generated from coulomb/repo-seed
tegwick
a4dde53fc3
Ship a starter model rate table at src/infospace_bench/model_rates.yaml (prompt_per_1k / completion_per_1k for the OpenRouter models we have actually touched: gpt-4o, gpt-4o-mini, gpt-4-turbo, claude 3.5 sonnet and haiku, claude 3 opus, gemini 1.5 flash/pro, llama 3.1 70b) and a load_rate_table() / estimate_cost_usd() pair that overlays an optional <workspace>/model-rates.yaml on top of the bundled defaults. generate run now passes a workspace-aware cost_resolver into record_run_usage, so cost_usd_estimated lands on every usage bucket whose model matches the table. Adapter-returned cost still wins (cost_status="known"); rate-table cost is reported under cost_status="estimated"; unmatched models are recorded as cost_status="unknown" rather than silently zeroed. Rate-table file is listed in pyproject.toml package-data so pip-installed users keep the defaults. 106 tests pass. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
418 lines
15 KiB
Python
418 lines
15 KiB
Python
import json
|
|
import os
|
|
import subprocess
|
|
import sys
|
|
import zipfile
|
|
from pathlib import Path
|
|
|
|
import yaml
|
|
|
|
from infospace_bench.budget import (
|
|
PLAN_RETENTION_DEFAULT,
|
|
PLANS_FILE,
|
|
PLANS_SCHEMA_VERSION,
|
|
read_plan_snapshots,
|
|
record_plan_snapshot,
|
|
)
|
|
from infospace_bench.generator import init_generation_infospace, plan_generation
|
|
|
|
|
|
CONTAINER_XML = """<?xml version="1.0"?>
|
|
<container version="1.0" xmlns="urn:oasis:names:tc:opendocument:xmlns:container">
|
|
<rootfiles>
|
|
<rootfile full-path="OEBPS/content.opf" media-type="application/oebps-package+xml"/>
|
|
</rootfiles>
|
|
</container>
|
|
"""
|
|
|
|
PACKAGE_OPF = """<?xml version="1.0" encoding="utf-8"?>
|
|
<package xmlns="http://www.idpf.org/2007/opf" version="3.0" unique-identifier="bookid">
|
|
<metadata xmlns:dc="http://purl.org/dc/elements/1.1/">
|
|
<dc:identifier id="bookid">urn:test:budget</dc:identifier>
|
|
<dc:title>Budget Test Book</dc:title>
|
|
<dc:creator>Author</dc:creator>
|
|
<dc:language>en</dc:language>
|
|
</metadata>
|
|
<manifest>
|
|
<item id="ch1" href="ch1.xhtml" media-type="application/xhtml+xml"/>
|
|
<item id="ch2" href="ch2.xhtml" media-type="application/xhtml+xml"/>
|
|
<item id="ch3" href="ch3.xhtml" media-type="application/xhtml+xml"/>
|
|
</manifest>
|
|
<spine>
|
|
<itemref idref="ch1"/>
|
|
<itemref idref="ch2"/>
|
|
<itemref idref="ch3"/>
|
|
</spine>
|
|
</package>
|
|
"""
|
|
|
|
|
|
def _write_three_chapter_epub(path: Path) -> None:
|
|
with zipfile.ZipFile(path, "w") as archive:
|
|
archive.writestr("mimetype", "application/epub+zip")
|
|
archive.writestr("META-INF/container.xml", CONTAINER_XML)
|
|
archive.writestr("OEBPS/content.opf", PACKAGE_OPF)
|
|
for idx, label in enumerate(("I", "II", "III"), start=1):
|
|
archive.writestr(
|
|
f"OEBPS/ch{idx}.xhtml",
|
|
f"<html><head><title>Book</title></head>"
