Max Agent
Articles

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.

MaxData EditorialPublished Updated

When two readings disagree, start by writing a testable question: “Under which documented settings should these observations match?” Our proposed workflow begins with a comparison record and ends with a limited, reproducible conclusion. Use it to investigate a chart, an alert or a data input without turning the exercise into a trading recommendation.

The source material for this article is the supplied documentation bundle. Product references below describe what that documentation says; they do not certify present availability. The worksheets, examples and acceptance rules are our editorial proposals, rather than documented product specifications.

Define the comparison before inspecting the difference

Our proposed first step is to name the exact question. For example, ask whether two providers report the same field for a specified instrument and interval. Avoid starting with a broad question such as which platform has better data. Choose a bounded period, a sample selection rule and an acceptable difference before inspecting the results.

Create a record for each reading. Include the metric definition, provider, instrument identifier, displayed value, unit, interval and retrieval time. Add separate fields for exchange coverage, market type, aggregation, time zone and observation status. For every definition, preserve the documentation address and a short explanation in your own words.

Treat unknown fields as unresolved questions. Our proposed worksheet uses “not established” rather than a guessed default. Give the investigation a stated horizon: the selected instrument, observations and period. Make any extension to other assets or intervals a separate task with its own evidence.

Our proposed workflow

Turn a mismatch into a test

  1. 1

    Frame

    Name the instrument, field and period you intend to compare.

  2. 2

    Capture

    Preserve each original reading and its settings.

  3. 3

    Question

    Turn missing definitions into specific provider questions.

  4. 4

    Predefine

    Choose the sample, tolerance and rejection conditions before judging the result.

An editorial investigation sequence, rather than a description of product behavior. Four steps for an investigation: define the question, capture both records, identify unknowns and set acceptance rules.

Check the unit and instrument identity

Ask for the full pair or contract identifier, including the quote asset. In our proposed convention, “BTC” alone is an incomplete instrument description for a comparison. Write the unit next to every value, and record any scaling factor. For a percentage, request the reference value and calculation window as well.

If conversion is necessary, make it an explicit stage. Preserve the original reading, the conversion input, its source and its timestamp. Define rounding and acceptable tolerance before calculating the comparison. If the conversion cannot be documented, leave the reconciliation unresolved instead of silently replacing the original value.

The Pro Alerts documentation lists altcoin pairs quoted in USDT and BTC, and describes selectable value sources including open, close, minimum, maximum and average. Treat these as documented descriptions when preparing questions about a configuration. Pro Alerts documentation.

As a proposed review question, ask: “Have I checked the identity of the input, or only matched the label shown on screen?” Do not approve the comparison until you can answer with the recorded fields.

Make exchange coverage explicit

Request the constituent exchanges and instruments for each series. Our proposed worksheet separates a single venue from a provider-defined group and asks for the membership of that group. Keep spot and derivative identifiers in separate fields. Record any filters, exclusions and eligibility rules supplied by the provider.

For a consistency test, choose an explicitly shared scope. For a coverage study, deliberately retain the different scopes and write a different question. For example: “What differences do I observe between group A and group B under the settings listed here?” This is our proposed experiment design, not an assertion about the cause of a particular discrepancy.

Keep several candidate explanations open. Mark “coverage difference” as a hypothesis until you have supporting documentation and a controlled comparison. A proposed rejection rule is straightforward: if the discrepancy remains after documented coverage alignment, do not close the case with coverage as its sole explanation. Record the remaining uncertainty.

Our proposed study design

Choose the question you want to answer

Consistency test
Scope
Specify a shared venue and instrument set
Question
Do matching definitions agree within my tolerance?
Conclusion
Limit to the tested observations and settings
Coverage study
Scope
Document each set separately
Question
What differences appear between these recorded scopes?
Conclusion
Describe differences without assigning an unsupported cause
Two editorial designs with different recording requirements; neither assigns a cause to a real discrepancy. Comparison of a consistency test using a shared scope and a coverage study retaining different scopes.

Reconstruct the aggregation recipe

Ask for the input observations, calculation function, weighting, interval boundaries and treatment of missing values. Our proposed rule is to reject “aggregated data” as a complete methodological description. Request the actual recipe or a worked example and preserve it with the comparison record.

The Max Terminal documentation describes multiple chart intervals and editing a metric line to change its appearance or data source. It does not need to serve as a substitute for your own written comparison record. Max Terminal documentation.

