cctally 1.81.0 → 1.82.1

package/bin/_lib_dashboard_sources.py

@@ -297,12 +297,41 @@ def reuse_coherent_source_state(
     return prior if _coherent_provider(prior) and prior.data_version == data_version else None
 
 
+def _hero_cycle_is_stale(state: SourceDashboardState) -> bool:
+    """Whether a provider's hero is bounded by STALE quota evidence (#350 §3.4).
+
+    ``hero.cycle_freshness`` is additive and OMITTED while the cycle is fresh, so
+    this is false for every provider and every generation that predates #350.
+    """
+    if not isinstance(state.data, Mapping):
+        return False
+    hero = state.data.get("hero")
+    return isinstance(hero, Mapping) and hero.get("cycle_freshness") == "stale"
+
+
+def _stale_cycle_providers(
+    claude: SourceDashboardState,
+    codex: SourceDashboardState,
+) -> tuple[str, ...]:
+    return tuple(
+        label for label, state in (("Claude", claude), ("Codex", codex))
+        if _hero_cycle_is_stale(state)
+    )
+
+
 def _combined_metrics(
     claude: SourceDashboardState,
     codex: SourceDashboardState,
 ) -> Mapping[str, object] | None:
     if not (_coherent_provider(claude) and _coherent_provider(codex)):
         return None
+    # #350 spec §3.5: a stale-cycle hero is NOT combinable. Retaining it would
+    # let a sum over stale evidence be published while `compose_all_state` marks
+    # the result fresh — and the All hero carries no Snapshot row or staleness
+    # marker at all, so the staleness would be silently undisclosed. The
+    # combined NUMBER therefore behaves exactly as it does today.
+    if _stale_cycle_providers(claude, codex):
+        return None
     for state in (claude, codex):
         hero_capability = state.capabilities.get("hero")
         if hero_capability is None or hero_capability.status not in {"supported", "derived"}:
@@ -366,7 +395,24 @@ def compose_all_state(
         raise ValueError("all composition requires Claude and Codex provider states")
     combined = _combined_metrics(claude, codex)
     providers_coherent = _coherent_provider(claude) and _coherent_provider(codex)
+    # #350 spec §3.5: "All behaves exactly as today" is true only of the combined
+    # NUMBER. Under §3.4 both providers stay coherent, so All now publishes
+    # partial/fresh with no provider warning to explain it — the status chip
+    # would fall through to a generic `degraded` and the All hero fallback would
+    # claim a provider is degraded while both provider envelopes say otherwise.
+    # This All-LOCAL warning states the real reason without touching either
+    # provider envelope. It is emitted only when the providers are otherwise
+    # coherent; an incoherent provider already publishes its own reason.
+    all_local_warnings: tuple[SourceDashboardWarning, ...] = ()
     if providers_coherent:
+        stale_cycle_providers = _stale_cycle_providers(claude, codex)
+        if stale_cycle_providers:
+            all_local_warnings = (SourceDashboardWarning(
+                "combined_totals_withheld",
+                f"{' and '.join(stale_cycle_providers)} quota evidence is stale, "
+                "so combined totals are withheld.",
+                "hero",
+            ),)
         availability: Availability = (
             "partial"
             if combined is None or "partial" in (claude.availability, codex.availability)
@@ -395,7 +441,11 @@ def compose_all_state(
         source="all",
         availability=availability,
         freshness=freshness,
-        warnings=tuple((*claude.warnings, *codex.warnings)),
+        # All-LOCAL warnings lead: `warningForSource` on the client falls back to
+        # the FIRST warning, so a merely-partial provider warning (e.g.
+        # `codex_metadata_incomplete`) would otherwise pre-empt the chip label
+        # and hide the real reason combined totals are withheld.
