Skip to content

Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.

29.0 KB
Raw Blame
"""Tests for the durable savings event ledger and the `headroom savings` CLI."""

from __future__ import annotations

import json
import logging
from datetime import datetime, timedelta, timezone

import pytest

from headroom import savings_ledger as L
from headroom.log_safety import WarnOnce
from tests._pricing_models import anthropic_pricing_model

MODEL = anthropic_pricing_model()

UTC = timezone.utc


def _events_env(monkeypatch, tmp_path):
    path = tmp_path / "savings_events.jsonl"
    monkeypatch.setenv("HEADROOM_SAVINGS_EVENTS_PATH", str(path))
    return path


# --------------------------------------------------------------------------- #
# core ledger
# --------------------------------------------------------------------------- #


def test_unknown_model_uses_blended_fallback(monkeypatch, tmp_path):
    _events_env(monkeypatch, tmp_path)
    assert L.record_savings_event(tokens_before=1000, tokens_after=400, model=None, client="c")
    report = L.aggregate_savings()
    assert report.lifetime["tokens_saved"] == 600
    expected = round(600 * L.DEFAULT_FALLBACK_INPUT_COST_PER_TOKEN, 6)
    assert report.lifetime["cost_usd"] == pytest.approx(expected)
    assert any(row["model"] == "unknown" for row in report.by_model)


def test_estimate_cost_unknown_short_circuits_to_fallback():
    assert L.estimate_cost_usd("unknown", 1000, fallback_rate=1e-6) == pytest.approx(0.001)
    assert L.estimate_cost_usd(L.UNKNOWN, 0) == 0.0


def test_free_model_is_not_billed_at_fallback(monkeypatch):
    """A known but 0-priced (free) model must cost $0, not the $3/M fallback.

    _estimate_compression_savings_usd returns a legitimate 0.0 for free models;
    the ledger must trust that rather than re-billing it at the blended rate.
    """
    monkeypatch.setattr(L, "_estimate_compression_savings_usd", lambda model, tokens: 0.0)

    assert L.estimate_cost_usd("free-local-model", 1_000_000) == 0.0


def test_priced_model_uses_litellm_estimate(monkeypatch):
    monkeypatch.setattr(L, "_estimate_compression_savings_usd", lambda model, tokens: tokens * 2e-6)

    assert L.estimate_cost_usd("some-model", 1_000_000) == pytest.approx(2.0)


def test_explicit_cost_is_honored(monkeypatch, tmp_path):
    _events_env(monkeypatch, tmp_path)
    L.record_savings_event(tokens_before=100, tokens_after=10, model="x", client="c", cost_usd=1.25)
    assert L.aggregate_savings().lifetime["cost_usd"] == pytest.approx(1.25)


def test_zero_or_negative_savings_not_recorded(monkeypatch, tmp_path):
    path = _events_env(monkeypatch, tmp_path)
    assert L.record_savings_event(tokens_before=100, tokens_after=100) is False
    assert L.record_savings_event(tokens_before=50, tokens_after=80) is False
    assert not path.exists() or path.read_text().strip() == ""
    assert L.aggregate_savings().lifetime["calls"] == 0


def test_stateless_mode_does_not_create_ledger(monkeypatch, tmp_path):
    path = _events_env(monkeypatch, tmp_path)
    monkeypatch.setenv("HEADROOM_STATELESS", "true")
    assert L.record_savings_event(tokens_before=100, tokens_after=50) is False
    assert not path.exists()


def test_breakdowns_aggregate_by_dimension(monkeypatch, tmp_path):
    _events_env(monkeypatch, tmp_path)
    L.record_savings_event(tokens_before=1000, tokens_after=300, model=None, client="claude-code")
    L.record_savings_event(tokens_before=500, tokens_after=200, model=None, client="claude-code")
    L.record_savings_event(
        tokens_before=2000, tokens_after=600, model="gpt", client="proxy", cost_usd=0.5
    )
    report = L.aggregate_savings()

    clients = {row["client"]: row for row in report.by_client}
    assert clients["claude-code"]["calls"] == 2
    assert clients["proxy"]["tokens_saved"] == 1400


