Proper batch gui that lets you manage batch history and reuse batches

TODO.md

@@ -6,106 +6,6 @@
 - Add an optional **per-game custom prompt** in the UI that merges with the default (custom rules layered on top of shared instructions, not a full replacement unless explicitly chosen).
 - Surface this in settings / project setup so each game folder can carry its own tone, terminology notes, or content warnings without forking the global prompt.
 
-## Batch history tab
-
-- Add a **Batch** tab (or dedicated panel) in the GUI to manage past Anthropic batch runs saved locally.
-- Persist batch metadata locally (batch id, submit time, file set, status, cost estimate, etc.) alongside existing `log/batch_*.json` artifacts (`batch_state.json`, `batch_results.json`, `batch_requests.json`).
-- Actions:
-  - **List batches** — browse all batches the tool has submitted or tracked for this project.
-  - **Redownload** — fetch results again from Anthropic for a completed batch (e.g. after a crash before consume, or to recover results). **Not in sketch yet.**
-  - **Cancel batch** — cancel an in-flight batch where the API allows it (`in_progress` → `canceling` → `ended`).
-  - **Usage / cost** — sum real billed tokens from batch results (input, output, cache read/write, thinking).
-  - Resume / link into the existing collect → submit → consume flow when a batch is still pending or fetched but not consumed.
-
-### Sketch: CLI batch manager (adapt for DazedMTLTool)
-
-Reference from sibling tooling (`batches.py`). Port to this repo using `_get_anthropic_client()` / `fetchTranslationBatches()` in `util/translation.py` and `log/batch_state.json` (maps to sketch's `tl/_batch_state.json`).
-
-```python
-#!/usr/bin/env python3
-# -*- coding: utf-8 -*-
-"""
-batches.py — list and cancel Anthropic Message Batches (the ones DazedMTLTool
-submits). Handy when a run is stuck polling or you submitted by mistake.
-
-    set ANTHROPIC_API_KEY=sk-...   # or key= in .env
-
-    python tooling/batches.py                 # list recent batches (default)
-    python tooling/batches.py list --limit 50
-    python tooling/batches.py cancel <batch_id> [<batch_id> ...]
-    python tooling/batches.py cancel --all     # cancel every in-progress batch
-    python tooling/batches.py cancel --mine    # cancel batch(es) in log/batch_state.json
-    python tooling/batches.py usage [batch_id] # real billed tokens + estimated cost
-
-    # TODO: redownload <batch_id>  — re-fetch results into log/batch_results.json
-
-Only an in-progress batch can be cancelled; cancellation finishes any in-flight
-requests (in_progress -> canceling -> ended). Already-ended batches can't be
-cancelled.
-"""
-import os
-import sys
-import argparse
-import collections
-
-# Adapt imports to DazedMTLTool:
-#   from util.translation import _get_anthropic_client, BATCH_STATE_FILE, _read_batch_file, fetchTranslationBatches
-
-CANCELABLE = {"in_progress"}
-
-
-def _counts(b):
-    rc = getattr(b, "request_counts", None)
-    if not rc:
-        return ""
-    parts = []
-    for k in ("processing", "succeeded", "errored", "canceled", "expired"):
-        v = getattr(rc, k, None)
-        if v:
-            parts.append(f"{k[:4]}={v}")
-    return " ".join(parts)
-
-
-def _mine_batch_ids():
-    """Return batch ids from log/batch_state.json (DazedMTLTool format)."""
-    # state = _read_batch_file(BATCH_STATE_FILE) or {}
-    # return [b["id"] for b in state.get("batches", [])]
-    ...
-
-
-def cmd_list(client, limit):
-    mine = set(_mine_batch_ids())
-    print(f"{'BATCH ID':<28} {'STATUS':<12} {'CREATED':<22} COUNTS")
-    print("-" * 90)
-    n = 0
-    for b in client.messages.batches.list(limit=limit):
-        n += 1
-        mark = "  <- yours" if b.id in mine else ""
-        created = str(getattr(b, "created_at", "") or "")[:22]
-        print(f"{b.id:<28} {b.processing_status:<12} {created:<22} {_counts(b)}{mark}")
-    ...
-
-
-def cmd_cancel(client, ids, do_all, do_mine, limit):
-    # targets from ids, --mine (log/batch_state.json), or --all (in_progress only)
-    # client.messages.batches.cancel(bid)
-    ...
-
-
-def cmd_usage(client, bid, model):
-    # Sum usage across client.messages.batches.results(bid)
-    # Price with getPricingConfig(model) + 50% batch discount on I/O
-    ...
-
-
-# cmd_redownload(client, bid):
-#     # TODO: client.messages.batches.results(bid) -> merge into log/batch_results.json
-#     # Wire custom_id map from batch_state.json entries
-#     ...
-```
-
-**GUI tab (later):** same operations as buttons on a saved local history table — list, cancel, usage, redownload, resume consume.
-
 ## Local translation cache (game retranslate)
 
 - Persist every successful translation locally so a full retranslate from scratch can reuse prior results instead of paying again.
@@ -117,5 +17,3 @@ def cmd_usage(client, bid, model):
 - **Retranslate workflow:** when the user re-imports fresh JSON and runs Translate again, cache hits should fill `translated/` automatically for unchanged source lines; only new or edited source text should incur API cost.
 - **UI (later):** cache stats (hits / misses / estimated savings), export/import, clear cache for one game, optional “prefer cache” toggle.
 - **Invalidation:** optional metadata (model id, prompt version) in the key or entry if we need to avoid reusing stale translations after prompt or model changes.
-
-## According to someone from f95, the patcher breaks if a folder's name has an apostrophe

