transcript-viewer 0.5.0__tar.gz → 0.6.1__tar.gz

transcript_viewer-0.5.0/README.md → transcript_viewer-0.6.1/PKG-INFO

@@ -1,17 +1,40 @@
+Metadata-Version: 2.5
+Name: transcript-viewer
+Version: 0.6.1
+Summary: Browse agent transcripts in a local, dependency-free web viewer.
+License: MIT
+Requires-Python: >=3.12
+Requires-Dist: atif-make>=0.5.0
+Provides-Extra: ai
+Requires-Dist: anthropic>=0.40; extra == 'ai'
+Provides-Extra: parquet
+Requires-Dist: atif-make[parquet]>=0.5.0; extra == 'parquet'
+Description-Content-Type: text/markdown
+
 # transcript-viewer
 
-Browse [ATIF v1.7](https://github.com/harbor-framework/harbor/blob/main/rfcs/0001-trajectory-format.md)
-trajectories in a local web viewer.
+Browse agent transcripts in a local web viewer — what Claude Code, Codex and
+Copilot write as they work.
 
-Conversion lives in [`atif-make`](https://github.com/jammastergirish/atif-make); this package depends on it and
-adds only the browser interface.
+Logs are converted on the way in to
+[ATIF v1.7](https://github.com/harbor-framework/harbor/blob/main/rfcs/0001-trajectory-format.md)
+by [`atif-make`](https://github.com/jammastergirish/atif-make), which this package depends on and which does
+all of the reading. What is added here is only the browser interface, so a
+format this cannot open is a parser missing from `atif-make` rather than
+anything to change in the viewer.
 
 ## Install
 
 ```sh
-uv tool install transcript-viewer        # pulls atif-make automatically
+uv tool install transcript-viewer                    # pulls atif-make automatically
+uv tool install "transcript-viewer[parquet]"         # + datasets published as Parquet
+uv tool install "transcript-viewer[ai,parquet]"      # + the optional Claude features
 ```
 
+The extras are only needed for what they name: `parquet` for a dataset that
+ships as Parquet rather than JSON, `ai` for the summarise and ask features. The
+viewer works without either.
+
 ```sh
 transcript-viewer                        # the library, empty on a first run
 transcript-viewer path/to/session.jsonl  # one log
@@ -137,6 +160,14 @@ Files are deleted only where they are the viewer's own copy under
 `~/.transcript-viewer/opened/`; a session found on this machine, or downloaded into a folder
 of yours, keeps its file.
 
+Some datasets ship as a single file holding many runs — ATBench publishes a
+thousand agent trajectories as one JSON array, METR's as Parquet shards
+(install with `uv tool install "transcript-viewer[parquet]"` to read those). Those are opened rather than
+listed: the file is split into a transcript apiece and each is indexed on its
+own, so a download of one file becomes a thousand sessions you can read. The
+pieces are kept beside what they came from, so removing the folder removes them
+too.
+
 This tool used to be called `atif-view` and kept all of that under `~/.atif`.
 If that directory is still there, the first run moves it to
 `~/.transcript-viewer` — index, annotations and stored keys together — and says
@@ -368,7 +399,7 @@ get.
 ```
 src/transcript_viewer/
   page.html    the whole front end — 2,232 lines of HTML, CSS and JavaScript
-  viewer.py    the HTTP server: fifteen endpoints over the page and the library
+  viewer.py    the HTTP server: sixteen endpoints over the page and the library
   corpus.py    the index — what is on this machine, and where it came from
   library.py   what you decide about a session: title, tags, stars, summaries
   fetch.py     bringing transcripts in from Hugging Face, GitHub, S3 or a link
@@ -376,13 +407,18 @@ src/transcript_viewer/
   config.py    settings and tokens
   store.py     small JSON files under ~/.transcript-viewer, written so a crash cannot truncate them
   cli.py       the command line
+  page.html    the whole interface: one page, no build step, no dependencies
 ```
 
 The page is a file rather than a string inside `viewer.py`, which is where it
 used to live. Two thirds of that module was CSS and JavaScript typed as though
 it were Python — a template expression once shipped inside static HTML and
 rendered its own source, which is harder to miss in a file that knows what it
-is. It is read once at import and served from memory, and the wheel carries it.
+is. It is re-read whenever it changes, so editing the interface and reloading
+shows the edit — twice it did not, and read as an edit that had failed. The
+cost is a stat per page load, and none on what the page then calls; for an
+installed copy the file never changes and the stat always says so. The wheel
+carries it.
 
 ## Tests
 
@@ -398,7 +434,8 @@ uv run --with-editable ../atif-make pytest
 ```
 
 Add `--extra ai` to either command to exercise the AI paths against a real SDK;
-the tests stub the model call, so this never contacts the API.
+the tests stub the model call, so this never contacts the API. `--extra parquet`
+does the same for the Parquet path, which is otherwise skipped.
 
 The suite starts a real server on an ephemeral port and exercises the endpoints,
 including that it binds loopback only. It is isolated from your own library,

transcript_viewer-0.5.0/PKG-INFO → transcript_viewer-0.6.1/README.md

@@ -1,28 +1,27 @@
-Metadata-Version: 2.5
-Name: transcript-viewer
-Version: 0.5.0
-Summary: Browse agent transcripts in a local, dependency-free web viewer.
-License: MIT
-Requires-Python: >=3.12
-Requires-Dist: atif-make>=0.2.0
-Provides-Extra: ai
-Requires-Dist: anthropic>=0.40; extra == 'ai'
-Description-Content-Type: text/markdown
-
 # transcript-viewer
 
-Browse [ATIF v1.7](https://github.com/harbor-framework/harbor/blob/main/rfcs/0001-trajectory-format.md)
-trajectories in a local web viewer.
+Browse agent transcripts in a local web viewer — what Claude Code, Codex and
+Copilot write as they work.
 
-Conversion lives in [`atif-make`](https://github.com/jammastergirish/atif-make); this package depends on it and
-adds only the browser interface.
+Logs are converted on the way in to
+[ATIF v1.7](https://github.com/harbor-framework/harbor/blob/main/rfcs/0001-trajectory-format.md)
+by [`atif-make`](https://github.com/jammastergirish/atif-make), which this package depends on and which does
+all of the reading. What is added here is only the browser interface, so a
+format this cannot open is a parser missing from `atif-make` rather than
+anything to change in the viewer.
 
