cctally 1.22.2 → 1.22.3

package/bin/_cctally_statusline.py

@@ -0,0 +1,630 @@
+"""`cctally statusline` command — one-line status summary for CC hooks.
+
+Holds `cmd_statusline` + its resolvers (`_resolve_statusline_tz`,
+`_resolve_context_window`, `_read_last_assistant_usage`,
+`_build_statusline_injections`) AND the two per-model context-window
+constant dicts (`CLAUDE_MODEL_CONTEXT_WINDOWS`,
+`CLAUDE_MODEL_CONTEXT_WINDOW_DEFAULT_FAMILY`), co-located with their sole
+consumer `_resolve_context_window`.
+
+Honest *name* imports are KERNEL-ONLY (`_cctally_core`: `eprint`, plus the
+kernel symbols `open_db` / `_command_as_of` per the kernel-extraction
+invariant in `tests/test_kernel_extraction_invariants.py`). `_lib_statusline`
+is the eagerly-preloaded library kernel (bin/cctally:416), imported
+qualified and referenced as a module object (`_lib_statusline.render_statusline`,
+`_lib_statusline.ParseError`, …). Every other sibling/kernel-homed symbol
+is reached via the call-time `_cctally()` accessor so test monkeypatches
+through `cctally`'s namespace are preserved (spec §3.2). The 5h seam —
+`_load_recorded_five_hour_windows`, `_maybe_swap_active_block_to_canonical`
+— lives in `_cctally_five_hour.py` and is reached via `c.<name>` (spec §3.3).
+
+bin/cctally re-exports `cmd_statusline` (parser `c.cmd_statusline`) plus the
+resolvers/dicts; `_resolve_statusline_tz` is retrieved by `tests/test_statusline.py`
+off the `cctally` namespace.
+
+Spec: docs/superpowers/specs/2026-05-30-extract-five-hour-statusline-cmd-design.md
+"""
+from __future__ import annotations
+
+import argparse
+import datetime as dt
+import json
+import os
+import pathlib
+import sys
+from zoneinfo import ZoneInfo, ZoneInfoNotFoundError
+
+import _lib_statusline
+from _cctally_core import _command_as_of, eprint, open_db
+
+
+def _cctally():
+    """Resolve the current `cctally` module at call-time (spec §3.2)."""
+    return sys.modules["cctally"]
+
+
+# Per-model context window (used by `cctally statusline` segment 4).
+# Keep in sync with Anthropic's docs:
+#   https://docs.anthropic.com/en/docs/about-claude/models
+# Unknown model id → segment renders `🧠 N/A` + one-shot stderr warn.
+CLAUDE_MODEL_CONTEXT_WINDOWS = {
+    # 1M-token variants (explicit IDs override the family default).
+    "claude-opus-4-8[1m]": 1_000_000,
+    "claude-opus-4-7[1m]": 1_000_000,
+    "claude-sonnet-4-5[1m]": 1_000_000,
+    # Default 200K for every other Sonnet/Opus/Haiku family member.
+    # The resolver does a substring match on the family token if the
+    # exact id is missing — see _resolve_context_window.
+}
+
+CLAUDE_MODEL_CONTEXT_WINDOW_DEFAULT_FAMILY = {
+    # Substring (case-insensitive) → window. Order matters; first hit wins.
+    "sonnet": 200_000,
+    "opus":   200_000,
+    "haiku":  200_000,
+}
+
+
+def _resolve_statusline_tz(cli_tz, cfg, warn_once):
+    """Resolve the IANA tz_name for cmd_statusline using the same 3-rung
+    precedence as every other reporting command:
+
+        CLI ``--timezone`` > ``config.display.tz`` > DISPLAY_TZ_DEFAULT ("local")
+
+    "local" is converted to a real IANA via ``_local_tz_name()`` before
+    returning. Unknown IANA names emit a one-shot warning and fall back
+    to ``"UTC"``. Returns a real IANA name (or ``"UTC"``) — never the
+    literal sentinel ``"local"``.
+
+    Prior to #86 G follow-up, this defaulted to ``"UTC"`` when no config
+    was set, so ``today`` computed on the UTC calendar day while
+    ``cctally daily`` (and every other reporting command) used the local
+    day — UTC-offset users saw a multi-hour lag between statusline and
+    daily. Regression: tests/test_statusline.py::TestTzResolution.