gui/batch_tab.py

@@ -0,0 +1,467 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""Batch history tab - list / cancel / usage / redownload / resume."""
+
+from __future__ import annotations
+
+import os
+from pathlib import Path
+
+from PyQt5.QtCore import Qt, QThread, pyqtSignal, QTimer
+from PyQt5.QtWidgets import (
+    QWidget,
+    QVBoxLayout,
+    QHBoxLayout,
+    QPushButton,
+    QLabel,
+    QTableWidget,
+    QTableWidgetItem,
+    QHeaderView,
+    QTextEdit,
+    QMessageBox,
+    QAbstractItemView,
+    QSplitter,
+)
+
+from dotenv import load_dotenv
+
+
+class _BatchOpsWorker(QThread):
+    """Run a blocking batch-history callable off the UI thread."""
+
+    done = pyqtSignal(bool, str, object)  # ok, message, payload
+    log = pyqtSignal(str)
+
+    def __init__(self, task, project_root: Path):
+        super().__init__()
+        self._task = task
+        self._project_root = project_root
+
+    def run(self):
+        old = os.getcwd()
+        try:
+            os.chdir(str(self._project_root))
+            load_dotenv()
+            ok, msg, payload = self._task(self.log.emit)
+            self.done.emit(bool(ok), str(msg), payload)
+        except Exception as exc:
+            import traceback
+
+            self.log.emit(traceback.format_exc())
+            self.done.emit(False, f"Error: {exc}", None)
+        finally:
+            try:
+                os.chdir(old)
+            except Exception:
+                pass
+
+
+class BatchTab(QWidget):
+    """Manage past / in-flight Anthropic Message Batches for this project."""
+
+    COLUMNS = [
+        "Batch ID",
+        "Status",
+        "Created",
+        "Requests",
+        "Model",
+        "Estimate",
+        "Actual $",
+        "Files",
+    ]
+
+    def __init__(self, parent=None):
+        super().__init__(parent)
+        self.parent_window = parent
+        self.project_root = Path(
+            getattr(parent, "project_root", None) or Path(__file__).resolve().parent.parent
+        )
+        self._worker: _BatchOpsWorker | None = None
+        self._rows: list[dict] = []
+        self._loaded_once = False
+        self._init_ui()
+
+    def showEvent(self, event):
+        super().showEvent(event)
+        # Refresh every time the Batches page is shown so live status stays current.
+        if self._worker is not None and self._worker.isRunning():
+            return
+        self._loaded_once = True
+        QTimer.singleShot(0, lambda: self.refresh_list(live=True))
+
+    def _init_ui(self):
+        layout = QVBoxLayout(self)
+        layout.setContentsMargins(16, 16, 16, 16)
+        layout.setSpacing(10)
+
+        title = QLabel("Batch History")
+        title.setStyleSheet("color:#ffffff;font-size:18px;font-weight:bold;")
+        layout.addWidget(title)
+
+        hint = QLabel(
+            "Local Anthropic Message Batch runs for this project. "
+            "Cancel / redownload / usage never submit a new batch."
+        )
+        hint.setWordWrap(True)
+        hint.setStyleSheet("color:#9d9d9d;font-size:12px;")
+        layout.addWidget(hint)
+
+        toolbar = QHBoxLayout()
+        toolbar.setSpacing(8)
+
+        self.refresh_btn = QPushButton("Refresh")
+        self.refresh_btn.setToolTip("Reload local history and refresh live status for active batches")
+        self.refresh_btn.clicked.connect(lambda: self.refresh_list(live=True))
+
+        self.cancel_btn = QPushButton("Cancel")
+        self.cancel_btn.setToolTip("Cancel selected in-progress batch(es)")
+        self.cancel_btn.clicked.connect(self.cancel_selected)
+
+        self.usage_btn = QPushButton("Usage")
+        self.usage_btn.setToolTip("Sum real billed tokens for the selected ended batch")
+        self.usage_btn.clicked.connect(self.usage_selected)
+
+        self.redownload_btn = QPushButton("Redownload")
+        self.redownload_btn.setToolTip("Re-fetch results into log/batch_results.json (no re-submit)")
+        self.redownload_btn.clicked.connect(self.redownload_selected)
+
+        self.resume_btn = QPushButton("Resume")
+        self.resume_btn.setToolTip("Activate this batch and continue poll/consume on the Translation tab")
+        self.resume_btn.clicked.connect(self.resume_selected)
+
+        for btn in (
+            self.refresh_btn,
+            self.cancel_btn,
+            self.usage_btn,
+            self.redownload_btn,
+            self.resume_btn,
+        ):
+            btn.setStyleSheet(
+                "QPushButton{background-color:#3c3c3c;color:#cccccc;border:1px solid #555555;"
+                "border-radius:4px;padding:6px 12px;}"
+                "QPushButton:hover{border-color:#007acc;}"
+                "QPushButton:disabled{color:#666666;border-color:#444444;}"
+            )
+            toolbar.addWidget(btn)
+        toolbar.addStretch()
+        layout.addLayout(toolbar)
+
+        splitter = QSplitter(Qt.Vertical)
+
+        self.table = QTableWidget(0, len(self.COLUMNS))
+        self.table.setHorizontalHeaderLabels(self.COLUMNS)
+        self.table.setSelectionBehavior(QAbstractItemView.SelectRows)
+        self.table.setSelectionMode(QAbstractItemView.ExtendedSelection)
+        self.table.setEditTriggers(QAbstractItemView.NoEditTriggers)
+        self.table.setAlternatingRowColors(True)
+        self.table.verticalHeader().setVisible(False)
+        self.table.setStyleSheet(
+            "QTableWidget{background-color:#1e1e1e;color:#cccccc;gridline-color:#3a3a3a;"
+            "alternate-background-color:#252526;}"
+            "QHeaderView::section{background-color:#2d2d30;color:#cccccc;padding:4px;"
+            "border:1px solid #3a3a3a;}"
+        )
+        header = self.table.horizontalHeader()
+        header.setSectionResizeMode(0, QHeaderView.Stretch)
+        for i in range(1, len(self.COLUMNS)):
+            header.setSectionResizeMode(i, QHeaderView.ResizeToContents)
+        self.table.itemSelectionChanged.connect(self._update_button_states)
+        splitter.addWidget(self.table)
+
+        self.log = QTextEdit()
+        self.log.setReadOnly(True)
+        self.log.setPlaceholderText("Operation log…")
+        self.log.setStyleSheet(
+            "QTextEdit{background-color:#1e1e1e;color:#cccccc;border:1px solid #3a3a3a;"
+            "font-family:monospace;font-size:12px;}"
+        )
+        splitter.addWidget(self.log)
+        splitter.setStretchFactor(0, 3)
+        splitter.setStretchFactor(1, 1)
+        layout.addWidget(splitter, 1)
+
+        self._update_button_states()
+
+    def _append_log(self, text: str):
+        if not text:
+            return
+        self.log.append(text.rstrip())
+
+    def _selected_entries(self) -> list[dict]:
+        rows = sorted({idx.row() for idx in self.table.selectedIndexes()})
+        out = []
+        for r in rows:
+            if 0 <= r < len(self._rows):
+                out.append(self._rows[r])
+        return out
+
+    def _update_button_states(self):
+        busy = self._worker is not None and self._worker.isRunning()
+        entries = self._selected_entries()
+        n = len(entries)
+        self.refresh_btn.setEnabled(not busy)
+        self.cancel_btn.setEnabled(
+            not busy and n >= 1 and any(
+                (e.get("api_status") == "in_progress") or (e.get("status") == "submitted")
+                for e in entries
+            )
+        )
+        self.usage_btn.setEnabled(not busy and n == 1)
+        self.redownload_btn.setEnabled(
+            not busy
+            and n == 1
+            and entries[0].get("status") in ("ended", "fetched", "consumed", "submitted", "canceling")
+        )
+        self.resume_btn.setEnabled(
+            not busy
+            and n == 1
+            and entries[0].get("status")
+            in ("submitted", "canceling", "ended", "fetched")
+        )
+
+    def _set_busy(self, busy: bool):
+        if busy:
+            for btn in (
+                self.refresh_btn,
+                self.cancel_btn,
+                self.usage_btn,
+                self.redownload_btn,
+                self.resume_btn,
+            ):
+                btn.setEnabled(False)
+        else:
+            self._update_button_states()
+
+    def _clear_worker(self):
+        if self.sender() is self._worker:
+            self._worker = None
+
+    def _run_task(self, task, on_done=None):
+        if self._worker is not None and self._worker.isRunning():
+            QMessageBox.information(self, "Busy", "A batch operation is already running.")
+            return
+
+        self._set_busy(True)
+        worker = _BatchOpsWorker(task, self.project_root)
+        self._worker = worker
+
+        def _finished(ok, msg, payload):
+            self._set_busy(False)
+            if msg:
+                self._append_log(msg)
+            if on_done:
+                try:
+                    on_done(ok, msg, payload)
+                except Exception as exc:
+                    self._append_log(f"[BATCH] UI update failed: {exc}")
+
+        worker.log.connect(self._append_log)
+        worker.done.connect(_finished)
+        # Release the reference only after run() returns. Clearing in the done
+        # slot aborts with "QThread: Destroyed while thread is still running".
+        worker.finished.connect(worker.deleteLater)
+        worker.finished.connect(self._clear_worker)
+        worker.start()
+
+    def _populate_table(self, entries: list[dict]):
+        self._rows = list(entries or [])
+        self.table.setRowCount(len(self._rows))
+        for r, entry in enumerate(self._rows):
+            est = entry.get("cost_estimate") or {}
+            est_s = ""
+            if isinstance(est, dict) and est.get("batch_cached_cost") is not None:
+                try:
+                    est_s = f"${float(est['batch_cached_cost']):.2f}"
+                except (TypeError, ValueError):
+                    est_s = ""
+            actual = entry.get("actual_cost")
+            actual_s = f"${actual:.4f}" if isinstance(actual, (int, float)) else ""
+            files = entry.get("file_set") or []
+            if not isinstance(files, list):
+                files = [str(files)]
+            files_s = ", ".join(str(f) for f in files[:3])
+            if len(files) > 3:
+                files_s += f" (+{len(files) - 3})"
+            values = [
+                str(entry.get("id") or ""),
+                str(entry.get("status") or ""),
+                str(entry.get("created_at") or "")[:19],
+                str(entry.get("request_count") or ""),
+                str(entry.get("model") or ""),
+                est_s,
+                actual_s,
+                files_s,
+            ]
+            for c, val in enumerate(values):
+                item = QTableWidgetItem(val)
+                if c == 0:
+                    item.setData(Qt.UserRole, entry.get("id"))
+                self.table.setItem(r, c, item)
+        self._update_button_states()
+
+    def refresh_list(self, live: bool = True):
+        def task(log):
+            from util.batch_history import list_local_batches
+
+            log("[BATCH] Loading local history...")
+            entries = list_local_batches(refresh_live=live)
+            log(f"[BATCH] {len(entries)} batch(es) in history.")
+            return True, f"Loaded {len(entries)} batch(es).", entries
+
+        def done(ok, _msg, payload):
+            if ok and isinstance(payload, list):
+                self._populate_table(payload)
+
+        self._run_task(task, on_done=done)
+
+    def cancel_selected(self):
+        entries = self._selected_entries()
+        if not entries:
+            return
+        ids = [e["id"] for e in entries if e.get("id")]
+        reply = QMessageBox.question(
+            self,
+            "Cancel Batch?",
+            "Cancel the selected in-progress batch(es)?\n\n"
+            "Anthropic may still finish and bill requests that were already "
+            "in flight when cancel is received.\n"
+            "This does not submit a new batch.",
+            QMessageBox.Yes | QMessageBox.No,
+        )
+        if reply != QMessageBox.Yes:
+            return
+
+        def task(log):
+            from util.batch_history import cancel_batches
+
+            log(f"[BATCH] Canceling {len(ids)} batch(es)…")
+            results = cancel_batches(ids)
+            lines = []
+            for r in results:
+                if r.get("ok"):
+                    lines.append(f"  canceled {r['id']} -> {r.get('api_status')}")
+                else:
+                    lines.append(f"  failed {r['id']}: {r.get('error')}")
+            for line in lines:
+                log(line)
+            return True, "Cancel finished.", results
+
+        def done(ok, _msg, _payload):
+            self.refresh_list(live=False)
+
+        self._run_task(task, on_done=done)
+
+    def usage_selected(self):
+        entries = self._selected_entries()
+        if len(entries) != 1:
+            return
+        bid = entries[0]["id"]
+
+        def task(log):
+            from util.batch_history import usage_for_batch
+
+            log(f"[BATCH] Computing usage for {bid}…")
+            info = usage_for_batch(bid)
+            u = info.get("usage") or {}
+            log(
+                f"[BATCH] tokens: in={u.get('input_tokens', 0)} out={u.get('output_tokens', 0)} "
+                f"cache_read={u.get('cache_read_input_tokens', 0)} "
+                f"cache_write={u.get('cache_creation_input_tokens', 0)} "
+                f"thinking={u.get('thinking_tokens', 0)}"
+            )
+            cost = info.get("actual_cost")
+            if cost is not None:
+                log(f"[BATCH] estimated billed cost (batch 50% off): ${cost:.4f}")
+            return True, "Usage updated.", info
+
+        def done(ok, _msg, _payload):
+            self.refresh_list(live=False)
+
+        self._run_task(task, on_done=done)
+
+    def redownload_selected(self):
+        entries = self._selected_entries()
+        if len(entries) != 1:
+            return
+        bid = entries[0]["id"]
+        reply = QMessageBox.question(
+            self,
+            "Redownload Results?",
+            f"Re-fetch results for {bid} into log/batch_results.json?\n\n"
+            "This does not create a new batch or re-collect.",
+            QMessageBox.Yes | QMessageBox.No,
+        )
+        if reply != QMessageBox.Yes:
+            return
+
+        def task(log):
+            from util.batch_history import redownload_batch
+
+            log(f"[BATCH] Redownloading {bid}…")
+            info = redownload_batch(bid)
+            log(
+                f"[BATCH] redownload ok={info.get('succeeded')} err={info.get('errored')} "
+                f"cost≈{info.get('actual_cost')}"
+            )
+            return True, "Redownload finished.", info
+
+        def done(ok, msg, payload):
+            self.refresh_list(live=False)
+            if not ok:
+                return
+            reply = QMessageBox.question(
+                self,
+                "Resume Consume?",
+                "Results are ready locally. Switch to Translation and resume consume now?",
+                QMessageBox.Yes | QMessageBox.No,
+            )
+            if reply == QMessageBox.Yes:
+                self._resume_with_state("fetched", entries[0])
+
+        self._run_task(task, on_done=done)
+
+    def resume_selected(self):
+        entries = self._selected_entries()
+        if len(entries) != 1:
+            return
+        entry = entries[0]
+        bid = entry["id"]
+
+        def task(log):
+            from util.batch_history import activate_for_resume
+
+            log(f"[BATCH] Activating {bid} for resume…")
+            state = activate_for_resume(bid)
+            log(f"[BATCH] active resume state: {state}")
+            return True, f"Activated ({state}).", {"state": state, "entry": entry}
+
+        def done(ok, _msg, payload):
+            self.refresh_list(live=False)
+            if not ok or not payload:
+                return
+            self._resume_with_state(payload["state"], payload.get("entry") or entry)
+
+        self._run_task(task, on_done=done)
+
+    def _resume_with_state(self, resume_state: str, entry: dict):
+        parent = self.parent_window
+        if parent is None or not hasattr(parent, "translation_tab"):
+            QMessageBox.warning(self, "Resume", "Translation tab is not available.")
+            return
+        tt = parent.translation_tab
+        file_set = entry.get("file_set") or []
+        if file_set and hasattr(tt, "select_files_by_name"):
+            tt.select_files_by_name(file_set)
+        # Switch to Translation page (index 0).
+        if hasattr(parent, "switch_page"):
+            parent.switch_page(0)
+        reply = QMessageBox.question(
+            self,
+            "Start Resume?",
+            f"Start Batch Translate resume ({resume_state}) on the Translation tab?\n\n"
+            "This will not clear batch files or submit a new batch.",
+            QMessageBox.Yes | QMessageBox.No,
+        )
+        if reply != QMessageBox.Yes:
+            return
+        tt.start_translation(forced_resume_state=resume_state)