 ## Install
 
 ```sh
-uv tool install transcript-viewer        # pulls atif-make automatically
+uv tool install transcript-viewer                    # pulls atif-make automatically
+uv tool install "transcript-viewer[parquet]"         # + datasets published as Parquet
+uv tool install "transcript-viewer[ai,parquet]"      # + the optional Claude features
 ```
 
+The extras are only needed for what they name: `parquet` for a dataset that
+ships as Parquet rather than JSON, `ai` for the summarise and ask features. The
+viewer works without either.
+
 ```sh
 transcript-viewer                        # the library, empty on a first run
 transcript-viewer path/to/session.jsonl  # one log
@@ -148,6 +147,14 @@ Files are deleted only where they are the viewer's own copy under
 `~/.transcript-viewer/opened/`; a session found on this machine, or downloaded into a folder
 of yours, keeps its file.
 
+Some datasets ship as a single file holding many runs — ATBench publishes a
+thousand agent trajectories as one JSON array, METR's as Parquet shards
+(install with `uv tool install "transcript-viewer[parquet]"` to read those). Those are opened rather than
+listed: the file is split into a transcript apiece and each is indexed on its
+own, so a download of one file becomes a thousand sessions you can read. The
+pieces are kept beside what they came from, so removing the folder removes them
+too.
+
 This tool used to be called `atif-view` and kept all of that under `~/.atif`.
 If that directory is still there, the first run moves it to
 `~/.transcript-viewer` — index, annotations and stored keys together — and says
@@ -379,7 +386,7 @@ get.
 ```
 src/transcript_viewer/
   page.html    the whole front end — 2,232 lines of HTML, CSS and JavaScript
-  viewer.py    the HTTP server: fifteen endpoints over the page and the library
+  viewer.py    the HTTP server: sixteen endpoints over the page and the library
   corpus.py    the index — what is on this machine, and where it came from
   library.py   what you decide about a session: title, tags, stars, summaries
   fetch.py     bringing transcripts in from Hugging Face, GitHub, S3 or a link
@@ -387,13 +394,18 @@ src/transcript_viewer/
   config.py    settings and tokens
   store.py     small JSON files under ~/.transcript-viewer, written so a crash cannot truncate them
   cli.py       the command line
+  page.html    the whole interface: one page, no build step, no dependencies
 ```
 
 The page is a file rather than a string inside `viewer.py`, which is where it
 used to live. Two thirds of that module was CSS and JavaScript typed as though
 it were Python — a template expression once shipped inside static HTML and
 rendered its own source, which is harder to miss in a file that knows what it
-is. It is read once at import and served from memory, and the wheel carries it.
+is. It is re-read whenever it changes, so editing the interface and reloading
+shows the edit — twice it did not, and read as an edit that had failed. The
+cost is a stat per page load, and none on what the page then calls; for an
+installed copy the file never changes and the stat always says so. The wheel
+carries it.
 
 ## Tests
 
@@ -409,7 +421,8 @@ uv run --with-editable ../atif-make pytest
 ```
 
 Add `--extra ai` to either command to exercise the AI paths against a real SDK;
-the tests stub the model call, so this never contacts the API.
+the tests stub the model call, so this never contacts the API. `--extra parquet`
+does the same for the Parquet path, which is otherwise skipped.
 
 The suite starts a real server on an ephemeral port and exercises the endpoints,
 including that it binds loopback only. It is isolated from your own library,

{transcript_viewer-0.5.0 → transcript_viewer-0.6.1}/pyproject.toml

@@ -1,19 +1,23 @@
 [project]
 name = "transcript-viewer"
-version = "0.5.0"
+version = "0.6.1"
 description = "Browse agent transcripts in a local, dependency-free web viewer."
 readme = "README.md"
 requires-python = ">=3.12"
 license = { text = "MIT" }
 # The only dependency is the converter. The viewer itself is stdlib http.server
 # plus one self-contained HTML page.
-# 0.2.0 is the floor, not politeness: Entry gained the content key this
-# viewer addresses sessions by, so an older atif-make crashes on the index.
-dependencies = ["atif-make>=0.2.0"]
+# The floor is not politeness: this viewer imports atif_make.container, which
+# is how a benchmark published as one JSON array of a thousand runs becomes a
+# thousand sessions. An older atif-make fails to import at all.
+dependencies = ["atif-make>=0.5.0"]
 
 [project.optional-dependencies]
 # Claude-backed explanations are opt-in, so the default install stays
 # dependency-free: pip install "transcript-viewer[ai]"
+# Datasets published as Parquet — METR's transcripts are the one so far. The
+# reader lives in atif-make, which only needs it to split a shard.
+parquet = ["atif-make[parquet]>=0.5.0"]
 ai = ["anthropic>=0.40"]
 
 [project.scripts]
@@ -22,17 +26,6 @@ transcript-viewer = "transcript_viewer.cli:main"
 [dependency-groups]
 dev = ["pytest>=8.0"]
 
-# The two packages move together, so track atif-make's `main` rather than a
-# release. Once atif-make is on PyPI you can delete this block entirely: the
-# `dependencies` entry above already resolves from the index.
-#
-# uv still records an exact commit in uv.lock, which is what keeps a given
-# checkout reproducible. To pick up newer atif-make commits, re-resolve:
-#
-#     uv lock --upgrade-package atif-make
-#
-[tool.uv.sources]
-atif-make = { git = "https://github.com/jammastergirish/atif-make", branch = "main" }
 
 [build-system]
 requires = ["hatchling"]

{transcript_viewer-0.5.0 → transcript_viewer-0.6.1}/src/transcript_viewer/cli.py

@@ -7,7 +7,7 @@ import sys
 from pathlib import Path
 
 from . import corpus, store
-from atif_make.archive import is_archive
+from atif_make.container import is_container
 
 from .viewer import serve
 
@@ -39,10 +39,11 @@ def cmd_view(args: argparse.Namespace) -> int:
         if not path.exists():
             print(f"transcript-viewer: no such path: {path}", file=sys.stderr)
             return 2
-        # A directory or archive holds many sessions; a plain file holds one.
+        # A directory or a container holds many sessions; a plain file
+        # holds one.
         entries = (
             corpus.scan([path])
-            if path.is_dir() or is_archive(path)
+            if path.is_dir() or is_container(path)
             else _single_entry(path)
         )
         # An explicit path that holds nothing is a mistake worth reporting; an

{transcript_viewer-0.5.0 → transcript_viewer-0.6.1}/src/transcript_viewer/corpus.py

@@ -16,7 +16,7 @@ from dataclasses import fields as dataclass_fields
 from datetime import datetime, timezone
 from pathlib import Path
 
-from atif_make.archive import extract, is_archive
+from atif_make.container import is_container, open_container
 
 from . import store
 from atif_make.convert import AGENTS
@@ -32,6 +32,13 @@ INDEX_PATH = store.ROOT / "index.json"
 # whoever asks, so a plain rescan classifies them correctly on its own.
 OPENED_ROOT = store.ROOT / "opened"
 
+# Where a collection's pieces are kept. A benchmark published as one JSON array
+# becomes a transcript per file, and those files are what the index records, so
+# they need somewhere that outlives the process that made them. Keyed by the
+# source's content, so re-indexing the same download reuses the split and a
+# changed file gets a new one.
+SPLIT_ROOT = store.ROOT / "split"
+
 # A first line shorter than this is structural rather than identifying — a
 # pretty-printed JSON document opens with a bare "{".
 
