tools/phase1_gate.py

#!/usr/bin/env python3
"""Phase 1 gate — falsification test for the jsonl-loader.

Loads the fixture via engine.loaders.jsonl_loader.load() and asserts the
CORRECTED regression oracle (CLAUDE.md / verified facts). NO model. Deterministic.

Oracle (corrected, START-anchored turn rule):
  turns                == 16   (human == 14, system == 2, system indices == [11, 14])
  total tool calls     == 239
  session output       == 1_555_504
  session fresh input  == 52_295
  session cache_read   == 213_300_315
  session cache_create == 2_508_984

Never edit the oracle to match the parser — fix the parser. Run:
  python3 tools/phase1_gate.py
"""
from __future__ import annotations

import sys
from pathlib import Path

# Make the repo root importable so `engine` resolves regardless of CWD.
REPO = Path(__file__).resolve().parent.parent
if str(REPO) not in sys.path:
    sys.path.insert(0, str(REPO))

from engine.loaders.jsonl_loader import load  # noqa: E402

FIXTURE = REPO / "fixtures" / "sample-session.jsonl"

# --- the corrected oracle (ground truth) ---------------------------------- #
ORACLE = {
    "turns": 16,
    "human": 14,
    "system": 2,
    "system_indices": [11, 14],
    "tools": 239,
    "out": 1_555_504,
    "in": 52_295,
    "cacheRead": 213_300_315,
    "cacheCreate": 2_508_984,
}


def _line(label: str, got, want) -> bool:
    ok = got == want
    flag = "OK  " if ok else "DIFF"
    print(f"  [{flag}] {label:<22} got={got!r:>14}  want={want!r}")
    return ok


def main() -> int:
    if not FIXTURE.exists():
        print(f"FAIL — fixture missing: {FIXTURE}")
        return 1

    out = load(str(FIXTURE))
    turns = out["turns"]
    session = out["session"]

    # measured
    n_turns = len(turns)
    humans = [t for t in turns if t.origin == "human"]
    systems = [t for t in turns if t.origin == "system"]
    system_indices = [t.i for t in systems]
    total_tools = sum(len(t.tools) for t in turns)

    sess_in = sum(t.tokens.in_ for t in turns)
    sess_out = sum(t.tokens.out for t in turns)
    sess_cr = sum(t.tokens.cacheRead for t in turns)
    sess_cc = sum(t.tokens.cacheCreate for t in turns)

    print("Her · हेर — Phase 1 gate (jsonl-loader)")
    print("=" * 60)
    print(f"fixture   : {FIXTURE}")
    print(f"session   : cwd={session.get('cwd')!r}")
    print(f"            sessionId={session.get('sessionId')!r}")
    print(f"            gitBranch={session.get('gitBranch')!r} version={session.get('version')!r}")
    print("-" * 60)

    checks = [
        _line("turns", n_turns, ORACLE["turns"]),
        _line("human turns", len(humans), ORACLE["human"]),
        _line("system turns", len(systems), ORACLE["system"]),
        _line("system indices", system_indices, ORACLE["system_indices"]),
        _line("total tools", total_tools, ORACLE["tools"]),
        _line("session output", sess_out, ORACLE["out"]),
        _line("session fresh input", sess_in, ORACLE["in"]),
        _line("session cache_read", sess_cr, ORACLE["cacheRead"]),
        _line("session cache_create", sess_cc, ORACLE["cacheCreate"]),
    ]

    print("-" * 60)
    ok = all(checks)
    print("GATE:", "PASS" if ok else "FAIL")
    return 0 if ok else 1


if __name__ == "__main__":
    sys.exit(main())