All apps

Past Self Debate

Fight night · Manchester vs Mexico City

Argue with the person you were in 2015.

Import your old posts. Your past self talks back in your own words. A judge scores what changed, and why.

"I can't believe I posted that. I also can't remember why I stopped believing it."Every archive, ever

Round 1 preview

Sample archive: Noor Rahman is a fictional person.

Bring your archive

Every big platform lets you download your data as a .zip. It is parsed on this phone; only the posts you pick are sent to the model.

Upload my archive (.zip or .json)

2015 archive · 412 posts

Pick a fight with yourself

Work2015 Noor vs 2026 Noor

Round 1Round 2Round 3

Growth map · 2015 to 2026

How your views moved

Each line runs from what 2015 you said (bottom) to what you say now (top). Dots are real later posts from your archive. Tap one.

Biggest mover

The grand mission

Change your mind on purpose.

People rarely see how much they changed, so they think nobody can. Arguing with your own past self is the gentlest proof that minds move, and the best practice for being fair to people who disagree with you now.

Year 1

Debate your past self from any platform archive, judge scores, a growth map and the weekly "ten years ago" post.

Year 2

Every-year mode: 2018 you joins the ring as a referee between 2015 and today. Couples and old friends can debate their shared past posts, with both people's consent.

Year 3

A lifelong "mind changelog": the record of what changed your views and why, which you can choose to pass on to your kids as the most honest memoir there is.

Blueprint

Why this is new

"On this day" memory features and old-post resurfacing show you what you said and stop there. Chatbots can role-play anyone. Nothing builds a sparring partner from your own words at one point in time, lets you argue turn by turn, then has a neutral judge explain what changed in you and what caused it, using your later posts as evidence.

The data it runs on

Source: the user's own data export. X, Facebook and Instagram all offer a "download your information" archive (zip with JSON or HTML inside). Reddit and Tumblr offer exports too.
Access: the user uploads the zip. Unzipping and parsing run on device. Posts are bucketed by year and topic locally; only the chosen year's posts on one topic (about 10 to 40 short posts) go to the model.
Limits and gaps: exports take hours to days to arrive. Formats change without notice. Deleted posts are gone. Many people did not post opinions, only photos, so some years are too thin to debate.
Fallback: paste a handful of old posts by hand, or answer 10 quick "what did you believe at 22?" prompts to build a past self from memory.

1. The mission

Help people see how and why they changed, by letting them argue with who they were. Being fair to your past self is practice for being fair to everyone else.

2. Why people can't resist it

Cringe and affection. Old posts are embarrassing and tender at once. Hearing them talk back is irresistible.
Winning, sort of. It is a debate you can win, but the judge also rewards being kind to the loser.
Self-knowledge with a score. "What changed you" written as a verdict feels like a reading about yourself, and people share those.

3. How it works for real

Parse. On-device parsers for each export format (JSON for X and Facebook, HTML for older Instagram) produce dated posts.
Topic and stance. A fast cheap model (Gemini Flash or DeepSeek via OpenRouter) tags each post with a topic and a stance, and finds later posts on the same topic.
Past self persona. A frontier LLM (Claude or GPT) is prompted with only that year's posts plus style stats (slang, punctuation, confidence) and must stay in that year: no knowledge after it.
Judge. A second frontier-model call reads the whole debate plus the later-post timeline and writes the verdict, the cause of change and the scores.
Growth map. Stance scores per year are plotted from the tagged posts, no extra model calls.

4. Cost per use

One debate is about $0.05 (persona turns plus judge). Tagging a full archive once is about $0.02 with a cheap model.
Free: 3 debates a month and the weekly "ten years ago" post. Paid ($4 a month): unlimited debates, any year, growth map across all topics, letters to future you.

5. Growth loop

The share card pairs a cringe 2015 quote with a wise 2026 line, which is perfect for posting. Friends who were tagged in those years ask to debate their own past selves. Tagged old posts can invite the friend into a "debate our 2015" duo mode.

6. Risks and guardrails

Privacy: the archive is parsed on device and never stored on servers. Other people's names in posts are masked before any model call.
Consent: only the archive owner can debate it. No uploading someone else's archive; ownership is checked against the export's account name.
Harm: old posts can include painful periods (grief, illness, abuse). Those are detected and excluded from debate topics by default.
Hate: if old posts contain slurs or hateful views, the past self will not repeat them; the judge frames it as growth, not entertainment.

7. MVP build order

X and Facebook export parsers plus topic tagging.
Persona prompt locked to one year, with chat turns.
Judge call and verdict screen.
Share card, then the weekly "ten years ago" unlock and growth map.