PERPETUAL // AMC + GME CYCLE & FRACTAL ROADMAP

REAL-ONLY LOCK: Historical reference = real regular-session 1-hour bars built from 30-minute provider bars and anchored to the opening bell: 09:30, 10:30, 11:30, 12:30, 13:30, 14:30, 15:30 (last bar 15:30–16:00), all times America/New_York. Pre-market and after-hours prints are excluded. AMC matching is restricted to the real May 18–June 10, 2021 window; the archive extends to Aug 31, 2021 only to obtain real continuation bars. GME searches its real 2020–2025 archive. The model reads completed daily sessions only; an in-progress session is drawn grey and is not fed to the matcher.
ACCURATE DTW ENGINE: The last 60 completed daily log-closes are z-normalized and compared against every real H1 window — every start bar × every integer window length from 45 (GME) / 60 (AMC) to 180 H1 bars. Each window is uniformly rescaled to 60 points in log space (scaled-and-warped matching), z-normalized, then scored with banded DTW: Sakoe-Chiba band ±10%, symmetric-2 step pattern, distance = √(D / 2n) in z-units. LB_Keogh is computed for every window and exact DTW is run in ascending-bound order; since DTW ≥ LB_Keogh, the ranking is provably identical to running full DTW on every window. Only windows with a full 25-day verified continuation are eligible. Top-5 analogs are exact and strictly non-overlapping. The search runs in a background thread.
PROJECTED CANDLES: Anchored at the last completed close. Each day's close = last close × the analog's historical log-return since match end, scaled by the amplitude ratio σ(query)/σ(match) — the transform the z-normalized match implies, without regression shrinkage. High/Low = the mapped highs/lows of the real reference bars whose closes fall inside that day's slice. Open = prior candle's close. r is a shape correlation, not a probability. LEGACY toggles back to the original engine (f0f7fed math, verbatim).
ENSEMBLE: The purple dashed line is the per-day median close of the top-5 independent analogs; the grey dotted lines are their highest and lowest close. A wide envelope means the historical analogs disagree.
SML COHERENT GATE: Same formula as the mcp-x402 Coherent Gate and PRO-LO. Each independent analog votes up / down / flat (move under 1%) at day 25, weighted by its shape correlation r; coherence = |Σ w·d| / Σ w. PASS when the vote is not flat and coherence ≥ 0.6. Walk-forward validation reports how often PASS calls got the direction right.
NO LOOK-AHEAD: A historical window and its 25-day continuation must end before the first session of the query, so a match can never contain the period being projected.
WALK-FORWARD VALIDATION: Replays ~40 past dates. At each date the engine sees only the 60 sessions that had completed by then; its projection is scored against the closes that actually followed. Direction hit rate is tested against a coin flip (exact binomial, sample size reduced for overlapping horizons); average error is compared with a no-change forecast. Results are cached per data set.
DATA INTEGRITY: Freshness compares the last completed candle with the last NYSE session that should be complete. Second-Source Check re-pulls the last 10 sessions from another provider you have a key for and flags any close that differs by more than 0.5%. Input SHA-256 fingerprints the exact candles and reference bars used, and EXPORT RUN RECORD saves the full run (inputs summary, parameters, analogs, projection, validation) for audit.
NOT INVESTMENT ADVICE: This is a historical-analog scenario built from real market data. Past price patterns do not guarantee future prices.