365 days.
Every rule on the same tape.

How this year was made

The example is fully generated in advance. Switching controls loads another completed comparison; it does not change your bots, spend credits, make a payment, or call an AI service.

1. Synthetic BTC-USD path, 1 January–31 December 2025 UTC. The dates organize an invented 365-day example; no 2025 market history, candles, or headlines were downloaded.
2. Seed 73129 is fixed before the run. Twelve predefined regimes change daily drift and volatility; Gaussian innovations and six-hour shocks generate the path. News bias/heat use an independent seeded stream every 15 minutes and do not look at future returns.
3. All 12,614,400 quote samples are evaluated at the live desk’s nominal 2.5-second cadence. Two aligned windows run on the same price path: 105,120 five-minute windows and 35,040 fifteen-minute windows. Ending windows settle before the next boundary quote is ingested.
4. The replay calls the app’s actual strategy decision function, shared implied-price model, and rounded payout function. The last 180 prices and six-tick momentum match the continuous browser session’s definitions. It does not replace the simplified rules with textbook indicators.
5. Every strategy/window/cost case starts in its own $5,000 account, uses a fixed $5 stake per phase, and receives no capital top-ups. No plan limit is applied to this read-only educational comparison. It is not one account running 100 bots, and it is not a combined portfolio.
6. A phase fires at most once per window. Probe + Scale can spend $10 across two $5 entries. Full stake and any fee are deducted at entry; winning payout is stake divided by binary price, rounded to cents; losing payout is zero. Up wins a tie of final window average versus open.
7. App model uses zero fees and zero slippage, matching the live paper ledger. Cost stress separately simulates a 20-bp stake fee (1¢ per $5 ticket) and a 25-bp absolute adverse entry-price shift (0.25¢), capped below $1. Insufficient cash skips a ticket; no borrowing, liquidation, or negative balance is modeled.
8. The report includes all entries: 99 deterministic rules evaluated and Grok Director explicitly excluded from performance because it needs genuine user-triggered model responses. The four Maker variants have zero entries because Up plus Down equals exactly $1. No artificial underpriced books are inserted to manufacture trades.
9. Metrics use the full simulated ledger. Daily equity is sampled after all UTC day-end settlements; maximum drawdown is checked after every window settlement. Monthly P&L sums net ticket outcomes. Example trades are the first in each active month plus early distinct win/loss and direction examples, not hand-picked best trades.
10. Idealized continuous operation: one coherent composite feed, no missing ticks, browser suspension, duplicate tabs, execution delay, partial fills, venue restrictions, taxes, funding, market impact, or subscription charges. The current real app needs Desk open and does not reconstruct missed windows. The modeled book is not executable or calibrated to real probabilities.
11. A profitable synthetic result only demonstrates this model on these inputs. It is not real historical performance, an estimate of future returns, or evidence that a strategy works in real markets. This one seed is not statistical validation; there is no out-of-sample study or confidence interval.

The CFTC explains why simulated trading assumptions can overstate or understate real performance, particularly when execution and costs differ. Read the CFTC’s explanation of hypothetical results.