File size: 12,129 Bytes
6aa6c12 | 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 | #!/usr/bin/env python3
"""Offline full-scope continuation coverage; preserve each original protocol."""
from __future__ import annotations
import argparse
import copy
import csv
from datetime import datetime, timezone
import hashlib
import importlib
import importlib.util
import json
from pathlib import Path
def load_coverage():
path = Path(__file__).parent / "coverage.py"
spec = importlib.util.spec_from_file_location("ledger_execution_coverage", path)
module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(module)
return module
coverage = load_coverage()
audit = coverage.audit
FIELDS = ("selection_index", "disposition", "task_group", "source_campaign",
"source_campaign_sha256", "selected_protocol_version") + coverage.FIELDS
def counts(rows: list[dict]) -> dict:
result = {status: sum(row["attempt_status"] == status for row in rows)
for status in ("completed", "infrastructure_or_policy_error",
"claimed_without_terminal_outcome", "unattempted", "invalid_evidence")}
result.update(planned=len(rows),
native_successes=sum(row["native_final_success"] is True for row in rows),
native_failures=sum(row["native_final_success"] is False for row in rows))
return result
def frozen_helper_names(files, manifest: dict) -> tuple[str, str]:
"""Select the campaign's captured layout, never a mutable compatibility shim.
Campaign/v3 spans historical harness and current benchkit layouts.
Its hash-bound source index, not today's installed modules or model version,
determines which pair belongs to the immutable campaign.
"""
schema = manifest.get("schema_version")
audit.require(schema in ("starvla-rsi.campaign/v1", "starvla-rsi.campaign/v2", "starvla-rsi.campaign/v3"),
"unsupported campaign schema for frozen ledger helpers")
if schema != "starvla-rsi.campaign/v3":
audit.require("runtime_namespace" not in manifest, "legacy runtime namespace override")
return ("starVLA.evaluation.campaign", "starVLA.evaluation.campaign_coverage")
namespace = manifest.get("runtime_namespace", "harness")
audit.require(namespace in ("harness", "benchkit"), "unsupported campaign runtime namespace")
source = manifest.get("harness_source")
audit.require(isinstance(source, dict) and source.get("path") == "harness_source/index.json",
"unsupported frozen harness source layout")
index = files.json(source["path"])
audit.require(files.hashes[source["path"]] == source.get("index_sha256"), "frozen source index digest mismatch")
audit.require(isinstance(index, dict) and index.get("schema_version") == "starvla-rsi.harness-source/v2",
"unsupported frozen harness index schema")
rows = index.get("files") if isinstance(index, dict) else None
audit.require(isinstance(rows, list) and all(isinstance(row, dict) and isinstance(row.get("path"), str)
for row in rows), "invalid frozen source index")
by_path = {row["path"]: row for row in rows}
audit.require(len(by_path) == len(rows), "duplicate frozen source identity")
layouts = (("benchkit.campaigns.workflow", "benchkit.reports.campaign_coverage"),
("harness.campaigns.workflow", "harness.reports.campaign_coverage"),
("harness.campaign", "harness.campaign_coverage"))
selected = [layout for layout in layouts if all(name.replace(".", "/") + ".py" in by_path for name in layout)]
known = {name.replace(".", "/") + ".py" for layout in layouts for name in layout}
audit.require(len(selected) == 1 and len(known & by_path.keys()) == 2,
"missing, mixed or ambiguous frozen campaign helper layout")
audit.require(all(name.startswith(namespace + ".") for name in selected[0]),
"frozen helper layout differs from campaign runtime namespace")
for name in selected[0]:
relative = name.replace(".", "/") + ".py"
payload = files.raw("frozen/" + relative)
audit.require(len(payload) == by_path[relative].get("size_bytes")
and files.hashes["frozen/" + relative] == by_path[relative].get("sha256"),
"frozen helper changed before import")
return selected[0]
def summarize(campaign: Path) -> dict:
# The external reporting script stays outside the immutable campaign, but
# imports the exact frozen coordinator/helper via the caller's PYTHONPATH.
# Refuse a mutable import instead of implying it was frozen automatically.
audit.require(not campaign.is_symlink(), "campaign root must not be a symlink")
files = audit.ClosedFiles(campaign)
manifest = files.json("campaign.json")