|
|
f"<body><h2>{label}</h2>"
|
|
f"<p>Body of chapter {label} with " + " ".join(f"word{n}" for n in range(40)) + ".</p></body></html>",
|
|
)
|
|
|
|
|
|
def _write_minimal_fixture(path: Path) -> None:
|
|
data = {
|
|
"responses": [
|
|
{
|
|
"stage_id": "summarize-source",
|
|
"input_artifact_id": "*",
|
|
"markdown": "# Source Summary\n\nA stub summary.\n",
|
|
},
|
|
{
|
|
"stage_id": "extract-entities",
|
|
"input_artifact_id": "*",
|
|
"markdown": (
|
|
"# Stub Entity\n\n## Definition\n\nA stub.\n\n## Context\n\nFor a budget test.\n"
|
|
),
|
|
},
|
|
{
|
|
"stage_id": "extract-relations",
|
|
"input_artifact_id": "*",
|
|
"markdown": (
|
|
"# Stub Entity Practices Something\n\n## Subject\n\nStub Entity\n\n"
|
|
"## Predicate\n\npractices\n\n## Object\n\nSomething\n\n## Relation Type\n\nsupport\n\n"
|
|
"## Evidence\n\nA stub.\n"
|
|
),
|
|
},
|
|
{
|
|
"stage_id": "evaluate-entity",
|
|
"input_artifact_id": "*",
|
|
"markdown": (
|
|
"---\nartifact_id: entity/stub-entity.md\nevaluator: fixture\n"
|
|
"evaluated_at: '2026-05-17T00:00:00'\n"
|
|
"scores:\n - name: groundedness\n value: 4.0\n max_value: 5.0\n"
|
|
" - name: usefulness\n value: 4.0\n max_value: 5.0\n---\n\n"
|
|
"# Evaluation: entity/stub-entity.md\n"
|
|
),
|
|
},
|
|
]
|
|
}
|
|
path.write_text(yaml.safe_dump(data, sort_keys=False), encoding="utf-8")
|
|
|
|
|
|
def _build_infospace(tmp_path: Path) -> Path:
|
|
book = tmp_path / "book.epub"
|
|
_write_three_chapter_epub(book)
|
|
infospace = init_generation_infospace(
|
|
tmp_path, book, "budget-test", name="Budget Test", profile="general-knowledge"
|
|
)
|
|
return infospace.root
|
|
|
|
|
|
def test_record_plan_snapshot_writes_yaml_with_stable_id(tmp_path: Path) -> None:
|
|
root = _build_infospace(tmp_path)
|
|
|
|
summary = plan_generation(root, persist=False)
|
|
snapshot_id_1 = record_plan_snapshot(root, summary)
|
|
snapshot_id_2 = record_plan_snapshot(root, summary)
|
|
|
|
persisted = (root / PLANS_FILE).read_text(encoding="utf-8")
|
|
data = yaml.safe_load(persisted)
|
|
|
|
assert data["schema_version"] == PLANS_SCHEMA_VERSION
|
|
assert data["pruned_count"] == 0
|
|
assert snapshot_id_1 == snapshot_id_2, "same summary must yield same snapshot_id"
|
|
# Duplicate writes refresh recorded_at instead of stacking
|
|
assert len(data["snapshots"]) == 1
|
|
assert data["snapshots"][0]["snapshot_id"] == snapshot_id_1
|
|
|
|
|
|
def test_different_filters_produce_distinct_snapshots(tmp_path: Path) -> None:
|
|
root = _build_infospace(tmp_path)
|
|
|
|
full_plan = plan_generation(root, persist=False)
|
|
chapter_only = plan_generation(root, from_chapter=2, to_chapter=2, persist=False)
|
|
record_plan_snapshot(root, full_plan)
|
|
record_plan_snapshot(root, chapter_only)
|
|
|
|
snapshots = read_plan_snapshots(root)
|
|
assert len(snapshots) == 2
|
|
ids = {snap["snapshot_id"] for snap in snapshots}
|
|
assert len(ids) == 2
|
|