def test_windows_today_week_last30(monkeypatch, tmp_path):
    _events_env(monkeypatch, tmp_path)
    now = datetime(2026, 6, 17, 12, 0, tzinfo=UTC)
    L.record_savings_event(
        tokens_before=1000, tokens_after=500, model=None, client="c", timestamp=now
    )
    L.record_savings_event(
        tokens_before=1000,
        tokens_after=600,
        model=None,
        client="c",
        timestamp=now - timedelta(days=3),
    )
    L.record_savings_event(
        tokens_before=1000,
        tokens_after=700,
        model=None,
        client="c",
        timestamp=now - timedelta(days=20),
    )
    report = L.aggregate_savings(now=now)
    assert report.windows["today"]["tokens_saved"] == 500
    assert report.windows["last_7_days"]["tokens_saved"] == 500 + 400
    assert report.windows["last_30_days"]["tokens_saved"] == 500 + 400 + 300
    assert report.windows["last_30_days"]["calls"] == 3
    # 500 saved out of 1000 before today
    assert report.windows["today"]["savings_percent"] == pytest.approx(50.0)


def test_new_input_basis_pairs_compression_only_with_new_input(monkeypatch, tmp_path):
    _events_env(monkeypatch, tmp_path)
    # 400 saved, of which 100 is tool-schema deferral; 900 tokens newly entered
    # context. Compression-only 300 / (900 + 300) = 25%, same as /stats.
    L.record_savings_event(
        tokens_before=10_000,
        tokens_after=9_600,
        model=None,
        client="claude-code",
        new_input_tokens=900,
        deferred_tokens=100,
    )
    # An event without a cache breakdown (MCP tool, Bedrock) keeps writing the
    # old line and must not lend its savings to the new-input ratio.
    L.record_savings_event(tokens_before=1000, tokens_after=500, model=None, client="mcp")
    report = L.aggregate_savings()
    window = report.windows["today"]
    assert window["tokens_saved"] == 900
    assert window["new_input_tokens"] == 900
    assert window["new_input_savings_percent"] == pytest.approx(25.0)
    clients = {row["client"]: row for row in report.by_client}
    assert clients["mcp"]["new_input_tokens"] == 0
    assert clients["mcp"]["new_input_savings_percent"] == 0.0
    # The whole-wire ratio is untouched by the new field.
    assert window["savings_percent"] == pytest.approx(900 / 11_000 * 100, abs=0.1)


def test_savings_cli_prints_new_input_line_only_with_cache_data(monkeypatch, tmp_path):
    from click.testing import CliRunner

    from headroom.cli.savings import savings

    _events_env(monkeypatch, tmp_path)
    L.record_savings_event(tokens_before=1000, tokens_after=500, model=None, client="mcp")
    out = CliRunner().invoke(savings, []).output
    assert "of new input" not in out
    L.record_savings_event(
        tokens_before=10_000, tokens_after=9_600, model=None, client="c", new_input_tokens=900
    )
    out = CliRunner().invoke(savings, []).output
    assert "of new input 30.8%" in out
    assert "newly entered context" in out


def test_retention_hard_capped_at_30_days(monkeypatch, tmp_path):
    _events_env(monkeypatch, tmp_path)
    now = datetime(2026, 6, 17, 12, 0, tzinfo=UTC)
    L.record_savings_event(
        tokens_before=1000, tokens_after=500, model=None, client="c", timestamp=now
    )
    # 60 days old: within the requested 365-day window but past the 30-day cap.
    L.record_savings_event(
        tokens_before=1000,
        tokens_after=500,
        model=None,
        client="c",
        timestamp=now - timedelta(days=60),
    )
    # Caller asks for 365 days, but retention is hard-capped at 30.
    report = L.aggregate_savings(now=now, retention_days=365)
    assert report.lifetime["calls"] == 1
    assert report.windows["last_30_days"]["calls"] == 1


def test_appends_do_not_clobber_and_survive_restart(monkeypatch, tmp_path):
    path = _events_env(monkeypatch, tmp_path)
    for _ in range(5):
        L.record_savings_event(tokens_before=100, tokens_after=10, model=None, client="c")
    lines = [ln for ln in path.read_text().splitlines() if ln.strip()]
    assert len(lines) == 5
    # aggregate_savings holds no in-memory state — it reads purely from disk,
    # so this also proves durability across a process restart.
    assert L.aggregate_savings().lifetime["calls"] == 5
    assert L.aggregate_savings().lifetime["tokens_saved"] == 5 * 90


def test_corrupt_lines_are_skipped(monkeypatch, tmp_path):
    path = _events_env(monkeypatch, tmp_path)
    L.record_savings_event(tokens_before=1000, tokens_after=400, model=None, client="c")
    with open(path, "a", encoding="utf-8") as handle:
        handle.write("not json\n\n")
    assert L.aggregate_savings().lifetime["calls"] == 1


