Grey Mirror can measure text patterns. It cannot know the whole relationship.
Relationship messages are sensitive and incomplete. Grey Mirror can organize observable communication patterns, but it cannot diagnose, prove intent, replace professional help, or see events that never appear in the upload.
Grey Mirror cannot know private intent, clinical conditions, legal facts, off-platform events, deleted messages, sarcasm with certainty, or whether someone loves you. It can only analyze the message history it receives.
What is the intended use?
Grey Mirror is intended for reflection on communication patterns in user-supplied message histories. It is a decision-support and self-understanding surface, not a replacement for human judgment.
What uses are out of scope?
The product should not be used as a diagnostic, legal, emergency, stalking, harassment, or coercive decision tool.
No clinical diagnosis or mental-health labeling.
No legal conclusions or evidence handling advice.
No emergency response or abuse-safety planning substitute.
No proof of love, cheating, intent, or compatibility.
Why does missing context matter?
A message file may miss phone calls, in-person conflict, deleted lines, other apps, cultural nuance, shared jokes, or major life events. Those gaps can change the interpretation.
How can bias or error appear?
Errors can come from parser mistakes, wrong participant mapping, language ambiguity, sentiment misreads, uneven data, or metric thresholds that do not fit a specific relationship context.
How should users interpret confidence?
Confidence is a reliability label, not moral certainty. Even high-confidence text patterns still require human interpretation and real-world context.
What makes a limitation useful instead of vague?
A limitation is useful when it tells the user exactly what is missing and how that missing context could change the read. “AI can be wrong” is too generic; parser confidence, timestamp quality, participant mapping, and missing off-platform context are specific.
When should the report slow down?
Grey Mirror should slow down when the output could become more confident than the evidence. That means using neutral wording, lowering confidence, asking for participant confirmation, or routing the user to human support language when the topic is sensitive.
If the parser cannot preserve the timeline, timing metrics should be limited.
If only one side appears in the export, reciprocity and effort claims should be narrow.
If the thread includes self-harm, threat, or safety language, the product should prioritize safety resources over pattern commentary.
If a user wants a verdict about another person, the report should reframe toward observable communication behavior.
What makes a limitation useful instead of vague?
Limit
What it can distort
Better user read
Deleted or missing messages
What it can distortRecurrence, repair aftermath, and silence gaps.
Better user readTreat trends as partial when the thread has visible holes.
Wrong participant labels
What it can distortEffort balance, response timing, and power signals.
Better user readResolve speakers before trusting participant-specific claims.
Sarcasm or private jokes
What it can distortSentiment, pressure, affection, and humor.
Better user readCheck whether the surrounding pattern supports the tone read.
Off-platform conflict
What it can distortWhy a shift started and whether repair happened elsewhere.
Better user readUse the text as a signal record, not the whole relationship record.
Frequently Asked Questions
Can Grey Mirror diagnose abuse?
No. It can surface concerning text patterns and recommend appropriate support language, but it cannot diagnose abuse or replace crisis, legal, or clinical help.
Can Grey Mirror detect sarcasm?
Sometimes it can infer tone from context, but sarcasm, irony, and private jokes remain common failure modes.
Can Grey Mirror tell me what to do?
It can summarize patterns and suggest reflection points, but final decisions require human judgment and real-world context.
An unavailable result is better than a fabricated one
Visual guide updated
01
Supported input
A measure depends on the format, context and evidence it was designed to use.
02
Uncertain interpretation
Ambiguity, language variation and missing offline events limit what the output can explain.
03
Unavailable finding
When the required evidence is missing, a report should state that boundary rather than invent a confident result.
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
Read each limitation alongside the finding it qualifies. An empty result can be an honest reflection of the evidence available.