gui/main.py

@@ -595,6 +595,7 @@ from gui.config_tab import ConfigTab
 from gui.translation_tab import TranslationTab
 from gui.workflow_tab import WorkflowTab
 from gui.wolf_workflow_tab import WolfWorkflowTab
+from gui.batch_tab import BatchTab
 
 class DazedMTLGUI(QMainWindow):
     """Main GUI window for the DazedMTLTool."""
@@ -830,9 +831,16 @@ class DazedMTLGUI(QMainWindow):
         sidebar_layout.addWidget(btn_workflow)
         self.nav_buttons.append(btn_workflow)
 
-        # Configuration button (third)
+        # Batch history button (third)
+        btn_batches = self.create_nav_button("📦", "Batches")
+        btn_batches.setToolTip("Batches — Anthropic Message Batch history")
+        btn_batches.clicked.connect(lambda: self.switch_page(2))
+        sidebar_layout.addWidget(btn_batches)
+        self.nav_buttons.append(btn_batches)
+
+        # Configuration button (fourth)
         btn_config = self.create_nav_button("⚙️", "Configuration")
-        btn_config.clicked.connect(lambda: self.switch_page(2))
+        btn_config.clicked.connect(lambda: self.switch_page(3))
         sidebar_layout.addWidget(btn_config)
         self.nav_buttons.append(btn_config)
         
@@ -884,6 +892,8 @@ class DazedMTLGUI(QMainWindow):
         
     def setup_tabs(self):
         """Set up all the tabs in the interface."""
+        self.project_root = PROJECT_ROOT
+
         # Translation Execution Tab (index 0)
         self.translation_tab = TranslationTab(self)
         self.content_stack.addWidget(self.translation_tab)
@@ -892,7 +902,11 @@ class DazedMTLGUI(QMainWindow):
         # RPGMaker and Wolf guided panels while keeping a single sidebar button.
         self.content_stack.addWidget(self._create_workflow_container())
 
-        # Configuration Tab (index 2)
+        # Batch History Tab (index 2)
+        self.batch_tab = BatchTab(self)
+        self.content_stack.addWidget(self.batch_tab)
+
+        # Configuration Tab (index 3)
         self.config_tab = ConfigTab()
         self.config_tab.config_changed.connect(self.on_config_changed)
         self.content_stack.addWidget(self.config_tab)

modules/main.py

@@ -1,5 +1,6 @@
 import sys
 import os
+import time
 import traceback
 import datetime
 from pathlib import Path
@@ -128,7 +129,18 @@ def main():
         if resume_state:
             confirm = ""
             while confirm not in ("y", "n"):
-                confirm = input(f"A previous batch run was interrupted ({resume_state}). Resume it? (y/n)\n").strip().lower()
+                if resume_state == "queued":
+                    prompt = (
+                        "A previous collect finished but was not submitted (queued). "
+                        "Resume and submit that queue? (y/n)\n"
+                        "(n discards the queue and re-collects - live collect charges again)\n"
+                    )
+                else:
+                    prompt = (
+                        f"A previous batch run was interrupted ({resume_state}). Resume it? (y/n)\n"
+                        "(n discards it and can bill again)\n"
+                    )
+                confirm = input(prompt).strip().lower()
             if confirm == "n":
                 resume_state = None
 
@@ -322,21 +334,72 @@ files to translate are in the /files folder and that you picked the right game e
                 tqdm.write("[BATCH] No requests queued — nothing needed the API.")
                 run_consume = False
             else:
-                estimateBatchCost()
+                est = estimateBatchCost()
                 confirm = ""
                 while confirm not in ("y", "n"):
                     confirm = input("Submit batch? (y/n)\n").strip().lower()
                 if confirm == "n":
-                    tqdm.write("[BATCH] Not submitted. The queue is kept in log/batch_requests.json.")
+                    tqdm.write(
+                        "[BATCH] Not submitted. The queue is kept in log/batch_requests.json "
+                        "(resume Batch Translate later to submit without re-collecting)."
+                    )
                     return
-                runTranslationBatches(poll)
+                from util.translation import submitTranslationBatches, checkTranslationBatches, fetchTranslationBatches
+                if not submitTranslationBatches(cost_estimate=est):
+                    run_consume = False
+                else:
+                    tqdm.write(
+                        f"[BATCH] polling every {poll}s (Ctrl-C is safe - resume later)..."
+                    )
+                    while not checkTranslationBatches():
+                        time.sleep(poll)
+                    fetchTranslationBatches()
+        elif resume_state == "queued":
+            tqdm.write(
+                Fore.CYAN
+                + "[BATCH] Resuming queued requests (skipping re-collect)..."
+                + Fore.RESET
+            )
+            if pendingBatchRequests() == 0:
+                tqdm.write("[BATCH] Queue is empty - nothing to submit.")
+                run_consume = False
+            else:
+                est = estimateBatchCost()
+                confirm = ""
+                while confirm not in ("y", "n"):
+                    confirm = input("Submit batch? (y/n)\n").strip().lower()
+                if confirm == "n":
+                    tqdm.write("[BATCH] Not submitted. Queue kept.")
+                    return
+                from util.translation import submitTranslationBatches, checkTranslationBatches, fetchTranslationBatches
+                if not submitTranslationBatches(cost_estimate=est):
+                    run_consume = False
+                else:
+                    tqdm.write(
+                        f"[BATCH] polling every {poll}s (Ctrl-C is safe - resume later)..."
+                    )
+                    while not checkTranslationBatches():
+                        time.sleep(poll)
+                    fetchTranslationBatches()
         elif resume_state == "submitted":
             tqdm.write(Fore.CYAN + "[BATCH] Resuming the submitted batch..." + Fore.RESET)
             runTranslationBatches(poll)
-        else:  # "fetched" — results already downloaded, just write the files
+        else:  # "fetched" - results already downloaded, just write the files
             tqdm.write(Fore.CYAN + "[BATCH] Resuming from fetched results..." + Fore.RESET)
 
         if run_consume:
+            try:
+                from util.batch_history import missing_result_count
+                present, expected = missing_result_count()
+                if expected and present < expected:
+                    tqdm.write(
+                        Fore.YELLOW
+                        + f"[BATCH] WARNING: only {present}/{expected} results present. "
+                        "Missing keys fall back to the live API (full price)."
+                        + Fore.RESET
+                    )
+            except Exception:
+                pass
             # Pass 2 — write the translated files from the fetched results.
             # Anything the batch missed falls back to the live API.
             tqdm.write(Fore.CYAN + "[BATCH] Pass 2/2: writing translated files..." + Fore.RESET)

