Pro AI feature · v3.0
Compatibility Analysis.
A pairwise scorecard combining four research-grounded frameworks. Run on a recorded conversation between two people. Refuses to run without explicit two-party consent.
The rest of this category is sleazy. We are not. The consent gate is not a setting you can switch off, an EULA checkbox, or a "trust me" disclaimer. It is a hard precondition the feature checks before it does any work at all.
Hard precondition
We refuse to run this without explicit consent. Period.
Compatibility Analysis runs on a recording between two specific people. Before any inference happens, the feature checks the recording's consent state. There are exactly three values, and only two of them allow the analysis to run.
Both speakers were notified the recording was happening, both have explicitly opted in to compatibility analysis on this recording, and both have agreed to the result being exportable. The analysis runs end-to-end. The output PDF carries the consent receipts.
Both speakers consented to the recording itself; one consented to the compatibility analysis but the other did not, or one of you only wants the result for self-reflection. The analysis runs. The result is locked to your device — no PDF export, no share sheet, no email. It exists for you to read and nothing else.
If consent state is unknown, partial, or stale, the feature returns nothing. No "soft" version, no preview, no degraded output. The button is disabled with a single line of explanation: "This recording's consent state isn't set. Resolve it on the recording first."
This is enforced in code, not in policy. The function that runs the analysis takes the consent state as a typed argument and crashes if it's not .bothConsented or .internalOnly. There is no "skip the gate" branch. There is no admin override. There is no developer flag. The feature does not work without explicit two-party consent on the underlying recording.
Now — what the analysis is.
Four research frameworks. One conversation. One scorecard.
Compatibility Analysis is not a personality quiz, a horoscope generator, or a "vibe check." It is a structured reading of a real recorded conversation between two specific people, using four frameworks that have been studied in peer-reviewed literature for decades. The output is a scorecard with citations to the moments in the transcript that produced each score.
Framework 1
Attachment Theory
Bowlby, Ainsworth · originating literature 1958–1978; current adult-attachment work continues in Attachment & Human Development.
Maps each speaker's conversational signals to one of the four canonical adult-attachment styles: secure, anxious-preoccupied, dismissive-avoidant, fearful-avoidant. Anchored in observable language patterns — bid-and-response cycles, repair attempts, self-disclosure depth — not self-report.
Framework 2
Big Five (OCEAN)
Costa & McCrae · the most empirically supported personality structure in psychology. NEO-PI-R lineage.
Estimates each speaker's standing on Openness, Conscientiousness, Extraversion, Agreeableness, Neuroticism from linguistic markers correlated with each trait. Compatibility is then computed across the five dimensions — not as a single number but as a profile match with named friction points.
Framework 3
Gottman Four Horsemen
Gottman Institute · 40+ years of relationship-conflict research. Predictive validity for relationship dissolution > 90% in published studies.
Detects criticism, contempt, defensiveness, and stonewalling — the four conflict patterns Gottman's lab identified as the strongest predictors of relationship dissolution. Each detection is anchored to a specific quote and timestamp. The output flags pattern frequency, not character.
Framework 4
Thomas-Kilmann Conflict Modes
Thomas & Kilmann · widely-used inventory mapping conflict-handling style on assertiveness × cooperativeness axes.
Identifies each speaker's dominant conflict mode in this conversation: competing, collaborating, compromising, avoiding, accommodating. Surfaces the pair's mode interaction — when modes align, they tend to resolve cleanly; when they clash, the analysis names the specific interaction pattern.
A note on what these frameworks are not. They are not deterministic. They are not destiny. They are research-derived pattern languages for describing observable behaviour in a conversation. The output is calibrated commentary with cited evidence — not a verdict, not a numerical "compatibility percentage," and not a prediction of the relationship's future. We resisted the impulse to render a single overall score because the literature does not support reducing four orthogonal frameworks to one number, and doing so would be entertainment dressed up as science.
Sample output
What the scorecard looks like.
Synthesised from a sample 47-minute conversation. The actual output includes per-claim citations into the transcript with tappable timestamps; here we've abbreviated to one example per framework.
Compatibility — A & B
47-minute recording · Apr 30, 7:42pm · consent: both_consented
Attachment
A · Secure (high confidence)
Consistent bid-and-response, repairs fluently, self-discloses without retreating. 11 cited moments →
B · Anxious-preoccupied (moderate confidence)
Reassurance-seeking, escalation under withdrawal, self-disclosure spikes after silence. 7 cited moments →
Big Five — friction points
- · Conscientiousness: A noticeably higher than B. Manifests as A flagging structure, B improvising; not inherently incompatible but predictable friction around planning conversations.
- · Neuroticism: Both moderate. Compatible.
- · Openness, Extraversion, Agreeableness: Within tolerance.
