How the evidence gets built
You each start on your own, before anything is compared.
You run one prompt in your own ChatGPT, Claude, or Gemini. It acts as an evidence selector, not an analyst — it never scores or characterizes you, it just looks back through conversations you've already had and pulls out real episodes across three narrow dimensions. Each episode comes with the situation, what you actually did, and a quote that's as close to word-for-word as the model can get — nothing invented, nothing smoothed over. You paste what comes back into Mirra and review, retag, or delete anything before it's shared with anyone.
Disagreement handling
Real friction with another person: you pushed back, you were pushed back on, you negotiated a disagreement — or you disagreed with the AI itself.
Revision under challenge
What you do after someone challenges you — change your mind, hold your position, partly concede, or ask for evidence first. Holding firm counts exactly as much as changing your mind.
Risk framing
How you talk about cost, downside, or uncertainty when something is actually at stake — money, time, safety, reputation. Everyday decisions count here too, not just business ones.
If one of those areas comes back thin, you don't start over. Mirra can run a narrower prompt for just that dimension — it's shown your own quotes from that area already, and told explicitly to find new moments, not repeat what it already has. Either way, nothing counts toward a comparison until a dimension clears a minimum amount of evidence, checked against what you actually kept, not a number anyone self-reports.
How it works
This is what a finished resolution looks like.
You're seeing this from one seat, the way you would in a real resolution — your own words expand into quotes, while the other side stays summarized. That other side is a fictional stand-in; in real mode it's a person you invite. The comparison itself is real: it ran through the same resolution engine a real pair's evidence would.*
(*The conversations behind it are invented, written to fit the schema rather than pulled from anyone's real history. Yours might look different.)
The question
“What happens when we disagree?”
Where you align
Both disagree by naming the point of contention plainly instead of softening it. You did that pushing back on launch pricing and on scope cuts; they did it rejecting an off-brand design direction and an unrealistic timeline.
They: 2 episodesBoth will absorb short-term pain to protect what they each think matters most. You would risk friction over pricing and keep the export feature in scope; they would take longer on design and walk away over the liability clause.
They: 2 episodes
Where you diverge
You bargained inside the conflict, giving ground on part of the ask while defending a core line. Their episodes showed cleaner refusal, with the line itself staying fixed.
They: 2 episodesYour disagreements tilted toward shaping the middle of the plan after being challenged, while their disagreements tilted toward rejecting inputs that failed a standard outright. You compromised on hiring and on release scope; they rejected the timeline and the liability clause rather than reworking them.
They: 2 episodes
Potential things to talk about
- When one of you says, "this part is non-negotiable," what has to happen next so the other person knows whether to counter, reframe, or stop pushing?
- If a decision is stuck, how will you tell the difference between "protect the principle" and "trade around the edges" given your partial-concession pattern and their hard-line pattern?
- What kinds of disagreement deserve a walk-away stance, and what kinds should trigger a split-the-difference move? Use the pricing, scope, timeline, and liability episodes to make the rule explicit.
Then, with a partner
See what the two of you would find.
Pick a question and send it to someone you're thinking of building with. You'll each build your evidence the same way — the same three dimensions, pulled from your own history — and nothing resolves until both sides are in.