alcides

Aeon: 32-item 0–1 knapsack with inductive stdlib lists

Run on 2026-11-04T13:04:08Z, using Aeon 41a88a70 and Python 3.13.15.

Attempted 36 canonical backend/model IDs plus 5 aliases (41 runs), sequentially, with a hard 10-second wall-clock limit per run. Excluded all 12 single-step tactics as requested. The multi-step tactics search is included.

No independently verified complete solution was reported within the 10-second deadline. This is a measured timeout/interface result, not a claim that these algorithms cannot solve knapsack.

There is no synthesized best_solution.ae for this run. reference_solution.ae is clearly labeled as an independently computed reference, not a synthesis result.

alcides

Aeon timing: stack GGGP vs grammatical evolution

Problem: centimeters-to-meters
Settings: population 100, evaluation cap 1000, gene length 128, wall budget 20s, lexicase, early stop
Measured: wall clock for one synthesis run; evaluation = genotype→AST mapping + fitness over train cases

Phase	gp_stack	share	ge	share
Wall clock	48.1s	100%	16.6s	100%
Evaluation (map + fitness)	47.8s	99.5%	14.6s	88.0%

alcides

; centimeters-to-meters — Aeon/GeneticEngine synthesis grammar
; Hole type: (x: Int) → Tuple
; Start symbol: <Tuple>

<Tuple> ::= app_pair(<Int>, <Int>)
          | app(<Int → Tuple>, <Int>)

<Int> ::= x
        | lit0
        | lit100

alcides

# /// script
# requires-python = ">=3.11"
# dependencies = [
#   "matplotlib>=3.8",
#   "numpy>=2.0",
#   "pandas>=2.2",
# ]
# ///
"""Entry-point metadata for running the experiment scripts with uv."""

alcides

NodeName=localhost Name=gpu type=a30 File=/dev/nvidia[0-1]
NodeName=localhost Name=mps count=200 File=/dev/nvidia[0-1]

alcides

"""
This script flattens the structure of this project in the output folder.
"""
import os
import re
import sys
import shutil
from datetime import date
from pathlib import Path

alcides

My First Compiler

Overview

This project shows an example of a compiler pipeline.

The frontend is responsible for parsing the input file (example.lang) and returning an AST. The AST is designed using an OO approach, with two main types (Statement and Expression), and multiple options for each. A program is a list of statements. There is one semantic validation: we are detecting if variables are used before they are defined. If so, we raise an exception.

alcides

# Puzzle from: https://twitter.com/wtgowers/status/1581667648592826369?s=12&t=5CR5L4TsH2MJWDJwBO8Wmw

import z3
s = z3.Solver()
a, b, c, d, e = z3.Ints("a b c d e")

# order
s.add( a <= c  )
s.add( b <= c  )
s.add( c <= d  )

alcides

• [alcides@franklin:~/Code/srbench] (Competition2022) $ ~/bin/act
[CI/check] 🚀  Start image=ghcr.io/catthehacker/ubuntu:act-latest
[CI/check]   🐳  docker pull image=ghcr.io/catthehacker/ubuntu:act-latest platform= username= forcePull=false
[CI/check]   🐳  docker create image=ghcr.io/catthehacker/ubuntu:act-latest platform= entrypoint=["/usr/bin/tail" "-f" "/dev/null"] cmd=[]
[CI/check]   🐳  docker run image=ghcr.io/catthehacker/ubuntu:act-latest platform= entrypoint=["/usr/bin/tail" "-f" "/dev/null"] cmd=[]
[CI/check]   🐳  docker exec cmd=[mkdir -m 0777 -p /var/run/act] user=root workdir=
[CI/check]   🐳  docker cp src=/home/alcides/Code/srbench/. dst=/home/alcides/Code/srbench
[CI/check]   🐳  docker exec cmd=[mkdir -p /home/alcides/Code/srbench] user= workdir=
[CI/check] ⭐  Run Checkout code
[CI/check]   ✅  Success - Checkout code

alcides

import Language.Haskell.Liquid.ProofCombinators 


{-@ reflect map1 @-}
map1 :: (Int -> Int) -> [Int] -> [Int]
map1 f [] = []
map1 f (x:xs) = (f x):(map1 f xs)

{-@ reflect length1 @-}
length1 :: [Int] -> Int