New: AI Phoenix Evidence Intelligence

Phoenix Evidence Intelligence is the layer that analyses evidence submitted to a dispute on Escrozon. This post explains what it does, what it cannot do, and why the decision still belongs to a person.

The Problem It Addresses

When a dispute opens, both parties upload material — screenshots, chat exports, receipts, test output. Reviewing that manually is slow and inconsistent: reading a long chat history, cross-referencing timestamps, and judging whether an image has been altered are all tasks where human attention degrades over time and varies between reviewers.

The result was resolution measured in days, and outcomes that could differ depending on who reviewed them.

What the Analysis Covers

Signs of manipulation. Inconsistent font rendering within a single screenshot, mismatched compression artefacts between regions, timestamps that do not align with the interface they appear in, and content spliced from separate sources.

Reuse across cases. The same image submitted to multiple disputes is a strong signal, and one no individual reviewer could ever spot — it requires looking across cases rather than within one.

Internal consistency. Whether the evidence agrees with the listing as it stood, the deal chat, and the recorded timeline. Most weak claims fail here rather than on image forensics: the story does not fit the record.

Confidence scoring. Each item receives a score with the reasoning attached, so a moderator sees why something was rated as it was rather than being handed a number.

What It Explicitly Does Not Do

It does not decide. Every dispute resolution is reviewed and approved by a human moderator. The analysis is advisory and labelled as such.

It does not read private messages. Only material explicitly uploaded to a dispute is processed.

It cannot verify claims about the outside world. Whether a channel's subscribers are genuine, whether revenue figures are honest, whether the seller intends to reclaim an account later — none of that is visible in a screenshot. Those remain verification tasks for the buyer during the inspection window.

It is not infallible. Detection of manipulated media is an adversarial problem, and confident-looking output can be wrong. That is a substantial part of why a person makes the final call.

Why the Split Exists

Machine analysis is consistent and tireless on mechanical work: reading four hundred messages, comparing timestamps, checking one image against every other image ever submitted. It does that the same way at 3am as at midday.

Human judgement is better at the questions that actually decide a dispute: whether a difference is material, whether someone acted in good faith and simply misunderstood, and what a fair split looks like when both parties are partly right. Those carry consequences for real people and are not reducible to a score.

Giving the analysis to the system and the decision to a person plays to each.

What It Means for You

If you are honest, disputes resolve faster because straightforward cases are cleared quickly rather than waiting behind complex ones. And your evidence is assessed the same way every time.

If you are submitting fabricated evidence, reuse across cases and manipulation artefacts are the two things this catches most reliably.

As a moderator, the effect is more time on genuinely difficult cases and less on mechanical verification, with a logged analysis trail for every decision.

How to Submit Evidence That Holds Up

Capture at the time, not afterwards. Contemporaneous material scores better because it is consistent with the timeline.

Keep full context. A crop showing a number is weak. The same screen with the account name, URL bar and surrounding interface is strong.

Prefer screen recordings for anything about live metrics. A continuous recording — logging in, navigating, changing a date range — is far harder to fabricate than a still.

Quote the listing. Most disputes turn on the gap between what was promised and what arrived, so anchor your evidence to the listing text.

Do not edit for clarity. Cropping to highlight something reads as alteration. Submit the original and explain in words what to look at.

What Makes Evidence Actually Persuasive

Most evidence submitted in disputes is weak not because the person is wrong but because of how it was captured.

Continuous beats cropped. A single unbroken screen recording carries far more weight than a set of images, because images can be assembled from anywhere and a recording shows context around each moment.

Timestamps that are visible in the recording matter. A system clock or an in-app timestamp visible on screen ties the footage to a moment. Metadata alone does not survive most sharing methods.

Original files beat forwarded ones. Every time a screenshot is forwarded through another application it loses information and gains compression. Submit what you captured, not what you sent to someone else and retrieved later.

Complete conversations beat excerpts. A cropped exchange raises the immediate question of what sat either side of it. Submitting the whole thing, including the parts that are less favourable to you, is more convincing than a curated selection.

What to Expect After You Submit

• Analysis is a summary, not a verdict. It organises what was submitted and highlights what is inconsistent or missing. It does not decide the outcome, and it is not treated as proof by the person who does.
• A human reviews every dispute. The analysis exists so that the reviewer starts from an organised picture rather than an unsorted pile, which shortens the process for both sides.
• Missing evidence is noted, not assumed against you. If something obvious is absent, you will usually be asked for it rather than judged on its absence.
• Both sides submit. The process is not adversarial in the sense of one party presenting and the other responding. Both accounts are gathered before anything is decided.

The practical advice is simple: capture evidence as the deal happens rather than reconstructing it afterwards. A recording made at the moment of handover takes two minutes and settles almost every dispute that follows.

Frequently Asked Questions

Does this replace moderators? No. It prepares the case; moderators decide, and every resolution is human-approved.

What if the analysis is wrong about my evidence? Say so in the dispute. The moderator sees your response alongside the analysis and its reasoning, and can weigh both.

Is my evidence shown to the other party? Both parties can see the case and respond. Disputes are adversarial but not secret.

Does it analyse my listing or my chats generally? No. Only material uploaded to a dispute.

Can I request a human-only review? Every dispute already receives one. The analysis informs that review; it does not substitute for it.