Strategy reference

Wolfe-wave observations

What this helps you see

The classical five-pivot geometry and EPA projection provide a structure to investigate. Patearn combines detected structures with the founder’s research into symmetry, Fibonacci confluence and possible point-five zones.

What makes it different: inspect more than one structure, distinguish confirmed observations from developing candidates, walk the dated pivots, and compare the machine’s observation with your own drawing. A projected zone or line is not a promised destination or a profitable trade.

Historical findings are conditional on the tested detector and population; they must not be read as validation of every later geometry revision.

Explore Wolfe observations.

About this reference

This reference explains the reading, its evidence and its limits. Implementation and calibration notes are retained in the internal research archive. Published test definitions and unfavourable results remain available below.

4. Status, validation & honesty fence

Evidence update — 2026-09-10. The July selection studies below describe their

tested rule versions and populations. They do not validate every later detector

revision. The later 2026-08-27 post-§A2 outcome readout recorded 76,592 rows

walked, **4,586 resolved cases, expectancy −0.1348 R, win fraction 0.2150 and

median −1 R. This is an aggregate historical outcome readout, not a new

out-of-sample profitability result**. It supersedes the older pre-§A2 aggregate

−0.0797 R / 14,626 resolved figures, not every separate July selection study.

The full dated comparison remains in the strategy ledger.

The live scanner and the separate point-in-time research engine also have

different lifecycle/timeframe capabilities; weekly research must not be presented

as a live weekly/monthly scanner. No profitable trading claim follows from any

chart geometry or score.

DESCRIPTIVE-ONLY. The lens never trades or ranks as a book. This is a binding fence, not a caveat — misrepresenting it is a blocking error. (The §B *scoring* is now ratified and live — but a validated *score* is not a validated *strategy*; the descriptive-only status is about the trade, not the math.)

• The edge is SELECTION, not craft. The raw Wolfe trade has no market-neutral edge (median −2% net; a pure tail game). The only validated edge is the winner-profile selection (reachable EPA + strong point-1 + not-narrowest zone) — a true OOS derivation (fit 2004-14 / test 2015-26) + beta-control + placebo. Re-validated 2026-07-11 under the rebalance (§8): the fit re-derives the identical rule D≤1 · p1≥2 · F≤2 (the F 0–4 widening is neutral to the filter), and the placebo gap stays negative everywhere → the edge lives in *which name / direction / when*, not in the entry/stop/target trade-craft. But post the fractal gate the numbers are softer than the original: on the survivorship-aware primary universe the current verdict is IN-SAMPLE-ONLY (bear-driven); it clears the OOS bar only on the nifty500 sensitivity.
• Split by side: BULL = a regime-ROBUST long selection edge (test 2015-26 medNet +4.4%, residual α +5.07, CI-excludes-0; positive even when the market falls). BEAR = tail-only — regime-dependent (loses when the tape isn't already falling), decaying into 2021-26, and it now fails the primary OOS medNet ≥ 0 bar (−1.0% inclusive; only marginally positive on the survivorship-biased nifty500). On the surfaces, BULL rows read ✓ edge, BEAR rows read ⚠ tail.
• Even the winner profile stays a SCANNER — the thin blended median (~+1%, BULL-carried) can't carry costs, slippage and discretionary error mechanically. It sharpens the eye at the point-5 zone; it does not pull the trigger. Ledger: the strategy ledger (Wolfe row) — same descriptive/selection class as the PEAD and harmonic lenses.
• The "Open trades — remaining ROI" surface is descriptive-only too, with its own fences. It lists every *OPEN* winner-profile setup (point 5 printed, EPA 1-4 not yet touched) within ~1 trading year, ranked by remaining ROI from the current price (run%/risk%/R:R). The validated +edge is measured only on FRESH ≤15d entries — older open trades carry run left but no validated entry-edge, so they are badged ("open · judge the run"), never presented as the edge. The EPA is aggressive on old waves (the 1-4 line extended to today) and is triple-fenced: a hard 1-year population cap + coherence floor (older/incoherent waves are disclosed-and-held-out, never silently hidden — they still draw on the chart), ~-muting of far-extrapolated values, and a conservative run→T1 floor shown beside it. Not a buy/sell signal.

The revert + fractal gate (2026-07-10, , the desk-directed). After "I am really disappointed that the fractal has been ignored," the code was REVERTED to the baseline and the 2/3/4 fractal rule was promoted from a soft score component to a MANDATORY detection gate (his verbatim: "must, minimum 2 fractals; without a fractal do not consider"). Measured: 32% of surfaced waves violated it (e.g. TCS BULL 2005-01-07, degrees [0,0,0,0], still CONFIRMED before the gate). The entire – layer was REMOVED from code — STR/LND split, structure-watch, attention/recency ranking, §B2 withhold, lifecycle queues/progress chips/CLOSED chips, EPA cache wiring — to be re-applied methodically *with* him. §A geometry and the §B component math were left untouched (and the latter then ratified by ).

§B resolved (was the headline freeze). The A/C/F/G/H/I recode that used to gate all Wolfe work is decided and live — see the callout at the top. There is no longer a scoring freeze; the remaining open threads (§9) are point-4-strength and the – re-apply, none of which block §B.

6. Data & provenance

• Input: bhavcopy_rows (NSE daily bhav copy — trade_date, open, high, low, close, volume) per symbol, corporate-action-adjusted via the production adjuster (src.automation.adjust). Universe from nse_equity_list, default Nifty 500 daily; an inclusive (delisted-aware, survivorship-honest) universe is available on the scanner.
• Point-in-time / no-look-ahead (mandatory, the wolfe rules notes §C). Every scan takes an *as-of date*; the result as-of *t* is **byte-identical whether or not bars after *t* exist**. Naturally PIT-honest: a fractal-N pivot confirms only N bars after its candle (points 1-4 carry their real delay); point 5 = the live candle extreme (seen immediately — exactly why it has no fractal gate). This one capability powers both ad-hoc historical review and the §C backtest.