@@ -251,6 +258,20 @@ def merge(existing: list[Entry], new: list[Entry]) -> list[Entry]:
     return sorted(out, key=lambda e: e.modified, reverse=True)
 
 
+def _split_home(path: Path) -> Path:
+    """Where this file's pieces belong, if it turns out to hold many.
+
+    A download is unpacked beside itself on arrival, so prefer that: the pieces
+    are already there, they sit with the thing they came from, and they go when
+    the folder does. Anything else — a file being scanned where writing beside
+    it would be rude — goes under the viewer's own directory instead.
+    """
+    beside = path.with_suffix("")
+    if beside.is_dir() and any(beside.iterdir()):
+        return beside
+    return SPLIT_ROOT / content_key(path)[:16]
+
+
 def scan(roots: list[Path] | None = None, origin: str = "scanned") -> list[Entry]:
     """Find every convertible log under ``roots``. A root may be a single file."""
     entries: list[Entry] = []
@@ -260,10 +281,11 @@ def scan(roots: list[Path] | None = None, origin: str = "scanned") -> list[Entry
         if not root.exists():
             continue
         if root.is_file():
-            # A zip or tarball is a container of logs, not a log.
-            if is_archive(root):
+            # A zip, a tarball, or a benchmark published as one JSON array of a
+            # thousand runs: a container of logs, not a log.
+            if is_container(root):
                 try:
-                    root = extract(root)
+                    root = open_container(root, _split_home(root))
                 except (ValueError, OSError):
                     continue
             else:
@@ -275,11 +297,11 @@ def scan(roots: list[Path] | None = None, origin: str = "scanned") -> list[Entry
         # bucket of agent runs is mostly zips — so look inside them too. Each is
         # unpacked once and its contents scanned in place.
         roots_here = [root]
-        for archive in sorted(root.rglob("*")):
-            if not archive.is_file() or not is_archive(archive):
+        for item in sorted(root.rglob("*")):
+            if not item.is_file() or not is_container(item):
                 continue
             try:
-                roots_here.append(extract(archive))
+                roots_here.append(open_container(item, _split_home(item)))
             except (ValueError, OSError):
                 continue
 
@@ -289,11 +311,18 @@ def scan(roots: list[Path] | None = None, origin: str = "scanned") -> list[Entry
                 sorted(where.rglob("*.jsonl"))
                 + sorted(where.rglob("*.har"))
                 + sorted(where.rglob("*.json"))
+                # Not a log, but a container of them, opened above.
+                + sorted(where.rglob("*.parquet"))
             )
         for path in candidates:
             resolved = path.resolve()
             if resolved in seen:
                 continue
+            # A container was opened above and its contents are already in this
+            # list. Indexing it as well would offer a session that cannot be
+            # opened, since it holds a thousand transcripts rather than one.
+            if is_container(path):
+                continue
             # Subagent traces are reached through their parent, not indexed alone.
             if path.parent.name == "subagents":
                 continue

{transcript_viewer-0.5.0 → transcript_viewer-0.6.1}/src/transcript_viewer/fetch.py

@@ -59,8 +59,11 @@ HOSTS = {
 # session than loose JSONL, so excluding archives made S3 useless for exactly
 # the case it was added for.
 LOGS = (".jsonl", ".json", ".har")
+# A published dataset is a container of runs rather than a log, but it is still
+# a thing worth fetching: METR's transcripts come as Parquet shards.
+DATASETS = (".parquet",)
 ARCHIVES = (".zip", ".tgz", ".tar", ".tar.gz", ".tar.bz2", ".tar.xz", ".gz")
-SUFFIXES = LOGS + ARCHIVES
+SUFFIXES = LOGS + DATASETS + ARCHIVES
 
 # Downloads land beside where the viewer was launched, so they are visible and
 # usable by other tools rather than buried in a dot-directory.

{transcript_viewer-0.5.0 → transcript_viewer-0.6.1}/src/transcript_viewer/library.py

@@ -51,17 +51,49 @@ def _now() -> str:
     return datetime.now(UTC).isoformat(timespec="seconds").replace("+00:00", "Z")
 
 
-def load(path: Path | None = None) -> dict[str, dict]:
-    """Every annotation, by key. A missing or damaged file reads as empty."""
-    data = store.read_json(path or LIBRARY_PATH)
+# What was parsed, and the state of the file it came from. The library is read
+# far more often than it is written — a page load asks about every session in
+# it — and it grows with the corpus, so parsing a megabyte of JSON per question
+# is the difference between a page appearing and a page taking four seconds.
+_parsed: tuple[int, int, dict[str, dict]] | None = None
+
+
+def _read(path: Path) -> dict[str, dict]:
+    data = store.read_json(path)
     entries = data.get("entries")
     if not isinstance(entries, dict):
         return {}
     return {k: v for k, v in entries.items() if isinstance(v, dict)}
 
 
+def load(path: Path | None = None) -> dict[str, dict]:
+    """Every annotation, by key. A missing or damaged file reads as empty.
+
+    The real library is cached against its own timestamp and size, so a change
+    made by anything — this process or another — is picked up on the next read.
+    A caller gets its own outer dict, since `remove` pops from what it is given
+    and callers should not be able to edit the cache by accident.
+    """
+    global _parsed
+    path = path or LIBRARY_PATH
+    if path != LIBRARY_PATH:
+        return _read(path)
+
+    try:
+        info = path.stat()
+        stamp = (info.st_mtime_ns, info.st_size)
+    except OSError:
+        return _read(path)
+
+    if _parsed is None or (_parsed[0], _parsed[1]) != stamp:
+        _parsed = (stamp[0], stamp[1], _read(path))
+    return dict(_parsed[2])
+
+
 def save(entries: dict[str, dict], path: Path | None = None) -> None:
     """Write the whole library atomically."""
+    global _parsed
+    _parsed = None  # the next read re-stamps against the file just written
     store.write_json(path or LIBRARY_PATH, {"version": VERSION, "entries": entries})
 
 

{transcript_viewer-0.5.0 → transcript_viewer-0.6.1}/src/transcript_viewer/page.html

@@ -430,8 +430,33 @@ pre.json{line-height:1.45}
 .more button{border:1px solid var(--line);background:var(--panel);color:var(--dim);border-radius:999px;
   padding:8px 18px;font:inherit;font-size:12.5px;cursor:pointer}
 .more button:hover{background:var(--sunk);color:var(--ink)}
+
+/* Shown only if opening takes long enough to notice — a library of a few
+   thousand sessions is a couple of megabytes, and a blank window while that
+   arrives reads as broken rather than busy. */
+#boot{position:fixed;inset:0;z-index:60;display:flex;flex-direction:column;
+  align-items:center;justify-content:center;gap:14px;background:var(--bg)}
+#boot[hidden]{display:none}
+#bootbar{width:min(320px,52vw);height:3px;border-radius:3px;background:var(--line);
+  overflow:hidden}
+#bootbar i{display:block;height:100%;width:0;border-radius:3px;background:var(--accent);
+  transition:width .18s ease-out}
+/* Nothing to measure yet: sweep rather than sit at zero, which reads as stuck. */
+#bootbar.waiting i{width:35%;animation:sweep 1.1s ease-in-out infinite}
+@keyframes sweep{0%{transform:translateX(-100%)}100%{transform:translateX(320%)}}
+#boottext{margin:0;font:12px/1.5 var(--font-mono);color:var(--muted);
+  letter-spacing:.02em;text-align:center;padding:0 20px}
+@media (prefers-reduced-motion:reduce){
+  #bootbar i{transition:none}
+  #bootbar.waiting i{animation:none;width:100%;opacity:.5}
+}
 </style>
 
+<div id="boot" hidden>
+  <div id="bootbar"><i></i></div>
+  <p id="boottext">Opening your library…</p>
+</div>
+
 <header id="top">
   <span class="mark">Transcript Viewer</span>
   <div id="crumb"></div>
@@ -932,10 +957,77 @@ addEventListener("keydown",e=>{
   if(e.key==="\\"&&e.target.tagName!=="INPUT"){e.preventDefault();toggleSide()}
 });
 
