From alert to position closure: build an evidence trail
A proposed review method for separating alert triggers, order submissions, partial fills and position closure, then reconciling the records with an exchange.
Start with a question that evidence can answer
For your next automation review, choose one question: “Which stage can I document?” Our proposed method uses separate records for a rule firing, an order request, an exchange acknowledgement, executed quantities and the final reconciliation of a position. Treat this as an editorial workflow for reviewing evidence, rather than an exchange status specification or a description of a particular MaxData integration.
Begin with one instrument, one account and a bounded review period. Use a saved export or a deliberately invented test case for your first pass. Define success as a documented conclusion about the records. Leave the merits of buying or selling outside the exercise.
The MaxData Pro Alerts guide describes an Alert as a notification activated by one condition and a Pro Alert as a notification activated by several conditions. [EN1] Pro Alerts guide.
Our convention reserves “alert triggered” for that first stage. Require a different evidence field for each subsequent stage. When a document is missing, use “not established” and write down the missing item. Avoid an overall “successful” label until you have defined exactly what success means for the review.
Use a vocabulary tied to documents
We propose the following working glossary. Adapt it to the terminology in your exchange documentation before applying it to real records.
Triggered: attach the rule identifier, configuration version, instrument and trigger timestamp. Require a saved rule event before marking this stage as established.
Submitted: attach evidence of an attempt to send an order request. Record the requested side, quantity and any response available to you. Reserve a separate identifier for each attempt in your own journal.
Acknowledged: require an exchange response identifying the relevant order. If you have only the outgoing request, leave acknowledgement unresolved under this convention.
Partially filled: use this working label when the documented executed quantity is greater than zero but below the requested quantity for the order under review. Link each included execution record to that order.
Reconciled as closed: use this label only after accounting for the quantity assigned to the reviewed position under your written rules. Specify whether your review covers an order, an entry-and-exit cycle or a larger allocation.
Add two fields beside every label: “evidence used” and “remaining question”. For example, write “request recorded; exchange order identifier missing”. If you permit a residual quantity when signing off a review, state that tolerance explicitly and display the residual separately. Do not silently round it away in the presentation.
Proposed journal convention
Choose the evidence for the stage
- Identify
- Rule event
- Attach
- Rule and version
- If evidence is absent
- Leave unresolved
- Identify
- Request attempt
- Attach
- Outgoing request
- If evidence is absent
- Leave unresolved
- Identify
- Execution record
- Attach
- Quantity and order link
- If evidence is absent
- Leave unresolved
Write the rule specification before examining timing
The Pro Alerts guide describes a window for satisfying conditions, a sleep period after activation, activation choices and data-source choices such as open, close, minimum, maximum and average. [EN2] Configuration terminology.
Our proposed rule card records the metric definition, its data source, the instrument, the observation interval, the threshold and the configuration version. Add the condition window and any sleep setting you intend to investigate. Write the effective start and end of that configuration beside its version.
For timing, use separate columns for the event timestamp, the time your system received the record and the time you retrieved the history. Preserve the timezone and timestamp precision supplied with each field. If your material gives only minute precision, leave seconds unspecified.
This workflow assumes no typical delay or guaranteed execution time. For your own measurements, define the start and end of each measured interval and record the observation period. Keep “not measured” distinct from a measured zero. Set the analysis horizon before reviewing the outcome: for example, from a specified trigger to the final retrieval of records for its associated order.
Reconcile one chain at a time
Use the following as our proposed checklist, with field names taken from the exchange and exports you actually use.
- Freeze the scope. Record the account, instrument, market type and time boundaries. State whether the unit of review is one order or an entire position cycle. Save the retrieval date with the source files.
- Collect the evidence groups. Assemble rule events, outgoing requests and responses, then exchange order and execution records. Mark an unavailable file as missing instead of reconstructing its contents from memory.
- Make the links explicit. Build a mapping from rule event to request attempt, order and execution records. For every link, name the matching field. Label a match based only on a similar timestamp as a hypothesis requiring further verification.
- Account for quantities. Keep requested quantity, included executions and residual quantity in separate columns. Write the unit beside every number. If you introduce a monetary equivalent, document the conversion method separately.
- Check the meaning of status labels. Consult the documentation for the exchange used in the case. Record the export filters and retrieval boundaries. Preserve earlier versions when you refresh the comparison.
- Record the conclusion. Choose “established”, “partly established” or “unresolved”. Explain the choice and identify the next document needed. Keep any decision about sending another order outside this evidence-review exercise.
