Max Agent
Articles

Three alert case studies: investigate the process before the result

Build an evidence file around inputs, competing explanations and historical observations. Three worked investigations without a trade recommendation.

Editorial teamPublished Updated

A useful case study should make its reasoning inspectable. For this article, our proposed editorial convention is to finish with an evidence file: the question, the inputs, the competing explanations, the historical result and the conditions for withdrawing the interpretation. A decision to leave a question unresolved counts as a completed exercise.

The cases below are methodological exercises. One examines a numerical illustration published in the documentation. The other two propose investigations without inventing observations or returns. They are not reports of independently downloaded market datasets. Every checklist, research rule and review gate below is our proposed workflow, rather than a claim of demonstrated investment performance.

Open an evidence file before opening a chart

Start with a question that names an observable relationship. Our suggested wording is: “What happened to the specified price series after the defined event?” Keep a separate line for the proposed explanation. Leave any trading decision outside the file.

Record the instrument, trading pair, venue, market type, data provider and timezone. Define the price input precisely: for example, the closing value of a chosen candle series. For another metric, request its definition, unit and underlying source before selecting a threshold. In this workflow, an unexplained abbreviation is an incomplete input.

Next, record three different time fields: the observation interval, the publication delay and the assessment horizon. For a workshop exercise, we propose fifteen-minute observations and a four-hour horizon. These are exercise settings, not market parameters to adopt. Where typical publication delay is undocumented, write “unestablished” and leave the question of timely execution unresolved.

Also create fields for the sample period, event count, missing observations and rule version. Before seeing an outcome, state which missing information would stop the investigation. Our proposed stopping rule is simple: do not assign a directional interpretation to an input whose identity or timing remains unclear.

Our proposed research checklist

Prepare an inspectable question

  1. 1

    Question

    Name the relationship you intend to examine.

  2. 2

    Inputs

    Record instrument, provider, definition and unit.

  3. 3

    Timing

    Specify interval, delay and assessment horizon separately.

  4. 4

    Review gate

    Declare what missing evidence would stop interpretation.

An editorial worksheet for defining an investigation before reviewing its outcome. Define the question, identify inputs, specify timing and write a stopping rule.

Case 1: interrogating a ten-event illustration

The English Backtests guide gives an example involving ten triggers over eight months and a statistical price decline of 5% within four hours, followed by a rise. The supplied guide presents this as an example and does not provide the underlying event records. E1 — Backtests

Treat the assignment as a request to reconstruct that illustration. Ask for the ten timestamps, the instrument identifier, the relevant price series and the exact statistic represented by the percentage. Until those materials are supplied, label the candle interval, publication delay and sample composition as unestablished.

Our proposed reconstruction sheet has one row per event. Include the reference price, the price at the four-hour endpoint, the lowest price inside the window and a completeness flag. Keep an additional field for overlapping observation windows. Do not replace an unavailable event with an invented “typical” path.

Prepare competing explanations as questions. Would comparable windows without a trigger show a similar pattern? Were the events concentrated in one part of the eight-month period? Does the quoted decline refer to an endpoint or an intermediate low? Leave each question open until the required observations are available.

The historical result available for this case is therefore the guide’s illustrative description. Our editorial conclusion is narrower: withhold a judgment about the alert’s independent value. Abandon the reconstruction if the event list or outcome definition cannot be established. Do not convert the illustration into an annual return target.

Case 2: does the extra condition change the investigation?

The Pro Alerts guide distinguishes an Alert based on one condition from a Pro Alert combining several conditions. Its description includes a window for satisfying the conditions and a sleep period after triggering. E2 — Pro Alerts

Our proposed second case compares two research specifications. Specification A uses a defined price-change condition. Specification B uses the same condition plus a second metric. Select that second input only after completing its definition, unit, source and delay fields. Leave the numerical thresholds unset until the research specification is written.

For both versions, hold the instrument, date coverage, price source and assessment horizon constant. Record the input interval separately from the condition-combination window. State whether an event means the first qualifying observation or another occurrence after a defined pause. This is a proposed research convention, not a statement about which settings an account offers.

Write two competing hypotheses before reviewing results. One asks whether the additional condition identifies observations with a different subsequent price pattern. The other asks whether the selected dates or event-grouping rule account for the apparent difference. Neither explanation is established in this exercise.