# --------------------------------------------------------------------------- #
# CLI
# --------------------------------------------------------------------------- #


def test_cli_reset_deletes_ledger(monkeypatch, tmp_path):
    pytest.importorskip("click")
    from click.testing import CliRunner

    from headroom.cli.savings import savings

    path = _events_env(monkeypatch, tmp_path)
    L.record_savings_event(tokens_before=1000, tokens_after=300, model=None, client="claude-code")
    assert path.exists()

    result = CliRunner().invoke(savings, ["--reset"])
    assert result.exit_code == 0
    assert "reset" in result.output.lower()
    assert not path.exists()

    # second reset on missing file is a no-op
    result2 = CliRunner().invoke(savings, ["--reset"])
    assert result2.exit_code == 0
    assert "Nothing to reset" in result2.output


def test_cli_empty_state(monkeypatch, tmp_path):
    pytest.importorskip("click")
    from click.testing import CliRunner

    from headroom.cli.savings import savings

    _events_env(monkeypatch, tmp_path)
    result = CliRunner().invoke(savings, [])
    assert result.exit_code == 0
    assert "No savings recorded yet." in result.output


def test_cli_renders_sections_and_json(monkeypatch, tmp_path):
    pytest.importorskip("click")
    from click.testing import CliRunner

    from headroom.cli.savings import savings

    _events_env(monkeypatch, tmp_path)
    L.record_savings_event(tokens_before=1000, tokens_after=300, model=None, client="claude-code")
    runner = CliRunner()

    result = runner.invoke(savings, [])
    assert result.exit_code == 0
    # No redundant top-line headline; the windows lead the output.
    assert "cost avoided" not in result.output
    assert "Today" in result.output and "Last 30 days" in result.output
    assert "Savings by client" in result.output and "claude-code" in result.output
    assert "Per-repo totals" not in result.output

    result_json = runner.invoke(savings, ["--json"])
    assert result_json.exit_code == 0
    payload = json.loads(result_json.output)
    assert payload["lifetime"]["tokens_saved"] == 700
    assert payload["windows"]["last_30_days"]["calls"] == 1
    assert "by_repo" not in payload


# --------------------------------------------------------------------------- #
# MCP tool path
# --------------------------------------------------------------------------- #


def test_mcp_compress_records_durable_event(monkeypatch, tmp_path):
    pytest.importorskip("mcp", reason="MCP SDK required")
    from headroom.ccr import mcp_server

    _events_env(monkeypatch, tmp_path)
    monkeypatch.setenv("HEADROOM_MCP_CLIENT", "claude-code")

    server = mcp_server.HeadroomMCPServer(check_proxy=False)
    server._record_savings({"original_tokens": 1000, "compressed_tokens": 250})

    report = L.aggregate_savings()
    assert report.lifetime["tokens_saved"] == 750
    assert {row["client"] for row in report.by_client} == {"claude-code"}


def test_mcp_record_savings_ignores_noop(monkeypatch, tmp_path):
    pytest.importorskip("mcp", reason="MCP SDK required")
    from headroom.ccr import mcp_server

    _events_env(monkeypatch, tmp_path)
    server = mcp_server.HeadroomMCPServer(check_proxy=False)
    server._record_savings({"original_tokens": 500, "compressed_tokens": 500})
    assert L.aggregate_savings().lifetime["calls"] == 0


