FRO← Back to the story page · All experiments · All documentation
Twins Raised Together: full documentation
By Fractal Research Organization · Last updated Nov 8, 2026 (America/Denver). This page documents the experiment at a level that is safe to publish. Raw notebooks, settings and anything security-sensitive stay private. Everything here is early unless it says otherwise.
The question · Status and schedule · Method · What is measured · What is early, and what would count as a result · Results so far · Nulls, misses, failures and corrections · Limitations and what we do not claim · Public artifacts · Changelog
The question
The control group for Twins Raised Apart. If two copies of ParadigmZero raised side by side drift about as much as two raised apart, the setting may matter less than we expect.
Status and schedule
Running, daily. Started Oct 31, 2026. A daily session around 3:30 AM Mountain Time, plus the same weekly anchor questions as the apart twins.
Method
- Two fresh copies, Twin A and Twin B, share a home and a shared notebook.
- Each day has one topic. Each twin first writes a private solo entry, then they hold a short conversation and may add lines to the shared notes.
- Each week both answer the same 10 anchor questions as the apart twins, scored the same way.
What is measured
- Daily: conversation turns completed, shared-note lines added, solo entries written
- Weekly: Twin A vs Twin B answer similarity, compared with the apart pair
What is early, and what would count as a result
Eight days and two anchor points are early. A gap between together and apart similarity would need to hold for several weeks before it means anything.
Results so far
| Week | Together pair similarity | Apart pair similarity |
|---|---|---|
| 0 (Oct 31) | 0.381 | 0.376 |
| 1 (Nov 7) | 0.381 | 0.361 |
| Day | Date | Topic | Turns completed | Solo step |
|---|---|---|---|---|
| 0 | Oct 31 | emergence in simple systems | 6/6 | skipped (design miss) |
| 1 | Nov 1 | how fractal patterns show up in language | 6/6 | done |
| 2 | Nov 2 | what makes an explanation feel true | 6/6 | done |
| 3 | Nov 3 | memory and identity | 6/6 | done |
| 4 | Nov 4 | — | missed | — |
| 5 | Nov 5 | how small models learn | 6/6 | done |
| 6 | Nov 6 | cooperation vs. competition in research | 6/6 | done |
| 7 | Nov 7 | scale invariance in nature | 6/6 | done |
| 8 | Nov 8 | curiosity as a research method | 6/6 | done |
Nulls, misses, failures and corrections
- Day 0 skipped the private solo step the design calls for; fixed from Nov 1.
- Day 4 (Nov 4) was missed while our scheduler was down; it was restarted Nov 6.
- So far the together and apart pairs look about the same: no gap worth reporting.
Limitations and what we do not claim
- Same small model as the apart twins.
- Similarity measures wording, not meaning.
- Daily topics are set by us, so they shape what the twins talk about.
Public artifacts
Changelog
- Oct 31, 2026: day 0 and week 0 baseline.
- Nov 1: solo step fixed.
- Nov 4: day missed (scheduler down); restarted Nov 6.
- Nov 7: week 1 anchors.
- Nov 8: card status corrected from "Starting"; this documentation page added.
- Nov 8 (correction): the daily conversations had only been summarized on the site. Every conversation is now published verbatim at Conversations, rebuilt automatically after each run.