Memory / Recovery

An interactive research companion · 30 September 2026

When a local rule forgets the whole.

A frozen table of tiny decisions turns a five-cell glider into a trail painter. We damage the pattern, replay the control, and ask what “recovery” can mean when memory is not part of the rule.

Built for the saved Jev 1.13.0 recording and offline Life replays. The animated board below is a classical Conway control or a clearly labelled toy intervention; it never makes a provider call.

Same grid.
Different memory.

five live cells
one local rule
many futures
“It ultimately takes only very simple rules to produce behavior of great complexity.”

01 / The saved record

Before we call it recovery, we have to name the control.

These figures come from immutable offline evidence and controlled replay lanes. They are counts and replays of saved artifacts, not a live benchmark and not a biological claim.

Frozen Jev tablerecorded
231 / 512

Local neighborhoods matched Conway

The saved `jev-1.13.0` table answered 231 of 512 binary neighborhoods correctly. Separate count controls scored 18 / 18.

Source · life-rule-repair / REPORT.md
Damage censusreplayed
10 / 200

Moving cases under exact + ID9

In the fixed wave 3 sweep, the exact+ID9 intervention produced 10 moving gliders across 200 labelled damage cases: 2 removals and 8 additions. On the matched bitmap comparison, exact produced 8 / 200; all eight persisted and two more removals recovered.

Source · wave-3-recovery-results.md
Mechanism checkbounded
0 ↔ 1

One local input, two demands

The same neighborhood ID9 must stay dead on a healthy path yet birth a neighboring cell in one damaged recovery. A memoryless table cannot do both exactly.

Source · mechanism-clarification-2026-09-30.md

Levin / context, not a quotation. Michael Levin’s work makes it natural to ask how local agents coordinate toward a larger form. This lab stays smaller and stricter: a finite digital board, declared rules, fixed damage, and stored evidence.

02 / Replay the question

Damage the grid. Compare the next step.

Choose a rule or control, then use the controls. The board is a synchronous Conway replay unless the label says otherwise. “Stored target” is deliberately circular; “Local variant” is an illustrative rule modifier.

Classical repair · Conway B3/S23
Generation
0
Population
5
Damage
armed
Control match
exact
live celldamaged sitelocal variant

03 / Four ways to read the same word

“Recovery” changes meaning when the evidence lane changes.

Use the tabs as a small editorial index. The numbers stay tied to their saved source; the illustrative lanes stay labelled as illustrations.

“Every beautiful pattern has a receipt. Check the table.”
Original AlexAI aphorism · evidence card

04 / Receipts before romance

What is here, and what is still open.

This companion is intentionally local. It carries the measurements needed to read the paper, while leaving provider credentials, request bodies, and private runtime details out of the public surface.

Saved evidence

Every source label below points to an immutable artifact outside this standalone file. The demo uses no CDN, no API key, and no remote fetch.

  • Frozen Jev recordingjev-1.13.0 · sha256 65176b3c…
  • Wave 3 recovery census4,000 labelled one-cell edits
  • Mechanism clarificationID9 healthy 0 / recovery 1
  • Replay coverage7,312 validated trajectories
  • Replay statuszero new provider calls

Open questions

A fixed memoryless 3 × 3 table can recover a specific next phase in one saved case, but the local conflict blocks exact preservation and repair simultaneously. Larger spatial context, temporal memory, and independent held-out tests remain open.

“A tidy rule can still make a spectacular mess.”
Original AlexAI aphorism · closing card