Research record (experiment log)

A research record is the written log of every strategy experiment: the rule version, the data source and its coverage, the test dates, the costs and fill assumptions, the account rules applied, the checks that failed, and the decision taken next, kept so that a result can be reproduced and a rejected idea is not tried again by accident.

Senzoukria · Glossary · Updated September 2026

What goes in the record

  • Rule version: the exact script or written rule, with the parameter values used.
  • Data: provider, instrument, contract, session, bar construction, and the coverage report, including gaps and roll dates.
  • Test protocol: development and evaluation periods, chosen before any parameter was adjusted.
  • Execution model: bar-based entry at the next open, stop and target policy when both are touched in one bar, costs per contract and per order.
  • Account rules applied in the challenge model and the clauses it could not model.
  • Results, including the failed configurations and the number of variants tried.
  • The decision: rejected, changed one named thing, or promoted to replay.

Why the failures matter more than the wins

A striking result after many attempts needs stronger evidence than the same result after one. Without a count of what was tried, that count is unknowable, and the survivor looks better than it is. The futures backtesting guide makes the point directly: a gap is not a flat market, a touched limit price is not a fill, and a green equity curve answers only the narrow question of what happened under the assumptions that produced it. The record is where those assumptions are written down so that they can be challenged later.

The record also protects against silent re-optimisation. If the holdout period is written down before the first run, repeatedly optimising on it becomes visible as a protocol violation rather than a vague feeling that the test was somehow tainted.

Difference from a trade journal

A trade journal records executed trades with their context: the chart as it was, the note, the numbers. A research record documents experiments, most of which never produce a live trade. The two overlap when a strategy moves from backtest to replay to a simulated account, and in Senzoukria a zone analysis from the heatmap intelligence panel can be saved to the journal so the reasoning travels with the trades.

In Senzoukria

The Journal screen fills itself from fills, keeps notes and tags per trade, and can receive a zone analysis saved from the heatmap intelligence panel. The Auto backtest panel keeps the protocol for a run: the walk-forward option explains that parameters are chosen on the start of each slice and evaluated on the end, and the Gauntlet lists its criteria, from account survival to the deflated Sharpe and the probability of backtest overfitting. Its previous-sessions view, headed "Does this repeat?", reports how many trades the verdict rests on. Only the last three backtest runs keep their trades, and reloading the app clears them, so the record of a run has to be written down if it is to be compared later.

Common mistakes

  • Keeping a screenshot of the final P&L and nothing about the data or the costs.
  • Logging the rule but not the version, so that two results with the same name came from different code.
  • Omitting the account clauses the model could not check, then reading a challenge report as a certification.
  • Changing several things between two runs and attributing the difference to one of them.

This page in other languages

Frequently asked questions

How detailed does a research record need to be?
Detailed enough that someone with the same data could rerun the test and get the same table. In practice that means the rule version, the data coverage, the periods, the costs and fill assumptions, and the decision. Anything you would need to defend the result to a sceptical colleague belongs in it.
Should rejected ideas be kept?
Yes. A rejected idea with its reason is the entry most likely to save time later, and the count of rejected variants is what makes the surviving one interpretable. Deleting failures is the quickest way to overstate the evidence for what remains.

Keep reading

Strategy researchTest a trading strategy, from observation to challenge rulesDefine an observation, question it with the copilot, write a rule and explore costs, drawdown and fictional challenge limits in an interactive example.Quant · Start hereQuant trading for beginners: from idea to backtestDefine a testable rule, choose suitable data and evaluate later periods after costs.Order flow · HubOrder flow trading: the complete guideThe hub. What order flow is, what it shows — aggression, delta, imbalance, absorption, liquidity — and the tools that read it, each linked to a full guide.