-fetch("/api/index").then(r=>r.json()).then(d=>{
-  INDEX=d.sessions||[];GROUPS=d.groups||[];TAGS=d.tags||[];AI=d.ai||{};DOWNLOADS=d.downloads||DOWNLOADS;
-  count.textContent=INDEX.length+" sessions";drawList();showLibrary();
-});
+/* Opening the library has three parts worth naming: waiting for the server to
+   answer, reading the answer, and drawing it. Held back briefly so a fast open
+   does not flash a bar at you. */
+const boot={el:null,bar:null,text:null,timer:0};
+
+function bootShow(){
+  boot.el=document.getElementById("boot");
+  boot.bar=document.getElementById("bootbar");
+  boot.text=document.getElementById("boottext");
+  if(!boot.el)return;
+  boot.timer=setTimeout(()=>{if(boot.el.hidden)boot.el.hidden=false},160);
+}
+
+function bootSay(what,done,total){
+  if(boot.text)boot.text.textContent=what;
+  if(!boot.bar)return;
+  if(total>0){
+    boot.bar.classList.remove("waiting");
+    boot.bar.firstElementChild.style.width=Math.min(100,done/total*100)+"%";
+  }else{
+    boot.bar.classList.add("waiting");
+  }
+}
+
+function bootDone(){
+  clearTimeout(boot.timer);
+  if(boot.el)boot.el.hidden=true;
+}
+
+/* Read the response as it arrives, so the bar tracks real bytes rather than
+   guessing at them. Without a length to measure against it sweeps instead of
+   claiming progress it cannot know. */
+async function readIndex(){
+  const response=await fetch("/api/index");
+  if(!response.ok)throw new Error("the server answered "+response.status);
+  const total=Number(response.headers.get("Content-Length")||0);
+  if(!response.body||!total)return response.json();
+
+  const reader=response.body.getReader();
+  const chunks=[];let done=0;
+  for(;;){
+    const piece=await reader.read();
+    if(piece.done)break;
+    chunks.push(piece.value);done+=piece.value.length;
+    bootSay("Reading your library… "+bytes(done)+" of "+bytes(total),done,total);
+  }
+  const joined=new Uint8Array(done);let at=0;
+  for(const chunk of chunks){joined.set(chunk,at);at+=chunk.length}
+  bootSay("Making sense of it…",0,0);
+  return JSON.parse(new TextDecoder().decode(joined));
+}
+
+async function openLibrary(){
+  bootShow();
+  bootSay("Looking for your sessions…",0,0);
+  try{
+    const d=await readIndex();
+    INDEX=d.sessions||[];GROUPS=d.groups||[];TAGS=d.tags||[];AI=d.ai||{};DOWNLOADS=d.downloads||DOWNLOADS;
+    bootSay("Drawing "+INDEX.length+" sessions…",0,0);
+    // Let that line paint before the work that blocks the thread.
+    await new Promise(r=>requestAnimationFrame(()=>requestAnimationFrame(r)));
+    count.textContent=INDEX.length+" sessions";drawList();showLibrary();
+  }catch(err){
+    // Leave it on screen: a blank window would say nothing at all.
+    bootSay("Could not open your library — "+((err&&err.message)||err),0,0);
+    return;
+  }
+  bootDone();
+}
+
+openLibrary();
 q.oninput=drawList;
 
 /* Diwan's collectionTree builds a forest carrying depth, then flattens it
@@ -1328,17 +1420,35 @@ const clearSecret=name=>settings({name,clear:true});
    files, and nobody should find that out by pressing a button once. */
 let PLANNED=null;
 
+/* Places that really do hold transcripts, checked rather than guessed: each was
+   fetched and converted before it was listed. Gated ones are marked, since a
+   token you have not accepted terms for fails at the door rather than here. */
 const EXAMPLES=[
+  {name:"ATBench",
+   url:"https://huggingface.co/datasets/AI45Research/ATBench/tree/main/ATBench",
+   note:"1,000 labelled agent trajectories"},
+  {name:"ATBench-500",
+   url:"https://huggingface.co/datasets/AI45Research/ATBench/tree/main/ATBench500",
+   note:"the smaller config, 500 trajectories"},
   {name:"SLEIGHT-Bench",
-   url:"https://huggingface.co/datasets/sleightbench/SLEIGHT-Bench/tree/main/attacks"},
-  {name:"Redwood agent transcripts", url:"s3://rr-agent-transcripts"},
+   url:"https://huggingface.co/datasets/sleightbench/SLEIGHT-Bench/tree/main/attacks",
+   note:"gated — accept the terms on Hugging Face first"},
+  {name:"METR MALT",
+   url:"https://huggingface.co/datasets/metr-evals/malt-transcripts-public/tree/main/data",
+   note:"gated — accept the terms first; shards are large, so pick one"},
+  {name:"METR MALT (vague CoT)",
+   url:"https://huggingface.co/datasets/metr-evals/malt-transcripts-public/tree/main/vague_cot",
+   note:"the same runs with the chain of thought made vaguer"},
+  {name:"Redwood agent transcripts",
+   url:"s3://rr-agent-transcripts",
+   note:"needs an AWS profile set in Settings"},
 ];
 
 function drawExamples(){
   const host=document.getElementById("egs");
   if(!host)return;
   host.innerHTML=`<span>Try:</span>`+EXAMPLES.map((e,i)=>
-    `<button onclick="useExample(${i})" title="${esc(e.url)}">${esc(e.name)}</button>`
+    `<button onclick="useExample(${i})" title="${esc(e.note?e.note+" — "+e.url:e.url)}">${esc(e.name)}</button>`
   ).join("");
 }
 

{transcript_viewer-0.5.0 → transcript_viewer-0.6.1}/src/transcript_viewer/viewer.py

@@ -26,7 +26,7 @@ from urllib.parse import parse_qs, unquote, urlparse
 
 from atif_make.atif import ContentPart, Trajectory
 from . import corpus
-from atif_make.archive import extract, is_archive
+from atif_make.container import is_container, open_container
 from atif_make.convert import convert
 from transcript_viewer.corpus import Entry, scan
 
@@ -35,7 +35,33 @@ from . import ai, config, fetch, library
 # The page is a real .html file rather than a string in here: it is 2,232 lines
 # of CSS and JavaScript, which is not Python and should not be typed as though
 # it were. Read once at import, and served from memory.
-PAGE = (Path(__file__).parent / "page.html").read_text(encoding="utf-8")
+PAGE_PATH = Path(__file__).parent / "page.html"
+
+# What was read, and the state of the file it was read from.
+_page: tuple[float, int, str] | None = None
+
+
+def page() -> str:
+    """The interface, re-read when the file behind it changes.
+
+    The whole interface is that one file, so editing it and seeing nothing
+    change until the server is restarted reads as the edit not having worked.
+    Twice now it has. The cost is one stat per page load — not per request, and
+    not on anything the page then calls — and for an installed copy, where the
+    file never changes, it is a stat that always says the same thing.
+    """
+    global _page
+    try:
+        info = PAGE_PATH.stat()
+        stamp = (info.st_mtime, info.st_size)
+    except OSError:
+        # If it cannot be stat'd, what was already read is better than nothing.
+        if _page is not None:
+            return _page[2]
+        raise
+    if _page is None or (_page[0], _page[1]) != stamp:
+        _page = (stamp[0], stamp[1], PAGE_PATH.read_text(encoding="utf-8"))
+    return _page[2]
 
 
 def _point_images_at_server(trajectory: Trajectory, index: str) -> None:
@@ -305,7 +331,7 @@ class _Handler(BaseHTTPRequestHandler):
         return {
             e.key
             for e in self.entries
-            if (at := _group(e, library.get(e.key).get("source", "")))
+            if (at := _group(e, library.load().get(e.key, {}).get("source", "")))
             and (at == name or at.startswith(f"{name}/"))
         }
 