tests/test_batch_history.py

@@ -0,0 +1,265 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""Tests for durable batch history and spend-safe ops."""
+
+from __future__ import annotations
+
+import tempfile
+import unittest
+from pathlib import Path
+from types import SimpleNamespace
+from unittest import mock
+
+import util.translation as T
+import util.batch_history as BH
+
+
+class BatchHistoryTestBase(unittest.TestCase):
+    """Isolate batch JSON files to a temp dir."""
+
+    def setUp(self):
+        self._tmp = tempfile.TemporaryDirectory()
+        tmp = Path(self._tmp.name)
+        self._orig = {
+            "QUEUE": T.BATCH_QUEUE_FILE,
+            "STATE": T.BATCH_STATE_FILE,
+            "RESULTS": T.BATCH_RESULTS_FILE,
+            "LOCK": T.BATCH_LOCK_FILE,
+            "HISTORY": BH.BATCH_HISTORY_FILE,
+            "results_mem": T._batch_results,
+            "pending": dict(T._batch_queue_pending),
+        }
+        T.BATCH_QUEUE_FILE = tmp / "batch_requests.json"
+        T.BATCH_STATE_FILE = tmp / "batch_state.json"
+        T.BATCH_RESULTS_FILE = tmp / "batch_results.json"
+        T.BATCH_LOCK_FILE = tmp / "batch_files.lock"
+        BH.BATCH_HISTORY_FILE = tmp / "batch_history.json"
+        # batch_history imports queue/state/results paths at call time via T.* —
+        # but it also imported BATCH_* as names. Rebind module-level aliases.
+        BH.BATCH_QUEUE_FILE = T.BATCH_QUEUE_FILE
+        BH.BATCH_STATE_FILE = T.BATCH_STATE_FILE
+        BH.BATCH_RESULTS_FILE = T.BATCH_RESULTS_FILE
+        T._batch_results = None
+        T._batch_queue_pending = {}
+
+    def tearDown(self):
+        T.BATCH_QUEUE_FILE = self._orig["QUEUE"]
+        T.BATCH_STATE_FILE = self._orig["STATE"]
+        T.BATCH_RESULTS_FILE = self._orig["RESULTS"]
+        T.BATCH_LOCK_FILE = self._orig["LOCK"]
+        BH.BATCH_HISTORY_FILE = self._orig["HISTORY"]
+        BH.BATCH_QUEUE_FILE = T.BATCH_QUEUE_FILE
+        BH.BATCH_STATE_FILE = T.BATCH_STATE_FILE
+        BH.BATCH_RESULTS_FILE = T.BATCH_RESULTS_FILE
+        T._batch_results = self._orig["results_mem"]
+        T._batch_queue_pending = self._orig["pending"]
+        self._tmp.cleanup()
+
+
+class BatchRunStateTests(BatchHistoryTestBase):
+    def test_none_when_empty(self):
+        self.assertIsNone(T.batchRunState())
+
+    def test_queued_when_only_queue(self):
+        T._write_batch_file(T.BATCH_QUEUE_FILE, {"k1": {"payload": "x", "language": "English", "params": {}}})
+        self.assertEqual(T.batchRunState(), "queued")
+
+    def test_submitted_when_state_has_batches(self):
+        T._write_batch_file(
+            T.BATCH_STATE_FILE,
+            {"batches": [{"id": "msgbatch_1", "custom_ids": {"req-000000": "k1"}}]},
+        )
+        self.assertEqual(T.batchRunState(), "submitted")
+
+    def test_fetched_when_results_present(self):
+        T._write_batch_file(T.BATCH_RESULTS_FILE, {"k1": {"text": "hi"}})
+        self.assertEqual(T.batchRunState(), "fetched")
+
+    def test_fetched_when_state_status_fetched(self):
+        T._write_batch_file(
+            T.BATCH_STATE_FILE,
+            {"status": "fetched", "batch_ids": ["msgbatch_1"], "batches": []},
+        )
+        self.assertEqual(T.batchRunState(), "fetched")
+
+
+class HistorySurvivalTests(BatchHistoryTestBase):
+    def test_history_survives_fetch_marker_and_clear(self):
+        custom_ids = {"req-000000": "cachekey1", "req-000001": "cachekey2"}
+        BH.record_submit(
+            [{"id": "msgbatch_abc", "custom_ids": custom_ids}],
+            model="claude-sonnet-4-5",
+            file_set=["Map001.json"],
+            cost_estimate={"batch_cached_cost": 1.23, "model": "claude-sonnet-4-5"},
+        )
+        BH.record_fetch(["msgbatch_abc"], succeeded=2, errored=0, usage={"input_tokens": 10}, actual_cost=0.5)
+        T._write_batch_file(T.BATCH_RESULTS_FILE, {"cachekey1": {"text": "A"}, "cachekey2": {"text": "B"}})
+        T._write_batch_file(
+            T.BATCH_STATE_FILE,
+            {"status": "fetched", "batch_ids": ["msgbatch_abc"], "batches": []},
+        )
+
+        T.clearBatchFiles()
+
+        # Active files gone…
+        self.assertFalse(T.BATCH_RESULTS_FILE.exists())
+        self.assertFalse(T.BATCH_STATE_FILE.exists())
+        # …but history retains custom_ids and is marked consumed.
+        history = BH.read_history()
+        entry = history["batches"][0]
+        self.assertEqual(entry["id"], "msgbatch_abc")
+        self.assertEqual(entry["custom_ids"], custom_ids)
+        self.assertEqual(entry["status"], BH.STATUS_CONSUMED)
+        self.assertEqual(entry["file_set"], ["Map001.json"])
+
+    def test_clear_does_not_wipe_history_file(self):
+        BH.upsert_history_entry("msgbatch_keep", status=BH.STATUS_SUBMITTED, custom_ids={"a": "b"})
+        T.clearBatchFiles()
+        self.assertTrue(BH.BATCH_HISTORY_FILE.exists())
+        self.assertEqual(len(BH.read_history()["batches"]), 1)
+
+
+class RedownloadTests(BatchHistoryTestBase):
+    def test_redownload_rebuilds_results_from_custom_ids(self):
+        custom_ids = {"req-000000": "keyA", "req-000001": "keyB"}
+        BH.upsert_history_entry(
+            "msgbatch_rd",
+            status=BH.STATUS_ENDED,
+            model="claude-sonnet-4-5",
+            custom_ids=custom_ids,
+            request_count=2,
+        )
+
+        usage = SimpleNamespace(
+            input_tokens=100,
+            output_tokens=50,
+            cache_read_input_tokens=10,
+            cache_creation_input_tokens=20,
+            thinking_tokens=5,
+        )
+        msg = SimpleNamespace(
+            content=[SimpleNamespace(text='{"Line1":"Hi"}')],
+            usage=usage,
+        )
+        ok_result = SimpleNamespace(type="succeeded", message=msg)
+        row = SimpleNamespace(custom_id="req-000000", result=ok_result)
+        row2_usage = SimpleNamespace(
+            input_tokens=80,
+            output_tokens=40,
+            cache_read_input_tokens=0,
+            cache_creation_input_tokens=0,
+            thinking_tokens=0,
+        )
+        msg2 = SimpleNamespace(content=[SimpleNamespace(text='{"Line1":"Yo"}')], usage=row2_usage)
+        row2 = SimpleNamespace(custom_id="req-000001", result=SimpleNamespace(type="succeeded", message=msg2))
+
+        fake_batch = SimpleNamespace(id="msgbatch_rd", processing_status="ended")
+        client = mock.MagicMock()
+        client.messages.batches.retrieve.return_value = fake_batch
+        client.messages.batches.results.return_value = [row, row2]
+
+        with mock.patch.object(BH, "_get_anthropic_client", return_value=client):
+            with mock.patch.object(BH, "getPricingConfig", return_value={"inputAPICost": 3.0, "outputAPICost": 15.0}):
+                info = BH.redownload_batch("msgbatch_rd")
+
+        self.assertEqual(info["succeeded"], 2)
+        results = T._read_batch_file(T.BATCH_RESULTS_FILE)
+        self.assertIn("keyA", results)
+        self.assertIn("keyB", results)
+        self.assertEqual(results["keyA"]["text"], '{"Line1":"Hi"}')
+        state = T._read_batch_file(T.BATCH_STATE_FILE)
+        self.assertEqual(state.get("status"), "fetched")
+        self.assertEqual(T.batchRunState(), "fetched")
+        entry = BH.read_history()["batches"][0]
+        self.assertEqual(entry["status"], BH.STATUS_FETCHED)
+        self.assertEqual(entry["custom_ids"], custom_ids)
+
+
+class CancelTests(BatchHistoryTestBase):
+    def test_cancel_updates_history_and_active_state(self):
+        T._write_batch_file(
+            T.BATCH_STATE_FILE,
+            {
+                "batches": [
+                    {"id": "msgbatch_c1", "custom_ids": {"req-000000": "k"}},
+                    {"id": "msgbatch_c2", "custom_ids": {"req-000000": "k2"}},
+                ]
+            },
+        )
+        BH.upsert_history_entry("msgbatch_c1", status=BH.STATUS_SUBMITTED, custom_ids={"req-000000": "k"})
+        BH.upsert_history_entry("msgbatch_c2", status=BH.STATUS_SUBMITTED, custom_ids={"req-000000": "k2"})
+
+        before = SimpleNamespace(id="msgbatch_c1", processing_status="in_progress")
+        after = SimpleNamespace(id="msgbatch_c1", processing_status="canceling")
+        client = mock.MagicMock()
+        client.messages.batches.retrieve.return_value = before
+        client.messages.batches.cancel.return_value = after
+
+        with mock.patch.object(BH, "_get_anthropic_client", return_value=client):
+            results = BH.cancel_batches(["msgbatch_c1"])
+
+        self.assertTrue(results[0]["ok"])
+        entry = next(e for e in BH.read_history()["batches"] if e["id"] == "msgbatch_c1")
+        self.assertEqual(entry["status"], BH.STATUS_CANCELING)
+        state = T._read_batch_file(T.BATCH_STATE_FILE)
+        ids = [b["id"] for b in state.get("batches", [])]
+        self.assertNotIn("msgbatch_c1", ids)
+        self.assertIn("msgbatch_c2", ids)
+
+
+class UsageTests(BatchHistoryTestBase):
+    def test_usage_sums_cache_and_thinking(self):
+        BH.upsert_history_entry(
+            "msgbatch_u",
+            status=BH.STATUS_ENDED,
+            model="claude-sonnet-4-5",
+            custom_ids={"req-000000": "k"},
+        )
+        usage = SimpleNamespace(
+            input_tokens=1000,
+            output_tokens=200,
+            cache_read_input_tokens=500,
+            cache_creation_input_tokens=100,
+            thinking_tokens=50,
+        )
+        msg = SimpleNamespace(content=[SimpleNamespace(text="ok")], usage=usage)
+        row = SimpleNamespace(custom_id="req-000000", result=SimpleNamespace(type="succeeded", message=msg))
+        client = mock.MagicMock()
+        client.messages.batches.retrieve.return_value = SimpleNamespace(processing_status="ended")
+        client.messages.batches.results.return_value = [row]
+
+        with mock.patch.object(BH, "_get_anthropic_client", return_value=client):
+            with mock.patch.object(BH, "getPricingConfig", return_value={"inputAPICost": 3.0, "outputAPICost": 15.0}):
+                info = BH.usage_for_batch("msgbatch_u")
+
+        u = info["usage"]
+        self.assertEqual(u["input_tokens"], 1000)
+        self.assertEqual(u["output_tokens"], 200)
+        self.assertEqual(u["cache_read_input_tokens"], 500)
+        self.assertEqual(u["cache_creation_input_tokens"], 100)
+        self.assertEqual(u["thinking_tokens"], 50)
+        self.assertIsInstance(info["actual_cost"], float)
+        self.assertGreater(info["actual_cost"], 0)
+        entry = BH.read_history()["batches"][0]
+        self.assertEqual(entry["usage"]["thinking_tokens"], 50)
+
+
+class ActivateResumeTests(BatchHistoryTestBase):
+    def test_activate_submitted_restores_state(self):
+        custom_ids = {"req-000000": "k"}
+        BH.upsert_history_entry(
+            "msgbatch_act",
+            status=BH.STATUS_SUBMITTED,
+            custom_ids=custom_ids,
+            model="claude-sonnet-4-5",
+            file_set=["a.json"],
+        )
+        state = BH.activate_for_resume("msgbatch_act")
+        self.assertEqual(state, "submitted")
+        disk = T._read_batch_file(T.BATCH_STATE_FILE)
+        self.assertEqual(disk["batches"][0]["id"], "msgbatch_act")
+        self.assertEqual(disk["batches"][0]["custom_ids"], custom_ids)
+
+
+if __name__ == "__main__":
+    unittest.main()

