How Grey Mirror turns a message history into a report.
Grey Mirror is a relationship communication analysis product developed by Wentropy Labs and available through JustLayMe. It performs full-thread relationship text analysis with evidence-backed metrics, timeline windows, participant resolution, and visible confidence bands. No single-message guessing, no hidden interpretations, no fabricated precision.
Message order and participant attribution stay connected
Full thread
Important findings identify supporting windows and metric drivers
Evidence
Uncertainty remains visible when the record is incomplete
Limits
How does Grey Mirror analyze relationship texts?
Grey Mirror is developed by Wentropy Labs and available through JustLayMe. It preserves speaker sequence, timestamps, and full timeline from your exported conversation. It then measures observable communication patterns including repair quality, timing drift, reciprocity, effort balance, planning language, and recurring loops. Every metric is evidence-linked with confidence bands, never guessing private intent.
Communication analysis with published methods
Psychology-informed relationship analysis, with metrics and supporting messages you can examine. Grey Mirror is developed by Wentropy Labs and available through JustLayMe, the consumer web platform Wentropy Labs operates.
How much analysis goes into a report?
Grey Mirror has 30 analyzers built on up to 47 models. Full-history analysis compares both participants across communication patterns, with supporting messages and stated limitations.
Analyzer counts describe configured analysis components. They do not count unique trained models, psychological frameworks or report pages. The metrics library publishes every component and its scope.
What full-thread analysis can reveal
A complete conversation export reveals patterns, timing shifts, and repair cycles that isolated screenshots cannot show. Keep the dates and both sides of the exchange so you can check the comparison.
Response-time drift and silence patterns across the entire timeline
Initiation balance and who consistently drives the conversation
Repair attempts versus rupture events and their success rates
Recurring communication loops and their resolution outcomes
What sits behind a report
Proprietary analysis built for relationship messages, with evidence you can open.
Grey Mirror reads the complete history with proprietary analysis systems built for relationship communication, compares both people across many behavioral dimensions, and connects every supported finding to the messages behind it. Plain-English explanations describe measurements that already exist.
A taxonomy built for this job
Messages are read against a proprietary behavioral and emotional taxonomy covering communication acts, emotion, needs and boundaries, repair, risk, tactics and topic, rather than one general prompt.
Trained on relationship text
The analysis is built specifically for relationship messages, not adapted from a general-purpose chat assistant.
Evidence is found, not invented
Each finding is tied to the specific messages in your thread that support it, so you can open them and judge for yourself.
Measured against a real corpus
Reported benchmarks come from 4,600,611 messages across 29 de-duplicated relationship histories, published with the measures that were computed and deliberately withheld.
Where it is weakest, stated plainly
Per-label performance is uneven: frequent communication acts classify far more reliably than rare ones, and the model limitations page names the categories where confidence is low rather than hiding them.
Why screenshots are insufficient for relationship analysis
A single screenshot shows what someone said but hides the context that gives it meaning. Full-thread analysis reveals whether a message is an isolated incident or part of a measurable pattern.
Missing sequence
Cannot see what came before or after to understand the context
No timing patterns
Cannot detect response-time drift, silence gaps, or timing-based signals
Hidden repetition
Cannot identify recurring patterns without the full timeline
What inputs produce the strongest analysis?
The strongest input is a complete exported thread with stable message order. Platform exports from iMessage, WhatsApp, Messenger, Telegram, Signal, and SMS are supported.
Screenshots remove sequence, timing, and recurrence. Full exports preserve those signals when the source format contains them.
Message text or structured rows from your export
Participant labels that map to real people in the conversation
Timestamps precise enough to calculate response gaps
Enough coverage to compare early, middle, and recent periods
What happens during normalization?
Normalization converts platform-specific formats into a consistent analysis structure without losing evidence. This step preserves who said what, when, and in what sequence.
Parse the uploaded text, CSV, JSON, HTML, or structured file
Preserve every readable message row without sampling
Standardize timestamps across platforms and timezones
Resolve and verify participant sender labels
Flag malformed rows with diagnostics for user review
How participant resolution protects accuracy
Participant mapping is critical because sender labels drive all timing, effort, reciprocity, and power metrics. Incorrect mapping can invert the story entirely.
Sender labels determine who initiated, who responded, who repaired
Windowing divides long conversations into comparable phases. This allows Grey Mirror to detect trajectory: is warmth increasing, stable, or declining? Are repair attempts becoming more or less effective over time?
Compare several periods to see whether the pattern changed.
Compare early-relationship patterns versus late-relationship patterns
Detect timing drift: are responses getting faster or slower?
Identify repair cycles: do arguments resolve or repeat?
Measure reciprocity: is effort balanced or one-sided?
What Grey Mirror measures
Grey Mirror focuses on observable, measurable communication dimensions that can be extracted from message content and timing.
Evidence linking and confidence bands
Every metric is backed by specific evidence from your conversation. Grey Mirror shows you exactly which message windows support each claim and displays confidence bands instead of fake precision.
When evidence is thin, confidence bands widen. When ambiguity exists, uncertainty is flagged. Never hidden.