@@ -560,12 +586,13 @@ class _Handler(BaseHTTPRequestHandler):
         home = corpus.OPENED_ROOT / corpus.content_key(staged)
         try:
             home.mkdir(parents=True, exist_ok=True)
-            if is_archive(staged):
+            if is_container(staged):
                 # Keep what is inside, not the container: the logs are what get
                 # indexed, and unpacking here keeps their paths stable and any
-                # sibling images resolvable. Storing the zip would mean
-                # re-extracting to a temp directory on every start.
-                unpacked = extract(staged)
+                # sibling images resolvable. Storing the zip — or a benchmark's
+                # thousand-run JSON array — would mean opening it again on every
+                # start.
+                unpacked = open_container(staged)
                 for item in unpacked.iterdir():
                     shutil.move(str(item), home / item.name)
                 target = home
@@ -662,10 +689,10 @@ class _Handler(BaseHTTPRequestHandler):
             # names the archive happens to use internally.
             unpacked_from: dict[Path, Path] = {}
             for item in sorted(home.rglob("*")):
-                if not item.is_file() or not is_archive(item):
+                if not item.is_file() or not is_container(item):
                     continue
                 try:
-                    unpacked = extract(item)
+                    unpacked = open_container(item)
                 except (ValueError, OSError):
                     continue
                 beside = item.with_suffix("")
@@ -827,10 +854,13 @@ class _Handler(BaseHTTPRequestHandler):
         url = urlparse(self.path)
 
         if url.path == "/":
-            self._send(PAGE.encode(), "text/html; charset=utf-8")
+            self._send(page().encode(), "text/html; charset=utf-8")
             return
 
         if url.path == "/api/index":
+            # Read once. Asking per session made a page load parse the whole
+            # library as many times as there were sessions in it.
+            records = library.load()
             rows = [
                 {
                     "key": e.key,
@@ -844,7 +874,7 @@ class _Handler(BaseHTTPRequestHandler):
                     "size_bytes": e.size_bytes,
                     "subagents": e.subagents,
                     "session_title": e.session_title,
-                    "group": _group(e, library.get(e.key).get("source", "")),
+                    "group": _group(e, records.get(e.key, {}).get("source", "")),
                 }
                 for e in self.entries
                 # A downloaded folder the reader deleted, or a drive not mounted

{transcript_viewer-0.5.0 → transcript_viewer-0.6.1}/tests/page.test.js

@@ -1165,20 +1165,39 @@ test("the examples fill the field and look straight away", async () => {
       globalThis.__calls.push(url);
       return {ok:true,json:async()=>({nodes:[]})};
     };
-    return useExample(1);`);
-  assert.strictEqual(fields.urlin.value, "s3://rr-agent-transcripts");
+    const i = EXAMPLES.findIndex(e=>e.url.startsWith("s3://"));
+    globalThis.__url = EXAMPLES[i].url;
+    return useExample(i);`);
+  assert.strictEqual(fields.urlin.value, globalThis.__url);
   assert.ok(
     calls.includes("/api/browse"),
     "it filled the field but did not look",
   );
 });
 
-test("both examples are offered by name", () => {
-  const names = run(`return EXAMPLES.map(e=>e.name);`);
-  assert.deepStrictEqual(names, ["SLEIGHT-Bench", "Redwood agent transcripts"]);
-  const urls = run(`return EXAMPLES.map(e=>e.url);`);
-  assert.match(urls[0], /^https:\/\/huggingface\.co\/datasets\/sleightbench\//);
-  assert.match(urls[1], /^s3:\/\//);
+test("every example names a place the fetcher can actually reach", () => {
+  const egs = run(`return EXAMPLES;`);
+  assert.ok(egs.length >= 2, "there should be more than one thing to try");
+  for (const e of egs) {
+    assert.ok(e.name, "an example without a name is a blank button");
+    assert.match(
+      e.url,
+      /^(https:\/\/(huggingface\.co|github\.com)\/|s3:\/\/)/,
+      `${e.name}: not a scheme the fetcher handles`,
+    );
+  }
+  // Both kinds are worth offering: one needs no credentials, one does.
+  assert.ok(egs.some(e => e.url.startsWith("https://huggingface.co/")));
+  assert.ok(egs.some(e => e.url.startsWith("s3://")));
+});
+
+test("an example that needs credentials says so", () => {
+  const egs = run(`return EXAMPLES;`);
+  for (const e of egs) {
+    if (e.url.startsWith("s3://") || /sleightbench/.test(e.url)) {
+      assert.ok(e.note, `${e.name} needs something first and does not say so`);
+    }
+  }
 });
 
 test("a selection is measured rather than guessed at", async () => {
@@ -1405,3 +1424,59 @@ test("everything reads as All sessions", () => {
   console.log(`\n${tests.length - failed} passed, ${failed} failed`);
   process.exit(failed ? 1 : 0);
 })();
+
+test("opening the library reports each stage it goes through", async () => {
+  const said = [];
+  const el = { hidden: true };
+  const bar = {
+    classList: { add() {}, remove() {} },
+    firstElementChild: { style: {} },
+  };
+  const text = { textContent: "" };
+  globalThis.__boot = { boot: el, bootbar: bar, boottext: text, said };
+
+  const seen = run(`
+    document.getElementById = id => globalThis.__boot[id] || {textContent:"",style:{},classList:{add(){},remove(){}}};
+    globalThis.requestAnimationFrame = fn => fn();
+    globalThis.fetch = async () => ({
+      ok: true,
+      headers: { get: () => "0" },
+      body: null,
+      json: async () => ({sessions: [], groups: [], tags: [], ai: {}}),
+    });
+    const stages = [];
+    const realSay = bootSay;
+    bootSay = (what) => { stages.push(what); realSay(what, 0, 0); };
+    return openLibrary().then(() => stages);
+  `);
+
+  return seen.then((stages) => {
+    assert.ok(
+      stages.some((s) => /Looking for your sessions/.test(s)),
+      "it should say it is looking before it has anything",
+    );
+    assert.ok(
+      stages.some((s) => /Drawing/.test(s)),
+      "it should say it is drawing once it has the data",
+    );
+  });
+});
+
+test("a failure to open leaves the reason on screen", async () => {
+  const text = { textContent: "" };
+  const el = { hidden: false };
+  globalThis.__boot = {
+    boot: el,
+    bootbar: { classList: { add() {}, remove() {} }, firstElementChild: { style: {} } },
+    boottext: text,
+  };
+
+  await run(`
+    document.getElementById = id => globalThis.__boot[id] || {textContent:"",style:{},classList:{add(){},remove(){}}};
+    globalThis.fetch = async () => ({ ok: false, status: 500 });
+    return openLibrary();
+  `);
+
+  assert.match(text.textContent, /Could not open your library/);
+  assert.strictEqual(el.hidden, false, "a blank window would say nothing at all");
+});

