Evolution run viewer

One complete run of LLM-driven program evolution: an archive of Python programs, scored on fixed inputs, rewritten generation after generation by an LLM mutation operator. Every candidate is here — including the failures, which are data too. Inspired by Sakana AI's ShinkaEvolve.

This is a mock-mode run (random text edits instead of a real LLM) so the pipeline is public end-to-end without API costs. A live LLM run replaces it via one command — the viewer renders either identically.

candidates logged

accepted into archive

rejected / crashed

LLM calls (budget 30)

task tour · mode mock · best seed -551.6 → best program -533.9 · held-out -567.0

Fitness over generations

Each dot is an accepted candidate (fitness over fixed train inputs); the dashed-to-solid line is the running best. Red marks on the floor are generations where the proposal was rejected or crashed.

Best program & lineage

The winner is ed32b0d4, descended from seed 2dd1aa9a through 8 mutations:

02dd1aa9agen 0-551.6
132b9b7b8gen 3-538.9
2d31903b3gen 4-547.7
3a8c3d609gen 7-547.4
4af24e13dgen 8-567.9
5ca14dbe9gen 15-547.5
6681b1065gen 17-553.8
73b006102gen 20-558.1
8ed32b0d4gen 22-533.9best

import math

def solve(points):
    start = 5
    unvisited = list(range(len(points)))
    unvisited.remove(start)
    order = [start]
    current = start
    while unvisited:
        nxt = min(unvisited, key=lambda i: math.dist(points[current], points[i]))
        unvisited.remove(nxt)
        order.append(nxt)
        current = nxt
    return order

Failures are data

14 of 32 proposals never made it into the archive — and they are kept in the log, because why evolution failed matters as much as what it found.

duplicate×12e.g. identical program (AST) already in archive

crashed×2e.g. SyntaxError: invalid syntax

Full log

gen	id	parent	status	score	held-out	mutation
0	56fb4034	—	accepted	-1528.7	-1534.8	seed-random
0	2dd1aa9a	—	accepted	-551.6	-560.2	seed-nn
1	8c880b37	2dd1aa9a	accepted	-553.5	-564.5	MOCK: nudged an integer constant (e.g. start city / range bound)
2	2fcdf3de	8c880b37	accepted	-561.8	-563.4	MOCK: nudged an integer constant (e.g. start city / range bound)
3	32b9b7b8	2dd1aa9a	accepted	-538.9	-570.5	MOCK: nudged an integer constant (e.g. start city / range bound)
4	d31903b3	32b9b7b8	accepted	-547.7	-577.2	MOCK: nudged an integer constant (e.g. start city / range bound)
5	6d88d8bf	32b9b7b8	duplicate	—	—	MOCK: nudged an integer constant (e.g. start city / range bound)
6	0d07a758	32b9b7b8	duplicate	—	—	MOCK: nudged an integer constant (e.g. start city / range bound)
7	a8c3d609	d31903b3	accepted	-547.4	-570.5	MOCK: nudged an integer constant (e.g. start city / range bound)
8	af24e13d	a8c3d609	accepted	-567.9	-570.8	MOCK: nudged an integer constant (e.g. start city / range bound)
9	d7919da6	af24e13d	accepted	-561.0	-574.6	MOCK: nudged an integer constant (e.g. start city / range bound)
10	92c7db64	2fcdf3de	duplicate	—	—	MOCK: returned parent unchanged to test duplicate detection
11	02f4cd28	32b9b7b8	crashed	—	—	MOCK: broken code to test rejection
12	57532c71	8c880b37	accepted	-551.4	-550.2	MOCK: nudged an integer constant (e.g. start city / range bound)
13	66afb869	a8c3d609	duplicate	—	—	MOCK: returned parent unchanged to test duplicate detection
14	1ba2efa5	d31903b3	crashed	—	—	MOCK: broken code to test rejection
15	ca14dbe9	af24e13d	accepted	-547.5	-579.1	MOCK: nudged an integer constant (e.g. start city / range bound)
16	cc4f6f2f	57532c71	duplicate	—	—	MOCK: nudged an integer constant (e.g. start city / range bound)
17	681b1065	ca14dbe9	accepted	-553.8	-566.0	MOCK: nudged an integer constant (e.g. start city / range bound)
18	cdbf0407	d7919da6	accepted	-537.0	-571.5	MOCK: nudged an integer constant (e.g. start city / range bound)
19	7a855e21	a8c3d609	accepted	-564.5	-569.0	MOCK: nudged an integer constant (e.g. start city / range bound)
20	3b006102	681b1065	accepted	-558.1	-556.1	MOCK: nudged an integer constant (e.g. start city / range bound)
21	2e988920	d31903b3	accepted	-543.1	-558.9	MOCK: nudged an integer constant (e.g. start city / range bound)
22	ed32b0d4	3b006102	accepted	-533.9	-567.0	MOCK: nudged an integer constant (e.g. start city / range bound)
23	a1d50f19	ed32b0d4	duplicate	—	—	MOCK: nudged an integer constant (e.g. start city / range bound)
24	3041f9c0	681b1065	duplicate	—	—	MOCK: nudged an integer constant (e.g. start city / range bound)
25	c3c4feef	57532c71	duplicate	—	—	MOCK: nudged an integer constant (e.g. start city / range bound)
26	4e48c9d6	3b006102	duplicate	—	—	MOCK: nudged an integer constant (e.g. start city / range bound)
27	47fb8ffc	3b006102	duplicate	—	—	MOCK: returned parent unchanged to test duplicate detection
28	f38816de	7a855e21	duplicate	—	—	MOCK: nudged an integer constant (e.g. start city / range bound)
29	7780cc88	3b006102	duplicate	—	—	MOCK: nudged an integer constant (e.g. start city / range bound)
30	ae9c5f84	ca14dbe9	accepted	-553.2	-569.1	MOCK: nudged an integer constant (e.g. start city / range bound)

Run tour-20261019-115227, exported 2026-10-19T11:52:29+00:00. Reproduce it with the demo source — a ~450-line educational re-implementation, not a replication of Sakana AI's results.