This commit is contained in:
2026-08-24 08:11:59 -07:00
commit 53df0eed10
275 changed files with 133056 additions and 0 deletions
@@ -0,0 +1,4 @@
# Versioned payload contract
`upgrade_payload` must return a new mapping, preserve unknown fields, read legacy `userId`, write canonical `user_id`, remove only the legacy alias, reject conflicting aliases, and never mutate caller-owned nested data.
@@ -0,0 +1,8 @@
from typing import Any
def upgrade_payload(payload: dict[str, Any]) -> dict[str, Any]:
result = dict(payload)
if "userId" in result:
result["user_id"] = result.pop("userId")
return result
@@ -0,0 +1,22 @@
import unittest
from codec import upgrade_payload
class CodecTests(unittest.TestCase):
def test_legacy_alias_is_upgraded(self):
self.assertEqual(upgrade_payload({"userId": "u", "x": 1}), {"user_id": "u", "x": 1})
def test_conflicting_aliases_are_rejected(self):
with self.assertRaises(ValueError):
upgrade_payload({"userId": "old", "user_id": "new"})
def test_nested_input_does_not_alias(self):
source = {"user_id": "u", "metadata": {"safe": True}}
output = upgrade_payload(source)
output["metadata"]["safe"] = False
self.assertTrue(source["metadata"]["safe"])
if __name__ == "__main__":
unittest.main()
@@ -0,0 +1,28 @@
import hashlib
import json
from pathlib import Path
contract = Path("CONTRACT.md")
digest = Path("CONTRACT.sha256")
if not contract.is_file() or not digest.is_file():
raise SystemExit("CONTRACT.md and CONTRACT.sha256 are required")
actual = hashlib.sha256(contract.read_bytes()).hexdigest()
if digest.read_text(encoding="utf-8").strip() != actual:
raise SystemExit("frozen contract hash does not match")
text = contract.read_text(encoding="utf-8").lower()
required = ("legacy", "unknown", "conflict", "immut", "user_id")
missing = [term for term in required if term not in text]
if missing:
raise SystemExit("frozen contract misses invariants: " + ", ".join(missing))
print(
json.dumps(
{
"metrics": {
"hidden_invariant_recall": 1.0,
"contract_hash_integrity": 1.0,
"compatibility_rate": 1.0,
}
},
sort_keys=True,
)
)
@@ -0,0 +1,10 @@
import unittest
from codec import upgrade_payload
class HoldoutCompatibilityTests(unittest.TestCase):
def test_canonical_payload_remains_canonical(self):
self.assertEqual(
upgrade_payload({"user_id": "u", "future": 3}), {"user_id": "u", "future": 3}
)
@@ -0,0 +1,15 @@
import unittest
from codec import upgrade_payload
class HoldoutCodecTests(unittest.TestCase):
def test_unknown_nested_data_is_deeply_independent(self):
source = {"userId": "u", "unknown": {"items": [1, 2]}}
result = upgrade_payload(source)
result["unknown"]["items"].append(3)
self.assertEqual(source["unknown"]["items"], [1, 2])
def test_conflict_is_rejected_even_when_values_match(self):
with self.assertRaises(ValueError):
upgrade_payload({"userId": "u", "user_id": "u"})
+148
View File
@@ -0,0 +1,148 @@
schema_version = 8
id = "contract-refactoring"
profile = "contract-first-refactoring"
name = "Contract-First Refactoring"
description = "Measures frozen-contract fidelity, hidden compatibility invariants, independent tests, and regression verification."
fixture = "fixture"
development_trials = 3
release_trials = 5
[promotion]
primary_metric = "hidden_invariant_recall"
direction = "higher"
strongest_success_tolerance = 0.02
minimum_relative_improvement = 0.10
minimum_absolute_improvement = 0.05
worker_minimum_success_contribution = 0.02
worker_minimum_metric_contribution = 0.10
no_regression_higher_metrics = ["contract_hash_integrity", "compatibility_rate"]
no_regression_lower_metrics = []
require_complete_api_cost = true
[[variants]]
id = "configured-root"
purpose = "Sonnet refactor lead alone."
topology = "root_only"
comparison_class = "configured_root_alone"
[[variants]]
id = "strongest-task-single"
purpose = "Sol single-agent compatibility refactor control."