{transcript_viewer-0.5.0 → transcript_viewer-0.6.1}/tests/test_corpus.py

@@ -433,3 +433,66 @@ def test_only_claude_code_is_asked_for_a_name(tmp_path):
     entry = corpus.describe(log)
     assert entry.format.startswith("codex")
     assert entry.session_title is None
+
+
+def _atbench_rows(n: int) -> list[dict]:
+    """Rows shaped like the real dataset, enough to be recognised as one."""
+    return [
+        {
+            "id": i,
+            "label": i % 2,
+            "risk_source": "indirect_prompt_injection",
+            "failure_mode": "unauthorized_information_disclosure",
+            "tool_used": [{"name": "f", "description": "d", "parameters": {}}],
+            "contents": [[
+                {"role": "user", "content": f"do thing {i}"},
+                {"role": "agent", "thought": "", "action": 'Complete{"response": "done"}'},
+            ]],
+        }
+        for i in range(1, n + 1)
+    ]
+
+
+def test_a_file_of_many_transcripts_indexes_as_many(tmp_path):
+    """A benchmark published as one JSON array is a container, not a session."""
+    source = tmp_path / "test.json"
+    source.write_text(json.dumps(_atbench_rows(5)))
+
+    entries = corpus.scan([source])
+    assert len(entries) == 5
+    assert {e.format for e in entries} == {"atbench"}
+
+
+def test_the_container_itself_is_not_offered_as_a_session(tmp_path):
+    """It cannot be opened — it holds five transcripts, not one."""
+    source = tmp_path / "test.json"
+    source.write_text(json.dumps(_atbench_rows(5)))
+
+    entries = corpus.scan([tmp_path])
+    assert len(entries) == 5
+    assert str(source) not in {e.path for e in entries}
+
+
+def test_the_pieces_keep_the_same_paths_across_scans(tmp_path):
+    """Paths are recorded in the index, so a scan must not move them."""
+    source = tmp_path / "test.json"
+    source.write_text(json.dumps(_atbench_rows(4)))
+
+    first = {e.path for e in corpus.scan([source])}
+    second = {e.path for e in corpus.scan([source])}
+    assert first == second
+    assert all(Path(p).exists() for p in first)
+
+
+def test_pieces_unpacked_beside_a_download_are_used_as_they_are(tmp_path):
+    """A fetch unpacks beside the file; scanning must not split it again."""
+    source = tmp_path / "test.json"
+    source.write_text(json.dumps(_atbench_rows(3)))
+    beside = tmp_path / "test"
+    beside.mkdir()
+    for i, row in enumerate(_atbench_rows(3), start=1):
+        (beside / f"test-{i:05d}-{i}.json").write_text(json.dumps([row]))
+
+    entries = corpus.scan([tmp_path])
+    assert len(entries) == 3
+    assert all(str(beside) in e.path for e in entries)

{transcript_viewer-0.5.0 → transcript_viewer-0.6.1}/tests/test_readme.py

@@ -9,6 +9,7 @@ Deliberately shallow: presence of a marker, not a parse. A test that tried to
 verify the prose itself would be a worse copy of the code.
 """
 
+import re
 from pathlib import Path
 
 import pytest
@@ -40,6 +41,8 @@ CLAIMS = [
     ("quadratic", "HISTORY_TURNS", "src/transcript_viewer/ai.py"),
     ("links into the transcript", "function jumpToStep", "src/transcript_viewer/page.html"),
     ("--extra ai", "ai = [", "pyproject.toml"),
+    ("re-read whenever it changes", "def page()", "src/transcript_viewer/viewer.py"),
+    ("stat per page load", "PAGE_PATH.stat()", "src/transcript_viewer/viewer.py"),
 ]
 
 
@@ -138,3 +141,48 @@ def test_every_measured_number_is_labelled_as_measured():
     """Figures in the README came from running something, not from a guess."""
     for figure in ["3.8 ms", "58 are sent entire", "202 of 202"]:
         assert figure in README, f"a measured figure went missing: {figure}"
+
+
+def test_the_endpoint_count_is_the_real_one():
+    """A number in prose drifts silently; count it instead of trusting it."""
+    words = {
+        12: "twelve", 13: "thirteen", 14: "fourteen", 15: "fifteen",
+        16: "sixteen", 17: "seventeen", 18: "eighteen",
+    }
+    source = Path("src/transcript_viewer/viewer.py").read_text()
+    actual = len(set(re.findall(r'"(/api/[a-z_-]+)"', source)))
+    readme = Path("README.md").read_text()
+    assert f"{words[actual]} endpoints" in readme, (
+        f"the README does not say {words[actual]} endpoints, but there are {actual}"
+    )
+
+
+def test_every_module_appears_in_the_map():
+    """A file nobody documented is a file nobody knows to look at."""
+    readme = Path("README.md").read_text()
+    for path in sorted(Path("src/transcript_viewer").iterdir()):
+        if path.name.startswith("_") or path.suffix not in {".py", ".html"}:
+            continue
+        assert f"  {path.name}" in readme, f"{path.name} is missing from the module map"
+
+
+def test_every_extra_offered_in_the_readme_exists():
+    """An install command that names an extra nobody defined simply fails."""
+    project = Path("pyproject.toml").read_text()
+    defined = set(re.findall(r"^([a-z][a-z0-9-]*) = \[", project, re.MULTILINE))
+    readme = Path("README.md").read_text()
+    offered = set()
+    for group in re.findall(r'transcript-viewer\[([a-z,]+)\]', readme):
+        offered.update(group.split(","))
+    missing = offered - defined
+    assert not missing, f"the README offers extras that do not exist: {missing}"
+
+
+def test_every_extra_that_exists_is_mentioned():
+    """An extra nobody is told about may as well not be there."""
+    project = Path("pyproject.toml").read_text()
+    block = project.split("[project.optional-dependencies]")[1].split("\n[")[0]
+    defined = set(re.findall(r"^([a-z][a-z0-9-]*) = \[", block, re.MULTILINE))
+    readme = Path("README.md").read_text()
+    unmentioned = {name for name in defined if name not in readme}
+    assert not unmentioned, f"extras the README never mentions: {unmentioned}"

{transcript_viewer-0.5.0 → transcript_viewer-0.6.1}/tests/test_viewer.py

@@ -911,3 +911,37 @@ def test_clearing_an_empty_library_is_harmless(server):
     _delete(server + "/api/library?all=1")
     status, payload = _delete(server + "/api/library?all=1")
     assert status == 200 and payload["removed"] == 0
+
+
+def test_the_page_is_re_read_when_it_changes(tmp_path, monkeypatch):
+    """Editing the interface and seeing nothing change reads as a broken edit."""
+    from transcript_viewer import viewer
+
+    page_file = tmp_path / "page.html"
+    page_file.write_text("<p>first</p>")
+    monkeypatch.setattr(viewer, "PAGE_PATH", page_file)
+    monkeypatch.setattr(viewer, "_page", None)
+
+    assert viewer.page() == "<p>first</p>"
+
+    # A same-size rewrite is the hard case: only the timestamp separates them.
+    page_file.write_text("<p>secnd</p>")
+    import os
+    stamp = page_file.stat().st_mtime + 10
+    os.utime(page_file, (stamp, stamp))
+
+    assert viewer.page() == "<p>secnd</p>"
+
+
+def test_an_unreadable_page_serves_what_was_already_read(tmp_path, monkeypatch):
+    """A file being written to should not take the interface down mid-save."""
+    from transcript_viewer import viewer
+
+    page_file = tmp_path / "page.html"
+    page_file.write_text("<p>kept</p>")
+    monkeypatch.setattr(viewer, "PAGE_PATH", page_file)
+    monkeypatch.setattr(viewer, "_page", None)
+    assert viewer.page() == "<p>kept</p>"
+
+    page_file.unlink()
+    assert viewer.page() == "<p>kept</p>"