+    """
+    c = _cctally()
+    tz_name = cli_tz
+    if not tz_name:
+        tz_name = c.get_display_tz_pref(cfg)
+    if tz_name in ("local", "LOCAL"):
+        try:
+            tz_name = c._local_tz_name() or "UTC"
+        except Exception:
+            tz_name = "UTC"
+    try:
+        ZoneInfo(tz_name)
+    except (ZoneInfoNotFoundError, Exception):
+        warn_once(
+            f"cctally statusline: invalid timezone {tz_name!r}; using 'UTC'"
+        )
+        tz_name = "UTC"
+    return tz_name
+
+
+def cmd_statusline(args: argparse.Namespace) -> int:
+    """`cctally statusline` — one-line status summary for CC hooks.
+
+    See docs/superpowers/specs/2026-05-28-issue-86-session-g-statusline-design.md
+    for the full design.
+
+    Exit codes:
+        0  success (every absent stdin field degrades gracefully)
+        1  stdin is not parseable JSON OR root is not a JSON object
+        2  argparse rejected a flag (e.g. --cost-source ccusage), OR
+           --config PATH unreadable
+    """
+    c = _cctally()
+    # NOTE: `--cost-source ccusage` is rejected at argparse-time by
+    # `_CostSourceAction` in `_build_statusline_parser`; it exits 2 with
+    # the rename hint before we get here, so no explicit re-check is
+    # needed in this function.
+
+    # Validate `--context-{low,medium}-threshold` BEFORE reading stdin
+    # so a misconfigured invocation fails fast without consuming the
+    # CC hook's stdin payload.
+    low = args.context_low_threshold
+    med = args.context_medium_threshold
+    if not isinstance(low, int) or low < 0 or low > 100:
+        eprint(
+            "cctally statusline: --context-low-threshold must be in [0, 100]"
+        )
+        return 2
+    if not isinstance(med, int) or med < 0 or med > 100:
+        eprint(
+            "cctally statusline: --context-medium-threshold must be in [0, 100]"
+        )
+        return 2
+    if low >= med:
+        eprint(
+            "cctally statusline: --context-low-threshold must be < "
+            "--context-medium-threshold"
+        )
+        return 2
+
+    # Silently clamp `--refresh-interval` to [0, 600]. The flag is a
+    # no-op alias for ccusage drop-in compat; users never observe the
+    # effect, but the spec mandates the clamp for forward-compat (when
+    # we promote it to a real flag, the clamped value should be the one
+    # propagated downstream).
+    try:
+        args.refresh_interval = max(0, min(600, int(args.refresh_interval)))
+    except (TypeError, ValueError):
+        args.refresh_interval = 1
+
+    # Read stdin once.
+    raw = sys.stdin.buffer.read()
+    parse_result = _lib_statusline.parse_statusline_stdin(raw)
+    if isinstance(parse_result, _lib_statusline.ParseError):
+        eprint(f"cctally statusline: {parse_result.message}")
+        return 1
+    inp = parse_result
+
+    # Resolve effective config: CLI > config.json > built-in default.
+    # `_load_claude_config_for_args` honors `--config PATH` (issue #88
+    # plumbing); a missing/invalid PATH raises SystemExit(2) inside
+    # `_load_config_from_explicit_path` so this call already enforces
+    # exit-2 on a bad --config.
+    cfg = c._load_claude_config_for_args(args)
+    sl_cfg = (cfg.get("statusline") or {}) if isinstance(cfg, dict) else {}
+    if not isinstance(sl_cfg, dict):
+        sl_cfg = {}
+
+    # Validate config values; on invalid, one-shot stderr warn + use default.
+    _warned: set = set()
+
+    def warn_once(msg: str) -> None:
+        if msg in _warned:
+            return
+        _warned.add(msg)
+        eprint(msg)
+
+    def _resolve(cli_val, cfg_key, default):
+        if cli_val is not None:
+            return cli_val
+        cv = sl_cfg.get(cfg_key)
+        if cv is None:
+            return default
+        return cv
+
+    vbr = _resolve(args.visual_burn_rate, "visual_burn_rate", "off")
+    if vbr not in ("off", "emoji", "text", "emoji-text"):
+        warn_once(
+            f"cctally statusline: invalid statusline.visual_burn_rate={vbr!r}; "
+            f"using 'off'"
+        )
+        vbr = "off"
+
+    cs = _resolve(args.cost_source, "cost_source", "auto")
+    if cs not in ("auto", "cctally", "cc", "both"):
+        warn_once(
+            f"cctally statusline: invalid statusline.cost_source={cs!r}; "
+            f"using 'auto'"
+        )
+        cs = "auto"
+
+    ext_on = _resolve(args.cctally_extensions, "cctally_extensions", True)
+    if not isinstance(ext_on, bool):
+        warn_once(
+            f"cctally statusline: invalid statusline.cctally_extensions="
+            f"{ext_on!r}; using True"
+        )
+        ext_on = True
+
+    tz_name = _resolve_statusline_tz(getattr(args, "timezone", None), cfg, warn_once)
+
+    # Color: explicit CLI > NO_COLOR env > TTY detect.