util/batch_history.py

@@ -0,0 +1,620 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""Durable Anthropic Message Batch history and spend-safe management ops.
+
+Active run files (queue / state / results) still live in util.translation.
+This module keeps an append-only index so batch ids and custom_id maps survive
+fetch and clear, enabling cancel / usage / redownload / resume without
+re-submitting (and re-billing) work.
+"""
+
+from __future__ import annotations
+
+import copy
+import time
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any, Iterable, Optional
+
+from util.translation import (
+    BATCH_LOCK,
+    BATCH_QUEUE_FILE,
+    BATCH_RESULTS_FILE,
+    BATCH_STATE_FILE,
+    _batch_file_lock,
+    _get_anthropic_client,
+    _read_batch_file,
+    _write_batch_file,
+    getPricingConfig,
+)
+
+BATCH_HISTORY_FILE = Path("log/batch_history.json")
+
+# Local lifecycle statuses (not 1:1 with Anthropic processing_status).
+STATUS_QUEUED = "queued"
+STATUS_SUBMITTED = "submitted"
+STATUS_CANCELING = "canceling"
+STATUS_ENDED = "ended"
+STATUS_FETCHED = "fetched"
+STATUS_CONSUMED = "consumed"
+STATUS_CANCELED = "canceled"
+STATUS_ERROR = "error"
+
+TERMINAL_STATUSES = frozenset({
+    STATUS_CONSUMED,
+    STATUS_CANCELED,
+    STATUS_ERROR,
+})
+CANCELABLE_API = frozenset({"in_progress"})
+
+
+def _utc_now() -> str:
+    return datetime.now(timezone.utc).replace(microsecond=0).isoformat()
+
+
+def _empty_history() -> dict:
+    return {"batches": []}
+
+
+def _read_history_unlocked() -> dict:
+    """Read history under an existing batch file lock."""
+    data = _read_batch_file(BATCH_HISTORY_FILE)
+    if not data:
+        return _empty_history()
+    if not isinstance(data.get("batches"), list):
+        print(f"[BATCH] Corrupt history (missing batches list): {BATCH_HISTORY_FILE}", flush=True)
+        return _empty_history()
+    return data
+
+
+def _write_history_unlocked(data: dict) -> None:
+    _write_batch_file(BATCH_HISTORY_FILE, data)
+
+
+def read_history() -> dict:
+    """Return the full history document (copy-safe for callers)."""
+    with _batch_file_lock():
+        return copy.deepcopy(_read_history_unlocked())
+
+
+def _find_entry(history: dict, batch_id: str) -> Optional[dict]:
+    for entry in history.get("batches", []):
+        if entry.get("id") == batch_id:
+            return entry
+    return None
+
+
+def upsert_history_entry(batch_id: str, **fields: Any) -> dict:
+    """Create or update one history row. Returns the updated entry (copy)."""
+    with BATCH_LOCK:
+        with _batch_file_lock():
+            history = _read_history_unlocked()
+            entry = _find_entry(history, batch_id)
+            now = _utc_now()
+            if entry is None:
+                entry = {
+                    "id": batch_id,
+                    "created_at": now,
+                    "updated_at": now,
+                    "status": STATUS_SUBMITTED,
+                    "model": "",
+                    "request_count": 0,
+                    "file_set": [],
+                    "cost_estimate": None,
+                    "custom_ids": {},
+                    "usage": None,
+                    "actual_cost": None,
+                    "notes": "",
+                    "request_counts": None,
+                    "api_status": None,
+                }
+                history.setdefault("batches", []).append(entry)
+            for key, value in fields.items():
+                if value is not None or key in ("notes", "usage", "actual_cost", "cost_estimate", "request_counts"):
+                    entry[key] = value
+            entry["updated_at"] = now
+            _write_history_unlocked(history)
+            return copy.deepcopy(entry)
+
+
+def record_submit(
+    batches: list[dict],
+    *,
+    model: str = "",
+    file_set: Optional[list] = None,
+    cost_estimate: Optional[dict] = None,
+) -> None:
+    """Record newly submitted Anthropic batches into durable history."""
+    file_set = list(file_set or [])
+    for info in batches:
+        bid = info.get("id")
+        if not bid:
+            continue
+        custom_ids = dict(info.get("custom_ids") or {})
+        upsert_history_entry(
+            bid,
+            status=STATUS_SUBMITTED,
+            model=model or (cost_estimate or {}).get("model") or "",
+            request_count=len(custom_ids),
+            file_set=file_set,
+            cost_estimate=copy.deepcopy(cost_estimate) if cost_estimate else None,
+            custom_ids=custom_ids,
+            api_status="in_progress",
+            notes="",
+        )
+
+
+def record_fetch(
+    batch_ids: Iterable[str],
+    *,
+    succeeded: int = 0,
+    errored: int = 0,
+    usage: Optional[dict] = None,
+    actual_cost: Optional[float] = None,
+) -> None:
+    """Mark batches as fetched after results land locally."""
+    note = f"fetched ok={succeeded} err={errored}"
+    for bid in batch_ids:
+        fields = {
+            "status": STATUS_FETCHED,
+            "api_status": "ended",
+            "notes": note,
+        }
+        if usage is not None:
+            fields["usage"] = copy.deepcopy(usage)
+        if actual_cost is not None:
+            fields["actual_cost"] = actual_cost
+        upsert_history_entry(bid, **fields)
+
+
+def mark_batches_consumed(batch_ids: Iterable[str]) -> None:
+    """Mark history rows consumed after a successful consume + clear."""
+    for bid in batch_ids:
+        upsert_history_entry(bid, status=STATUS_CONSUMED, notes="consumed")
+
+
+def mark_batches_canceled(batch_ids: Iterable[str], *, api_status: str = "canceling") -> None:
+    local = STATUS_CANCELED if api_status == "ended" else STATUS_CANCELING
+    for bid in batch_ids:
+        upsert_history_entry(
+            bid,
+            status=local,
+            api_status=api_status,
+            notes=f"cancel requested ({api_status})",
+        )
+
+
+def list_local_batches(*, refresh_live: bool = False) -> list[dict]:
+    """Return history entries newest-first. Optionally refresh non-terminal via API."""
+    history = read_history()
+    entries = list(history.get("batches") or [])
+    if refresh_live:
+        for entry in entries:
+            status = entry.get("status")
+            if status in TERMINAL_STATUSES or status == STATUS_CONSUMED:
+                continue
+            if status in (STATUS_FETCHED, STATUS_QUEUED):
+                continue
+            try:
+                refresh_batch_status(entry["id"])
+            except Exception as exc:
+                upsert_history_entry(entry["id"], notes=f"refresh failed: {exc}")
+        history = read_history()
+        entries = list(history.get("batches") or [])
+    entries.sort(key=lambda e: e.get("created_at") or "", reverse=True)
+    return entries
+
+
+def refresh_batch_status(batch_id: str) -> dict:
+    """Retrieve live Anthropic status and update the history row."""
+    client = _get_anthropic_client()
+    b = client.messages.batches.retrieve(batch_id)
+    api_status = getattr(b, "processing_status", None) or ""
+    counts = getattr(b, "request_counts", None)
+    counts_dict = None
+    if counts is not None:
+        counts_dict = {
+            k: getattr(counts, k, 0) or 0
+            for k in ("processing", "succeeded", "errored", "canceled", "expired")
+        }
+
+    entry = None
+    with _batch_file_lock():
+        history = _read_history_unlocked()
+        entry = _find_entry(history, batch_id)
+        prev_status = (entry or {}).get("status")
+
+    # Map API status onto local lifecycle without clobbering fetched/consumed.
+    local_status = None
+    if prev_status in (STATUS_FETCHED, STATUS_CONSUMED):
+        local_status = prev_status
+    elif api_status == "in_progress":
+        local_status = STATUS_SUBMITTED
+    elif api_status == "canceling":
+        local_status = STATUS_CANCELING
+    elif api_status == "ended":
+        # Ended after cancel vs normal end - inspect counts when available.
+        if counts_dict and (counts_dict.get("canceled") or 0) > 0 and (counts_dict.get("succeeded") or 0) == 0:
+            local_status = STATUS_CANCELED
+        else:
+            local_status = STATUS_ENDED
+
+    fields: dict[str, Any] = {
+        "api_status": api_status,
+        "request_counts": counts_dict,
+    }
+    if local_status:
+        fields["status"] = local_status
+    return upsert_history_entry(batch_id, **fields)
+
+
+def cancel_batches(batch_ids: Iterable[str]) -> list[dict]:
+    """Cancel in-progress batches. Never submits new work.
+
+    Also updates active batch_state so resume does not keep polling canceled ids.
+    """
+    client = _get_anthropic_client()
+    results = []
+    canceled_active = []
+    for bid in batch_ids:
+        try:
+            b = client.messages.batches.retrieve(bid)
+            api_status = getattr(b, "processing_status", "") or ""
+            if api_status not in CANCELABLE_API:
+                results.append({"id": bid, "ok": False, "error": f"not cancelable ({api_status})"})
+                upsert_history_entry(bid, notes=f"cancel skipped: {api_status}", api_status=api_status)
+                continue
+            b = client.messages.batches.cancel(bid)
+            new_status = getattr(b, "processing_status", "canceling") or "canceling"
+            mark_batches_canceled([bid], api_status=new_status)
+            canceled_active.append(bid)
+            results.append({"id": bid, "ok": True, "api_status": new_status})
+        except Exception as exc:
+            results.append({"id": bid, "ok": False, "error": str(exc)})
+            upsert_history_entry(bid, notes=f"cancel failed: {exc}", status=STATUS_ERROR)
+
+    if canceled_active:
+        _remove_active_batch_ids(canceled_active)
+    return results
+
+
+def _remove_active_batch_ids(batch_ids: list[str]) -> None:
+    """Drop canceled ids from active state so poll/resume does not wait on them."""
+    id_set = set(batch_ids)
+    with BATCH_LOCK:
+        with _batch_file_lock():
+            state = _read_batch_file(BATCH_STATE_FILE)
+            batches = [b for b in (state.get("batches") or []) if b.get("id") not in id_set]
+            if not batches:
+                # No active submitted batches left.
+                if state.get("status") == "fetched" or state.get("batch_ids"):
+                    state["batches"] = []
+                    _write_batch_file(BATCH_STATE_FILE, state)
+                else:
+                    try:
+                        if BATCH_STATE_FILE.exists():
+                            BATCH_STATE_FILE.unlink()
+                    except Exception:
+                        pass
+            else:
+                state["batches"] = batches
+                _write_batch_file(BATCH_STATE_FILE, state)
+
+
+def _usage_from_message(u) -> dict:
+    cr = getattr(u, "cache_read_input_tokens", 0) or 0
+    cw = getattr(u, "cache_creation_input_tokens", 0) or 0
+    inp = getattr(u, "input_tokens", 0) or 0
+    out = getattr(u, "output_tokens", 0) or 0
+    # Adaptive thinking may surface under several names depending on SDK version.
+    thinking = (
+        getattr(u, "thinking_tokens", None)
+        or getattr(u, "output_thinking_tokens", None)
+        or 0
+    ) or 0
+    return {
+        "input_tokens": inp,
+        "output_tokens": out,
+        "cache_read_input_tokens": cr,
+        "cache_creation_input_tokens": cw,
+        "thinking_tokens": thinking,
