A public, pre-registered experiment

The Honest Bot Experiment

Five trading bots, rules frozen in advance, results published win or lose.

DRY RUN: pipeline shakeout. Ledger resets on July 6, 2026.Read the pre-registration

Scoreboard

Trendline

CASH

$100,000

0.00%

vs Benchmark: 0.00 pp

Snapback

CASH

$100,000

0.00%

vs Benchmark: 0.00 pp

Oracle

CASH

$100,000

0.00%

vs Benchmark: 0.00 pp

Benchmark

CASH

The bar to clear

$100,000

0.00%

Coinflip

CASH

$100,000

0.00%

vs Benchmark: 0.00 pp

Equity curves

$99k$99.7k$100.3k$101kJun 11
TrendlineSnapbackOracleBenchmark(the bar)Coinflip

All five sleeves start from $100,000 in simulated cash. Fills are at official opening prices with a 0.01% cost per trade.

The numbers

BotTotal returnSharpe (ann.)Max drawdownTrades
Trendline0.00%0.00%0
Snapback0.00%0.00%0
Oracle0.00%0.00%0
Benchmark0.00%0.00%0
Coinflip0.00%0.00%0

Backtest, for the record (Jan 2000 to Jun 2026, QuantConnect)

Trendline: 413.67% · Sharpe 0.263 · maxDD 24.2% · 173 orders

Snapback: 287.87% · Sharpe 0.183 · maxDD 34.5% · 605 orders

The forward test exists to measure the gap from these numbers. Oracle has no backtest, by design.

Oracle's corner

Its answer, verbatim

"CASH"

Resulting signal:CASH
The frozen prompt
You are deciding a one-day position in SPY (the S&P 500 ETF) based solely on this morning's headlines. Below are the ten most recent headlines from a major financial news feed, fetched at 9:00 AM Eastern on {DATE}.

{HEADLINES}

Based only on these headlines, decide whether to be LONG SPY for today's session or to hold CASH. Reply with exactly one word: LONG or CASH.

Daily log

Most recent: June 12, 2026

TrendlineLONG
close
737.76
sma200
685.8166
SnapbackCASH
rsi2
62.2997
OracleCASH
source
cnbc-markets
headline_count
10
BenchmarkLONG

always LONG

CoinflipLONG
draw
3877220648
DateSignalsActivity
Jun 12TLSCOCBLCL·
Jun 11T·S·O·B·C··

Each chip: bot letter, then stance (L long, C cash). Bots: T Trendline · S Snapback · O Oracle · B Benchmark · C Coinflip

Changelog

No amendments. The rules have not changed since pre-registration.

15 pipeline runs, 10 errors. All public, including the errors.

Every table behind this page is publicly readable: prices, signals, trades, the daily ledger, the run log, and this changelog.

A bot wins only by beating Benchmark on both return and Sharpe over the full window. Six months can't distinguish skill from luck; that's what Coinflip is for.

Nothing here is investment advice. All money is simulated.

Follow by email

A machine-composed reading of the ledger every two weeks, win or lose. Double opt-in; unsubscribe in one click.