+        warnings=tuple((*all_local_warnings, *claude.warnings, *codex.warnings)),
         data_version=data_version,
         last_success_at=last_success_at,
         capabilities={

package/bin/_lib_pricing.py

@@ -53,7 +53,7 @@ def _chip_for_model(name: str) -> str:
 # Date the embedded pricing snapshots below were last verified against
 # vendor sources. Bump whenever CLAUDE_MODEL_PRICING / CODEX_MODEL_PRICING
 # is synced. Read by `pricing-check` + the release pre-flight staleness nudge.
-PRICING_SNAPSHOT_DATE = "2026-07-19"
+PRICING_SNAPSHOT_DATE = "2026-07-24"
 PRICING_STALENESS_DAYS = 60  # release pre-flight WARNs past this age
 
 # Canonical machine-readable pricing source (Claude values + Codex values).
@@ -99,7 +99,7 @@ PRICING_DRIFT_ALLOWLIST: list[dict] = [
 
 # Anthropic API pricing snapshot:
 # - Source: https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json
-# - Captured: 2026-07-19 (see PRICING_SNAPSHOT_DATE)
+# - Captured: 2026-07-24 (see PRICING_SNAPSHOT_DATE)
 # - Verified by maintainer against docs.claude.com/en/docs/about-claude/pricing;
 #   update in PRs touching this table.
 #   2026-06-10: added claude-fable-5 ($10/$50 per MTok; 1M context, no
@@ -114,6 +114,22 @@ PRICING_DRIFT_ALLOWLIST: list[dict] = [
 #   value_drift on all four cost fields. Suppressed via PRICING_DRIFT_ALLOWLIST
 #   above (the non-vacuity guard forces removal once LiteLLM reverts to the
 #   standard rate after 2026-08-31).
+#   2026-07-24: added claude-opus-5 ($5/$25 per MTok — identical to Opus
+#   4.5/4.6/4.7/4.8; $6.25 5-minute cache write and $0.50 cache read at the
+#   standard 1.25x/0.1x multipliers). 1M context at standard pricing, so NO
+#   above-200k tier. Dateless pinned id with no dated twin (matches Anthropic's
+#   published ID/alias table). LiteLLM has no opus-5 entry yet, so this table is
+#   simply ahead of it — `ahead_of_litellm` is never a drift finding, so no
+#   PRICING_DRIFT_ALLOWLIST entry is needed.
+#
+# Known gap — Claude *fast mode* is not modelled. Anthropic bills fast mode at a
+# premium ($10/$50 per MTok on Opus 5 and Opus 4.8; $30/$150 on Opus 4.7), and
+# Claude Code records the tier per assistant entry at `message.usage.speed`
+# ("standard" | "fast"), so it IS derivable from the JSONL we already ingest —
+# but this table is one rate per model and `_calculate_entry_cost` ignores speed,
+# so a fast-mode entry is priced at the standard rate (2x undercount on Opus 5).
+# The Codex side already carries the shape for this
+# (`_calculate_codex_entry_cost(..., speed=...)` + its fast-tier multiplier).
 CLAUDE_MODEL_PRICING: dict[str, dict[str, Any]] = {
     "claude-3-5-haiku-20241022": {
         "input_cost_per_token": 8e-07,
@@ -269,6 +285,12 @@ CLAUDE_MODEL_PRICING: dict[str, dict[str, Any]] = {
         "cache_creation_input_token_cost": 6.25e-06,
         "cache_read_input_token_cost": 5e-07,
     },
+    "claude-opus-5": {
+        "input_cost_per_token": 5e-06,
+        "output_cost_per_token": 2.5e-05,
+        "cache_creation_input_token_cost": 6.25e-06,
+        "cache_read_input_token_cost": 5e-07,
+    },
     "claude-sonnet-4-20250514": {
         "input_cost_per_token": 3e-06,
         "output_cost_per_token": 1.5e-05,
@@ -321,7 +343,9 @@ _unknown_model_warnings: set[str] = set()
 #
 # Codex (OpenAI) API pricing snapshot:
 # - Source: https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json
-# - Captured: 2026-07-19 (see PRICING_SNAPSHOT_DATE)
+# - Captured: 2026-07-19 — the last full Codex sync. PRICING_SNAPSHOT_DATE moved
+#   ahead to 2026-07-24 for the Claude-side opus-5 sync; these Codex values were
+#   NOT re-verified that day.