Consider an invented exercise: worksheet A says “closing value,” while worksheet B says “average.” Under our proposed protocol, classify the pair as different definitions and prepare a second test using a documented common field. Do not declare a provider error from that exercise.

If you can reproduce a calculation from inputs, specify how many observations you used and why you chose them. Preserve excluded records and the reasons for exclusion. Do not select only the observations that make your preferred explanation appear convincing.

Align the clock and the interval boundary

Our proposed reporting convention is to use UTC in the comparison worksheet while retaining each original timestamp and its stated zone. This convention is an editorial choice, not a claim about a platform's internal clock. Ask what the timestamp denotes: interval opening, interval closing or another reference event.

Record the exact start and end of the interval and ask how boundary observations are assigned. Keep event time, publication time and receipt time in separate fields whenever the provider can identify them. If a field is unavailable, preserve that limitation rather than constructing a timestamp from an assumption.

Choose whether your experiment covers completed observations, provisional observations or both in separate groups. Define the acceptance rule for each group. As our proposed discipline, avoid extending a result from one boundary convention to another without a fresh comparison. In the final report, state the time range, interval and unresolved clock questions beside the numerical result.

Observe delivery and revisions separately

Create two fields for delay: “typical according to documentation” and “observed during this test.” Our proposed rule is to leave the first unresolved without an attributable specification. Choose a monitoring schedule for the second and record the reason for that schedule. Do not present the chosen interval as a service guarantee.

For revision testing, select an observation and retain successive copies. Record its identifier, value, receipt time and any version information. Ask the provider about provisional status, historical corrections, publication schedules and change notices. Keep an unchanged result as an observation from your sample, not a declaration that revisions never occur.

Use neutral language until the mechanism is established. Our proposed labels include “value changed between retrievals,” “new observation not received within the test window” and “cause unresolved.” Reserve a specific explanation for evidence that supports it. Make the experiment repeatable by preserving the retrieval settings and the planned follow-up times.

Our proposed observation schedule

Keep a history of the same observation

  1. Initial capture

    Store the observation identifier, value and receipt time.

  2. Scheduled repeat

    Retrieve the same observation and describe any change without assuming its cause.

  3. Later review

    Compare retained versions and request an explanation for unresolved changes.

Choose and record the intervals yourself. This schedule specifies no provider latency or revision frequency. An initial capture, a scheduled repeat and a later review of one observation, with settings preserved throughout.

Test alert interpretation with a separate record

The Pro Alerts documentation describes a time window for multiple conditions, a sleep period after activation and choices about when a condition activates an alert. Documented alert settings.

Our proposed alert review copies the input definition, threshold unit, interval, condition window and repeat rule into a separate record. Write the expected result before observing the test. Use an observation-only exercise, with no transactions, to compare the recorded input against the recorded alert outcome. Leave the result unresolved when the necessary settings cannot be established.

The Backtests documentation describes checking historical trigger counts and viewing average and median price changes around alert events. Backtests documentation.

If you add a historical study, our proposed worksheet requires the sample size, selection period and evaluation horizon. Keep that historical result separate from future expectations. Before tuning, reserve a separate evaluation period as our proposed check against selecting settings solely for their historical fit. If the study includes simulated execution, explicitly specify fees, slippage, liquidity assumptions and order size; otherwise, leave execution conclusions outside its scope.

Close the case with a bounded conclusion

Our proposed closing statuses are “documented definition difference,” “agreement within the specified test” and “unresolved discrepancy.” Attach the two records, original observations, settings, sample selection and remaining questions. State the condition that would reopen the case, such as a corrected definition, changed scope or a failed repetition.

For a support request, ask one concrete question about the unresolved field and include enough detail to reproduce it. The Help Hub introduces MaxData Info Bot as an assistant for questions about MaxData and the workspace. MaxData Help Hub.

End the investigation with a result limited to what you tested. As our proposed review convention, do not convert a matching sample into blanket approval of a data feed or automation. Commercial invitation: the account link below is separate from the diagnostic conclusion and does not establish data specifications or validate the outcome of your experiment.

Sources

  1. MaxData Help Hub – Pomoc i przewodniki Mój workspace — agent.maxdata.app; accessed November 6, 2026
  2. Max Terminal — agent.maxdata.app; accessed November 6, 2026
  3. Pro Alerts — agent.maxdata.app; accessed November 6, 2026
  4. Backtests — agent.maxdata.app; accessed November 6, 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
Crypto data discrepancies: units, timing and revisions