+    if args.color is True or args.color is False:
+        color = args.color
+    else:
+        color = (os.environ.get("NO_COLOR", "") == "") and sys.stdout.isatty()
+
+    sargs = _lib_statusline.StatuslineArgs(
+        visual_burn_rate=vbr,
+        cost_source=cs,
+        context_low_threshold=int(args.context_low_threshold),
+        context_medium_threshold=int(args.context_medium_threshold),
+        cctally_extensions=bool(ext_on),
+        color=bool(color),
+        display_tz_name=tz_name,
+        debug=bool(args.debug),
+    )
+
+    # Build injections (DB + transcript file IO).
+    inj = _build_statusline_injections(warn_once)
+
+    # `_command_as_of()` honors the `CCTALLY_AS_OF` testing hook so the
+    # golden harness can pin "now" for deterministic block-remaining and
+    # 5h/7d countdown numbers. Falls back to wall-clock UTC otherwise.
+    now = _command_as_of()
+    try:
+        line = _lib_statusline.render_statusline(inp, sargs, inj, now)
+    except Exception as exc:  # pragma: no cover — defensive
+        eprint(f"cctally statusline: render failed: {exc}")
+        return 1
+    print(line)
+    return 0
+
+
+def _resolve_context_window(model_id, warn_once) -> "int | None":
+    """Look up ``model_id`` in ``CLAUDE_MODEL_CONTEXT_WINDOWS``; fall back
+    to a family-substring match against
+    ``CLAUDE_MODEL_CONTEXT_WINDOW_DEFAULT_FAMILY``. Unknown id → ``None`` +
+    one-shot stderr warning.
+    """
+    if not model_id:
+        return None
+    if model_id in CLAUDE_MODEL_CONTEXT_WINDOWS:
+        return CLAUDE_MODEL_CONTEXT_WINDOWS[model_id]
+    mid_lower = model_id.lower()
+    for family, window in CLAUDE_MODEL_CONTEXT_WINDOW_DEFAULT_FAMILY.items():
+        if family in mid_lower:
+            return window
+    warn_once(
+        f"cctally statusline: unknown model {model_id!r}; context % unavailable"
+    )
+    return None
+
+
+def _read_last_assistant_usage(transcript_path):
+    """Tail-walk the transcript JSONL backwards to the most recent
+    ``type=assistant`` line carrying ``message.usage``. Returns the usage
+    dict or ``None``.
+
+    Reads in 64 KB chunks from the end so multi-MB transcripts don't
+    block the hot statusline path with a full-file parse.
+    """
+    if not transcript_path:
+        return None
+    path = pathlib.Path(transcript_path)
+    if not path.exists():
+        return None
+    try:
+        with path.open("rb") as fh:
+            fh.seek(0, 2)
+            size = fh.tell()
+            tail = b""
+            chunk = 65536
+            # Read backwards in chunks until we have at least one full line
+            # and the tail starts with a newline (so the first line is whole).
+            while size > 0 and tail.count(b"\n") < 2:
+                read_at = max(0, size - chunk)
+                fh.seek(read_at)
+                tail = fh.read(size - read_at) + tail
+                size = read_at
+            lines = tail.split(b"\n")
+    except OSError:
+        return None
+    for line in reversed(lines):
+        line = line.strip()
+        if not line:
+            continue
+        try:
+            obj = json.loads(line)
+        except Exception:
+            continue
+        if not isinstance(obj, dict):
+            continue
+        if obj.get("type") != "assistant":
+            continue
+        msg = obj.get("message") or {}
+        usage = msg.get("usage") if isinstance(msg, dict) else None
+        if isinstance(usage, dict):
+            return usage
+    return None
+
+
+def _build_statusline_injections(warn_once):
+    """Wire DB- and FS-backed implementations for the kernel's injection ports.