{transcript_viewer-0.5.0 → transcript_viewer-0.6.1}/uv.lock

@@ -44,8 +44,17 @@ wheels = [
 
 [[package]]
 name = "atif-make"
-version = "0.4.0"
-source = { git = "https://github.com/jammastergirish/atif-make?branch=main#42672b0ac3ff5b559dee46f0656f352e560b6877" }
+version = "0.5.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/60/2c/9e5d1b1ba278fde5135b8066487e029dd0620161b6a220f5cb2a118a0333/atif_make-0.5.0.tar.gz", hash = "sha256:d4c3122a63b553099f0843d5d1c666136b0bccb1c11e8ebd2b4f9c4a9200cb5e", size = 215842, upload-time = "2026-08-23T05:15:04.439Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/00/47/f9a09d10b3d53600c43fa1b22c1d01cb731e628b3572c2e40109de195cfb/atif_make-0.5.0-py3-none-any.whl", hash = "sha256:b52fac5489f5cbc34f947f5d8b21b51a6a04276420d908a96f938e1bf7443791", size = 49769, upload-time = "2026-08-23T05:15:03.265Z" },
+]
+
+[package.optional-dependencies]
+parquet = [
+    { name = "pyarrow" },
+]
 
 [[package]]
 name = "colorama"
@@ -217,6 +226,42 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/54/20/4d324d65cc6d9205fabedc306948156824eb9f0ee1633355a8f7ec5c66bf/pluggy-1.6.0-py3-none-any.whl", hash = "sha256:e920276dd6813095e9377c0bc5566d94c932c33b27a3e3945d8389c374dd4746", size = 20538, upload-time = "2025-05-15T12:30:06.134Z" },
 ]
 
