Skip to the evidence
AI Trading SignalsDefinition / Population / Proof

AI claims / Consumer evidence

AI Market Summaries: Citations, Versions and Timing

Review a market explanation by the exact documents and timestamps it used, not by how confidently the prose describes a later move.

A summary is a separate product object

A generated market explanation may help a reader organise information without being the engine that chose an earlier signal. Retain its output, production time, source list and relationship to the original decision. Do not call every persuasive paragraph predictive evidence. The original algorithmic-versus-AI protocol separates these components; this chapter follows a cited news summary through a versioned source packet. No hidden model reasoning or private prompt is purportedly inspected here.

A working citation establishes a destination, not necessarily the exact passage, vintage or input that produced the summary. Ask which document version was retrieved and when. If a cited page later changes, retain the original observation or mark it unavailable. A later revised article cannot become the text an earlier decision received by relabelling the same URL.

Publication, retrieval and output are different clocks

A news report's original time, the system's retrieval time and the explanation's output time can differ. Preserve them independently. If the explanation mentions a fact first published after the signal, that fact cannot establish the earlier signal's information set. It may legitimately belong to a later commentary product. The review should identify that role rather than asserting a false contemporaneous rationale or treating all post-event commentary as dishonest.

Check supported claims individually

Map material assertions to the inspected source and its scope. A source saying that an event occurred is not evidence that it caused a price change, which model feature used it or that a subscriber made money. Unknown factual or causal links should stay unresolved. The CFTC advisory provides caution around expansive AI profit claims; it does not establish that a particular summary or seller is fraudulent. This is a consumer evidence method, not an automated truth certificate.

A usable conclusion without a forecast badge

State which cited documents and versions were inspected, which statements they support and what chronology remains missing. The reader may value an explanation product while still requiring separate model-validation and account evidence for a stronger purchase claim. Preserve the correction history when an explanation changes. Searchable prose and machine-readable records should retain the same qualifications instead of converting a documentation review into an independently tested financial endorsement.

Completed fictional evidence case

A fictional decision is recorded at 09:00, a cited article adds a new detail at 09:30 and an AI summary is produced at 09:40. That summary may describe the new detail accurately, but cannot prove the 09:00 decision used it. The packet keeps the three objects and clocks separate. If the earlier article version is unavailable, its content stays unknown. The completed case does not identify a real provider or claim that a language model caused a trade; it demonstrates why output eloquence and a live citation do not establish original input timing.

Evidence objectRecorded valueClaim boundary
Decision09:00Original record
Added source detail09:30Later information
Summary09:40Commentary, not earlier evidence
An original claim, its available references and the bounded review
Illustrative record flow; no vendor experiment or customer account is pictured.

Direct sources and reviewed scope

Public documents, not private vendor experiments. The worked cases are invented; source dates do not establish a model-validation engagement.

Related evidence questions