Message-level evidence
Specific lines and references when payload allows them
Window-level evidence
Time ranges where patterns appear repeatedly
Metric-level evidence
Counts, deltas, and confidence labels explaining each score
Synthetic example: How patterns emerge from sequence
Consider this fabricated teaching example: after each synthetic argument, one participant sends a repair message within 2 hours, and the other responds within 4 hours. The pattern repeats 7 times across 3 synthetic months, with 6 exchanges returning to a calmer baseline.
Pattern detected: repair-and-response cycle
Frequency: 7 occurrences over 3 months
Accepted repairs: 6 of 7 synthetic examples
Timeline: consistent 2-4 hour response window
Confidence: high due to repetition and clarity
Psychological research behind the questions
Repair, reciprocity and language coordination are established relationship-research topics. Grey Mirror uses these concepts to frame questions about a conversation, then evaluates its own text-based implementations separately.
These sources provide research context. They do not establish endorsement, clinical validation or an individual review of this product by the cited researchers.
Repair and conflict: Gottman research
The <a href="https://www.gottman.com/about/research/" target="_blank" rel="noopener noreferrer">Gottman Institute research overview</a> describes studies of couples and interaction patterns. It informs the questions behind repair and positive/negative signal measures. A text classifier is not the same instrument as direct observation in those studies.
Language coordination: Ireland and colleagues, 2011
<a href="https://doi.org/10.1177/0956797610392928" target="_blank" rel="noopener noreferrer">Ireland et al., Psychological Science (2011)</a> studied language-style matching in speed dates and couples’ instant messages. This provides context for investigating language coordination; it does not validate Grey Mirror’s scores or predict the outcome of an individual relationship.
Our own measures need their own evidence
The metrics library defines the configured analyzers and measurement limits. Our public benchmarks identify the evaluated task, dataset and withheld measures. Research findings, synthetic detection tests and model-retrieval results answer different questions and must remain separate.
Route-aware interpretation
The same communication pattern can mean different things in different relationship contexts. Grey Mirror adjusts interpretation based on whether the conversation is romantic, platonic or family.
Full-thread analysis is powerful because Grey Mirror measures observable communication behavior with confidence labels, evidence windows, and traceable report outputs.
Grey Mirror reveals what the text shows and how strongly the thread supports the finding.
Evidence-linked communication metrics
Confidence labels for sensitive or sparse findings
Participant mapping and parser quality checks
No fabricated evidence when data is missing or ambiguous
Timeline windows for repair, timing, effort, and recurrence
Trust-page routing for safety or professional-support context
Why it mattersReveals engagement shifts and priority changes
Initiation balance
What it measuresWho starts conversations more often
Why it mattersShows effort distribution and interest level
Repair attempts
What it measuresFrequency and quality of repair messages
Why it mattersIndicates conflict resolution ability
Reciprocity
What it measuresResponse rate to questions and statements
Why it mattersMeasures mutual engagement and respect
Silence gaps
What it measuresDuration of non-response periods
Why it mattersReveals disengagement or external pressures
Planning language
What it measuresFuture-oriented statements
Why it mattersSignals commitment and forward focus
Turn-taking
What it measuresMessage exchange rhythm
Why it mattersIndicates conversation flow and engagement quality
Frequently Asked Questions
Does Grey Mirror read every message in my export?
Yes, the goal is full-fidelity analysis of every countable parsed message. Damaged or unreadable rows are flagged for review, never silently sampled into a weaker report.
Why does participant mapping sometimes pause processing?
Because every finding compares two people. If it is not clear who sent what, the report could be misleading, so Grey Mirror asks you to confirm rather than guess.
What makes Grey Mirror different from a chatbot or summary tool?
Grey Mirror combines proprietary relationship analysis, full-history measurement and evidence-linked reports. General assistants and other products have different input limits and outputs; compare their current documentation and results.
Can Grey Mirror detect sarcasm or jokes?
Sarcasm and jokes create ambiguity that reduces confidence scores. The system flags uncertainty rather than guessing intent. Cultural context and inside jokes are particularly challenging.
How does Grey Mirror handle deleted or missing messages?
Deleted messages create gaps that reduce confidence in timing-based metrics. The system flags missing context and reduced evidential support rather than fabricating data.
How long does Grey Mirror retain my conversation?
Retention depends on the current upload, report, and account workflow. Review the privacy and data-deletion pages for the current policy and available deletion controls.
What platforms does Grey Mirror support?
iMessage, WhatsApp, Messenger, Telegram, Signal, SMS, and other text-based platforms. We support TXT, CSV, JSON, HTML exports, and ZIP archives.
How do I export my messages for analysis?
iPhone Settings does not include a native conversation-export command. Use a lawful export method that preserves sender labels, timestamps, text, and sequence; WhatsApp provides an in-app Export Chat flow, while Android export steps depend on the chosen tool.
Read the supplied format while keeping sender evidence, chronology and available coverage intact.
02
Measure the pattern
Use defined counts, comparisons and time windows rather than treating generated prose as a measurement.
03
Explain with limits
Tie conclusions to evidence and disclose uncertainty, missing context and unavailable results.
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
Inspect the method behind the result you want to use. A component inventory tells you what exists, not that every conclusion is equally reliable.