Datasets:
Formats:
json
Languages:
English
Size:
< 1K
Tags:
artificial-intelligence
recursive-self-improvement
continual-learning
representation-learning
world-models
cognitive-architecture
File size: 9,723 Bytes
1c1abed | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 | #!/usr/bin/env python3
"""Data-only JSON adapter for the exact AXIOMESH research backend.
This adapter does not call an LLM, execute proposed Python, or acquire evidence.
Every result concerns the declared finite linear family and trusted input data.
"""
from __future__ import annotations
import argparse
import json
from pathlib import Path
import re
import sys
ROOT = Path(__file__).resolve().parent
sys.path.insert(0, str(ROOT / "src"))
import numpy as np
from axiomesh.core import Capsule, FrozenModel, Memory, close_operators, hole_certificate
from axiomesh.esre_v01 import require_prime
MAX_BYTES = 1_048_576
MAX_DIM = 32
MAX_ROWS = 128
MAX_ACTIONS = 16
MAX_WORD = 256
ACTION_NAME = re.compile(r"[A-Za-z0-9_.+\-]{1,64}\Z")
def integer(value, label, low=0, high=250):
if type(value) is not int or not low <= value <= high:
raise ValueError(f"{label} must be an integer in {low}..{high}")
return value
def keys(obj, required, optional=()):
if not isinstance(obj, dict):
raise ValueError("expected a JSON object")
missing, extra = set(required) - obj.keys(), obj.keys() - set(required) - set(optional)
if missing or extra:
raise ValueError(f"object keys invalid: missing={sorted(missing)}, extra={sorted(extra)}")
def field(value):
p = integer(value, "field p", 2, 251)
require_prime(p)
return p
def rows(value, label, p, width=None, minimum=0):
if not isinstance(value, list) or not minimum <= len(value) <= MAX_ROWS:
raise ValueError(f"{label}: expected {minimum}..{MAX_ROWS} rows")
if width is None:
if not value or not isinstance(value[0], list):
raise ValueError(f"{label}: a nonempty matrix is required")
width = integer(len(value[0]), f"{label} width", 1, MAX_DIM)
for row in value:
if not isinstance(row, list) or len(row) != width:
raise ValueError(f"{label}: matrix width mismatch")
for entry in row:
integer(entry, f"{label} entry", 0, p - 1)
return np.asarray(value, dtype=np.int64).reshape(len(value), width)
def operators(value, p, width):
if not isinstance(value, dict) or len(value) > MAX_ACTIONS:
raise ValueError(f"operators: expected at most {MAX_ACTIONS} actions")
result = {}
for name, data in value.items():
if not isinstance(name, str) or not ACTION_NAME.fullmatch(name):
raise ValueError("invalid action name")
result[name] = rows(data, f"operator {name}", p, width)
if result[name].shape != (width, width):
raise ValueError("operators must be square")
return result
def word(value, ops):
if not isinstance(value, list) or len(value) > MAX_WORD:
raise ValueError(f"word: expected at most {MAX_WORD} chronological actions")
if any(not isinstance(a, str) or a not in ops for a in value):
raise ValueError("word contains an undeclared action")
return value
def load_json(path):
with path.open("rb") as handle:
data = handle.read(MAX_BYTES + 1)
if len(data) > MAX_BYTES:
raise ValueError("JSON input exceeds 1 MiB")
return json.loads(data)
def envelope(task, status, **values):
return {
"format": "axiomesh-agent-response-v1",
"task": task,
"status": status,
"scope": "declared small prime-field linear model; trusted input observations",
"neural_rsi_demonstrated": False,
**values,
}
def recover_and_run(request, root):
keys(request, ["task", "capsule", "word"])
relative = request["capsule"]
if not isinstance(relative, str) or not 1 <= len(relative) <= 256 or Path(relative).is_absolute():
raise ValueError("capsule must be a repository-relative JSON path")
path = (root / relative).resolve()
if not path.is_relative_to(root.resolve()) or path.suffix.lower() != ".json":
raise ValueError("capsule must remain inside the repository")
obj = load_json(path)
if not isinstance(obj, dict) or "program" not in obj or "memory" not in obj:
raise ValueError("capsule requires program and memory objects")
program, memory = obj["program"], obj["memory"]
keys(program, ["p", "rank", "operators", "query"])
keys(memory, ["p", "rank", "shares"])
p = field(program["p"])
rank = integer(program["rank"], "rank", 1, MAX_DIM)
if field(memory["p"]) != p or integer(memory["rank"], "memory rank", 1, MAX_DIM) != rank:
raise ValueError("program and memory fields/ranks disagree")
ops = operators(program["operators"], p, rank)
rows(program["query"], "query", p, rank, 1)
