TRADESTACK / RESEARCH LAB

Study every win.
Investigate every loss.

We build tools to understand automated decisions, not to manufacture a perfect equity curve. A trustworthy session review checks the recording, the signal, the exit and the costs.

DETERMINISTIC REPLAYFILL MODELINGNET P&LRISK ANALYSIS
FEATURED / CASE STUDY 001

A smaller loss is not a proven edge.

See the October 16 controlled replay findings and the evidence required before promoting any strategy rule.

Read the case study
01 / RESEARCH WORKFLOW

Test the assumptions.
Then test the result.

Our code contains session analyzers, recording validation, research reports and an observation workflow for investigating what actually happened.

01 / CAPTURE

Record market context

Store tick/depth events and session metadata in date-partitioned Parquet data. Check recording completeness before drawing conclusions.

02 / REPLAY

Reproduce decisions

Replay captured events through a simulation clock and use depth-aware fill modeling to evaluate order assumptions and timing.

03 / ATTRIBUTION

Analyze each trade

Examine entries, exit reasons, market regime, setup scores, trading costs and whether a small gross gain survives charges.

04 / COMPARISON

Compare cohorts

Group behavior by strategy setup, direction, risk state and session conditions rather than judging isolated trade screenshots.

05 / OBSERVATION

Document hypotheses

Persist or report findings separately from trading records. Distinguish an observation from a confirmed causal explanation.

06 / VALIDATION

Retest before release

Check changes against recorded sessions, fill assumptions and out-of-sample behavior. An attractive backtest alone is insufficient.

02 / DATA DISCIPLINE

What we will report.
And what we won't invent.

Context alongside metrics

Future public reports should state execution mode, dates, traded instrument, sample size, net results after estimated charges, drawdown, limitations and recording coverage.

Known uncertainties

Replay fidelity, partial fills, latency, stale quotes and inconsistent exit labels can change apparent outcomes. They must be investigated rather than hidden.

Signals aren't outcomes

Multi-stage filters and strategy scores help structure decisions, but their existence does not establish positive expectancy or guarantee profitability.

AI remains experimental

The engine includes an optional AI validator, currently documented as disabled by default. We do not present AI decisions as proven, live or latency-free trading intelligence.

Public performance reporting is not launched. Do not interpret hypothetical research charts, simulated trades or planned analytics as real-money results.
03 / PROJECT STATUS

Progress over promises.
Evidence over hype.

Session and trade research helpersIMPLEMENTED IN CODE
Recorded market replayIMPLEMENTED; FIDELITY UNDER VALIDATION
Multi-day automated tear sheetsNOT YET COMPLETE
Validated public performance pagePLANNED
Guaranteed daily incomeNOT CLAIMED