If a report makes a strong claim, you should be able to check it. Grey Mirror pairs supported findings with counts, dates and message evidence. This page explains what we require and when the available history is too thin for a confident answer.
Evidence is any report support that connects a claim to the uploaded thread: message references, timeline windows, participant counts, metric deltas, parser diagnostics, and confidence labels. Evidence is not a mood, hunch, or unsupported summary.
What are the evidence levels?
Grey Mirror evidence can exist at multiple levels. The right level depends on access, privacy, and what the stored result payload actually contains.
Message-level evidence
Specific message references or excerpts when the report payload and access gating allow it.
Window-level evidence
A time range where a signal appears repeatedly, such as repair after conflict or a cooling response pattern.
Metric-level evidence
Counts, deltas, confidence, and participant splits behind a score or tile.
Report-level evidence
Cross-metric agreement that explains the trajectory rather than a single isolated signal.
Why would evidence be withheld or marked low confidence?
A responsible report should hold back when the evidence is not safe to show or not strong enough to support the claim.
The user does not have access to full evidence.
The payload predates a feature or does not include evidence references.
The parser could not preserve enough timestamps or participant labels.
The metric is based on too few messages or too few repeated examples.
Showing raw evidence would expose sensitive private content in the wrong surface.
How should confidence labels work?
Confidence labels should explain reliability, not decorate the UI. A high-confidence label needs enough data and supporting evidence; a low-confidence label should tell the user what is missing.
How is evidence protected?
Evidence belongs inside gated report surfaces, not public pages. Public pages can explain evidence standards, but they should never expose a user upload, private report artifact, or full-only evidence link.
Evidence standards and authoritative references
Grey Mirror's evidence approach aligns with established principles from digital forensics, legal evidence standards, and information quality frameworks.
These references inform our evidence standards but do not constitute legal advice. Grey Mirror is a pattern analysis tool, not a legal or clinical diagnostic.
Digital forensics chain of custody
Our evidence preservation follows principles from <a href="https://www.nist.gov/itl/sed/nist-special-publication-800-86" target="_blank" rel="noopener noreferrer">NIST SP 800-86: Guide to Integrating Forensic Techniques into Incident Response</a>, focusing on maintaining data integrity and traceability from export to analysis.
Legal evidence relevance and reliability
Our confidence bands and evidence thresholds draw from <a href="https://www.law.cornell.edu/rules/fre/rule_401" target="_blank" rel="noopener noreferrer">Federal Rules of Evidence Rule 401</a> (relevance) and <a href="https://www.law.cornell.edu/rules/fre/rule_402" target="_blank" rel="noopener noreferrer">Rule 402</a> (admissibility), adapted for pattern analysis rather than legal proceedings.
Information quality principles
Our evidence quality checks align with information quality frameworks that emphasize accuracy, completeness, and traceability of data sources. We apply these principles to message exports and parser outputs.
How should confidence labels work?
Label
Meaning
User interpretation
High confidence
MeaningMultiple evidence sources agree.
User interpretationUseful signal, still not a verdict.
Moderate confidence
MeaningSome support exists but context may be incomplete.
User interpretationRead with caveats.
Low confidence
MeaningThe signal is thin, ambiguous, or parser-limited.
User interpretationDo not over-weight this claim.
Insufficient evidence
MeaningThe report cannot support the claim.
User interpretationTreat as absent, not negative.
Frequently Asked Questions
What is a why-this-score explanation?
It is a short explanation of which evidence and metric families drove a score, including confidence and evidence quality.
Can public samples show private report evidence?
No. Public samples explain the format without exposing user uploads, private report artifacts, or account-scoped evidence identifiers.
Is missing evidence the same as a bad relationship signal?
No. Missing evidence means the report cannot support a conclusion.
Identify the supplied messages, sender mapping and date window supporting the observation.
02
Measure
Explain what was counted or compared, including the relevant denominator.
03
Interpretation
State confidence, limitations and plausible alternative context before drawing a conclusion.
Conceptual reading guide. This diagram explains what to inspect; it is not a customer report, diagnostic test or measured relationship score.Use this guide on the question you came with
If a report cannot show this trail, treat the claim cautiously. Specific wording does not compensate for missing evidence.