Spaces:
Paused
Paused
| """Build the page where a visitor pastes a diff and three checks run in the browser: `python -m itemator.tryit --out public/try`. | |
| web/try.html is the page, web/try_checks.js holds the checks (the same rules as moving.py, shapes.py and untrusted.py; eval/try_agreement.py | |
| compares the two) and web/try_examples.json the example diffs. The numbers beside each check are read from the results files at build | |
| time, so the page cannot say more than the files do. The page forbids itself every network request with a Content-Security-Policy. | |
| Standard library only.""" | |
| from __future__ import annotations | |
| import argparse | |
| import hashlib | |
| import json | |
| import pathlib | |
| ROOT = pathlib.Path(__file__).resolve().parent.parent | |
| WEB = ROOT / "web" | |
| RESULTS = ROOT / "results" | |
| REPO = "https://gitlab.com/gitlab-ai-hackathon/transcend-october-2026/39194174/showcase/-/blob/main/" | |
| def safe_json(data) -> str: | |
| """JSON for inside a <script> tag: no sequence in the data can close the tag or open a comment.""" | |
| return json.dumps(data, ensure_ascii=True).replace("<", "\\u003c").replace(">", "\\u003e").replace("&", "\\u0026") | |
| def _load(name: str) -> dict: | |
| return json.loads((RESULTS / name).read_text(encoding="utf-8")) | |
| def _ratio(row: dict) -> str: | |
| return f"{row['spoke']} of {row['read']}" | |
| def evidence(mesa: dict, agents: dict, incidents: dict, agreement: dict) -> dict: | |
| """The table beside the checks, from the results files only.""" | |
| by_check: dict[str, list[dict]] = {} | |
| for i in incidents["incidents"]: | |
| by_check.setdefault(i["expected_check"], []).append(i) | |
| reads = {"moving": "Added lines of CI files and Dockerfiles", "shapes": "Added lines of any file but tests, docs and lock files", | |
| "untrusted": "Added lines of CI files and shell scripts"} | |
| rows = [] | |
| for c in ("shapes", "untrusted", "moving"): | |
| inc = by_check.get(c, []) | |
| rows.append({"check": c, "reads": reads[c], "ordinary": _ratio(mesa["ordinary"]["checks"][c]), "agents": _ratio(agents["checks"][c]), | |
| "incidents": f"{sum(1 for i in inc if i['fired'])} of {len(inc)}" if inc else "none in the set"}) | |
| units = sum(agreement[k]["units"] for k in ("agent_commits", "mesa_commits", "line_corpus", "page_examples")) | |
| notes = [ | |
| {"text": f"Ordinary commits: {mesa['ordinary']['checks']['moving']['read']} commits of the Mesa project, 2024 to 2026, drawn at random. Agent-written commits: commits of AI agents' pull requests in the public AIDev data set, read from the pull requests that had human review comments, so riskier than average."}, | |
| {"text": "Public incident diffs: the Amazon Q extension, the Nx workflow and the xz backdoor were read before the rules were written, so this column shows that the rules can see them, not that they will find the next one."}, | |
| {"text": "Mesa and agent-commit results", "href": REPO + "results/mesa_noise_v1.json"}, | |
| {"text": "Agent-written commits", "href": REPO + "results/agent_noise_v1.json"}, | |
| {"text": "Incident diffs", "href": REPO + "results/incident_replay_v1.json"}, | |
| {"text": f"This page's checks agree with the Python on all {units} units tried", "href": REPO + "results/try_agreement_v2.json"}, | |
| ] | |
| return {"rows": rows, "notes": notes, "units": units} | |
| def page() -> str: | |
| mesa, agents = _load("mesa_noise_v1.json"), _load("agent_noise_v1.json") | |
| incidents, agreement = _load("incident_replay_v1.json"), _load("try_agreement_v2.json") | |
| if not agreement["all_identical"]: | |
| raise SystemExit("itemator.tryit: the browser checks and the Python checks disagree (results/try_agreement_v2.json); not building") | |
| checks = (WEB / "try_checks.js").read_text(encoding="utf-8") | |
| ev = evidence(mesa, agents, incidents, agreement) | |
| payload = {"examples": json.loads((WEB / "try_examples.json").read_text(encoding="utf-8")), "evidence": ev, | |
| "footer": f"Checks version {hashlib.sha256(checks.encode()).hexdigest()[:12]} (web/try_checks.js). Standard HTML and JavaScript; no library, no network, no model."} | |
| html = (WEB / "try.html").read_text(encoding="utf-8") | |
| for marker, text in (("/*PAYLOAD*/", safe_json(payload)), ("/*CHECKS*/", checks.replace("</script", "<\\/script"))): | |
| assert html.count(marker) == 1 | |
| html = html.replace(marker, text, 1) | |
| return html | |
| def main(argv=None) -> int: | |
| ap = argparse.ArgumentParser(prog="itemator.tryit", description=(__doc__ or "").split("\n\n")[0]) | |
| ap.add_argument("--out", default="public/try") | |
| a = ap.parse_args(argv) | |
| out = pathlib.Path(a.out) | |
| out.mkdir(parents=True, exist_ok=True) | |
| (out / "index.html").write_text(page(), encoding="utf-8") | |
| print(f"wrote {out / 'index.html'}") | |
| return 0 | |
| if __name__ == "__main__": | |
| raise SystemExit(main()) | |