 # - As of the 2026-07-19 sync this carries every openai-provider
 #   gpt-5* model the LiteLLM snapshot lists, so `pricing-check`'s scope finds
 #   nothing missing. Models absent from this table still fall back to `gpt-5`

package/bin/_lib_readme_refresh.py

@@ -0,0 +1,401 @@
+#!/usr/bin/env python3
+"""Pure kernel for the public README auto-refresh (issue #354).
+
+Four stdlib-only responsibilities, each a pure function so the private
+release driver, a public pytest, and a `--check` CLI can all share one
+implementation:
+
+1. ``extract_highlights`` — pull up to N lint-clean highlight bullets from a
+   specific ``## [X.Y.Z] - YYYY-MM-DD`` CHANGELOG section (exact version match,
+   Added > Changed > Fixed priority, multiline flattening, terminal issue-ref
+   stripping, fenced-code skipping).
+2. ``normalize_highlight`` — deterministically rewrite a bullet so it satisfies
+   the copy lint by construction (em/en dashes with surrounding spaces become
+   ``, ``; doubled separators collapse; leading/trailing separators trim).
+3. ``splice_marker_block`` — replace only the region between the HTML-comment
+   markers, byte-preserving everything outside (CRLF included), failing closed
+   on missing/duplicated/malformed markers.
+4. ``lint_copy`` — scan README prose (outside fenced code) for em dash, en
+   dash, a defined emoji set, and mid-line spaced-hyphen dash substitutes.
+
+This module is PUBLIC (listed in `.mirror-allowlist` + `package.json` files[]),
+because its pytest lands on the public mirror and a public test must not import
+a private module. It contains nothing sensitive — Markdown parsing and linting.
+
+CLI: ``python3 bin/_lib_readme_refresh.py --check README.md`` runs the copy lint
+plus marker well-formedness and exits 0 (clean) / 1 (issues) / 2 (usage).
+"""
+from __future__ import annotations
+
+import re
+import sys
+
+MARKER_BEGIN = "<!-- cctally:latest-stable:begin -->"
+MARKER_END = "<!-- cctally:latest-stable:end -->"
+
+
+class ChangelogSectionMissing(ValueError):
+    """Raised when the requested version has no exact CHANGELOG section."""
+
+
+class MarkerError(ValueError):
+    """Raised when the latest-stable markers are missing/duplicated/malformed."""
+
+
+# --------------------------------------------------------------------------
+# Fence handling (shared by extraction + lint)
+# --------------------------------------------------------------------------
+# GFM code fences: a line whose content (after up to 3 leading spaces) begins
+# with a run of >= 3 backticks or >= 3 tildes. The closing fence must use the
+# same character, be at least as long as the opener, and (unlike the opener)
+# carry no info string.
+_FENCE_RE = re.compile(r"^[ ]{0,3}(`{3,}|~{3,})([^\n]*)$")
+
+
+class _UnclosedFence(Exception):
+    def __init__(self, line_no: int, excerpt: str) -> None:
+        super().__init__(f"unclosed fence opened on line {line_no}")
+        self.line_no = line_no
+        self.excerpt = excerpt
+
+
+def _fence_marker(line: str):
+    """Return ``(char, length, info)`` if *line* is a fence marker, else None."""
+    m = _FENCE_RE.match(line)
+    if not m:
+        return None
+    run = m.group(1)
+    return (run[0], len(run), m.group(2).strip())
+
+
+def _iter_content_lines(text: str):
+    """Yield ``(line_no, line)`` for lines OUTSIDE fenced code blocks.
