The gold evidence file changed. Can another researcher reconstruct the earlier verdict?

A working link is not a frozen record

A research article can retain the same evidence URL while the file behind it changes. New events may be appended, a provider may correct a candle, or an export may alter its ordering. The link still opens, but the reader may no longer have the inputs used for the original verdict. Reproducibility requires version identity as well as availability.

This article proposes a research workflow for preserving that identity. It is distinct from deduplicating webhook deliveries or deciding whether a target was reached. The question comes after those tasks: can another analyst reconstruct the same classification from the same evidence and rules? No production outcome or successful trade is asserted here.

Define what is being fingerprinted

RFC 8785 describes a JSON canonicalization scheme intended to produce a repeatable representation for hashing or signing. It uses defined serialization and property ordering rather than assuming that every visually similar JSON file has identical bytes. The RFC is an informational publication dated June 2020, freshly consulted for this article; it is not a certification of any gold dataset.

That matters because whitespace and object-property ordering can change a file’s byte-level fingerprint without changing the interpreted values. A project must specify whether it fingerprints the original received bytes, a canonicalized data object, or both. The choice should be recorded in the manifest. An unexplained digest is less useful than a clearly documented procedure another reader can run.

Source notes: [1] RFC Editor — RFC 8785, JSON Canonicalization Scheme

A digest cannot establish that the market event was true

A cryptographic fingerprint can help identify whether the chosen representation matches a retained version. It cannot, by itself, establish who created the data, whether a price was genuine, whether an event occurred at the claimed time, or whether a broker executed a trade. An invented dataset can be hashed just as consistently as an authentic one.

My proposed evidence statement therefore separates integrity from provenance. Integrity concerns whether the retained object matches the declared version. Provenance concerns its origin, collection and chain of handling. Independent market verification requires appropriate source evidence beyond a local digest. Conflating these layers can turn a technically sophisticated archive into an unjustified claim of authenticity.

Build a small manifest around the actual decision

I would retain the source or receiver identity, collection time, observation cutoff, instrument definition, event count, relevant date range and exact calculation version. Record the file’s fingerprint together with its algorithm and canonicalization policy. The manifest should also identify excluded rows and the reason for exclusion, without exposing credentials or personal data.

For a signal audit, original entry, stop and target fields belong with the original events. A separate derived table can contain normalized timestamps, reference distances and outcome classifications. Keeping those layers distinct lets a reader challenge the calculation without losing the underlying record. Sanitization should occur under a documented policy before publication, and the public fingerprint should describe the sanitized object that readers can actually obtain.

Treat corrections as new evidence rather than silent repairs

Imagine a hypothetical provider correcting one bar’s high after an article was published. The earlier classification may have been valid for the earlier dataset and invalid for the corrected one. Save both versions, explain the changed field, and rerun the declared rules. The research question is whether the conclusion survives the correction, not how to make the original page appear prescient.

A correction notice should identify the previous and replacement snapshots, the time the revision became known and which conclusions changed. If no conclusion changes, say that too. A revision history is useful precisely because corrections need not imply misconduct or a broken method. It provides a way to distinguish a data correction from a calculation correction and from an author’s changed interpretation.

Make reproduction a bounded task

An effective archive tells the next researcher where to start: which file to download, which version of the rules to apply, and which expected intermediate totals to compare. It should not require guessing which current dashboard state resembles an earlier screenshot. The goal is a finite, inspectable procedure that can fail clearly when an input is missing.

The Research Lab is the natural home for this discipline, while the Daily Gold Brief and weekly reports retain their own publication and source dates. None of these practices validates future profitability. They make claims easier to examine and correct. Fills, commissions, spread, slippage and P&L remain unverified unless separate execution evidence supports them; an attractive seal in the conceptual illustration is never a substitute for that record.