+
+    See ``_lib_statusline.StatuslineInjections`` for the contract. All
+    callables fast-fail to "no data" on any exception — statusline must
+    NEVER block the Claude Code hook tick.
+    """
+    def _cctally_session_cost(sid):
+        c = _cctally()
+        if not sid:
+            return None
+        try:
+            conn = c.open_cache_db()
+        except Exception:
+            return None
+        try:
+            # Walk all entries via session_files join; sum costs whose
+            # session_id matches. Stays read-only — does NOT call
+            # sync_cache (too heavy for the hot statusline path; the
+            # record-usage + hook-tick paths keep the cache warm).
+            sql = (
+                "SELECT se.timestamp_utc, se.model, "
+                "  se.input_tokens, se.output_tokens, "
+                "  se.cache_create_tokens, se.cache_read_tokens, "
+                "  se.cost_usd_raw, se.usage_extra_json "
+                "FROM session_entries se "
+                "LEFT JOIN session_files sf ON sf.path = se.source_path "
+                "WHERE sf.session_id = ?"
+            )
+            rows = list(conn.execute(sql, (sid,)))
+        except Exception:
+            return None
+        finally:
+            try:
+                conn.close()
+            except Exception:
+                pass
+        if not rows:
+            return None
+        total = 0.0
+        for r in rows:
+            usage = {
+                "input_tokens":                r[2] or 0,
+                "output_tokens":               r[3] or 0,
+                "cache_creation_input_tokens": r[4] or 0,
+                "cache_read_input_tokens":     r[5] or 0,
+            }
+            try:
+                if r[7]:
+                    extras = json.loads(r[7])
+                    if isinstance(extras, dict):
+                        usage.update(extras)
+            except Exception:
+                pass
+            try:
+                total += c._calculate_entry_cost(
+                    r[1], usage, mode="auto", cost_usd=r[6],
+                )
+            except Exception:
+                continue
+        return total
+
+    def _today_cost(tz_name, now):
+        c = _cctally()
+        try:
+            tz = ZoneInfo(tz_name) if tz_name and tz_name != "UTC" else dt.timezone.utc
+        except Exception:
+            tz = dt.timezone.utc
+        local_now = now.astimezone(tz)
+        day_start_local = local_now.replace(
+            hour=0, minute=0, second=0, microsecond=0
+        )
+        day_end_local = day_start_local + dt.timedelta(days=1)
+        range_start = day_start_local.astimezone(dt.timezone.utc)
+        range_end = day_end_local.astimezone(dt.timezone.utc)
+        # Two-filter pattern (UTC half-open + display-tz date check):
+        # the UTC range fetches the candidate window cheaply via the
+        # cache's indexed `timestamp` column; the display-tz date check
+        # then trims any entries that fall outside today's local
+        # calendar day (the UTC window straddles two local dates when
+        # the display tz has any UTC offset, so the SQL range alone is
+        # slightly wider than the local day).
+        try:
+            entries = c.get_entries(range_start, range_end, skip_sync=True)
+        except Exception:
+            return 0.0
+        total = 0.0
+        for e in entries:
+            try:
+                # Filter to today in display tz (second-pass trim).
+                ts_local = e.timestamp.astimezone(tz)
+                if ts_local.date() != local_now.date():
+                    continue
+                total += c._calculate_entry_cost(
+                    e.model, e.usage, mode="auto", cost_usd=e.cost_usd,
+                )
+            except Exception:
+                continue
+        return total
+
+    def _active_block(now):
+        c = _cctally()
+        try:
+            # Look at last 24h — captures the full active 5h window.