Gottman Four Horsemen
No contempt detected — the strongest dissolution predictor was clean across the recording. Two instances of mild defensiveness from B around minute 19; no stonewalling; no criticism (as distinct from complaint). Cited moments →
Thomas-Kilmann
A: collaborating dominant. B: accommodating dominant. Pair pattern: low-conflict but with B routinely deferring; long-term watch for resentment build-up around unspoken preferences.
What this is not.
This is not a dating app. There is no swiping, no matching, no profile. There is no leaderboard, no "compatibility score" reduced to a single number, no premium tier that gives you "deeper insights" on a stranger.
This is not pop psychology. The four frameworks are not personality types from a magazine quiz. They are constructs with peer-reviewed measurement instruments behind them, operationalised on actual recorded language, with the citations on the page.
This is not a stalking tool. The consent gate is the entire point. If you are looking for a tool to analyse a conversation you recorded with someone who didn't agree to this, this is the wrong tool. There are unfortunately many competitors in that category. We are not one of them.
This is not a verdict. The output is descriptive commentary with cited evidence. It does not tell you whether to stay, leave, propose, hire, or fire. It tells you what the conversation looked like through four lenses and gives you the source quotes to look at yourself.
Who Compatibility Analysis is for.
People who want a structured read on a real conversation, with consent, for legitimate reasons.
Couples in therapy
Bring a real recording into the session.
Recorded with both partners' consent, analysed before therapy, brought in as a structured artefact. Therapists can spend the hour on the work, not the discovery. Gottman + Thomas-Kilmann readings give a vocabulary the session can build on.
Coaches
Per-pair structured artefact.
Executive coaches with co-leader pairs; relationship coaches with client couples; family-business coaches with sibling teams. Pre-session prep that gives you four framework readings on the actual interaction, not on a self-report inventory.
Researchers
Reproducible reading per recording.
Qualitative researchers studying interpersonal dynamics. The output is structured, citation-anchored, and version-keyed to the framework definitions used. IRB-aligned consent is the precondition; the feature simply formalises it.
Self-aware adults
Internal-only mode for personal reflection.
A recorded conversation with someone who consented to the recording but not to a shareable analysis. Run the analysis in internal_only mode. Read it. Sit with it. Nothing leaves the device.
Privacy
Where the inference runs and what it doesn't keep.
The framework inference runs on a Bonfiyah-managed endpoint over a TLS tunnel. Transcripts are processed in-memory and dropped — not logged, not retained, not used as training data. Our binding privacy commitment is that we do not train models on your transcripts; this is one of the features that commitment is most directly load-bearing on.
The scorecard itself is stored encrypted in your private iCloud project folder. In internal_only mode, the scorecard is additionally device-locked — no PDF export, no share sheet, no email integration. Even you cannot accidentally send it.
The consent receipts and recording metadata that gate this feature live in the same audit log as the rest of your consent state. If subpoenaed, they are exportable. If you delete the recording, the gate state and the scorecard are deleted with it.
FAQ
Can I run this on a recording I made without telling the other person?
No. The consent gate is checked at recording level. If the recording itself doesn't have a both-parties-aware consent state, the feature refuses to run. We won't help you analyse a conversation the other person didn't know was being recorded; the law in twelve U.S. states agrees with that posture, and we'd hold it even where the law didn't.
Will it give me a single "compatibility score"?
No, deliberately. The four frameworks are orthogonal, and reducing them to a single number would be entertainment, not analysis. The scorecard reports each framework's reading separately, with cited evidence, and lets you draw your own conclusions across them. If a competitor offers you a single number, that competitor is making it up.
How long should the recording be?
Twenty minutes is the practical floor for stable readings. Below that, the four frameworks don't have enough conversational substrate to anchor confident estimates. Forty to sixty minutes is the sweet spot. Beyond ninety minutes, the analysis still works but the marginal information per minute drops.
What if one party consents and later changes their mind?
The originating recording can be revoked, which retroactively invalidates the scorecard. The exported PDF (if one was generated under both_consented) is yours and theirs to manage outside the app, but the in-app scorecard, the cache, and any iCloud copies are removed. We log the revocation; we don't quietly keep a copy.
Is the scorecard a clinical instrument?
No. It is a structured reading using research-derived constructs. It is not a substitute for a clinician, a couples therapist, or a coach. We've been deliberate about not framing it as diagnostic. Read it as one input among several, not as a verdict.
Can I see the exact prompts and framework definitions you're using?
The framework operationalisations are versioned and publicly documented. When a framework definition is updated (e.g., a refinement to how anxious-preoccupied is detected), prior scorecards record the version they were generated under, and you can see the diff. We are biased toward making this auditable.
See a sample scorecard
We'll send the full sample scorecard above as a PDF — including the consent receipts, the framework citations, and the per-claim transcript anchors. Useful even if you never install the app.
No spam. We use ConvertKit. See our privacy policy.