actions = word(request["word"], ops)
shares = memory["shares"]
if not isinstance(shares, dict) or len(shares) > MAX_ROWS:
raise ValueError("shares: expected at most 128 points")
for point, value in shares.items():
if not isinstance(point, str) or not point.isascii() or not point.isdigit():
raise ValueError("share points must be canonical decimal strings")
if str(int(point)) != point:
raise ValueError("share points must be canonical decimal strings")
integer(int(point), "share point", 0, p - 1)
integer(value, "share value", 0, p - 1)
status, seed = Capsule.from_json(memory).recover()
if status != "exact":
return envelope(request["task"], status, output=None, source_state_used=False)
model = FrozenModel.from_json(program, seed=seed)
return envelope(request["task"], "exact", output=model.run(actions).tolist(),
quotient_rank=rank, source_state_used=False,
damage_contract="known erasures; no general corruption certificate")
def linear_request(request):
task = request["task"]
optional = ["product_budget"] if task == "revalidate" else ["product_budget", "erasure_budget", "word"]
keys(request, ["task", "p", "query", "operators", "memory"], optional)
p = field(request["p"])
query = rows(request["query"], "query", p, minimum=1)
ops = operators(request["operators"], p, query.shape[1])
keys(request["memory"], ["matrix", "rhs"])
matrix = rows(request["memory"]["matrix"], "memory", p, query.shape[1])
rhs = request["memory"]["rhs"]
if not isinstance(rhs, list) or len(rhs) != len(matrix):
raise ValueError("memory rhs length mismatch")
for value in rhs:
integer(value, "memory rhs", 0, p - 1)
memory = Memory(matrix, np.asarray(rhs, dtype=np.int64), p)
budget = integer(request.get("product_budget", MAX_DIM * MAX_ACTIONS), "product budget", 0, 512)
closure = close_operators(query, ops, p, product_budget=budget)
status, seed, _ = memory.recall(closure.basis)
missing = None if status == "inconsistent" else memory.supplement(closure.basis).tolist()
common = {
"quotient_rank": closure.rank,
"row_operator_products": closure.row_operator_products,
"required_extra_scalar_observations": None if missing is None else len(missing),
"suggested_observation_rows": missing,
"observations_acquired": 0,
"certificate": hole_certificate(closure, memory),
}
if task == "revalidate":
return envelope(task, status, **common)
erasures = integer(request.get("erasure_budget", 0), "erasure budget", 0, 32)
actions = word(request.get("word", []), ops)
if status != "exact" or closure.rank == 0:
outcome = "zero_observable" if status == "exact" else status
return envelope(task, outcome, output=None, **common)
model = closure.compile(seed)
capsule = Capsule.encode(seed, p, erasures)
return envelope(task, "committed", output=model.run(actions).tolist(),
capsule={"format": "axiomesh-frozen-v0.2",
"program": model.json_object(include_seed=False),
"memory": capsule.json_object()}, **common)
def dispatch(request, root=ROOT):
if not isinstance(request, dict) or request.get("task") not in {"revalidate", "compile", "recover_and_run"}:
raise ValueError("task must be revalidate, compile, or recover_and_run")
if request["task"] == "recover_and_run":
return recover_and_run(request, root)
return linear_request(request)
def main(argv=None):
parser = argparse.ArgumentParser(description=__doc__)
source = parser.add_mutually_exclusive_group()
source.add_argument("--request", type=Path, help="read a JSON request file; default: stdin")
source.add_argument("--schema", action="store_true", help="print the request JSON Schema")
source.add_argument("--capabilities", action="store_true", help="print machine-readable capabilities")
args = parser.parse_args(argv)
try:
if args.schema or args.capabilities:
name = "agent_request.schema.json" if args.schema else "AGENT_CAPABILITIES.json"
result = load_json(ROOT / name)
else:
if args.request:
request = load_json(args.request)
else:
data = sys.stdin.buffer.read(MAX_BYTES + 1)
if len(data) > MAX_BYTES:
raise ValueError("JSON input exceeds 1 MiB")
request = json.loads(data)
result = dispatch(request)
print(json.dumps(result, ensure_ascii=False, allow_nan=False))
return 2 if result.get("status") in {"ambiguous", "inconsistent", "zero_observable"} else 0
except (OSError, ValueError, KeyError, TypeError, RuntimeError, OverflowError) as exc:
print(json.dumps(envelope("request", "error", detail=str(exc)), allow_nan=False))
return 1
if __name__ == "__main__":
raise SystemExit(main())
|