You acted on a number. You cannot remember where the number came from. That is the starting point for a large proportion of flawed reviews — not bad analysis, not weak discipline, but a missing chain of custody. A figure you trusted had no traceable origin, and when the situation did not behave the way you expected, you had no way to find out whether the figure was wrong, stale, misremembered, or accurate all along.

This article teaches you how to keep a data audit trail: a record that lets you trace any number you relied on back to its source, the date it was captured, and the method used to capture it. By the end you will be able to reconstruct the evidence behind any decision you made, and you will understand why the absence of that record is not a minor inconvenience — it is a structural blind spot in how you learn from experience.

Why provenance matters

Every number you use in your analysis has a history. It was collected at a specific moment, from a specific source, using a specific method. When none of that history is recorded, the number becomes a belief — something you hold with confidence but cannot verify. Beliefs are hard to correct. A number with a source attached is much easier to challenge and update.

The problem is subtle because most numbers feel solid when you first encounter them. You read a figure in a report, a summary, or a discussion. It fits your existing picture of the situation. You file it away. Two weeks later you recall it — but what you recall is the figure as it felt, not the figure as it was stated, and certainly not the source it came from. Memory degrades precision. It rounds numbers, conflates sources, and strips away the caveats that surrounded the original figure.

When you later review a decision and something does not add up, you are working from a reconstructed version of the evidence. You cannot tell whether your analysis was sound but your data was wrong, or your data was fine but your logic was flawed. Without provenance, you cannot separate those two failure modes. Without separating them, you cannot fix the right thing.

The mental model: every number has a birth certificate

Think of every figure you rely on as having a birth certificate — a document that records where it came from, when it was issued, and what method produced it. In the legal chain of custody for physical evidence, an item without a documented chain is inadmissible: not because it is necessarily false, but because its reliability cannot be verified. The same logic applies to the data you use to make decisions.

A number without a birth certificate is not automatically wrong. But it is unverifiable. And in your own review process — where the goal is to isolate what went right and what went wrong — unverifiable is almost as bad as wrong, because you cannot use it to build a reliable pattern library. You end up learning from reconstructed history rather than from actual history, which means the patterns you identify are partly patterns in your own memory, not patterns in the market.

Keeping a data audit trail is the practice of issuing birth certificates to your numbers before you act on them. It takes less than a minute per figure. The cost is low. The cost of not doing it compounds quietly over every review you ever run.

What an audit trail captures

A data audit trail does not need to be elaborate. Four fields are sufficient to make any figure traceable:

  1. Source. Where did this number come from? Name the specific document, database, data provider, or publication. "I read it somewhere" is not a source.
  2. Date captured. When did you record this figure? Not when it was published — when you wrote it down. A figure that was accurate six months ago may be outdated now. Knowing when you captured it tells you how stale it might be.
  3. Method. How was this figure produced? Was it a direct reading from a primary data source? A calculation you performed? A figure quoted in secondary analysis that itself cited a primary source? Each layer of remove adds a layer of potential distortion.
  4. Raw value alongside derived value. If you took a raw figure and transformed it — adjusted for something, averaged it, compared it to another figure — keep both. The raw figure is what the source actually said. The derived figure is what you made from it. If the derived figure later seems wrong, you need the raw figure to check your math.

These four fields fit in a single row of a plain spreadsheet, a notebook margin, or a notes app. The format is not the constraint. The habit of filling in all four fields, every time, is the constraint.

How unsourced numbers corrupt review

When you sit down to review a decision and one of your inputs is unsourced, you face a choice: treat the number as reliable and proceed, or acknowledge that you cannot verify it and stop. Most reviewers, under time pressure and motivated by the desire to reach a conclusion, treat it as reliable and proceed. This is the mechanism by which unsourced numbers corrupt review.

The corruption is not dramatic. It does not produce obviously wrong conclusions. It produces conclusions that are plausible but fragile — built on a foundation that has not been checked. If you later encounter contradictory information, you have no way to know which figure was right. If you try to replicate the analysis with different inputs, you cannot reproduce your original result. The review looks complete, but it is not anchored.

Over time, a practice of reviewing against unverified inputs produces a pattern library full of confident-sounding conclusions that cannot be cross-referenced or stress-tested. You have learned something, but you cannot say precisely what, from which episodes, or whether those episodes were described accurately in your memory. That is a weak foundation for any behavioral correction.

A hypothetical example

Consider a learner — call her Maya — who was reviewing a simulation session and trying to understand why a particular decision had underperformed. Her stated reasoning at the time had referenced a figure she described in her notes as "the typical drawdown range for this kind of setup." She had not recorded where the figure came from.

Working backward, she tried to reconstruct the source. She thought she remembered reading it in a summary of historical scenarios. But when she searched her notes, she found two different figures that were both plausible — one roughly 30% higher than the other. She did not know which one she had actually used in her original reasoning, or whether it was a third figure she had remembered imprecisely from reading. She spent forty minutes trying to reconstruct information that would have taken thirty seconds to record at the time.