# Filter values are echoed back into the snapshot
|
|
chapter_snapshot = next(s for s in snapshots if s["selected_chunk_count"] == 1)
|
|
assert chapter_snapshot["filters"]["from_chapter"] == 2
|
|
assert chapter_snapshot["filters"]["to_chapter"] == 2
|
|
|
|
|
|
def test_plan_generation_persists_snapshot_by_default(tmp_path: Path) -> None:
|
|
root = _build_infospace(tmp_path)
|
|
|
|
result = plan_generation(root, from_chapter=1, to_chapter=2)
|
|
|
|
assert "snapshot_id" in result
|
|
assert (root / PLANS_FILE).is_file()
|
|
snapshots = read_plan_snapshots(root)
|
|
assert len(snapshots) == 1
|
|
assert snapshots[0]["snapshot_id"] == result["snapshot_id"]
|
|
|
|
|
|
def test_plan_generation_persist_false_skips_write(tmp_path: Path) -> None:
|
|
root = _build_infospace(tmp_path)
|
|
|
|
plan_generation(root, persist=False)
|
|
|
|
assert not (root / PLANS_FILE).exists()
|
|
|
|
|
|
def test_plan_snapshot_retention_prunes_old_entries(tmp_path: Path) -> None:
|
|
root = _build_infospace(tmp_path)
|
|
|
|
# Produce 5 distinct snapshots and cap retention at 3.
|
|
for chapter in (1, 2, 3, None, None):
|
|
kwargs = {"from_chapter": chapter, "to_chapter": chapter} if chapter else {}
|
|
summary = plan_generation(root, persist=False, **kwargs)
|
|
if not chapter:
|
|
# vary another field to avoid duplicate refresh
|
|
summary["max_calls"] = (summary.get("max_calls") or 0) + 1
|
|
summary["exceeds_max_calls"] = False
|
|
record_plan_snapshot(root, summary, retention=3)
|
|
|
|
data = yaml.safe_load((root / PLANS_FILE).read_text(encoding="utf-8"))
|
|
assert len(data["snapshots"]) == 3
|
|
assert data["pruned_count"] >= 1
|
|
|
|
|
|
def test_record_run_usage_aggregates_by_workflow_stage_provider_model(tmp_path: Path) -> None:
|
|
root = _build_infospace(tmp_path)
|
|
from infospace_bench.budget import record_run_usage, read_usage_runs
|
|
|
|
workflow_results = [
|
|
{
|
|
"run_id": "run-1",
|
|
"workflow_id": "generic-source-entities",
|
|
"status": "completed",
|
|
"stages": [
|
|
{
|
|
"stage_id": "extract-entities",
|
|
"provider": "openrouter",
|
|
"metadata": {
|
|
"model": "openai/gpt-4o-mini",
|
|
"usage": {"prompt_tokens": 1000, "completion_tokens": 200, "total_tokens": 1200},
|
|
},
|
|
},
|
|
{
|
|
"stage_id": "extract-entities",
|
|
"provider": "openrouter",
|
|
"metadata": {
|
|
"model": "openai/gpt-4o-mini",
|
|
"usage": {"prompt_tokens": 800, "completion_tokens": 150, "cost": 0.0012},
|
|
},
|
|
},
|
|
{"stage_id": "split-entities", "message": "split 3 entities"},
|
|
],
|
|
}
|
|
]
|
|
|
|
entry = record_run_usage(root, workflow_results, snapshot_id="abc123", duration_seconds=4.2)
|
|
|
|
assert entry["rollup"]["total_calls"] == 2
|
|
assert entry["rollup"]["total_prompt_tokens"] == 1800
|
|
assert entry["rollup"]["total_completion_tokens"] == 350
|
|
assert entry["rollup"]["total_cost_usd_known"] == 0.0012
|
|
assert entry["snapshot_id"] == "abc123"
|
|
assert entry["duration_seconds"] == 4.2