+            range_start = now - dt.timedelta(hours=24)
+            entries = c.get_entries(range_start, now, skip_sync=True)
+        except Exception:
+            return None
+        if not entries:
+            return None
+        try:
+            recorded_windows, block_start_overrides, canonical_intervals = (
+                c._load_recorded_five_hour_windows(
+                    range_start - c.BLOCK_DURATION, now + c.BLOCK_DURATION,
+                )
+            )
+        except Exception:
+            recorded_windows, block_start_overrides, canonical_intervals = (
+                [], {}, {},
+            )
+        try:
+            blocks = c._group_entries_into_blocks(
+                entries,
+                mode="auto",
+                recorded_windows=recorded_windows,
+                block_start_overrides=block_start_overrides,
+                canonical_intervals=canonical_intervals,
+                now=now,
+            )
+        except Exception:
+            return None
+        for b in blocks:
+            if not b.is_gap and b.is_active:
+                remaining_s = int((b.end_time - now).total_seconds())
+                elapsed_s = int((now - b.start_time).total_seconds())
+                return (float(b.cost_usd or 0.0), remaining_s, elapsed_s)
+        return None
+
+    def _hwm_clamp(five_resets, seven_resets):
+        c = _cctally()
+        five_hwm = None
+        seven_hwm = None
+        try:
+            conn = open_db()
+        except Exception:
+            return (None, None)
+        try:
+            if five_resets is not None:
+                try:
+                    key = c._canonical_5h_window_key(int(five_resets))
+                    row = conn.execute(
+                        "SELECT MAX(five_hour_percent) "
+                        "FROM weekly_usage_snapshots "
+                        "WHERE five_hour_window_key = ?",
+                        (key,),
+                    ).fetchone()
+                    if row and row[0] is not None:
+                        five_hwm = float(row[0])
+                except Exception:
+                    pass
+            if seven_resets is not None:
+                try:
+                    # Seven-day window bounds from the resets_at epoch:
+                    # week_end = reset; week_start = reset - 7 days. The
+                    # date form is the snapshot lookup key (week_start_date
+                    # is deliberately NOT re-anchored across a mid-week
+                    # reset — see _apply_reset_events_to_subweeks).
+                    week_end_dt = dt.datetime.fromtimestamp(
+                        int(seven_resets), tz=dt.timezone.utc,
+                    )
+                    week_start_dt = week_end_dt - dt.timedelta(days=7)
+                    week_start_date = week_start_dt.date().isoformat()
+                    # Reset-aware floor. An Anthropic mid-week reset / in-
+                    # place credit leaves the pre-reset peak snapshots in
+                    # this SAME week_start_date bucket (the boundary the
+                    # snapshots carry does not change). A naive bucket-wide
+                    # MAX(weekly_percent) would clamp the post-reset value
+                    # UP to that stale peak — the statusline would show the
+                    # pre-reset 7d %. Mirror the CLI/dashboard segmentation
+                    # (_apply_reset_events_to_subweeks: post-reset window
+                    # start_ts := effective_reset_at_utc) by flooring the
+                    # MAX to snapshots captured at/after the latest reset
+                    # effective WITHIN this window. unixepoch() on both
+                    # sides — reset rows carry mixed offset spellings
+                    # (+00:00 / +03:00) while captured_at_utc uses 'Z', so a
+                    # lexical compare would misorder them (same rule as the
+                    # 5h-block cross-reset flag).
+                    floor_row = conn.execute(
+                        "SELECT MAX(effective_reset_at_utc) "
+                        "FROM week_reset_events "
+                        "WHERE unixepoch(effective_reset_at_utc) >= unixepoch(?) "
+                        "  AND unixepoch(effective_reset_at_utc) <  unixepoch(?)",
+                        (week_start_dt.isoformat(), week_end_dt.isoformat()),
+                    ).fetchone()
+                    floor_iso = (
+                        floor_row[0] if floor_row and floor_row[0] else None
+                    )
+                    if floor_iso is not None:
+                        row = conn.execute(
+                            "SELECT MAX(weekly_percent) "
+                            "FROM weekly_usage_snapshots "
+                            "WHERE week_start_date = ? "
+                            "  AND unixepoch(captured_at_utc) >= unixepoch(?)",
+                            (week_start_date, floor_iso),
+                        ).fetchone()
+                    else:
+                        row = conn.execute(
+                            "SELECT MAX(weekly_percent) "
+                            "FROM weekly_usage_snapshots "
+                            "WHERE week_start_date = ?",
+                            (week_start_date,),
+                        ).fetchone()
+                    if row and row[0] is not None:
+                        seven_hwm = float(row[0])
+                except Exception:
+                    pass
+        finally:
+            try:
+                conn.close()
+            except Exception:
+                pass
+        return (five_hwm, seven_hwm)
+
+    def _db_latest_rate_limits():
+        try:
+            conn = open_db()
+        except Exception:
+            return None
+        try:
+            # Prefer `week_end_at` (ISO timestamp; sub-day precision) over
+            # `week_end_date` (date-only; UTC-midnight). Older snapshots
+            # may have `week_end_at` NULL — fall back to the date column
+            # in that case. See the neighbor query in `pick_week_selection`
+            # (bin/cctally:3849) for the precedent.