# --------------------------------------------------------------------------- #
# proxy path
# --------------------------------------------------------------------------- #


def test_proxy_record_request_appends_ledger_event(tmp_path, monkeypatch):
    pytest.importorskip("fastapi")
    import asyncio

    from fastapi.testclient import TestClient

    from headroom.proxy.server import ProxyConfig, create_app

    monkeypatch.setenv("HEADROOM_SAVINGS_PATH", str(tmp_path / "proxy_savings.json"))
    monkeypatch.setenv("HEADROOM_SAVINGS_EVENTS_PATH", str(tmp_path / "savings_events.jsonl"))
    monkeypatch.setattr(
        "headroom.proxy.server.CostTracker._get_cache_prices",
        # (cache_read, cache_write_5m, cache_write_1h, uncached). **kwargs so
        # the stub keeps standing in as the real signature grows -- it takes a
        # keyword-only `long_context` tier selector.
        lambda self, model, **kwargs: (0.001, 0.0015, 0.0024, 0.002),
    )

    config = ProxyConfig(cache_enabled=False, rate_limit_enabled=False, log_requests=False)
    with TestClient(create_app(config)) as client:
        proxy = client.app.state.proxy
        # identified harness -> recorded as that client
        asyncio.run(
            proxy.metrics.record_request(
                provider="openai",
                model="gpt-4o",
                input_tokens=120,
                output_tokens=24,
                tokens_saved=40,
                latency_ms=15.0,
                client="claude-code",
            )
        )
        # unidentified harness -> falls back to "proxy"
        asyncio.run(
            proxy.metrics.record_request(
                provider="openai",
                model="gpt-4o",
                input_tokens=80,
                output_tokens=10,
                tokens_saved=20,
                latency_ms=15.0,
            )
        )

    report = L.aggregate_savings()
    assert report.lifetime["tokens_saved"] == 60
    clients = {row["client"] for row in report.by_client}
    assert "claude-code" in clients
    assert "proxy" in clients
    assert any(row["model"] == "gpt-4o" for row in report.by_model)


# --------------------------------------------------------------------------- #
# retention cap + stale-schema reset
# --------------------------------------------------------------------------- #


@pytest.mark.parametrize("bad_days", ["31", "60", "365"])
def test_cli_days_flag_capped_at_30(monkeypatch, tmp_path, bad_days):
    pytest.importorskip("click")
    from click.testing import CliRunner

    from headroom.cli.savings import savings

    _events_env(monkeypatch, tmp_path)
    result = CliRunner().invoke(savings, ["--days", bad_days])
    assert result.exit_code != 0
    assert "30" in result.output  # IntRange error mentions the allowed max


def test_compaction_backs_off_when_nothing_can_be_dropped(tmp_path, monkeypatch):
    """A busy install's whole retention window can exceed the size threshold.

    Compaction only removes events older than retention, so such a ledger has
    NOTHING to drop — and without a back-off every subsequent append re-read and
    rewrote the entire growing file under an exclusive lock, once per compressed
    request. The back-off is per-process and purely a work-saver: retention is
    enforced on read regardless.
    """
    import headroom.savings_ledger as sl

    path = tmp_path / "events.jsonl"
    # Tiny thresholds so a handful of in-retention events trips compaction.
    monkeypatch.setattr(sl, "_COMPACT_SIZE_BYTES", 200)
    monkeypatch.setattr(sl, "_COMPACT_GROWTH_BYTES", 10_000)
    monkeypatch.setattr(sl, "_compact_min_size", 200)

    for _ in range(5):
        sl.record_savings_event(
            tokens_before=1_000, tokens_after=600, model="unknown", source="mcp", path=path
        )

    assert path.stat().st_size > 200
    # Every event is fresh, so none aged out and the floor ratcheted above the
    # current size rather than staying at the threshold.
    assert sl._compact_min_size > 200

    # All five events survive: skipping the rewrite must never lose data.
    assert sl.aggregate_savings(path=path).lifetime["calls"] == 5


def test_compaction_still_drops_out_of_retention_events(tmp_path, monkeypatch):
    """The back-off must not disable compaction for ledgers that CAN shrink."""
    import json
    from datetime import timedelta

    import headroom.savings_ledger as sl

    path = tmp_path / "events.jsonl"
    stale = (sl._utc_now() - timedelta(days=sl.MAX_RETENTION_DAYS + 5)).isoformat()
    with path.open("w", encoding="utf-8") as fh:
        for _ in range(50):
            fh.write(
                json.dumps(
                    {
                        "v": 1,
                        "ts": stale,
                        "before": 1_000,
                        "after": 600,
                        "saved": 400,
                        "cost_usd": 0.0012,
                        "model": "unknown",
                        "client": "mcp",
                        "source": "mcp",
                    },
                    separators=(",", ":"),
                )
                + "\n"
            )
    before_size = path.stat().st_size
    monkeypatch.setattr(sl, "_COMPACT_SIZE_BYTES", 200)
    monkeypatch.setattr(sl, "_compact_min_size", 200)