+[[package]]
+name = "pyarrow"
+version = "25.0.1"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/3d/e3/27f57f80141379d60defe6703eb50a707325706f07fedfd1312c7a751995/pyarrow-25.0.1.tar.gz", hash = "sha256:9150a83248bfed9813ea3c3af74c3856c1984d444aa28e58bf7733b9750ddf6a", size = 1201653, upload-time = "2026-08-10T12:40:53.904Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/a6/e2/9ab15b88cbfac28e16419ce5439ec29234c5172cb8259301b4ba639bdec0/pyarrow-25.0.1-cp312-cp312-macosx_12_0_arm64.whl", hash = "sha256:df961f2e7ae9cf496459259d798652c70625f6c080650d6952f8c04053c58ee9", size = 35861559, upload-time = "2026-08-10T12:38:02.567Z" },
+    { url = "https://files.pythonhosted.org/packages/58/79/a0036dbe1eabe1f73127427342f1d99982584c4a2cde2651d6c93499c6f6/pyarrow-25.0.1-cp312-cp312-macosx_12_0_x86_64.whl", hash = "sha256:cc4aa407fde9fc660be3939e49ea31f50f3e9fec17c0ec63159f7711edd3efc9", size = 37628383, upload-time = "2026-08-10T12:38:09.083Z" },
+    { url = "https://files.pythonhosted.org/packages/13/49/d93a57d375f4bf0cf82913dd6bb54acafde83dd993be2282c81ac5616cad/pyarrow-25.0.1-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:4340f0ba6c1d2e13f21658de1d7c662ca2545018568d0030a1e9afca159d87e3", size = 46820190, upload-time = "2026-08-10T12:38:15.458Z" },
+    { url = "https://files.pythonhosted.org/packages/60/c9/711ca85d79f1ec98f29a5eae2b051e25b4ecec5de3e3c0e2d5c5dcb15664/pyarrow-25.0.1-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:5389cdf79447ed1515c9e31620e6e1e2302249564d603f2ad727d4f6d313e4c3", size = 50102437, upload-time = "2026-08-10T12:38:22.487Z" },
+    { url = "https://files.pythonhosted.org/packages/80/53/8fb8359ff17cfb6263a1cf3ebf7caec9fe197de118719e84fcb1d0618026/pyarrow-25.0.1-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:d51592cb7561e87877c506113e7adbf1342ab579e6c21f0ef44b8ba41cb74c80", size = 49942424, upload-time = "2026-08-10T12:38:28.755Z" },
+    { url = "https://files.pythonhosted.org/packages/e8/83/4e5ae02a9341571b18a6fca380ac7a58ce6ddae7ab3c060208c0a1e79f02/pyarrow-25.0.1-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:6109c94d8b9f3b17a041daca16cacb2f651ad8f1ef70a4232c2c0f37a23da2a8", size = 53144206, upload-time = "2026-08-10T12:38:34.862Z" },
+    { url = "https://files.pythonhosted.org/packages/65/ee/197cbf47e49f83e6ebeb946a5259a48a638dea27ac774db42fe78022179d/pyarrow-25.0.1-cp312-cp312-win_amd64.whl", hash = "sha256:8858d7bfc22e3f51529aeaa4077225029724623e4595dc9eff8c793935c34140", size = 27953934, upload-time = "2026-08-10T12:38:39.808Z" },
+    { url = "https://files.pythonhosted.org/packages/cc/8d/8f271a7a034c834910ec925d56fa4b29733b1380f5289419f5aaa3b02777/pyarrow-25.0.1-cp313-cp313-macosx_12_0_arm64.whl", hash = "sha256:c7c534ec03c358a76ea3e505e74c1b6aef290af90c444dfd092dbfe23e755b85", size = 35855328, upload-time = "2026-08-10T12:38:45.489Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/cd/5bac242f4e841b9971d5eb94fdfe2577e2b70be983e27401e72055786037/pyarrow-25.0.1-cp313-cp313-macosx_12_0_x86_64.whl", hash = "sha256:dda9470024204d7bbf2042b47c6e8a0e47a3eeb8e34405882dfaea6577e0c153", size = 37622415, upload-time = "2026-08-10T12:38:51.107Z" },
+    { url = "https://files.pythonhosted.org/packages/63/1f/96d03b4e1506524f7087adb0fd6b2f69f0c9c7aaff1ec36d8030082e15a5/pyarrow-25.0.1-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:44a9120ce5bd81936b8ab9a88076e3fd47c2c6838e0e43630fed83626aca81d9", size = 46813813, upload-time = "2026-08-10T12:38:57.773Z" },
+    { url = "https://files.pythonhosted.org/packages/98/d6/33a411115b61dbfc16ad6ad73e71730f6fea654ee3667673bc53ab0e2fe7/pyarrow-25.0.1-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:0befcf816e45a1af33ac775a9970b749e4868a230c7372f0ae5e932bee27039f", size = 50104452, upload-time = "2026-08-10T12:39:04.579Z" },
+    { url = "https://files.pythonhosted.org/packages/33/ae/b1b97c9ca87f9f9ddbb5230c798df94eccce61bd79b9b45458c69a478588/pyarrow-25.0.1-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:3f89685964f46e4216103c75483aac0c0692a5f72212d7ca835adba5ede56ce3", size = 49951343, upload-time = "2026-08-10T12:39:11.8Z" },
+    { url = "https://files.pythonhosted.org/packages/98/9e/a112df5cfd5a68cb1d9fc31cfe38c28d5aec9f10865ce37ecef2e4450873/pyarrow-25.0.1-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:6943e2fe7954d29d84de45d29d34c8dc36ce96570e67d89aa9976e650a4a9138", size = 53144784, upload-time = "2026-08-10T12:39:20.503Z" },
+    { url = "https://files.pythonhosted.org/packages/31/24/97e8bd98f1e3b07e2ba08bcdff690674fbe16d69a7d2712cc3884665e615/pyarrow-25.0.1-cp313-cp313-win_amd64.whl", hash = "sha256:31e49a7888fcdf3a835da33ae777f6bb9a866334e5a789282fc26dcf426f7f15", size = 27870159, upload-time = "2026-08-10T12:39:26.161Z" },
+    { url = "https://files.pythonhosted.org/packages/36/4c/b525824ad3094076919273cd97db61fb3d78252dee76fa3b8dc8f76774aa/pyarrow-25.0.1-cp314-cp314-macosx_12_0_arm64.whl", hash = "sha256:bf0b672390cdcb640d7288f96b826d71ff4e9abb254a86c89890baf51a29cee6", size = 35885255, upload-time = "2026-08-10T12:39:32.366Z" },
+    { url = "https://files.pythonhosted.org/packages/08/62/448bb0e940de41aec31d1a956e63ad9c54afdf122a103cc3ab20c2a3ce33/pyarrow-25.0.1-cp314-cp314-macosx_12_0_x86_64.whl", hash = "sha256:38a9a4b4b9613380e200641891495a56c3d5a98a092db4a870af9975e220471d", size = 37644461, upload-time = "2026-08-10T12:39:38.142Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/9a/13587e38bd4806fd218f50fd13b8903fab60588a699ff0c406372e5b4043/pyarrow-25.0.1-cp314-cp314-manylinux_2_28_aarch64.whl", hash = "sha256:0b726ad7e7b669be982b0c71c07fe4b037d654354130da79a7902a669e93a66b", size = 46877146, upload-time = "2026-08-10T12:39:43.722Z" },
+    { url = "https://files.pythonhosted.org/packages/8d/61/1c5d1229fa21da4cff5365e41e57177aaac57c563c727f35419b8513d1c1/pyarrow-25.0.1-cp314-cp314-manylinux_2_28_x86_64.whl", hash = "sha256:9171748cdf796972d85a4b60157c279913e242992e350c90c7450182a9838b2a", size = 50131616, upload-time = "2026-08-10T12:39:49.304Z" },
+    { url = "https://files.pythonhosted.org/packages/43/20/291e1d65cc0b09aa19f03cf25cf51a2f5fa94b5db315178f2d254ed5cad4/pyarrow-25.0.1-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:b7a296aac7a71fa0886c08e155ddb6c636a50013f801f6178daafa0f9e726188", size = 50008879, upload-time = "2026-08-10T12:39:56.891Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/7c/1b7c9ec28e76576337e4f97b31141c9a181b89b6d1d6221e9d8205621a58/pyarrow-25.0.1-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:0fe7c8b6c03969b49c8c66182e4a18e3819ab92d07cfab5d8370c531b9369ef0", size = 53170864, upload-time = "2026-08-10T12:40:04.918Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/75/f3d789dc06011a765d14d86bda799cf72ac1d715b6a6edecaa0d73d95062/pyarrow-25.0.1-cp314-cp314-win_amd64.whl", hash = "sha256:f729cfdbd36fd99d543b67a914d2de044c84ebe45be8b34902b299b608c15c8f", size = 28620729, upload-time = "2026-08-10T12:40:51.41Z" },
+    { url = "https://files.pythonhosted.org/packages/fc/05/647a8ee6f7c2662feb6921315617bc04dcd6034763fb61b1199720bf6162/pyarrow-25.0.1-cp314-cp314t-macosx_12_0_arm64.whl", hash = "sha256:59a2de54c0cbd954da861eee4d1d330f8e909c45b53455baef696380f2c55033", size = 36130288, upload-time = "2026-08-10T12:40:11.014Z" },
+    { url = "https://files.pythonhosted.org/packages/93/f8/c9ee997554d7bea94520667dd1933f109ac1da3ee3556d2b49381e023484/pyarrow-25.0.1-cp314-cp314t-macosx_12_0_x86_64.whl", hash = "sha256:35935cd5de130aa5cf4dea052a63e6bf2e17006c35c3a468194242b9b2bf5956", size = 37762187, upload-time = "2026-08-10T12:40:16.592Z" },
+    { url = "https://files.pythonhosted.org/packages/a2/08/a28c01c7fe9e96e8233ce2d13df1d402f4f999f848f51d2daacd6bb4c036/pyarrow-25.0.1-cp314-cp314t-manylinux_2_28_aarch64.whl", hash = "sha256:f3831aaa25c67a99f99dc8b05873cb9d64560390372e2aa197ce9dd4a3f06a44", size = 46888003, upload-time = "2026-08-10T12:40:23.242Z" },
+    { url = "https://files.pythonhosted.org/packages/1b/b9/58612e977d28dc58c878448866838369ee8da2f1e7cc8ed2c84b952aafee/pyarrow-25.0.1-cp314-cp314t-manylinux_2_28_x86_64.whl", hash = "sha256:6a1fdfc6659b6b19022f2e50627fb5cf7156a66c46bf4299379955cbe742382a", size = 50079036, upload-time = "2026-08-10T12:40:29.169Z" },
+    { url = "https://files.pythonhosted.org/packages/72/13/66e1402dcc860e1dc2760b1e0292c9a569b62b3bccab69def1b3e907d006/pyarrow-25.0.1-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:169d3429d5be7c752125890620f75a60776d38b0035eddae939651640822332e", size = 50040226, upload-time = "2026-08-10T12:40:35.186Z" },
+    { url = "https://files.pythonhosted.org/packages/78/10/3f1a5497a7ef732ab0f03ecca3e66d89d9c0f57fdc61b4794c456b781f01/pyarrow-25.0.1-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:119297a6dc197e45d9c6d4415f7814a67ffa36c180d26f68c154c58067ae782d", size = 53149035, upload-time = "2026-08-10T12:40:41.454Z" },
+    { url = "https://files.pythonhosted.org/packages/93/c0/37d4a7e8e2f7a6076283673d5298018ca26478b934c6ee369e10505ab32c/pyarrow-25.0.1-cp314-cp314t-win_amd64.whl", hash = "sha256:4288f27577352d608ca08553b0865e4a9b3aa14820c5d95b53337218d609835b", size = 28753071, upload-time = "2026-08-10T12:40:46.623Z" },
+]
+
 [[package]]
 name = "pydantic"
 version = "2.13.4"
@@ -343,7 +388,7 @@ wheels = [
 
 [[package]]
 name = "transcript-viewer"
-version = "0.5.0"
+version = "0.6.1"
 source = { editable = "." }
 dependencies = [
     { name = "atif-make" },
@@ -353,6 +398,9 @@ dependencies = [
 ai = [
     { name = "anthropic" },
 ]
+parquet = [
+    { name = "atif-make", extra = ["parquet"] },
+]
 
 [package.dev-dependencies]
 dev = [
@@ -362,9 +410,10 @@ dev = [
 [package.metadata]
 requires-dist = [
     { name = "anthropic", marker = "extra == 'ai'", specifier = ">=0.40" },
-    { name = "atif-make", git = "https://github.com/jammastergirish/atif-make?branch=main" },
+    { name = "atif-make", specifier = ">=0.5.0" },
+    { name = "atif-make", extras = ["parquet"], marker = "extra == 'parquet'", specifier = ">=0.5.0" },
 ]
-provides-extras = ["ai"]
+provides-extras = ["parquet", "ai"]
 
 [package.metadata.requires-dev]
 dev = [{ name = "pytest", specifier = ">=8.0" }]