+
+    Raises ``_UnclosedFence`` (at the end of the walk) if a fence is opened but
+    never closed. Operates on ``str.split("\\n")`` so a CRLF file keeps its
+    trailing ``\\r`` on each line — harmless for the char scans below.
+    """
+    in_fence = False
+    fence_char = ""
+    fence_len = 0
+    fence_open_line = 0
+    fence_open_text = ""
+    for idx, line in enumerate(text.split("\n"), start=1):
+        marker = _fence_marker(line)
+        if in_fence:
+            if (
+                marker is not None
+                and marker[0] == fence_char
+                and marker[1] >= fence_len
+                and marker[2] == ""
+            ):
+                in_fence = False
+            continue
+        if marker is not None:
+            in_fence = True
+            fence_char, fence_len = marker[0], marker[1]
+            fence_open_line = idx
+            fence_open_text = line.strip()
+            continue
+        yield idx, line
+    if in_fence:
+        raise _UnclosedFence(fence_open_line, fence_open_text)
+
+
+# --------------------------------------------------------------------------
+# Copy lint
+# --------------------------------------------------------------------------
+_EM_DASH = "—"
+_EN_DASH = "–"
+_BULLET_PREFIX_RE = re.compile(r"^\s*[-*] ")
+
+
+def _has_emoji(s: str) -> bool:
+    for ch in s:
+        cp = ord(ch)
+        if (
+            0x1F000 <= cp <= 0x1FBFF
+            or 0x2600 <= cp <= 0x27BF
+            or 0x2B00 <= cp <= 0x2BFF
+            or cp == 0xFE0F
+            or cp == 0x200D
+        ):
+            return True
+    return False
+
+
+def lint_copy(readme_text: str) -> "list[tuple[int, str, str]]":
+    """Return copy-lint issues as ``(1-based line_no, code, excerpt)`` tuples.
+
+    Empty list means clean. Codes: ``em-dash``, ``en-dash``, ``emoji``,
+    ``spaced-hyphen``. Scans only outside fenced code blocks; an unclosed fence
+    collapses to a single ``unclosed-fence`` issue.
+    """
+    try:
+        content = list(_iter_content_lines(readme_text))
+    except _UnclosedFence as uf:
+        return [(uf.line_no, "unclosed-fence", uf.excerpt)]
+
+    issues: "list[tuple[int, str, str]]" = []
+    for line_no, line in content:
+        excerpt = line.strip()
+        if _EM_DASH in line:
+            issues.append((line_no, "em-dash", excerpt))
+        if _EN_DASH in line:
+            issues.append((line_no, "en-dash", excerpt))
+        if _has_emoji(line):
+            issues.append((line_no, "emoji", excerpt))
+        # Line-leading "- "/"* " bullets are fine; only a REMAINING mid-line
+        # " - " is a dash substitute.
+        remainder = _BULLET_PREFIX_RE.sub("", line, count=1)
+        if " - " in remainder:
+            issues.append((line_no, "spaced-hyphen", excerpt))
+    return issues
+
+
+# --------------------------------------------------------------------------
+# Highlight normalization
+# --------------------------------------------------------------------------
+def normalize_highlight(bullet: str) -> str:
+    """Rewrite a bullet so it satisfies the copy lint deterministically."""
