3 with the bias taken from H4 instead of H1 | FAIL 0 of 3 |
Ten ideas, eleven runs, no pass. The one bright spot — EURUSD under v3 — is exactly the kind of result pre-registration exists to keep in proportion: one instrument, one variant, chosen after the fact, is what a search over ten variants is expected to produce by chance.
It says the pullback entry, as specified in any of these ten ways, does not turn the ZTD bias into a tradable edge on these instruments at real spreads. It does not say the bias itself is wrong; ZTD's own record is separate and shorter (see Zero-Touch Divergence — what has been verified). And it does not say a different entry could not work — only that these ten did not, and that any next one starts from a pre-registration, not from this table.
ZTD Entry runs live as a tracked engine: the same tracker, scanner and simulated broker the research ran, reporting its bias, working order and paper position on the instruments members watch. Every card carries this record in its advisory and links here. Each position it reports is graded to its outcome like any other read, so the record keeps building in public rather than stopping at this table.