For duplicate-looking records, our proposed convention is to retain the raw rows and document the identity rule used in the working table. Do not delete a row solely because its timestamp and quantity resemble another row. Where you cannot establish identity, leave the ambiguity visible.
Our proposed checklist
Build a reviewable chain
- 1
Bound the case
Specify the account, instrument, period and unit of review.
- 2
Preserve evidence
Collect event, request, order and execution records.
- 3
Map and account
Document identifier links, quantities and residuals.
- 4
Write the finding
State what is established and name the missing evidence.
Practise with a deliberately incomplete case
Consider an invented training file. Event S belongs to rule version 2. Request R asks for 12 units of asset X. A response assigns order O. Two execution records linked to O contain 2 units and 5 units. A later record describes the remaining requested quantity as cancelled. These are fictional assumptions for the exercise, not observations about an exchange.
Apply the glossary: document the trigger, submission and acknowledgement separately. Enter 7 units as the executed quantity and 5 units as the unexecuted part of the request. Leave position closure unresolved until the exercise provides evidence accounting for the acquired quantity.
Now add a fictional exit record for 4 units. Under the exercise's simple quantity convention, carry forward 3 units for further reconciliation. If you want the exercise to include fees charged in the asset itself, add an explicit fee assumption and revise the quantity worksheet accordingly.
Write an invalidation condition beside each proposed link. Examples include a different instrument in the response, an account mismatch or an execution identifier absent from the supporting file. If one appears, withdraw your own conclusion and return the affected stage to unresolved.
Finally, draft two alternative explanations to investigate, without asserting either: “the retrieved period excludes the response” and “the request was not acknowledged”. List the document that would distinguish them. This is our proposed diagnostic practice, not a finding about the fictional request.
Fictional training file
Follow the documents, then stop at the gap
Event S
Attach the event to rule version 2.
Request R
Record the fictional 12-unit request and response identifying order O.
Two executions
Account for the assumed 2-unit and 5-unit records.
Incomplete exit
After the assumed 4-unit exit, carry 3 units into the next reconciliation step.
Keep historical signal analysis in a separate worksheet
The English MaxData backtest guide describes checking historical trigger counts and displaying average and median price changes before and after an alert. [EN3] Backtests guide.
Our recommendation is to maintain a historical-analysis worksheet alongside the execution-review worksheet. For the historical analysis, record the rule definition, sample period, number of events, observation interval and horizon after each event. For the execution review, record requests, responses, executed quantities, prices and the costs included in your accounting.
When comparing an analytical model with your own records, add explicit assumptions for commissions, your chosen slippage benchmark and liquidity. Mark unknown costs as unknown. Record which settings were selected using the same historical sample and which were evaluated on a separate period. Keep historical results under a historical heading; do not relabel them as an expected future return.
Before judging a discrepancy, state the question narrowly. Our preferred wording is: “Do these documents support this link and quantity?” If you want to assess performance as a separate task, first write its calculation method, inclusion rules and limitations.
Finish with a precise support question
Our proposed handoff package contains the rule version, time range, relevant identifiers, expected stage, observed stage and missing evidence. Include only what is needed to reproduce the question, and remove authentication details.
The Help Hub introduces MaxData Info Bot as an assistant for questions about MaxData and the workspace. [EN4] MaxData Help Hub.
Ask a bounded question such as: “I have event S and request R; which record should I inspect for the response identifying order O?” For your review log, accept either a supported conclusion or a clearly described evidence gap. Make the unresolved item, its owner and the next retrieval step explicit before signing off.
Sources
- MaxData Help Hub – Pomoc i przewodniki Mój workspace — agent.maxdata.app; accessed November 17, 2026
- Pro Alerts — agent.maxdata.app; accessed November 17, 2026
- Backtests — agent.maxdata.app; accessed November 17, 2026
Start working with your own agent
The agent knows this article context and still follows the assigned department checklist.
Related articles
Give Every Alert a Review: Usefulness, Missed Events and Attention Cost
Build an alert review around a defined task, an independent event register and a time budget. Our proposed workflow includes explicit rules for changing or retiring notifications.
From backtest to live observation: a review process for trading automation
Build an evidence file with a reserved test period, walk-forward reviews, paper trading logs, and explicit conditions for allowing or stopping automation.
Turn a question into an alert specification
Define the event, condition, interval and invalidation rule, then write an explicit policy for repeat notifications and test it.
Why two crypto readings disagree: a practical reconciliation workflow
Build a reproducible comparison of units, exchange coverage, aggregation, timestamps, delivery delays and revisions before judging a discrepancy.