+    }
+
+
+def _price_usage(usage: dict, model: str) -> float:
+    """Price real batch usage with cache multipliers and 50% batch discount."""
+    pricing = getPricingConfig(model)
+    br = pricing["inputAPICost"] / 1_000_000
+    orr = pricing["outputAPICost"] / 1_000_000
+    cr = usage.get("cache_read_input_tokens", 0) or 0
+    cw = usage.get("cache_creation_input_tokens", 0) or 0
+    inp = usage.get("input_tokens", 0) or 0
+    out = usage.get("output_tokens", 0) or 0
+    thinking = usage.get("thinking_tokens", 0) or 0
+    # Thinking tokens are billed as output on Anthropic.
+    raw = cr * br * 0.10 + cw * br * 2.00 + inp * br + (out + thinking) * orr
+    return raw * 0.50
+
+
+def _sum_usage_from_results(client, batch_id: str, custom_ids: dict) -> tuple[dict, int, int]:
+    """Stream batch results and sum usage. Returns (usage, succeeded, errored)."""
+    totals = {
+        "input_tokens": 0,
+        "output_tokens": 0,
+        "cache_read_input_tokens": 0,
+        "cache_creation_input_tokens": 0,
+        "thinking_tokens": 0,
+    }
+    succeeded = 0
+    errored = 0
+    for r in client.messages.batches.results(batch_id):
+        res = r.result
+        if res.type != "succeeded":
+            errored += 1
+            continue
+        succeeded += 1
+        part = _usage_from_message(res.message.usage)
+        for k, v in part.items():
+            totals[k] = totals.get(k, 0) + (v or 0)
+    # custom_ids unused for summing but kept for API symmetry / future filters
+    _ = custom_ids
+    return totals, succeeded, errored
+
+
+def usage_for_batch(batch_id: str, model: Optional[str] = None) -> dict:
+    """Sum real billed tokens from Anthropic results and persist onto history."""
+    history = read_history()
+    entry = _find_entry(history, batch_id)
+    if entry is None:
+        raise ValueError(f"Unknown batch id (not in local history): {batch_id}")
+
+    client = _get_anthropic_client()
+    # Ensure batch has ended before streaming results.
+    b = client.messages.batches.retrieve(batch_id)
+    api_status = getattr(b, "processing_status", "") or ""
+    if api_status != "ended":
+        raise ValueError(f"Batch {batch_id} is not ended (status={api_status})")
+
+    custom_ids = dict(entry.get("custom_ids") or {})
+    usage, succeeded, errored = _sum_usage_from_results(client, batch_id, custom_ids)
+    use_model = model or entry.get("model") or ""
+    cost = _price_usage(usage, use_model) if use_model else None
+    updated = upsert_history_entry(
+        batch_id,
+        usage=usage,
+        actual_cost=cost,
+        api_status="ended",
+        notes=f"usage ok={succeeded} err={errored}",
+    )
+    return {
+        "id": batch_id,
+        "usage": usage,
+        "actual_cost": cost,
+        "succeeded": succeeded,
+        "errored": errored,
+        "model": use_model,
+        "entry": updated,
+    }
+
+
+def _result_entry_from_message(msg) -> dict:
+    text = "".join(getattr(b, "text", "") or "" for b in msg.content)
+    u = msg.usage
+    part = _usage_from_message(u)
+    cr = part["cache_read_input_tokens"]
+    cw = part["cache_creation_input_tokens"]
+    inp = part["input_tokens"]
+    out = part["output_tokens"]
+    entry = {
+        "text": text,
+        "prompt_tokens": inp + cr + cw,
+        "completion_tokens": out,
+        "cache_read_input_tokens": cr,
+        "cache_creation_input_tokens": cw,
+    }
+    if part.get("thinking_tokens"):
+        entry["thinking_tokens"] = part["thinking_tokens"]
+    return entry
+
+
+def download_batch_results(
+    batch_id: str,
+    custom_ids: dict,
+    *,
+    client=None,
+) -> tuple[dict, list, dict]:
+    """Download one ended batch into a cache-key -> result map.
+
+    Returns (results, errored_list, usage_totals).
+    """
+    client = client or _get_anthropic_client()
+    results, errored = {}, []
+    usage_totals = {
+        "input_tokens": 0,
+        "output_tokens": 0,
+        "cache_read_input_tokens": 0,
+        "cache_creation_input_tokens": 0,
+        "thinking_tokens": 0,
+    }
+    for r in client.messages.batches.results(batch_id):
+        key = custom_ids.get(r.custom_id)
+        if key is None:
+            continue
+        res = r.result
+        if res.type != "succeeded":
+            detail = res.type
+            err = getattr(res, "error", None)
+            if err is not None:
+                inner = getattr(err, "error", err)
+                detail = (
+                    f"{res.type} | {getattr(inner, 'type', '')}: "
+                    f"{str(getattr(inner, 'message', '') or err)[:200]}"
+                )
+            errored.append((r.custom_id, detail))
+            continue
+        msg = res.message
+        results[key] = _result_entry_from_message(msg)
+        part = _usage_from_message(msg.usage)
+        for k, v in part.items():
+            usage_totals[k] = usage_totals.get(k, 0) + (v or 0)
+    return results, errored, usage_totals
+
+
+def redownload_batch(batch_id: str) -> dict:
+    """Re-fetch results into batch_results.json using stored custom_ids. No re-submit."""
+    history = read_history()
+    entry = _find_entry(history, batch_id)
+    if entry is None:
+        raise ValueError(f"Unknown batch id (not in local history): {batch_id}")
+    custom_ids = dict(entry.get("custom_ids") or {})
+    if not custom_ids:
+        raise ValueError(f"Batch {batch_id} has no stored custom_ids - cannot redownload")
+
+    client = _get_anthropic_client()
+    b = client.messages.batches.retrieve(batch_id)
+    api_status = getattr(b, "processing_status", "") or ""
+    if api_status != "ended":
+        raise ValueError(f"Batch {batch_id} is not ended (status={api_status})")
+
+    results, errored, usage = download_batch_results(batch_id, custom_ids, client=client)
+    model = entry.get("model") or ""
+    cost = _price_usage(usage, model) if model else None
+
+    import util.translation as T
+
+    with T.BATCH_LOCK:
+        with _batch_file_lock():
+            merged = _read_batch_file(BATCH_RESULTS_FILE)
+            merged.update(results)
+            _write_batch_file(BATCH_RESULTS_FILE, merged)
+            # Activate fetched state without re-submit; keep custom_ids in history.
+            _write_batch_file(
+                BATCH_STATE_FILE,
+                {"status": "fetched", "batch_ids": [batch_id], "batches": []},
+            )
+            # Queue is no longer needed for consume.
+            try:
+                if BATCH_QUEUE_FILE.exists():
+                    BATCH_QUEUE_FILE.unlink()
+            except Exception:
+                pass
+        T._batch_results = None
+
+    record_fetch([batch_id], succeeded=len(results), errored=len(errored), usage=usage, actual_cost=cost)
+    return {
+        "id": batch_id,
+        "succeeded": len(results),
+        "errored": len(errored),
+        "errors": errored[:20],
+        "usage": usage,
+        "actual_cost": cost,
+    }
+
+
+def activate_for_resume(batch_id: str) -> str:
+    """Ensure active files match a history entry for Translation-tab resume.
+
+    Returns the batchRunState string to pass as batch_resume_state:
+    'submitted' | 'fetched'.
+    """
+    history = read_history()
+    entry = _find_entry(history, batch_id)
+    if entry is None:
+        raise ValueError(f"Unknown batch id: {batch_id}")
+
+    status = entry.get("status")
+    custom_ids = dict(entry.get("custom_ids") or {})
+
+    if status in (STATUS_FETCHED, STATUS_ENDED, STATUS_CONSUMED):
+        # Prefer local results; redownload if missing.
+        with _batch_file_lock():
+            results = _read_batch_file(BATCH_RESULTS_FILE)
+        if not results:
+            if status == STATUS_CONSUMED:
+                raise ValueError(
+                    f"Batch {batch_id} was already consumed and local results are gone. "
+                    "Use Redownload first."
+                )
+            redownload_batch(batch_id)
+        else:
+            with BATCH_LOCK:
+                with _batch_file_lock():
+                    _write_batch_file(
+                        BATCH_STATE_FILE,
+                        {"status": "fetched", "batch_ids": [batch_id], "batches": []},
+                    )
+        return "fetched"
+
+    if status in (STATUS_SUBMITTED, STATUS_CANCELING, STATUS_ENDED):
+        # Restore active state so poll/fetch can continue.
+        with BATCH_LOCK:
+            with _batch_file_lock():
+                state = _read_batch_file(BATCH_STATE_FILE)
+                batches = list(state.get("batches") or [])
+                existing = {b.get("id") for b in batches}
+                if batch_id not in existing:
+                    batches.append({"id": batch_id, "custom_ids": custom_ids})
+                state = {
+                    "batches": batches,
+                    "submitted_at": state.get("submitted_at") or entry.get("created_at"),
+                    "model": entry.get("model") or state.get("model") or "",
+                    "file_set": entry.get("file_set") or state.get("file_set") or [],
+                    "cost_estimate": entry.get("cost_estimate") or state.get("cost_estimate"),
+                }
+                _write_batch_file(BATCH_STATE_FILE, state)
+        if status == STATUS_ENDED:
+            # Already ended on Anthropic - fetch path via resume submitted → poll sees ended.
+            return "submitted"
+        return "submitted"
+
+    if status == STATUS_CANCELED:
+        raise ValueError(f"Batch {batch_id} was canceled - nothing to resume")
+
+    raise ValueError(f"Batch {batch_id} cannot be resumed from status={status}")
+
+
+def missing_result_count() -> tuple[int, int]:
+    """Return (present, expected) result counts for the active fetched run.
+
+    expected is summed from history request_count for active fetched batch ids.
+    """
+    with _batch_file_lock():
+        results = _read_batch_file(BATCH_RESULTS_FILE)
+        state = _read_batch_file(BATCH_STATE_FILE)
+        history = _read_history_unlocked()
+
+    present = len(results)
+    ids = list(state.get("batch_ids") or [])
+    if not ids:
+        ids = [b.get("id") for b in (state.get("batches") or []) if b.get("id")]
+    expected = 0
+    for bid in ids:
+        entry = _find_entry(history, bid)
+        if entry:
+            expected += int(entry.get("request_count") or len(entry.get("custom_ids") or {}))
+    if not expected:
+        expected = present
+    return present, expected
+
+
+def active_fetched_batch_ids() -> list[str]:
+    """Batch ids associated with the current fetched active run."""
+    with _batch_file_lock():
+        state = _read_batch_file(BATCH_STATE_FILE)
+    ids = list(state.get("batch_ids") or [])
+    if ids:
+        return ids
+    return [b.get("id") for b in (state.get("batches") or []) if b.get("id")]
+
+
+def on_clear_active_files(*, had_results: bool, fetched_ids: Optional[list] = None) -> None:
+    """Called from clearBatchFiles - mark fetched ids consumed, never wipe history."""
+    if not had_results and not fetched_ids:
+        return
+    ids = list(fetched_ids or [])
+    if not ids and had_results:
+        # Fall back: mark any currently-fetched history rows as consumed.
+        history = read_history()
+        ids = [e["id"] for e in history.get("batches", []) if e.get("status") == STATUS_FETCHED]
+    if ids:
+        mark_batches_consumed(ids)