|
|
assert len(entry["per_bucket"]) == 1
|
|
bucket = entry["per_bucket"][0]
|
|
assert bucket["workflow_id"] == "generic-source-entities"
|
|
assert bucket["stage_id"] == "extract-entities"
|
|
assert bucket["provider"] == "openrouter"
|
|
assert bucket["model"] == "openai/gpt-4o-mini"
|
|
assert bucket["calls"] == 2
|
|
|
|
runs = read_usage_runs(root)
|
|
assert len(runs) == 1
|
|
assert runs[0]["run_index"] == 1
|
|
|
|
|
|
def test_record_run_usage_handles_fixture_runs_without_aborting(tmp_path: Path) -> None:
|
|
root = _build_infospace(tmp_path)
|
|
from infospace_bench.budget import record_run_usage
|
|
|
|
workflow_results = [
|
|
{
|
|
"run_id": "fix-1",
|
|
"workflow_id": "generic-source-summary",
|
|
"stages": [
|
|
{"stage_id": "summarize-source", "provider": "fixture"},
|
|
{"stage_id": "summarize-source", "provider": "fixture"},
|
|
],
|
|
}
|
|
]
|
|
|
|
entry = record_run_usage(root, workflow_results)
|
|
|
|
fixture_bucket = next(b for b in entry["per_bucket"] if b["provider"] == "fixture")
|
|
assert fixture_bucket["calls"] == 2
|
|
assert fixture_bucket["prompt_tokens"] == 0
|
|
assert fixture_bucket["cost_status"] == "unknown"
|
|
assert entry["rollup"]["total_cost_usd_known"] == 0.0
|
|
|
|
|
|
def test_run_generation_writes_usage_yaml_with_plan_snapshot_id(tmp_path: Path) -> None:
|
|
root = _build_infospace(tmp_path)
|
|
from infospace_bench.budget import USAGE_FILE, read_usage_runs
|
|
from infospace_bench.generator import run_generation
|
|
|
|
fixture = tmp_path / "responses.yaml"
|
|
_write_minimal_fixture(fixture)
|
|
|
|
plan_payload = plan_generation(root)
|
|
run_generation(root, fixture_responses=fixture)
|
|
|
|
runs = read_usage_runs(root)
|
|
assert (root / USAGE_FILE).is_file()
|
|
assert len(runs) == 1
|
|
assert runs[0]["snapshot_id"] == plan_payload["snapshot_id"]
|
|
assert runs[0]["duration_seconds"] is not None and runs[0]["duration_seconds"] >= 0
|
|
assert runs[0]["rollup"]["total_calls"] >= 0
|
|
# Fixture mode runs should not claim any known cost
|
|
assert runs[0]["rollup"]["total_cost_usd_known"] == 0.0
|
|
|
|
|
|
def test_rate_table_known_model_resolves_cost(tmp_path: Path) -> None:
|
|
from infospace_bench.budget import estimate_cost_usd, load_rate_table
|
|
|
|
rates = load_rate_table()
|
|
|
|
assert "openai/gpt-4o-mini" in rates
|
|
cost = estimate_cost_usd("openai/gpt-4o-mini", 1000, 500, rates)
|
|
# gpt-4o-mini: prompt 0.00015/1k, completion 0.0006/1k → 0.00015 + 0.0003 = 0.00045
|
|
assert cost is not None
|
|
assert abs(cost - 0.00045) < 1e-9
|
|
|
|
|
|
def test_rate_table_unknown_model_returns_none(tmp_path: Path) -> None:
|
|
from infospace_bench.budget import estimate_cost_usd, load_rate_table
|
|
|
|
rates = load_rate_table()
|
|
|
|
assert estimate_cost_usd("acme/no-such-model", 1000, 500, rates) is None
|
|
|
|
|
|
def test_workspace_rate_table_overrides_package_default(tmp_path: Path) -> None:
|
|
