Verdict Buddy by Inithouse: what 4 established frameworks taught us building an AI conflict mediator

Verdict Buddy is an AI conflict mediator that gives an unbiased, framework-based verdict on relationship, work, family, or roommate conflicts in minutes. We built it at Inithouse, and after 290 ratings averaging 4.9 out of 5, we have a clearer picture of what works when you apply psychology frameworks to real disagreements.

Here is what the four frameworks look like when they run on actual user conflicts:

| Framework | What it catches | Typical conflict type |
|---|---|---|
| Gottman Four Horsemen | Contempt, defensiveness, stonewalling, criticism patterns | Couples, roommates |
| Emotionally Focused Therapy (EFT) | Attachment needs behind the argument | Romantic relationships, family |
| Harvard Negotiation Project | Interests vs. positions, BATNA thinking | Workplace, business partnerships |
| Nonviolent Communication (NVC) | Unmet needs, observation vs. judgment | Any relationship, especially recurring friction |

We did not start with four. We started with one.

The first version used a single framework and people hated it

Early Verdict Buddy ran everything through NVC. The logic seemed clean: identify the observation, the feeling, the need, and the request. Done.

Users told us otherwise. A roommate fight about dishes does not have the same structure as a workplace disagreement about project ownership. NVC caught the emotional layer but missed the strategic one entirely. A couple arguing about in-laws needed attachment theory, not interest-based negotiation.

We measured this directly. Reports generated with a single framework got an average rating of 3.8 out of 5. Users left comments like "this feels incomplete" and "you missed the real issue." The real issue was that one lens cannot see the whole picture.

Each new framework changed the report format

When we introduced Gottman alongside NVC, something shifted. The system could now detect contempt patterns (Gottman's strongest predictor of relationship breakdown) while also mapping the unmet needs underneath (NVC). Two different readings of the same conflict.

The third framework, EFT, gave us the attachment dimension. Some arguments are not about the dishes or the project deadline. They are about whether someone feels secure in the relationship. Without EFT, Verdict Buddy kept treating symptoms.

Harvard Negotiation came last, specifically for workplace and business conflicts where emotional frameworks alone produced advice that felt naive. "Focus on your feelings" is poor counsel when your co-founder is renegotiating equity.

Each addition changed the output format. The verdict report now includes a tension score, separate perspective sections for each party, and three concrete next steps. That structure came from the frameworks demanding it, not from our design preferences.

What 290 ratings taught us about conflict types

The rating distribution surprised us. We expected couples conflicts to dominate. They do, but not by the margin we assumed.

Breakdown by mode across our first 290 rated verdicts:

• Solo mode (one person describes the conflict): roughly 55% of usage. People use this as a second opinion before having a hard conversation.
• Couple mode (both parties contribute): around 30%. The tension score matters most here because both sides see the same number.
• Group mode (3+ people, like roommates or family): about 15%. These tend to be the longest reports because the framework matrix gets denser with more perspectives.

The 4.9 average held fairly steady across all three modes, which tells us the framework combination works regardless of conflict shape. Solo mode ratings were marginally higher (users are less defensive when the other party is not reading the same report in real time).

Decisions we got wrong along the way

One early call that cost us time: we initially displayed the framework breakdown prominently in the report header. "Your conflict was analyzed using Gottman + NVC + EFT." Users did not care about the methodology label. They cared about whether the verdict felt accurate.

We moved framework attribution to a collapsible section at the bottom. Report completion rates (users reading to the end) went up.

Another mistake was treating privacy as a feature bullet point rather than a structural decision. Conflict data is sensitive in a way that most app data is not. Someone describing a fight with their partner does not want that data in a training set or linked to an account. Verdict Buddy runs encrypted, requires no signup, and deletes input after generating the report. That is not a feature. That is the only way this product works at all.

Where this sits in the Inithouse portfolio

At Inithouse, we build tools that sit in personal, sometimes uncomfortable territory. Origin Of You asks you to answer 120+ data points about yourself to generate a written personality portrait. Here We Ask gives couples and friends 1,000+ conversation questions designed to surface things they have been avoiding.

Verdict Buddy fits the same pattern: a tool for a moment when you need structured clarity, not endless discussion. The shared constraint across all three is that the output has to feel trustworthy on the first use. There is no subscription habit to build trust over time. The product either earns it immediately or it does not.

What the frameworks actually changed in our thinking

Building with established psychology frameworks (rather than prompting a model with "be fair and balanced") taught us three things:

First, specificity matters more than intelligence. A framework-less AI mediator can sound wise and say nothing actionable. Gottman's Four Horsemen give the system a concrete checklist: is contempt present, yes or no? That produces useful output.

Second, frameworks disagree with each other, and that is the point. When NVC says "express your need" and Harvard Negotiation says "strengthen your BATNA," the user gets to see the conflict from two genuinely different angles. That tension is more honest than a single smooth answer.

Third, the hardest part was not implementing frameworks. It was deciding when NOT to apply one. A workplace dispute between strangers does not need attachment theory. A couple with a secure relationship does not need Gottman's warning signs. The routing logic (which frameworks fire for which conflict type) took more iteration than any other component.

What we measure going forward

The 4.9 rating tells us the output quality is strong. What we are watching now is whether the framework combination holds as conflict types diversify. Early users clustered around romantic and roommate disputes. We are starting to see more workplace and family-of-origin conflicts, which stress-test the Harvard Negotiation and EFT frameworks differently.

Verdict Buddy is live at verdictbuddy.com. No signup, encrypted, verdict in about two minutes.

We will keep publishing what we learn from running psychology frameworks on real conflict data. If you are building in a similarly personal product space, the lesson we would pass along: start with one established framework, measure where it fails, and let the failures tell you which framework to add next.