01 / GPT SLOTLane A
GPT × GEMINI / EXPERIMENT STUDIO
Audit the decision.
Follow the capital.
Compare two frozen selections on the same price path. Inspect entry costs, cash, drawdown, and the contribution of every position.
SYNTHETIC REPLAY
Loading fixture…
No model calls · no account connections · no orders02 / GEMINI SLOTLane B
03 / SHARED BASELINEEqual-weight universe
MARKED EQUITY / SAME ENTRY ASSUMPTIONS
One timeline. Three portfolios.
Lane ALane BUniverse baseline
Selection is frozen at the first observation. Fractional-share fills occur at the next observation’s close plus slippage. All subsequent marks use supplied closes; no exit costs or cash interest.
POSITION ATTRIBUTION
Every dollar has a source.
| Asset | Shares | Entry fill | Cost + fees | Final mark | Contribution ($) |
|---|
Bring a reproducible experiment / inspect the fixture
Paste schema-version 1 JSON with provenance, frozen selections and a complete dated price matrix. Tick the attestation only if the selections were fixed before execution. Price rows must have the same universe; missing data is rejected. Data stays in this browser.
A comparison is only as good
as its experiment design.
The included lanes occupy the original GPT and Gemini slots, but contain hand-authored selections. The invented tickers and deterministic price paths are a software demonstration, not actual model outputs or evidence that either model outperforms.
For a real study, freeze prompts, model versions, news cutoffs, responses and selections before the execution timestamp. Use a documented adjusted-price source, retain rejected runs, and evaluate many independent periods. A favorable replay alone does not validate a strategy.