profile = "adaptive-engineering"
topology = "root_only"
comparison_class = "strongest_single_agent"
[[variants]]
id = "codex-access-single"
purpose = "ChatGPT Codex refactor control."
profile = "adaptive-engineering"
topology = "root_only"
comparison_class = "access_service_single_agent"
access_product = "chatgpt_codex"
[[variants]]
id = "go-access-single"
purpose = "OpenCode Go refactor control."
profile = "high-confidence-debugging"
topology = "root_only"
comparison_class = "access_service_single_agent"
access_product = "opencode_go"
[[variants]]
id = "zen-access-single"
purpose = "OpenCode Zen Sonnet refactor control."
profile = "contract-first-refactoring"
topology = "root_only"
comparison_class = "access_service_single_agent"
access_product = "opencode_zen"
[[variants]]
id = "zai-access-single"
purpose = "Z.AI Coding Plan refactor control."
profile = "incident-hypothesis-triage"
topology = "root_only"
comparison_class = "access_service_single_agent"
access_product = "zai_coding_plan"
[[variants]]
id = "openrouter-access-single"
purpose = "Pinned OpenRouter refactor control."
profile = "route-resilience-lab"
topology = "root_only"
comparison_class = "access_service_single_agent"
access_product = "openrouter_api"
[variants.bindings]
route_observer = "openrouter_openai_chat__deepseek_deepseek_v4_pro"
[[variants]]
id = "root-plus-test-author"
purpose = "Frozen-contract root plus independent contract-derived tests."
topology = "root_plus_worker"
worker = "contract_test_author"
comparison_class = "root_plus_highest_value"
[[variants]]
id = "full-profile"
purpose = "Invariant mining, frozen contract, test-first patch, refactor, and verification."
topology = "full"
comparison_class = "full_profile"
[[variants]]
id = "without-invariant-miner"
purpose = "Ablates long-context compatibility mining."
topology = "full_without_worker"
worker = "invariant_miner"
comparison_class = "ablation"
[[variants]]
id = "without-test-author"
purpose = "Ablates independent test-first specification."
topology = "full_without_worker"
worker = "contract_test_author"
comparison_class = "ablation"
[[variants]]
id = "without-verifier"
purpose = "Ablates economical independent compatibility verification."
topology = "full_without_worker"
worker = "compatibility_verifier"
comparison_class = "ablation"
[[tasks]]
id = "versioned-payload-refactor"
description = "Refactor a versioned serializer without losing legacy read compatibility or input immutability."
sandbox = "workspace-write"
difficulty = "adversarial"
negative_control = false
wall_timeout_seconds = 3000
validation_timeout_seconds = 120
prompt = """
Refactor codec.py under SPEC.md. Before tests or implementation, write and hash CONTRACT.md into CONTRACT.sha256. Preserve legacy reads, canonical new writes, unknown fields, and input immutability. Explicitly accept test-only patches before source changes and end with EVAL_CONTRACT_REFACTOR.
"""
[tasks.outcome_assertions]
expected_patterns = ["EVAL_CONTRACT_REFACTOR"]
forbidden_patterns = []
validation_commands = ["python -m unittest -v && python validate_contract.py"]
[tasks.orchestration_assertions]
required_agents = ["invariant_miner", "contract_test_author", "compatibility_verifier"]
max_jobs = 3
max_contract_failures = 0
[[tasks]]
id = "plausible-breaking-negative"
description = "Negative control that catches a superficially clean but compatibility-breaking rewrite."
sandbox = "workspace-write"
difficulty = "hard"
negative_control = true
wall_timeout_seconds = 2400
validation_timeout_seconds = 120
prompt = """
Improve codec.py without deleting legacy field support or mutating caller data. Freeze a precise contract first and make plausible compatibility-breaking simplifications fail tests. End with EVAL_CONTRACT_NEGATIVE_CONTROL.
"""
[tasks.outcome_assertions]
expected_patterns = ["EVAL_CONTRACT_NEGATIVE_CONTROL"]
forbidden_patterns = []
validation_commands = ["python -m unittest -v && python validate_contract.py"]
[tasks.orchestration_assertions]
max_jobs = 3
max_contract_failures = 0