util/translation.py

@@ -756,9 +756,12 @@ def _read_batch_file(path):
         if path.exists():
             with open(path, "r", encoding="utf-8") as f:
                 data = json.load(f)
-                return data if isinstance(data, dict) else {}
-    except Exception:
-        pass
+                if isinstance(data, dict):
+                    return data
+                print(f"[BATCH] Corrupt batch file (not a JSON object): {path}", flush=True)
+    except Exception as exc:
+        # State/history corruption must not look like an empty run - log it.
+        print(f"[BATCH] Failed to read {path}: {exc}", flush=True)
     return {}
 
 
@@ -840,30 +843,54 @@ def pendingBatchRequests():
 
 
 def batchRunState():
-    """'submitted' when a batch is still in flight, 'fetched' when results are
-    waiting to be consumed, else None. Lets an interrupted batch run resume
-    instead of re-collecting and paying for a second submission."""
+    """Resume detector for an interrupted batch run.
+
+    Returns:
+      'submitted' - Anthropic batch(es) in flight (or ended but not fetched)
+      'fetched'   - results on disk waiting for consume
+      'queued'    - collect finished / submit declined; queue still on disk
+      None        - nothing to resume
+    """
     with _batch_file_lock():
-        if _read_batch_file(BATCH_STATE_FILE).get("batches"):
+        state = _read_batch_file(BATCH_STATE_FILE)
+        if state.get("batches"):
             return "submitted"
-        if _read_batch_file(BATCH_RESULTS_FILE):
+        if state.get("status") == "fetched" or _read_batch_file(BATCH_RESULTS_FILE):
             return "fetched"
+        if _read_batch_file(BATCH_QUEUE_FILE):
+            return "queued"
     return None
 
 
 def clearBatchFiles():
-    """Remove queue/state/results left over from any previous batch run."""
+    """Remove active queue/state/results. Never deletes durable batch history."""
     global _batch_results, _batch_queue_pending
     with BATCH_LOCK:
-        _batch_results = None
-        _batch_queue_pending = {}
+        fetched_ids = []
+        had_results = False
         with _batch_file_lock():
+            state = _read_batch_file(BATCH_STATE_FILE)
+            had_results = bool(_read_batch_file(BATCH_RESULTS_FILE)) or state.get("status") == "fetched"
+            if had_results:
+                fetched_ids = list(state.get("batch_ids") or [])
+                if not fetched_ids:
+                    fetched_ids = [b.get("id") for b in (state.get("batches") or []) if b.get("id")]
             for path in (BATCH_QUEUE_FILE, BATCH_STATE_FILE, BATCH_RESULTS_FILE):
                 try:
                     if path.exists():
                         path.unlink()
                 except Exception:
                     pass
+        _batch_results = None
+        _batch_queue_pending = {}
+    # Only mark history consumed when clearing after a successful fetch/consume,
+    # never when discarding a still-submitted or queued run.
+    if had_results:
+        try:
+            from util.batch_history import on_clear_active_files
+            on_clear_active_files(had_results=True, fetched_ids=fetched_ids or None)
+        except Exception as exc:
+            print(f"[BATCH] history consume mark failed: {exc}", flush=True)
 
 
 def _get_anthropic_client():
@@ -968,11 +995,13 @@ def estimateBatchCost(model=None):
     }
 
 
-def submitTranslationBatches():
+def submitTranslationBatches(file_set=None, cost_estimate=None):
     """Submit the queued requests to the Anthropic Message Batches API.
 
     Splits at the API limits and saves the custom_id -> cache-key mapping so
-    fetchTranslationBatches can route results back. Returns the batch ids."""
+    fetchTranslationBatches can route results back. Also appends durable
+    history entries (custom_ids survive later fetch/clear). Returns the batch ids.
+    """
     flush_batch_queue()
     with _batch_file_lock():
         queue = _read_batch_file(BATCH_QUEUE_FILE)
@@ -983,6 +1012,16 @@ def submitTranslationBatches():
     client = _get_anthropic_client()
     batches = []
     requests, id_map, size = [], {}, 0
+    models = set()
+    for entry in queue.values():
+        m = (entry.get("params") or {}).get("model")
+        if m:
+            models.add(m)
+    model = (
+        (cost_estimate or {}).get("model")
+        or next(iter(models), None)
+        or os.getenv("model", "")
+    )
 
     def _submit():
         nonlocal requests, id_map, size
@@ -1003,48 +1042,134 @@ def submitTranslationBatches():
             _submit()
     _submit()
 
