Some checks failed
CI / build (push) Failing after 3s
PHASE 2.3 (AUDIT-2026-Q3.md):
Task 1 — Wire-in Wave 3 (1 router mounted, 2 deferred):
- app.routers.unified_scanner_router mounted at /api/v2/scanner/* (2 routes:
POST /api/v2/scanner/token/scan, POST /api/v2/scanner/wallet/scan).
Refactored prefix from /api/v2 -> /api/v2/scanner to avoid future conflicts
with the v1 /api/v1/scanner/ stub.
- app.routers.unified_wallet_scanner DEFERRED (no router APIRouter attribute;
library module consumed by unified_scanner_router via get_wallet_scanner()).
- app.routers.admin_extensions DEFERRED (DORMANT per audit; 25 routes at
/api/v1/admin/* would shadow /api/v1/admin/alerts_webhook).
Task 2 — Archive 136 dead-code files to app/_archive/legacy_2026_07/:
- 73 routers in app/routers/ (reach graph showed zero reach into mount.py).
- 63 flat app/*.py (domain modules never imported by live code).
- 1 file RESTORED post-archive: app/routers/x402_bridge_health.py (caught by
tests/unit/test_bridge_health.py which directly imports it; reach graph
considered tests/ only as transitive reach — to be patched in next cycle).
Forced-LIVE (NOT archived per user directive):
- app/ai_pipeline_v3.py (3 importers in audit window, importers themselves DEAD)
- app/splade_bm25.py (LIVE via app.rag_service)
- app/wallet_manager_v2.py (LIVE via x402_enforcement, x402_tools, sweep_all, sweep_now)
- app/crypto_embeddings.py (NOT in audit ARCHIVE list; heavy import graph)
Verification (forward-import closure from mount.py + main.py + factory.py + lifespan.py):
- imports = 348 app.* modules
- reached = 194 files reachable from roots
- archive set = audit_dead (186) - reached - forced_live (4) - test_live (1) = 136
- Net delta: 136 files moved, 44,932 LOC reduction, 293->295 active routes (+2 from Wave 3)
pyproject.toml updates:
- setuptools.packages.find: added exclude for app._archive*
- ruff.extend-exclude: added "app/_archive/"
- mypy.exclude: added "app/_archive/"
Smoke test: pytest tests/ — 817 passed, 3 pre-existing failures unchanged
(0 new failures; 0 routes lost; all 4 forced-LIVE files still importable).
Restoration: git mv app/_archive/legacy_2026_07/<name>.py <original-path>
and add the import to app/mount.py ROUTER_MODULES.
Refs: AUDIT-2026-Q3.md /home/dev/pry/rmi-final-deadcode-2026-07-06.md
154 lines
5.1 KiB
Python
154 lines
5.1 KiB
Python
"""A/B Testing Framework - split traffic, measure accuracy, auto-promote winners."""
|
|
|
|
import hashlib
|
|
from datetime import UTC, datetime
|
|
|
|
from fastapi import APIRouter, Query
|
|
|
|
router = APIRouter(prefix="/api/v1/ab-testing", tags=["ab-testing"])
|
|
|
|
_experiments: dict[str, dict] = {}
|
|
_results: dict[str, list] = {}
|
|
|
|
|
|
@router.post("/experiment")
|
|
async def create_experiment(
|
|
name: str,
|
|
variants: str = Query(..., description="Comma-separated variant names, e.g. 'prompt_v1,prompt_v2'"),
|
|
traffic_split: str = Query("50,50", description="Traffic split percentages"),
|
|
metric: str = Query("accuracy", description="Metric to evaluate"),
|
|
):
|
|
"""Create a new A/B test experiment."""
|
|
variant_list = [v.strip() for v in variants.split(",")]
|
|
split_list = [int(s.strip()) for s in traffic_split.split(",")]
|
|
|
|
if len(variant_list) != len(split_list):
|
|
return {"error": "Variants and splits must have same length"}
|
|
if sum(split_list) != 100:
|
|
return {"error": "Traffic splits must sum to 100"}
|
|
|
|
exp_id = hashlib.md5(name.encode()).hexdigest()[:8]
|
|
_experiments[exp_id] = {
|
|
"id": exp_id,
|
|
"name": name,
|
|
"variants": variant_list,
|
|
"traffic_split": split_list,
|
|
"metric": metric,
|
|
"created_at": datetime.now(UTC).isoformat(),
|
|
"status": "active",
|
|
}
|
|
_results[exp_id] = []
|
|
return _experiments[exp_id]
|
|
|
|
|
|
@router.get("/assign/{experiment_id}")
|
|
async def assign_variant(experiment_id: str, user_id: str = "anonymous"):
|
|
"""Assign a user to a variant deterministically (same user always gets same variant)."""