More importantly: when she finally identified the most likely source, she found the figure had applied to a narrower set of conditions than she had been applying it to. Her original reasoning had been built on a figure that was technically real but was being used outside its valid range. The decision was not wrong because her logic was flawed. It was wrong because her data was misapplied — and she could only discover that because she eventually traced it back. Without the trace, she would have concluded that the logic itself was at fault, and she would have changed the wrong thing.

If she had recorded the source, date, and method at the time of her original analysis, the review would have taken five minutes and produced a precise, actionable conclusion. Instead it took forty-five minutes and produced a tentative one.

Reconciling before acting

One of the highest-value moments to apply the audit trail habit is immediately before you make a decision based on data. Reconciliation means doing a brief check: is the figure I am about to rely on the same figure I originally captured, from the same source, at the same point in time?

This matters because figures change. A metric that was valid when you first recorded it may have been revised, restated, or updated since. If you are working from a cached version of a number and the underlying reality has shifted, your analysis is operating on stale data — and you may not know it. A one-minute reconciliation before acting is the practice that catches this.

Reconciliation also catches arithmetic errors. If your derived figure does not match what you get when you re-apply your method to the current raw figure, something changed: either the raw figure was updated, or your original calculation contained an error. Either way, you want to know before you act, not after.

Common mistakes

  • Recording the derived figure but not the raw. If you averaged three figures and recorded only the average, you cannot check your averaging later. Keep the inputs alongside the output.
  • Treating "I remember reading" as a source. Memory is not a source. It is a reconstruction. If you cannot name a specific document, the figure is unsourced, and it should be labeled as such until you can trace it.
  • Backdating captures. Writing "captured on [date]" when you are actually filling in the field days later defeats the purpose. The value of the timestamp is precisely that it reflects when you actually encountered the figure. An approximate date noted honestly is more useful than an invented precise date.
  • Assuming a figure is still current. Data ages. A figure that was accurate at capture may have been revised. Before acting on any figure you captured more than a short time ago, check whether the source has updated it.
  • Maintaining the trail only for numbers that seem important. The numbers that seem unimportant are often the ones that turn out to matter. Apply the trail to all figures, even the background ones you are treating as context rather than inputs.

Simulator exercise: the debrief source check

Abu Terminal's Speed Run and replay modes are built on a verified historical price registry — every figure in every scenario is traceable to a specific data source. That architecture makes the simulator a model of what an audit trail looks like in practice. Each number in a debrief panel is reproducible: you can replay the same event and get the same figure every time.

Use that property deliberately. After any Speed Run session, open the debrief panel and pick three figures that the session referenced — a price movement, a percentage shift, a duration. For each one, note in your journal:

  1. What the figure was.
  2. What scenario or event it came from.
  3. What you would need to look up if you were citing this figure in a review.

Then compare that three-field entry to any figures you are currently holding in memory from readings outside the simulator — market statistics, historical anecdotes, "rules of thumb" you have absorbed. Can you fill in the same three fields for those external figures? If not, they are currently unverifiable, and you should treat them accordingly: as working hypotheses, not as established inputs.

The exercise is not about the simulator data specifically. It is about building the muscle of tracing figures before relying on them. The simulator gives you a safe environment where the right answers are verifiable, so you can practice the trace without consequence. That habit then transfers to every context where the right answers are harder to check.

Reflection prompts

At the end of a review session, three questions help you assess the quality of your data foundation:

  1. If someone asked me to prove that any one figure in today's review was accurate, which figures could I verify and which could I not?
  2. Are there conclusions I reached today that depended on a number I cannot trace? If so, how confident should I actually be in those conclusions?
  3. What would I need to do differently at capture time to make the next review easier to anchor?

Self-check

Three questions to check your understanding before moving on:

  1. What are the four fields a data audit trail captures? Source, date captured, method, and raw value alongside any derived value. All four together make a figure traceable; dropping any one weakens the chain.
  2. Why does recording only the derived figure cause problems? Because if the derived figure later looks wrong, you need the raw input to check whether the error was in the calculation or in the raw data itself. Without the raw figure, you cannot isolate the failure.
  3. What does reconciliation mean, and when should it happen? Reconciliation means checking that the figure you are about to act on matches the figure you originally captured, from the same source, and has not been revised or updated since. It should happen immediately before acting, not days later during review.

Closing

The gap between what you remember and what was actually true is not a character flaw. It is the normal behavior of memory under time pressure and accumulated experience. The data audit trail is not a corrective for a personal weakness — it is a structural response to a universal limitation. Provenance is not bureaucracy. It is the mechanism that keeps your review process anchored to reality rather than to a cleaned-up reconstruction of it. A number with a birth certificate can be challenged, updated, and corrected. A number without one can only be believed or doubted. For the work of improving decision-making over time, belief is too fragile a foundation.

Educational simulator content, not financial advice.