Myna · Lab · Çoruh Şimşek

A reflective diary for understanding yourself over time. The 2022 app ran a closed beta and was paused; v2 is a research-backed design that asks questions instead of giving verdicts, tested for now through my own brain-dump sessions.

Question

Can a diary help someone understand themselves over time without telling them who they are?

I started Myna so people could understand themselves better, and so they would not feel alone with what they go through. The anonymous shared feed came from group therapy: you see that others have been in the same place, and you get to look at your own situation through their eyes. Myna was the first product of a psychology AI company I co-founded, a reflective diary built with psychologists and academics.

Approach

FIG. 1Myna v2: three questions, then it stops. The v2 design, not built. After the third question Myna stops asking. The reading never appears on screen unless you ask, and each question should leave room for an answer that proves it wrong (a proposed guard).
Source: Myna sheet, Approach; v2 decision register (sources.md)

Myna 2022. A phone app. You wrote an entry, picked the emotion you felt, and the app charted your picks over time. Each day it asked one question, and only after answering could you read how others had answered, anonymously.

We ran a closed beta, then paused. The models of the time could not read a person precisely enough, and a wrong self-analysis could hurt someone.

Myna v2. I reopened it research first and made the design calls in a decision register:

	Myna 2022 app	Myna v2 (designed, not built)
Who names the emotion	You, from a fixed list of seven	You, for how you feel right now
What the system reads	Your picks, counted	Each person, event or idea in the entry, with an emotion, a thought and a tendency
What you get back	A mood chart	At most three open questions, then it closes

If someone asks about its reading, Myna explains it as a check on itself: "when you said X, I took it as Y, and I didn't want to get you wrong." Its stance: I'm listening, not judging, trying to understand.

Results

FIG. 2The proof of concept so far: brain-dump notes. 51 dated notes from the brain-dump sessions: I write freely, the assistant keeps my words as they are and analyses only when I ask. Counts only; no note is quoted.
Source: Myna sheet, Results; count of the notes folder on 2026-11-05 (sources.md)

FIG. 3Why the public dataset was dropped. The research suggested mapping Myna's emotion words onto a widely used public dataset of labelled comments; two checks of that dataset say its labels are too often wrong.
Source: Myna sheet, Results; v2 decision register (sources.md)

Myna v2 is on hold and has no code yet. Its working proof of concept is the brain-dump sessions I run with my own AI assistant.

One recommendation from the research failed against our own past: mapping Myna's emotion words onto a widely used public dataset of labelled comments. Our 2022 emotion model was trained on it, and we had found too many of its labels wrong, so I dropped it.

What I learned

The sessions showed me the cost of too much analysis. I got stuck looking at my own life in the third person, from outside, and felt cut off from myself. The research points toward distance: the strongest evidence it found for Myna's goal is a third-person diary study. But that study ran for a week or a month. It says little about distance that does not end.

That is part of why Myna v2 keeps its reading to itself. People tend to take a reading on screen as fact, and I have felt what a steady supply of readings does. Its questions let the person reach the point and say it themselves.

Open questions

For someone who builds evals: if nearly every hidden reading is confirmed by the person's answers, the questions were probably leading. Is the share of revised readings a usable measure of that?
For a designer of reflective tools: what in someone's entries would show that reflection has turned into watching themselves from outside, and what should the diary do then?