    sl.record_savings_event(
        tokens_before=1_000, tokens_after=600, model="unknown", source="mcp", path=path
    )

    assert path.stat().st_size < before_size
    # Only the fresh event remains; the floor reset so compaction stays armed.
    assert sl.aggregate_savings(path=path).lifetime["calls"] == 1
    assert sl._compact_min_size == 200


def test_v1_events_still_aggregate_alongside_v2(tmp_path):
    """A ledger written across the upgrade must stay legible, not silently blend.

    A v1 event has no cache mix and never can have one, so it contributes its
    list dollar to both columns — and drags the reported basis down to `list`,
    which is how a reader learns the total is not fully cache-aware.
    """
    import json

    import headroom.savings_ledger as sl

    path = tmp_path / "events.jsonl"
    with path.open("w", encoding="utf-8") as fh:
        fh.write(
            json.dumps(
                {
                    "v": 1,
                    "ts": sl._utc_now().isoformat(),
                    "before": 10_000,
                    "after": 6_000,
                    "saved": 4_000,
                    "cost_usd": 0.012,
                    "model": MODEL,
                    "client": "claude-code",
                    "source": "proxy",
                }
            )
            + "\n"
        )
    sl.record_savings_event(
        tokens_before=108_000,
        tokens_after=100_000,
        model=MODEL,
        client="claude-code",
        source="proxy",
        saved_compression=1_000,
        saved_tool_schema=7_000,
        cache_read_tokens=95_000,
        cache_write_tokens=4_000,
        cache_write_5m_tokens=4_000,
        uncached_input_tokens=1_000,
        provider="anthropic",
        path=path,
    )

    lifetime = sl.aggregate_savings(path=path).lifetime
    assert lifetime["calls"] == 2
    # The v2 event is discounted, so the effective total sits below the list one.
    assert lifetime["cost_effective_usd"] < lifetime["cost_usd"]
    # ...but the v1 event's list dollar is still counted, not dropped.
    assert lifetime["cost_effective_usd"] > 0.012
    assert lifetime["basis"] == "list"


def test_zero_saving_request_keeps_its_place_in_the_new_input_denominator(monkeypatch, tmp_path):
    """100 saved on 100 new input, then 0 saved on 10,000 new input. The
    new-input basis is 100 / (10,100 + 100), under 1 percent, not the 50
    percent that only counting requests that saved would report. The
    saved-event figures do not see the second request at all."""
    _events_env(monkeypatch, tmp_path)
    assert L.record_savings_event(
        tokens_before=200, tokens_after=100, model=None, client="claude-code", new_input_tokens=100
    )
    assert L.record_savings_event(
        tokens_before=10_000,
        tokens_after=10_000,
        model=None,
        client="claude-code",
        new_input_tokens=10_000,
    )
    # Without a new-input figure a zero-saving request is still not an event.
    assert not L.record_savings_event(
        tokens_before=500, tokens_after=500, model=None, client="claude-code"
    )
    window = L.aggregate_savings().windows["today"]
    assert window["tokens_saved"] == 100
    assert window["new_input_tokens"] == 10_100
    assert window["new_input_savings_percent"] == pytest.approx(1.0, abs=0.05)
    assert window["savings_percent"] == pytest.approx(50.0)


# --------------------------------------------------------------------------- #
# failures are logged, not swallowed
# --------------------------------------------------------------------------- #


class _ListHandler(logging.Handler):
    def __init__(self) -> None:
        super().__init__(level=logging.DEBUG)
        self.records: list[logging.LogRecord] = []

    def emit(self, record: logging.LogRecord) -> None:
        self.records.append(record)


@pytest.fixture
def ledger_records():
    # Attach directly: caplog misses `headroom.*` loggers once the proxy has
    # configured logging earlier in the session.
    logger = logging.getLogger("headroom.savings_ledger")
    handler = _ListHandler()
    previous = logger.level
    logger.addHandler(handler)
    logger.setLevel(logging.DEBUG)
    yield handler.records
    logger.removeHandler(handler)
    logger.setLevel(previous)