+    b = re.sub(r"\s*[—–]\s*", ", ", bullet)
+    b = re.sub(r"(, )+", ", ", b)
+    b = b.strip()
+    b = b.strip(", ")
+    return b.strip()
+
+
+# --------------------------------------------------------------------------
+# CHANGELOG extraction
+# --------------------------------------------------------------------------
+_ISSUE_REF_RE = re.compile(r"\s*\(#\d+[^)]*\)\s*$")
+_SUBSECTIONS = (("### Added", "added"), ("### Changed", "changed"), ("### Fixed", "fixed"))
+_SUBSECTION_KEYS = {name: key for name, key in _SUBSECTIONS}
+_BULLET_RE = re.compile(r"^[-*] ")
+
+
+def _section_heading_re(version: str) -> "re.Pattern[str]":
+    return re.compile(
+        r"^## \[" + re.escape(version) + r"\] - (\d{4}-\d{2}-\d{2})\s*$"
+    )
+
+
+def _section_body_lines(changelog_text: str, version: str) -> "list[str]":
+    lines = changelog_text.split("\n")
+    heading_re = _section_heading_re(version)
+    start = None
+    for i, line in enumerate(lines):
+        if heading_re.match(line):
+            start = i
+            break
+    if start is None:
+        raise ChangelogSectionMissing(version)
+    end = len(lines)
+    for j in range(start + 1, len(lines)):
+        if lines[j].startswith("## "):
+            end = j
+            break
+    return lines[start + 1:end]
+
+
+def changelog_release_date(changelog_text: str, version: str) -> str:
+    """Return the ``YYYY-MM-DD`` release date for *version*'s section."""
+    heading_re = _section_heading_re(version)
+    for line in changelog_text.split("\n"):
+        m = heading_re.match(line)
+        if m:
+            return m.group(1)
+    raise ChangelogSectionMissing(version)
+
+
+def _collect_raw_bullets(body_lines: "list[str]") -> "dict[str, list[str]]":
+    """Group raw (un-normalized) bullets by subsection, flattening wraps.
+
+    Raises ``ValueError`` if a fence is opened inside the section and never
+    closed before the section ends.
+    """
+    buckets: "dict[str, list[str]]" = {"added": [], "changed": [], "fixed": []}
+    subsection: "str | None" = None
+    current: "str | None" = None
+    in_fence = False
+    fence_char = ""
+    fence_len = 0
+
+    def finalize() -> None:
+        nonlocal current
+        if current is not None and subsection in buckets:
+            buckets[subsection].append(current.rstrip())
+        current = None
+
+    for line in body_lines:
+        marker = _fence_marker(line)
+        if in_fence:
+            if (
+                marker is not None
+                and marker[0] == fence_char
+                and marker[1] >= fence_len
+                and marker[2] == ""
+            ):
+                in_fence = False
+            continue
+        if marker is not None:
+            finalize()
+            in_fence = True
+            fence_char, fence_len = marker[0], marker[1]
+            continue
+        stripped = line.rstrip()
+        if stripped.startswith("## "):
+            finalize()
+            subsection = None
+            continue
+        if stripped.startswith("### "):
+            finalize()
+            subsection = _SUBSECTION_KEYS.get(stripped)
+            continue
+        if _BULLET_RE.match(line):
+            finalize()
+            current = line[2:].strip() if subsection in buckets else None
+            continue
+        if not stripped:
+            finalize()
+            continue
+        # Continuation of the current bullet.
+        if current is not None:
+            current = current + " " + line.strip()
+    if in_fence:
+        raise ValueError("unclosed code fence inside CHANGELOG section")
+    finalize()
+    return buckets
+
+
+def extract_highlights(
+    changelog_text: str, version: str, *, limit: int = 3
+) -> "list[str]":
+    """Return up to ``limit`` normalized, lint-clean highlight bullets.
+
+    Exact heading match (never prefix), Added > Changed > Fixed priority,
+    multiline flattening, one terminal issue-ref stripped, then normalized.
+    A bullet that still fails the copy lint after normalization is dropped and
+    the next candidate is taken. Raises ``ChangelogSectionMissing`` if the
+    version section is absent.
+    """
+    body = _section_body_lines(changelog_text, version)
+    buckets = _collect_raw_bullets(body)
+    candidates = buckets["added"] + buckets["changed"] + buckets["fixed"]
+    kept: "list[str]" = []
+    for raw in candidates:
+        if len(kept) >= limit:
+            break
+        stripped = _ISSUE_REF_RE.sub("", raw)
+        normalized = normalize_highlight(stripped)
+        if not normalized:
+            continue
+        if lint_copy("- " + normalized):
+            continue  # drop-bullet fallback
+        kept.append(normalized)
+    return kept
+
+
+def render_latest_stable_block(
+    version: str, date: str, bullets: "list[str]"
+) -> str:
+    """Render the auto-maintained latest-stable block (no trailing newline
+    after the last bullet; version+date only when *bullets* is empty)."""