+    submitted_at = time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
+    state_doc = {
+        "batches": batches,
+        "submitted_at": submitted_at,
+        "model": model,
+        "file_set": list(file_set or []),
+        "cost_estimate": cost_estimate,
+        "request_count": sum(len(b.get("custom_ids") or {}) for b in batches),
+    }
     with BATCH_LOCK:
         with _batch_file_lock():
-            _write_batch_file(BATCH_STATE_FILE, {"batches": batches})
+            _write_batch_file(BATCH_STATE_FILE, state_doc)
+    try:
+        from util.batch_history import record_submit
+        record_submit(
+            batches,
+            model=model,
+            file_set=file_set,
+            cost_estimate=cost_estimate,
+        )
+    except Exception as exc:
+        print(f"[BATCH] history record_submit failed: {exc}", flush=True)
     return [b["id"] for b in batches]
 
 
 def checkTranslationBatches():
-    """Print the processing status of submitted batches. True when all ended."""
+    """Print the processing status of submitted batches. True when all ended.
+
+    Also returns a structured status list as the second value when called as
+    ``ended, statuses = checkTranslationBatchStatuses()`` - prefer that helper
+    for UI work. Kept for CLI/print compatibility.
+    """
+    ended, _statuses = checkTranslationBatchStatuses(print_status=True)
+    return ended
+
+
+def checkTranslationBatchStatuses(print_status=True):
+    """Return (all_ended, statuses) for submitted batches.
+
+    Each status dict: id, api_status, counts{processing,succeeded,errored,canceled,expired}.
+    """
     with _batch_file_lock():
         state = _read_batch_file(BATCH_STATE_FILE)
     if not state.get("batches"):
-        print("[BATCH] No submitted batches — submit the queue first.", flush=True)
-        return False
+        if print_status:
+            print("[BATCH] No submitted batches - submit the queue first.", flush=True)
+        return False, []
     client = _get_anthropic_client()
     all_ended = True
+    statuses = []
     for info in state["batches"]:
-        b = client.messages.batches.retrieve(info["id"])
-        counts = getattr(b, "request_counts", None)
-        suffix = f"  counts: {counts}" if counts else ""
-        print(f"[BATCH] {time.strftime('%H:%M:%S')}  {b.id}: {b.processing_status}{suffix}", flush=True)
-        if b.processing_status != "ended":
+        bid = info["id"]
+        b = client.messages.batches.retrieve(bid)
+        api_status = getattr(b, "processing_status", "") or ""
+        counts_obj = getattr(b, "request_counts", None)
+        counts = {
+            k: int(getattr(counts_obj, k, 0) or 0)
+            for k in ("processing", "succeeded", "errored", "canceled", "expired")
+        } if counts_obj is not None else {
+            "processing": 0, "succeeded": 0, "errored": 0, "canceled": 0, "expired": 0,
+        }
+        statuses.append({
+            "id": bid,
+            "api_status": api_status,
+            "counts": counts,
+            "request_count": len(info.get("custom_ids") or {}),
+        })
+        if print_status:
+            parts = [f"{k[:4]}={v}" for k, v in counts.items() if v]
+            suffix = ("  " + " ".join(parts)) if parts else ""
+            print(
+                f"[BATCH] {time.strftime('%H:%M:%S')}  {bid}: {api_status}{suffix}",
+                flush=True,
+            )
+        if api_status != "ended":
             all_ended = False
-    return all_ended
+    return all_ended, statuses
 
 
-def fetchTranslationBatches():
+def fetchTranslationBatches(batches=None):
     """Download finished batch results into the local results store.
 
     Successes are stored keyed by the payload cache key for the consume pass;
     errored/expired requests are reported and simply fall back to the live API
-    during consume. Returns (succeeded, errored) counts."""
+    during consume. Durable history retains custom_ids for later redownload.
+
+    batches: optional list of {id, custom_ids} (defaults to active batch_state).
+    Returns (succeeded, errored) counts.
+    """
     global _batch_results
     with _batch_file_lock():
         state = _read_batch_file(BATCH_STATE_FILE)
-    if not state.get("batches"):
-        print("[BATCH] No submitted batches — nothing to fetch.", flush=True)
+    batch_list = batches if batches is not None else (state.get("batches") or [])
+    if not batch_list:
+        print("[BATCH] No submitted batches - nothing to fetch.", flush=True)
         return 0, 0
+
+    try:
+        from util.batch_history import download_batch_results, record_fetch, _price_usage
+    except Exception:
+        download_batch_results = None
+        record_fetch = None
+        _price_usage = None
+
     client = _get_anthropic_client()
     results, errored = {}, []
-    for info in state["batches"]:
+    usage_totals = {
+        "input_tokens": 0,
+        "output_tokens": 0,
+        "cache_read_input_tokens": 0,
+        "cache_creation_input_tokens": 0,
+        "thinking_tokens": 0,
+    }
+    batch_ids = []
+    for info in batch_list:
+        bid = info.get("id")
+        if bid:
+            batch_ids.append(bid)
         id_map = info.get("custom_ids", {})
-        for r in client.messages.batches.results(info["id"]):
+        if download_batch_results is not None:
+            part, err_part, usage_part = download_batch_results(bid, id_map, client=client)
+            results.update(part)
+            errored.extend(err_part)
+            for k, v in usage_part.items():
+                usage_totals[k] = usage_totals.get(k, 0) + (v or 0)
+            continue
+        # Fallback if batch_history import failed - preserve prior behaviour.
+        for r in client.messages.batches.results(bid):
             key = id_map.get(r.custom_id)
             if key is None:
                 continue
@@ -1066,26 +1191,61 @@ def fetchTranslationBatches():
             out = getattr(u, "output_tokens", 0) or 0
             results[key] = {
                 "text": text,
-                # prompt_tokens matches _AnthropicCompat: total incl. cache fields.
                 "prompt_tokens": inp + cr + cw,
                 "completion_tokens": out,
                 "cache_read_input_tokens": cr,
                 "cache_creation_input_tokens": cw,
             }
+
+    model = state.get("model") or os.getenv("model", "")
+    actual_cost = None
+    if _price_usage is not None and model:
+        try:
+            actual_cost = _price_usage(usage_totals, model)
+        except Exception:
+            actual_cost = None
+
     with BATCH_LOCK:
         with _batch_file_lock():
             merged = _read_batch_file(BATCH_RESULTS_FILE)
             merged.update(results)
             _write_batch_file(BATCH_RESULTS_FILE, merged)
-            # Queue and state are consumed; only the results store remains.
-            for path in (BATCH_QUEUE_FILE, BATCH_STATE_FILE):
-                try:
-                    if path.exists():
-                        path.unlink()
-                except Exception:
-                    pass
+            # Drop the queue; keep a lightweight fetched marker (ids for consume→history).
+            # custom_ids stay in durable history - do not destroy recovery maps.
+            try:
+                if BATCH_QUEUE_FILE.exists():
+                    BATCH_QUEUE_FILE.unlink()
+            except Exception:
+                pass
+            _write_batch_file(
+                BATCH_STATE_FILE,
+                {
+                    "status": "fetched",
+                    "batch_ids": batch_ids,
+                    "batches": [],
+                    "model": model,
+                    "file_set": state.get("file_set") or [],
+                    "cost_estimate": state.get("cost_estimate"),
+                    "fetched_at": time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime()),
+                },
+            )
         _batch_results = None
+
+    if record_fetch is not None:
+        try:
+            record_fetch(
+                batch_ids,
+                succeeded=len(results),
+                errored=len(errored),
+                usage=usage_totals,
+                actual_cost=actual_cost,
+            )
+        except Exception as exc:
+            print(f"[BATCH] history record_fetch failed: {exc}", flush=True)
+
     print(f"[BATCH] fetched {len(results)} results ({len(errored)} errored).", flush=True)
+    if actual_cost is not None:
+        print(f"[BATCH] actual batch usage cost (est.): ${actual_cost:.4f}", flush=True)
     for cid, why in errored[:20]:
         print(f"[BATCH]   ! {cid}: {why}", flush=True)
     if len(errored) > 20:
@@ -2626,6 +2786,29 @@ def translateAI(text, history, config, filename=None, pbar=None, lock=None, mism
                 with lock:
                     pbar.update(len(tItem) if isinstance(tItem, list) else 1)
 
+            # Consume pass: still record what was applied. Cache hits used to
+            # skip the log entirely, which left the GUI Translation Log empty
+            # after a resume even though translated/ was written.
+            if batch_phase == "consume" and not config.estimateMode:
+                try:
+                    if isinstance(cached_result, list):
+                        out_payload = {
+                            f"Line{i+1}": string for i, string in enumerate(cached_result)
+                        }
+                        formatted_output = json.dumps(out_payload, indent=4, ensure_ascii=False)
+                    else:
+                        formatted_output = json.dumps(
+                            {"Line1": cached_result}, indent=4, ensure_ascii=False
+                        )
+                    Path(config.logFilePath).parent.mkdir(parents=True, exist_ok=True)
+                    with open(config.logFilePath, "a", encoding="utf-8") as logFile:
+                        logFile.write("[CACHE] Applied cached translation (no new API call)\n")
+                        logFile.write(f"Input:\n{subbedT}\n")
+                        logFile.write(f"Output:\n{formatted_output}\n")
+                        logFile.flush()
+                except Exception:
+                    pass
+
             continue
 
         # Create context — static_system is the stable prompt.txt content;
@@ -2701,6 +2884,7 @@ def translateAI(text, history, config, filename=None, pbar=None, lock=None, mism
         max_retries = 2  # 1 initial attempt + 2 retries
         final_translations = None
         last_raw_translation = ""
+        from_batch = False
         numLines = len(clean_tItem) if isinstance(tItem, list) else 1
 
         for attempt in range(max_retries + 1):
@@ -2726,7 +2910,7 @@ def translateAI(text, history, config, filename=None, pbar=None, lock=None, mism
                 if pbar:
                     pbar.write(f"Retrying translation... (Attempt {attempt + 1}/{max_retries + 1})")
 
-            # Translate — the consume pass tries the fetched batch result first;
+            # Translate - the consume pass tries the fetched batch result first;
             # a missing or invalid result falls through to the live API.
             from_batch = False
             if batch_phase == "consume" and attempt == 0:
@@ -2944,6 +3128,8 @@ def translateAI(text, history, config, filename=None, pbar=None, lock=None, mism
             # Only open and write to log file when we have something to log
             try:
                 with open(config.logFilePath, "a", encoding="utf-8") as logFile:
+                    if from_batch:
+                        logFile.write("[BATCH] Applied Anthropic batch result\n")
                     logFile.write(f"Input:\n{subbedT}\n")
                     logFile.write(f"Output:\n{formatted_output}\n")
                     logFile.flush()  # Ensure data is written to disk immediately