@pytest.fixture
def fresh_append_warnings(monkeypatch):
    monkeypatch.setattr(L, "_append_warnings", WarnOnce(64, "unwritable savings ledgers"))


def _warnings(records: list[logging.LogRecord]) -> list[str]:
    return [r.getMessage() for r in records if r.levelno == logging.WARNING]


def test_unreadable_ledger_warns_instead_of_reporting_zero_silently(tmp_path, ledger_records):
    readable = tmp_path / "savings_events.jsonl"
    assert L.record_savings_event(
        tokens_before=1000, tokens_after=400, model=None, client="c", path=readable
    )
    assert L.aggregate_savings(path=readable).lifetime["tokens_saved"] == 600
    assert _warnings(ledger_records) == []

    # A directory where the ledger file should be: unreadable.
    report = L.aggregate_savings(path=tmp_path)

    assert report.lifetime["tokens_saved"] == 0
    warnings = _warnings(ledger_records)
    assert len(warnings) == 1
    assert "could not read" in warnings[0]
    assert str(tmp_path) in warnings[0]


def test_failed_append_warns_once_per_ledger_path(tmp_path, ledger_records, fresh_append_warnings):
    first = tmp_path / "a"
    second = tmp_path / "b"
    first.mkdir()
    second.mkdir()

    # A directory where the ledger file should be: every append fails.
    assert not L.record_savings_event(tokens_before=1000, tokens_after=400, path=first)
    assert not L.record_savings_event(tokens_before=1000, tokens_after=400, path=first)
    assert not L.record_savings_event(tokens_before=1000, tokens_after=400, path=second)

    warnings = _warnings(ledger_records)
    # One WARNING per path; the repeat on `first` goes to DEBUG only.
    assert len(warnings) == 2
    assert str(first) in warnings[0] and "append to" in warnings[0]
    assert str(second) in warnings[1]
    assert any(r.levelno == logging.DEBUG and "append to" in r.getMessage() for r in ledger_records)


def test_successful_append_rearms_the_warning(tmp_path, ledger_records, fresh_append_warnings):
    ledger = tmp_path / "savings_events.jsonl"
    ledger.mkdir()

    assert not L.record_savings_event(tokens_before=1000, tokens_after=400, path=ledger)
    ledger.rmdir()
    assert L.record_savings_event(tokens_before=1000, tokens_after=400, path=ledger)
    ledger.unlink()
    ledger.mkdir()
    assert not L.record_savings_event(tokens_before=1000, tokens_after=400, path=ledger)

    # The failure after a recovery is a new episode and warns again.
    assert len(_warnings(ledger_records)) == 2


def test_append_warning_names_only_the_type_of_a_non_os_error(
    monkeypatch, tmp_path, ledger_records, fresh_append_warnings
):

    def fail(*_args, **_kwargs):
        raise TypeError("secret-looking event payload")

    # Shadow open() inside the module only, so nothing else is affected.
    monkeypatch.setattr(L, "open", fail, raising=False)
    assert not L.record_savings_event(
        tokens_before=1000, tokens_after=400, path=tmp_path / "savings_events.jsonl"
    )

    warnings = _warnings(ledger_records)
    assert len(warnings) == 1
    assert "TypeError" in warnings[0]
    formatted = [logging.Formatter().format(record) for record in ledger_records]
    assert all("secret-looking" not in line for line in formatted)

    # The repeat goes to DEBUG, and DEBUG carries no payload either.
    assert not L.record_savings_event(
        tokens_before=1000, tokens_after=400, path=tmp_path / "savings_events.jsonl"
    )
    formatted = [logging.Formatter().format(record) for record in ledger_records]
    assert any(r.levelno == logging.DEBUG and "TypeError" in r.getMessage() for r in ledger_records)
    assert all("secret-looking" not in line for line in formatted)


def test_many_unwritable_ledgers_do_not_warn_on_every_cycle(
    tmp_path, ledger_records, fresh_append_warnings
):
    # A directory where each ledger file should be: every append fails.
    ledgers = [tmp_path / f"ledger-{i}" for i in range(65)]
    for ledger in ledgers:
        ledger.mkdir()

    for _cycle in range(3):
        for ledger in ledgers:
            assert not L.record_savings_event(tokens_before=1000, tokens_after=400, path=ledger)

    warnings = _warnings(ledger_records)
    # 64 per-path warnings plus one overflow notice, over all three cycles.
    assert len(warnings) == 65
    assert sum("append to" in w for w in warnings) == 64
    assert "More than 64 warnings about unwritable savings ledgers" in warnings[-1]


