#!/usr/bin/env python3 """Independent O3 REST check — no repo clone, no mpmath. Download this file next to math-o3-tables.json (or a capability reproduce.json) and compare the public REST result to the tabulated expected_f64 within ≤2 ULP. python3 check-math-o3-rest.py tan python3 check-math-o3-rest.py --reproduce reproduce.json python3 check-math-o3-rest.py tan --url https://www.calculatorx.com/api/v1/calc/tan """ from __future__ import annotations import argparse import json import math import struct import sys import urllib.error import urllib.request from pathlib import Path SITE = "https://www.calculatorx.com" THRESHOLD_ULP = 2 FAMILY_TOOL = { "tan": "tan", "sin": "sin", "cos": "cos", "atan": "arctan", "asin": "arcsin", "ln": "ln", "log10": "log", "sqrt": "square-root", "exponent": "exponent", "antilog": "antilog", "root": "root", "exp_growth": "exponential-growth", "divide": "division", } def ulp_distance(a: float, b: float) -> int: if not (math.isfinite(a) and math.isfinite(b)): return 2**31 if a == b: return 0 def bits(x: float) -> int: return struct.unpack(" int: return u if (u >> 63) == 0 else (~u) & ((1 << 64) - 1) return abs(order(bits(a)) - order(bits(b))) def result_number(payload: dict, family: str): r = payload.get("result") if isinstance(r, dict): if family in {"atan", "asin"} and r.get("deg") is not None: return float(r["deg"]) if family == "sqrt": if r.get("principal") is not None: return float(r["principal"]) if r.get("imag") is not None: return float(r["imag"]) if r.get("value") is not None: return float(r["value"]) if isinstance(r, (int, float)): return float(r) detail = payload.get("result_detail") or {} if isinstance(detail, dict) and detail.get("value") is not None: v = detail["value"] if isinstance(v, dict): if family == "sqrt": if v.get("principal") is not None: return float(v["principal"]) if v.get("imag") is not None: return float(v["imag"]) if v.get("value") is not None: return float(v["value"]) return float(v) raise ValueError(f"no numeric result in REST payload keys={list(payload)}") def load_vectors(table_path: Path, reproduce_path: Path | None, family: str): if reproduce_path: doc = json.loads(reproduce_path.read_text(encoding="utf-8")) fam = doc.get("capability_id") or family vectors = [] for row in doc.get("vectors") or []: vectors.append( { "id": row.get("id"), "inputs": row.get("inputs") or {}, "y_f64": row.get("expected_f64"), "threshold_ulp": row.get("threshold_ulp"), } ) return fam, vectors, int(doc.get("result_summary", {}).get("threshold_ulp") or THRESHOLD_ULP) tables = json.loads(table_path.read_text(encoding="utf-8")) fam = tables.get("families", {}).get(family) if not fam: raise SystemExit(f"family {family!r} not in {table_path}") return family, fam.get("vectors") or [], THRESHOLD_ULP def post_json(url: str, body: dict) -> dict: req = urllib.request.Request( url, data=json.dumps(body).encode("utf-8"), headers={"Content-Type": "application/json", "Accept": "application/json"}, method="POST", ) with urllib.request.urlopen(req, timeout=30) as resp: return json.loads(resp.read().decode("utf-8")) def main() -> int: here = Path(__file__).resolve().parent parser = argparse.ArgumentParser(description="Compare public REST results to the O3 table.") parser.add_argument("family", nargs="?", help="O3 family name, e.g. tan") parser.add_argument("--table", type=Path, default=here / "math-o3-tables.json") parser.add_argument("--reproduce", type=Path, help="Capability reproduce.json (preferred when present)") parser.add_argument("--url", help="Override REST URL") parser.add_argument("--threshold", type=int, default=None) args = parser.parse_args() if not args.family and not args.reproduce: parser.error("pass a family (tan) or --reproduce reproduce.json") family_key = args.family or "tan" family, vectors, default_thr = load_vectors(args.table, args.reproduce, family_key) if args.reproduce and not args.family: cap = str(json.loads(args.reproduce.read_text(encoding="utf-8")).get("capability_id") or "") family = cap.split(".", 1)[-1] if cap.startswith("math.") else family if family == "arctan": family = "atan" if family == "arcsin": family = "asin" if family == "square_root": family = "sqrt" global_thr = args.threshold tool = FAMILY_TOOL.get(family, family) url = args.url or f"{SITE}/api/v1/calc/{tool}" if not vectors: print("no vectors", file=sys.stderr) return 1 failed = 0 max_ulp = 0 max_thr_used = default_thr for row in vectors: expected = row.get("y_f64") row_thr = row.get("threshold_ulp") thr = global_thr if global_thr is not None else (int(row_thr) if row_thr is not None else default_thr) max_thr_used = max(max_thr_used, thr) try: payload = post_json(url, row["inputs"]) if payload.get("status") not in (None, "success", "ok"): raise ValueError(payload.get("error") or payload.get("status")) if expected is None: result = payload.get("result") or {} q = result.get("quotient") r = result.get("remainder") ok = str(q) == str(row.get("quotient")) and str(r) == str(row.get("remainder")) if not ok: failed += 1 print(f"{'PASS' if ok else 'FAIL'} {row.get('id')} exact q,r expected={row.get('quotient')} R {row.get('remainder')} actual={q} R {r}") continue actual = result_number(payload, family) except (urllib.error.URLError, ValueError, KeyError, TypeError) as exc: failed += 1 print(f"FAIL {row.get('id')} REST {exc}") continue err = ulp_distance(float(expected), actual) max_ulp = max(max_ulp, err) ok = err <= thr if not ok: failed += 1 print(f"{'PASS' if ok else 'FAIL'} {row.get('id')} ulp={err} thr={thr} expected={expected} actual={actual}") total = len(vectors) print(f"{total - failed}/{total} within per-vector ULP (summary default ≤{default_thr}, max declared ≤{max_thr_used}) · max {max_ulp} ULP · {url}") return 1 if failed else 0 if __name__ == "__main__": sys.exit(main())