+    base = f"**Latest stable: v{version}** ({date})\n"
+    if bullets:
+        return base + "\n" + "\n".join(f"- {b}" for b in bullets)
+    return base
+
+
+# --------------------------------------------------------------------------
+# Marker splice
+# --------------------------------------------------------------------------
+def _marker_alone_on_line(text: str, idx: int, marker: str) -> bool:
+    before = text[:idx]
+    if before and not before.endswith("\n"):
+        return False
+    tail_start = idx + len(marker)
+    nl = text.find("\n", tail_start)
+    tail = text[tail_start:] if nl == -1 else text[tail_start:nl]
+    return tail.strip() == ""
+
+
+def _find_marker_pair(text: str) -> "tuple[int, int]":
+    if text.count(MARKER_BEGIN) != 1:
+        raise MarkerError(
+            f"expected exactly one {MARKER_BEGIN!r}, found "
+            f"{text.count(MARKER_BEGIN)}"
+        )
+    if text.count(MARKER_END) != 1:
+        raise MarkerError(
+            f"expected exactly one {MARKER_END!r}, found {text.count(MARKER_END)}"
+        )
+    begin_idx = text.index(MARKER_BEGIN)
+    end_idx = text.index(MARKER_END)
+    if begin_idx >= end_idx:
+        raise MarkerError("begin marker does not precede end marker")
+    if not _marker_alone_on_line(text, begin_idx, MARKER_BEGIN):
+        raise MarkerError("begin marker is not alone on its line")
+    if not _marker_alone_on_line(text, end_idx, MARKER_END):
+        raise MarkerError("end marker is not alone on its line")
+    return begin_idx, end_idx
+
+
+def splice_marker_block(readme_text: str, block: str) -> str:
+    """Replace the interior between the markers with ``\\n`` + block + ``\\n``.
+
+    Every byte outside the region is preserved (CRLF included). Fails closed on
+    missing, duplicated, or malformed markers.
+    """
+    begin_idx, end_idx = _find_marker_pair(readme_text)
+    region_start = begin_idx + len(MARKER_BEGIN)
+    return (
+        readme_text[:region_start]
+        + "\n"
+        + block
+        + "\n"
+        + readme_text[end_idx:]
+    )
+
+
+# --------------------------------------------------------------------------
+# CLI (--check)
+# --------------------------------------------------------------------------
+def _marker_wellformed_issues(text: str) -> "list[tuple[int, str, str]]":
+    try:
+        _find_marker_pair(text)
+    except MarkerError as exc:
+        line_no = 1
+        if MARKER_BEGIN in text:
+            line_no = text[: text.index(MARKER_BEGIN)].count("\n") + 1
+        return [(line_no, "markers", str(exc))]
+    return []
+
+
+def _check(path: str) -> int:
+    with open(path, "r", encoding="utf-8") as fh:
+        text = fh.read()
+    issues = lint_copy(text) + _marker_wellformed_issues(text)
+    for line_no, code, excerpt in issues:
+        print(f"{path}:{line_no}: {code}: {excerpt}", file=sys.stderr)
+    return 1 if issues else 0
+
+
+def main(argv: "list[str] | None" = None) -> int:
+    import argparse
+
+    parser = argparse.ArgumentParser(
+        description="Copy-lint + marker well-formedness check for README.md.",
+    )
+    parser.add_argument(
+        "--check",
+        metavar="FILE",
+        required=True,
+        help="Markdown file to lint (copy rules + latest-stable markers).",
+    )
+    args = parser.parse_args(argv)
+    return _check(args.check)
+
+
+if __name__ == "__main__":
+    sys.exit(main())