+            row = conn.execute(
+                "SELECT five_hour_percent, five_hour_window_key, "
+                "  weekly_percent, week_end_at, week_end_date "
+                "FROM weekly_usage_snapshots "
+                "ORDER BY captured_at_utc DESC, id DESC LIMIT 1"
+            ).fetchone()
+            if not row:
+                return None
+            five_pct = float(row[0]) if row[0] is not None else None
+            five_resets = int(row[1]) if row[1] is not None else None
+            seven_pct = float(row[2]) if row[2] is not None else None
+            seven_resets = None
+            week_end_at = row[3]
+            week_end_date = row[4]
+            if week_end_at:
+                try:
+                    # `datetime.fromisoformat` accepts the trailing `Z`
+                    # only on Python 3.11+; normalize to `+00:00` so 3.10
+                    # checkouts (and any odd Z-suffixed snapshot) parse.
+                    raw_iso = str(week_end_at)
+                    if raw_iso.endswith("Z"):
+                        raw_iso = raw_iso[:-1] + "+00:00"
+                    end_dt = dt.datetime.fromisoformat(raw_iso)
+                    if end_dt.tzinfo is None:
+                        end_dt = end_dt.replace(tzinfo=dt.timezone.utc)
+                    seven_resets = int(end_dt.timestamp())
+                except Exception:
+                    seven_resets = None
+            if seven_resets is None and week_end_date:
+                try:
+                    end_dt = dt.datetime.fromisoformat(str(week_end_date))
+                    if end_dt.tzinfo is None:
+                        end_dt = end_dt.replace(tzinfo=dt.timezone.utc)
+                    # week_end_date is exclusive — that's the reset moment.
+                    seven_resets = int(end_dt.timestamp())
+                except Exception:
+                    seven_resets = None
+            return (five_pct, five_resets, seven_pct, seven_resets)
+        except Exception:
+            return None
+        finally:
+            try:
+                conn.close()
+            except Exception:
+                pass
+
+    def _context_pct(transcript_path, model_id):
+        if not transcript_path or not model_id:
+            return None
+        window = _resolve_context_window(model_id, warn_once)
+        if window is None:
+            return None
+        try:
+            usage = _read_last_assistant_usage(transcript_path)
+        except Exception:
+            return None
+        if not isinstance(usage, dict):
+            return None
+        try:
+            ctx_tokens = (
+                int(usage.get("input_tokens", 0) or 0)
+                + int(usage.get("cache_read_input_tokens", 0) or 0)
+                + int(usage.get("cache_creation_input_tokens", 0) or 0)
+            )
+        except (TypeError, ValueError):
+            return None
+        if window <= 0:
+            return None
+        return ctx_tokens / window * 100.0
+
+    return _lib_statusline.StatuslineInjections(
+        cctally_session_cost=_cctally_session_cost,
+        today_cost=_today_cost,
+        active_block=_active_block,
+        hwm_clamp=_hwm_clamp,
+        db_latest_rate_limits=_db_latest_rate_limits,
+        context_pct=_context_pct,
+        warn_once=warn_once,
+    )
+
+

package/bin/_cctally_sync_week.py

@@ -6,11 +6,12 @@ JSONL-aggregates the week's cost + inserts a `weekly_cost_snapshots`
 row + emits the success line (or `--json` envelope).
 
 Every helper this command calls — `load_config`, `get_week_start_name`,
-`open_db`, `pick_week_selection`, `compute_week_cost`,
-`format_local_iso`, `insert_cost_snapshot`, `make_week_ref`,
+`open_db`, `pick_week_selection`, `format_local_iso`, `make_week_ref`,
 `get_latest_usage_for_week` — stays in `bin/cctally` (they're shared
-with the rest of the subcommand surface), reached via the `_cctally()`
-call-time accessor (spec §5.2 / §5.5 pattern).
+with the rest of the subcommand surface); `compute_week_cost` /
+`insert_cost_snapshot` now live in `_cctally_milestones.py` (re-exported on
+the ns). All are reached via the `_cctally()` call-time accessor (spec
+§5.2 / §5.5 pattern).
 
 bin/cctally re-exports `cmd_sync_week` so the two non-extracted internal
 callers (`cmd_record_usage`'s milestone-cost-sync path and the