ForexBestRobots V231 — Awesome Oscillator + Alligator H1 first saucer

Retrospective educational simulation, not a broker statement or a trading promise.
Uses the same five unchanged HistData EURUSD ASCII M1 Bid OHLC archives as the
preceding studies. Fixed EST plus five hours gives UTC; no DST adjustment.
Only exact duplicate minutes are removed. Conflicting duplicates abort.
Gaps, weekends and partial observed H1 candles remain; no synthetic quotes.
2022 initializes indicators. Entries are UTC 2023–2025 only; January 2026 exits.

Requires Python 3, numpy and pandas. In this extracted folder run in order:
python prepare_data.py
python run_research.py
python verify_research.py

Results and validation are written under research/, public data under assets/.
Read fixed-protocol.json, its SHA-256 and source/source-manifest.json first.
AO uses SMA5(median) minus SMA34(median), with exact integer numerators.
The histogram is displayed in pips, not percentages. Green/red means change,
not positive/negative sign. The first 33 AO values are unavailable.
Long saucer: three strictly positive values, lower middle, higher third.
Short: all strictly negative, higher middle, lower third. No zero or ties.
Three completed ordered Alligator candles are required. From the third close
inclusive, only the first matching saucer is considered. Failed price filters
or skipped executions consume that episode. These episode/price/exit choices
are the study protocol, not a full system prescribed by the platform.

The separate reference reaggregates original CSVs, uses floating SMMA and
rolling AO, vectorized patterns, completed episodes and separate M1 replay.
Agreement still uses the same provider. Missing price paths stay unknown.
Source M1 Bid OHLC is not a full executable Bid/Ask tick feed. Source 2023 has
notably fewer observed minutes than 2024–2025. Gaps may hide level crossings.
Adverse stop gaps fill at Open; favorable target gaps are capped at target.
Both levels in one minute: stop first. At timeout use Open only, never later
minute range; opening stop/target gaps precede the time exit.

First long and first short examples are chronological, not outcome selections.
Last chart candle excludes remaining exit-minute range. Because AO is unshifted,
its last chart column is independently recalculated from truncated High/Low
and prior completed medians; it is partial, not a new completed signal.
The first visible AO color uses its true previous value outside the crop.
The table's three signal AO values remain completed pre-entry values.

Cost presets: 0/0, 1.5/0.2 and 3/0.5 pips per round trip/per UTC midnight.
Allowances, not measured spreads, slippage or swaps. No swap income or triple
rollover. Net R subtracts allowance/risk_pips; stopped trades can lose over 1R.
Sum R is not cash-account percentage; no lot/leverage/margin model.
303 trades lose -28.42R before costs, -88.13R base and -148.39R stress.
No optimisation or post-result parameter replacement. Later 2025 is also
retrospective, not a blind out-of-sample or live forward result.

Primary definitions: official MetaTrader 4 Awesome Oscillator / Alligator help;
source specification: official HistData detailed file specification.
