Historical AAPL data

The normalized AAPL-2015-2020.csv contains 1,327 daily raw OHLCV bars, 2015-02-17 through 2020-05-22. The original 506 Plotly rows are preserved exactly; 821 later rows extend the experiment. The loader validates ascending unique dates and OHLC bounds.

Sources retained beside the normalized file:

• Plotly original, retrieved 2026-09-14.
• Extended historical CSV, retrieved 2026-09-15; 1,258 raw rows, 2015-05-27 through 2020-05-22.
• The 437 overlapping closing prices agree within $0.005007 (rounding). All split factors in the extension are 1. Only raw open/high/low/close/volume fields enter the replay, never adjusted prices or future labels.

The original 222 rows through 2015-12-31 are unchanged. Training remains 16 matched pairs of 24-day native reward trials (768 decisions; 180 distinct 2015 action dates). Both sides of a pair start with identical cash and dates. Reward is net portfolio log return after modeled costs and dividends; cash-only has zero reward. There are no training action labels.

The frozen historical test now contains 1,105 decisions, 2016-01-04 through 2020-05-22, in 18 automatic chapters: seventeen 63-day chapters and a final 34-day chapter. Cash, shares, dividend entitlements, random comparison accounts, and learned weights carry across boundaries. Development has inspected these dates; neither replaying them nor this single run establishes predictive skill.

A decision uses the previously observed daily close and earlier prices, current portfolio state, and time since filling an order. It executes at the next recorded open. The next candle's close/high/low/volume is revealed afterward, not fed into the decision. Previous observed volume controls modeled liquidity. Intraday ticks are not invented.

Dividend amounts, ex-dates and payable dates: Apple investor relations. Entitlement accrues before an ex-date opening transaction, is included in equity as a receivable, and becomes cash on payment. The extended schedule includes thirteen additional 2017–2020 payments.

The $1,000 paper account is long-only, whole-share, without leverage. Orders request about 10% of initial cash, with an affordable one-share minimum; new buys respect a 25% equity allocation cap. The passive comparison buys once at that cap. Costs: $0.005/share commission with $0.35 minimum, $0.005 assumed half-spread, 2 basis points adverse impact plus participation adjustment. These are assumptions, not historical bid/ask records. Taxes, interest, settlement restrictions and auction depth are omitted.

Default 4× playback takes about 47 minutes for the test; 12× takes about 16 minutes, plus training and transition time. Actual speed depends on native GPU performance. A separate automated verification temporarily lowers the bar timer while retaining every native decision; it is recorded as a test override, not a user playback speed.

The result screen is computed from the final ledger, with net return, costs, drawdown, ending holdings and passive comparison. It does not force trading, profit or loss.