from infospace_bench.budget import estimate_cost_usd, load_rate_table
|
|
|
|
override = tmp_path / "model-rates.yaml"
|
|
override.write_text(
|
|
yaml.safe_dump(
|
|
{
|
|
"schema_version": 1,
|
|
"rates": {
|
|
"openai/gpt-4o-mini": {
|
|
"prompt_per_1k": 1.0,
|
|
"completion_per_1k": 2.0,
|
|
},
|
|
"acme/bespoke": {
|
|
"prompt_per_1k": 0.1,
|
|
"completion_per_1k": 0.2,
|
|
},
|
|
},
|
|
}
|
|
),
|
|
encoding="utf-8",
|
|
)
|
|
|
|
rates = load_rate_table(tmp_path)
|
|
|
|
overridden = estimate_cost_usd("openai/gpt-4o-mini", 1000, 1000, rates)
|
|
bespoke = estimate_cost_usd("acme/bespoke", 1000, 1000, rates)
|
|
|
|
assert overridden == round(1.0 + 2.0, 6)
|
|
assert bespoke == round(0.1 + 0.2, 6)
|
|
|
|
|
|
def test_record_run_usage_fills_estimated_cost_via_resolver(tmp_path: Path) -> None:
|
|
root = _build_infospace(tmp_path)
|
|
from infospace_bench.budget import make_cost_resolver, record_run_usage
|
|
|
|
workflow_results = [
|
|
{
|
|
"run_id": "run-cost",
|
|
"workflow_id": "generic-source-entities",
|
|
"stages": [
|
|
{
|
|
"stage_id": "extract-entities",
|
|
"provider": "openrouter",
|
|
"metadata": {
|
|
"model": "openai/gpt-4o-mini",
|
|
"usage": {"prompt_tokens": 2000, "completion_tokens": 1000},
|
|
},
|
|
},
|
|
{
|
|
"stage_id": "extract-entities",
|
|
"provider": "openrouter",
|
|
"metadata": {
|
|
"model": "openai/gpt-4o-mini",
|
|
"usage": {
|
|
"prompt_tokens": 1000,
|
|
"completion_tokens": 500,
|
|
"cost": 0.123,
|
|
},
|
|
},
|
|
},
|
|
],
|
|
}
|
|
]
|
|
|
|
entry = record_run_usage(
|
|
root,
|
|
workflow_results,
|
|
cost_resolver=make_cost_resolver(tmp_path),
|
|
)
|
|
|
|
bucket = entry["per_bucket"][0]
|
|
# The first call has no adapter cost so it gets estimated:
|
|
# 2000/1000*0.00015 + 1000/1000*0.0006 = 0.0003 + 0.0006 = 0.0009
|
|
assert bucket["cost_usd_estimated"] == round(0.0009, 6)
|
|
assert bucket["cost_usd_known"] == 0.123
|
|
assert bucket["cost_status"] == "known" # at least one call returned cost
|
|
assert entry["rollup"]["total_cost_usd_known"] == 0.123
|
|
assert entry["rollup"]["total_cost_usd_estimated"] == round(0.0009, 6)
|
|
|
|
|
|
def test_plan_cli_writes_snapshot(tmp_path: Path) -> None:
|
|
root = _build_infospace(tmp_path)
|
|
env = os.environ.copy()
|
|
env["PYTHONPATH"] = "src:/home/worsch/markitect-tool/src"
|
|
|
|
result = subprocess.run(
|
|
[
|
|
sys.executable,
|
|
"-m",
|
|
"infospace_bench",
|
|
"generate",
|
|
"plan",
|
|
str(root),
|
|
"--from-chapter",
|
|
"1",
|
|
"--to-chapter",
|
|
"2",
|
|
"--cost-per-1k",
|
|
"0.5",
|
|
],
|
|
check=False,
|
|
env=env,
|
|
text=True,
|
|
capture_output=True,
|
|
)
|
|
|
|
assert result.returncode == 0, result.stderr
|
|
payload = json.loads(result.stdout)
|
|
assert "snapshot_id" in payload
|
|
snapshots = read_plan_snapshots(root)
|
|
assert len(snapshots) == 1
|
|
assert snapshots[0]["filters"]["from_chapter"] == 1
|
|
assert snapshots[0]["filters"]["to_chapter"] == 2
|