Request an event list, a count, a distribution of endpoint changes and a comparison with explicitly selected control windows. The historical result here is “not calculated”: no event dataset is being presented. Our proposed invalidation conditions include inconsistent price definitions, unequal date coverage or an unknown availability time for the additional metric.

Proposed comparative exercise

Compare specifications before outcomes

Specification A
Input rule
Defined price change
Study boundaries
Shared instrument and dates
Timing record
Event definition and interval
Historical finding
Not calculated
Specification B
Input rule
Same price change plus second metric
Study boundaries
Shared instrument and dates
Timing record
Event definition, interval and combination window
Historical finding
Not calculated
Two research designs with shared boundaries; no performance comparison has been calculated. Specification A uses a price condition; specification B adds a documented second metric while retaining the same study boundaries.

Case 3: reviewing a selected configuration

The Polish backtesting guide describes testing take-profit, stop-loss and timed-exit configurations. It also describes a heatmap showing results for different TP and SL settings. E3 — Backtesting guide

Separately, the English guide describes historical trigger counts and average and median price-change lines before and after an alert. E4 — Backtests

Our third assignment is a report audit. Request the full configuration list, the period used to choose parameters and a record of the selection rule. Propose another historical segment for evaluating the frozen choice. Under our convention, changing parameters after reviewing that segment creates a new study version rather than silently updating the original.

Investigate alternative explanations through explicit questions. Was the selected configuration isolated from its neighbours? Did a few events dominate the reported result? What happened in other portions of the sample? Include historical overfitting as a review question, without declaring that it has been detected in an alert we have not tested.

Leave the historical outcome blank until the report and observations are supplied. Our proposed conclusion for this case is a request for evidence, not a ranking of configurations. Stop the review if the report does not distinguish the data used for parameter selection from the data reserved for evaluation.

Keep execution assumptions on a separate page

The Polish guide identifies slippage and commissions among execution differences to consider when a historical test and market use diverge. E5 — Backtesting guide

Our proposed execution sheet asks for the commission assumption, slippage assumption, modelled order size and liquidity constraints. Mark any unspecified item as unknown. Reserve the label “after costs” for a report whose stated calculation includes the listed costs. Keep an event’s price change separate from a simulated sequence of fills.

We also propose recording several explicitly named execution scenarios. Do not assign probabilities without supporting research. Keep the event set and exit rules unchanged while varying the declared assumptions. Add sample size, missing-data treatment and parameter-selection history to the review page before writing the final interpretation.

Turn the exercise into supervised observation

Our proposed final stage is observation without order placement. Freeze the research rule and log new qualifying events. Record the event timestamp, the time the input became available and the notification receipt time. Assign a reviewer and define pause conditions, including unexplained gaps, changed metric definitions or inconsistent timestamps.

MaxData Help Hub presents an assistant inviting questions about MaxData and the workspace. E6 — Help Hub

For a focused support request, our suggested questions concern input definitions, intervals, typical delays and report assumptions. Keep the answers in the evidence file and identify anything still unresolved. This article is educational; the account invitation below is a commercial element. Under our proposed completion rule, publish the question, the evidence obtained, the historical finding or explicit missing result, the competing explanations and the conditions for withdrawing the conclusion.

Our proposed observation sequence

From a written rule to a reviewed log

  1. Specify

    Freeze inputs, definitions and pause conditions.

  2. Reconstruct

    Record historical findings or explicitly missing evidence.

  3. Observe

    Log new events and data and notification timestamps.

  4. Review

    Explain discrepancies before creating another rule version.

A staged educational workflow without order placement. Freeze the rule, document historical evidence, observe new events and review discrepancies.

Sources

  1. MaxData Help Hub – Pomoc i przewodniki Mój workspace — agent.maxdata.app; accessed November 3, 2026
  2. Pro Alerts — agent.maxdata.app; accessed November 3, 2026
  3. Backtests — agent.maxdata.app; accessed November 3, 2026
  4. Backtesty Automatyczne Weryfikowanie Strategii — agent.maxdata.app; accessed November 3, 2026

Start working with your own agent

The agent knows this article context and still follows the assigned department checklist.

Create a MaxData account
Max Agent