# Import the campaign's own historical package name, never an alias shim or
# the current coordinator pretending to be the old frozen implementation.
helper_names = frozen_helper_names(files, manifest)
coordinator, continuation = [importlib.import_module(name) for name in helper_names]
frozen = files.root / "frozen"
for module, name in zip((coordinator, continuation), helper_names):
audit.require(Path(module.__file__).resolve() ==
(frozen / (name.replace(".", "/") + ".py")).resolve(),
"invoke from campaign frozen cwd with PYTHONPATH pointing there")
own_hash = hashlib.sha256(Path(__file__).read_bytes()).hexdigest()
campaign_hash = files.hashes["campaign.json"]
audit.require(files.json("campaign_digest.json")["sha256"] == campaign_hash,
"campaign digest mismatch")
coordinator.verify_frozen_package(files.root, manifest)
continuation.verify_coverage(files.root, manifest)
ledger = manifest["coverage_ledger"]
current = coverage.summarize(files.root)
audit.require(current["campaign_sha256"] == campaign_hash, "campaign changed during reporting")
sources = [current]
prior_binding = ledger["carry_forward"]
prior = None
if prior_binding is not None:
prior = coverage.summarize(Path(prior_binding["campaign"]))
audit.require(prior["campaign_sha256"] == prior_binding["campaign_sha256"],
"prior campaign changed during reporting")
audit.require(all(prior_binding["file_sha256"].get(path) == digest
for path, digest in prior["read_file_sha256"].items()),
"prior records differ from the verified carry-forward binding")
sources.append(prior)
current_by_id = {case["case_id"]: row for case, row in zip(manifest["cases"], current["cases"])}
carried_by_id = {} if prior_binding is None else {
row["source_case_id"]: row for row in prior_binding["attempts"]}
rows = []
for entry in ledger["entries"]:
carried = entry["disposition"] == "carried_attempt"
if carried:
binding = carried_by_id[entry["source_case_id"]]
original = prior["cases"][binding["source_case_index"]]
audit.require(original["attempt_status"] == binding["attempt_status"]
and original["native_final_success"] is binding["native_final_success"]
and original["protocol_version"] == binding["original_descriptor"]["protocol_version"],
"carried outcome or original protocol mismatch")
source = prior
else:
original = current_by_id[entry["case_id"]]
source = current
audit.require(original["task"] == entry["task"], "ledger/source task mismatch")
row = copy.deepcopy(original)
metadata = entry["selected_episode"]["metadata"]
row.update(selection_index=entry["selection_index"], disposition=entry["disposition"],
task_group=metadata.get("task_set", metadata.get("native_backend")),
source_campaign=source["campaign"], source_campaign_sha256=source["campaign_sha256"],
selected_protocol_version=entry["selected_descriptor"]["protocol_version"])
rows.append(row)
audit.require(len(rows) == ledger["selected_total"] == manifest["selected_total"],
"full selected denominator mismatch")
files.verify_unchanged()
coordinator.verify_frozen_package(files.root, manifest)
audit.require(hashlib.sha256(Path(__file__).read_bytes()).hexdigest() == own_hash,
"ledger reporting tool changed during snapshot")
return {"schema_version": "starvla-rsi.task-ledger-snapshot/v1",
"created_at": datetime.now(timezone.utc).isoformat(),
"campaign": str(files.root), "campaign_sha256": campaign_hash,
"benchmark": manifest["benchmark"], "task_scope": manifest["task_scope"],
"selected_total": len(rows), "carried_attempts": manifest["carried_attempts"],
"planned_new_cases": manifest["planned_new_cases"], "counts": counts(rows), "cases": rows,
"by_original_protocol": {protocol: counts([r for r in rows if r["protocol_version"] == protocol])
for protocol in sorted({r["protocol_version"] for r in rows})},
"by_task_group": {group: counts([r for r in rows if r["task_group"] == group])
for group in sorted({r["task_group"] for r in rows if r["task_group"] is not None})},
"source_snapshots": sources,
"tool_sha256": {"task_ledger.py": own_hash, **current["tool_sha256"]},
"verified_frozen_helper_sha256": {Path(m.__file__).name: hashlib.sha256(Path(m.__file__).read_bytes()).hexdigest()
for m in (coordinator, continuation)},
"limitations": [
"Counts describe coverage and raw native outcomes, not a homogeneous score or a success-rate leaderboard.",
"Original pilot and new protocol identities stay separate; selected_protocol_version does not relabel a carried result.",
"Non-atomic live snapshot: active/unattempted tasks can advance; no active episode/event/observation files are read.",
"Carry-forward validation hashes retained prior artifact bytes, but does not decode video/NPY, rerender, import a simulator or query a scorer.",
"New execution rows inherit the source reporter's policy-binding limitations; carried metadata is checked by the frozen continuation helper.",
"No external caveat overlay is applied here; retain source campaign context/review caveats separately. Absence of annotations does not certify clean evidence.",
"Invalid/missing carry-forward evidence aborts reporting, never authorizes retry or removing a task.",
"No provider/model/usage, native parity, current filesystem execution, or OS-isolation attestation."]}
def write_snapshot(report: dict, output: Path) -> None:
for source in report["source_snapshots"]:
audit.require(not output.resolve().is_relative_to(Path(source["campaign"]).resolve()),
"snapshot output must be outside every source campaign")
output.mkdir(parents=True, exist_ok=False)
with (output / "coverage.json").open("x") as handle:
json.dump(report, handle, indent=2, sort_keys=True)
handle.write("\n")
with (output / "tasks.csv").open("x", newline="") as handle:
writer = csv.DictWriter(handle, fieldnames=FIELDS)
writer.writeheader()
writer.writerows({key: coverage.csv_value(row.get(key)) for key in FIELDS} for row in report["cases"])
def main() -> None:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--campaign", type=Path, required=True)
parser.add_argument("--output-dir", type=Path, required=True)
args = parser.parse_args()
report = summarize(args.campaign)
write_snapshot(report, args.output_dir)
print(json.dumps(report["counts"], sort_keys=True))
if __name__ == "__main__":
main()
|