_ROUTING = {  # what routemegood publishes, trimmed to the routing_stats contract
    "decisions": 11,
    "window": "lifetime",
    "pairs": [
        {
            "requested": "claude-opus-5-5",
            "served": "claude-opus-5-5",
            "direction": "same",
            "count": 5,
            "enforced": 0,
            "holdout": 0,
            "measured": 0,
            "savings_usd": 0.0,
        },
        {
            "requested": "claude-sonnet-5-5",
            "served": "claude-haiku-5-5",
            "direction": "downgrade",
            "count": 5,
            "enforced": 4,
            "holdout": 1,
            "measured": 4,
            "savings_usd": 0.057101,
        },
        {
            "requested": "gpt-5.6-sol",
            "served": "gpt-5.6-terra",
            "direction": "downgrade",
            "count": 2,
            "enforced": 2,
            "holdout": 0,
            "measured": 0,
            "savings_usd": 0.0,
        },
    ],
}


def test_cli_shows_the_proxys_routing_savings_beside_compression(monkeypatch, tmp_path):
    from click.testing import CliRunner

    from headroom.cli import savings as cli

    _events_env(monkeypatch, tmp_path)
    L.record_savings_event(tokens_before=1000, tokens_after=300, model=None, client="claude-code")
    monkeypatch.setattr(cli, "_proxy_routing", lambda: _ROUTING)

    out = CliRunner().invoke(cli.savings, []).output
    # The headline is the sum of the pairs; a holdout stayed on Sonnet and is not moved.
    assert "Model routing (lifetime): 6 of 11 requests moved · $0.0571 saved" in out
    assert "claude-sonnet-5-5 → claude-haiku-5-5" in out
    assert "not priced (no usage reported)" in out  # Sol -> Terra: a saving we cannot claim
    assert "claude-opus-5-5 → claude-opus-5-5" not in out

    data = json.loads(CliRunner().invoke(cli.savings, ["--json"]).output)
    assert data["routing"]["pairs"][1]["savings_usd"] == 0.057101


def test_cli_shows_routing_savings_before_any_compression(monkeypatch, tmp_path):
    from click.testing import CliRunner

    from headroom.cli import savings as cli

    _events_env(monkeypatch, tmp_path)
    monkeypatch.setattr(cli, "_proxy_routing", lambda: _ROUTING)
    out = CliRunner().invoke(cli.savings, []).output
    assert "No savings recorded yet." not in out
    assert "$0.0571 saved" in out


def test_proxy_routing_is_none_without_a_router(monkeypatch):
    from headroom.cli import savings as cli

    monkeypatch.setattr(
        "headroom.install.health.probe_json", lambda url, timeout=2.0: {"routing": None}
    )
    assert cli._proxy_routing() is None
    monkeypatch.setattr("headroom.install.health.probe_json", lambda url, timeout=2.0: None)
    assert cli._proxy_routing() is None


def test_cli_names_the_routing_window_the_provider_reports(monkeypatch, tmp_path):
    from click.testing import CliRunner

    from headroom.cli import savings as cli

    _events_env(monkeypatch, tmp_path)
    for window, label in (("session", "this proxy session"), (None, "all decisions on record")):
        monkeypatch.setattr(cli, "_proxy_routing", lambda w=window: {**_ROUTING, "window": w})
        out = CliRunner().invoke(cli.savings, []).output
        assert f"Model routing ({label}): 6 of 11 requests moved" in out


def test_cli_without_a_router_never_mentions_routing(monkeypatch, tmp_path):
    from click.testing import CliRunner

    from headroom.cli import savings as cli

    _events_env(monkeypatch, tmp_path)
    L.record_savings_event(tokens_before=1000, tokens_after=300, model=None, client="claude-code")
    monkeypatch.setattr(cli, "_proxy_routing", lambda: None)
    assert "routing" not in CliRunner().invoke(cli.savings, []).output.lower()
    assert "routing" not in json.loads(CliRunner().invoke(cli.savings, ["--json"]).output)