|
|
if experiment_id not in _experiments:
|
|
return {"variant": "control", "note": "Experiment not found"}
|
|
|
|
exp = _experiments[experiment_id]
|
|
# Deterministic assignment based on user_id hash
|
|
h = int(hashlib.md5(f"{experiment_id}:{user_id}".encode()).hexdigest(), 16)
|
|
bucket = h % 100
|
|
|
|
cumulative = 0
|
|
for variant, split in zip(exp["variants"], exp["traffic_split"], strict=False):
|
|
cumulative += split
|
|
if bucket < cumulative:
|
|
return {"experiment": experiment_id, "variant": variant, "user": user_id}
|
|
|
|
return {"experiment": experiment_id, "variant": exp["variants"][-1], "user": user_id}
|
|
|
|
|
|
@router.post("/record/{experiment_id}")
|
|
async def record_result(experiment_id: str, variant: str, correct: bool = False):
|
|
"""Record a result for an A/B test variant."""
|
|
if experiment_id not in _experiments:
|
|
return {"error": "Experiment not found"}
|
|
|
|
_results[experiment_id].append(
|
|
{
|
|
"variant": variant,
|
|
"correct": correct,
|
|
"timestamp": datetime.now(UTC).isoformat(),
|
|
}
|
|
)
|
|
|
|
# Check if we have enough data to declare a winner (30+ samples per variant)
|
|
results = _results[experiment_id]
|
|
variant_counts = {}
|
|
variant_correct = {}
|
|
for r in results:
|
|
v = r["variant"]
|
|
variant_counts[v] = variant_counts.get(v, 0) + 1
|
|
if r["correct"]:
|
|
variant_correct[v] = variant_correct.get(v, 0) + 1
|
|
|
|
# Check if all variants have enough samples
|
|
min_samples = 30
|
|
all_ready = all(count >= min_samples for count in variant_counts.values())
|
|
|
|
if all_ready:
|
|
scores = {v: variant_correct.get(v, 0) / variant_counts[v] for v in variant_counts}
|
|
best = max(scores, key=scores.get)
|
|
worst = min(scores, key=scores.get)
|
|
improvement = scores[best] - scores[worst]
|
|
|
|
result = {
|
|
"experiment": experiment_id,
|
|
"status": "complete" if improvement > 0.05 else "inconclusive",
|
|
"winner": best if improvement > 0.05 else None,
|
|
"scores": {v: round(s, 3) for v, s in scores.items()},
|
|
"samples": variant_counts,
|
|
"confidence": round(improvement * 100, 1),
|
|
}
|
|
|
|
if improvement > 0.05:
|
|
_experiments[experiment_id]["status"] = "complete"
|
|
_experiments[experiment_id]["winner"] = best
|
|
|
|
return result
|
|
|
|
return {
|
|
"experiment": experiment_id,
|
|
"status": "collecting",
|
|
"samples": variant_counts,
|
|
"min_needed": min_samples,
|
|
}
|
|
|
|
|
|
@router.get("/experiments")
|
|
async def list_experiments():
|
|
return {
|
|
"experiments": list(_experiments.values()),
|
|
"active": sum(1 for e in _experiments.values() if e["status"] == "active"),
|
|
}
|
|
|
|
|
|
@router.get("/experiment/{experiment_id}")
|
|
async def get_experiment(experiment_id: str):
|
|
if experiment_id not in _experiments:
|
|
return {"error": "Experiment not found"}
|
|
exp = _experiments[experiment_id]
|
|
results = _results.get(experiment_id, [])
|
|
return {
|
|
**exp,
|
|
"total_trials": len(results),
|
|
"results": _summarize_results(results, exp["variants"]),
|
|
}
|
|
|
|
|
|
def _summarize_results(results: list, variants: list) -> dict:
|
|
counts = dict.fromkeys(variants, 0)
|
|
correct = dict.fromkeys(variants, 0)
|
|
for r in results:
|
|
v = r["variant"]
|
|
if v in counts:
|
|
counts[v] += 1
|
|
if r["correct"]:
|
|
correct[v] += 1
|
|
return {
|
|
v: {"trials": counts[v], "correct": correct[v], "accuracy": round(correct[v] / max(counts[v], 1), 3)}
|
|
for v in variants
|
|
}
|