ablation-workbench 0.3.0__tar.gz → 0.3.2__tar.gz

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/CHANGELOG.md

@@ -2,6 +2,22 @@
 
 <!-- version list -->
 
+## v0.3.2 (2026-10-30)
+
+### Bug Fixes
+
+- 固定回放书签位置并保持悬停背景不透明 ([#14](https://github.com/yiminspace/ablation/pull/14),
+  [`4b6c727`](https://github.com/yiminspace/ablation/commit/4b6c727127cdc4e633e4320bef39846d21465da6))
+
+
+## v0.3.1 (2026-10-29)
+
+### Bug Fixes
+
+- 固定 GUI 默认端口并使用 /ablation 路径 ([#13](https://github.com/yiminspace/ablation/pull/13),
+  [`5259162`](https://github.com/yiminspace/ablation/commit/525916204ec73f0d7bfc25952867cbd9b595ba54))
+
+
 ## v0.3.0 (2026-10-28)
 
 ### Features

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: ablation-workbench
-Version: 0.3.0
+Version: 0.3.2
 Summary: A local CLI and review GUI for reproducible prompt experiments
 Project-URL: Homepage, https://github.com/yiminspace/ablation
 Project-URL: Repository, https://github.com/yiminspace/ablation
@@ -54,6 +54,8 @@ ablation --workspace ./experiments import request.json --format request
 ablation --workspace ./experiments gui
 ```
 
+The GUI starts at `http://127.0.0.1:8766/ablation/`. It tries subsequent ports only when the default port is occupied. `--port PORT` binds an explicit port without fallback; `--port 0` requests a random port.
+
 The example addresses are placeholders. Replace them with your own endpoints and supply credentials through the named environment variables. Configuration stores variable names, not secret values. Source retrieval and model inference use separate endpoint and credential settings.
 
 Choose an explicit import format:
@@ -119,7 +121,7 @@ The initial view fixes the imported original request on the left and shows the m
 
 The summary counts historical runs for the selected versions and scenario, so a later passing replay does not hide earlier failures. The response diff shows the latest completed outputs; inspect individual runs for earlier evidence.
 
-The sidebar starts collapsed and lists saved experiment sessions by title and full session ID, with search and replay counts. Each session retains all replay batches, scenarios, repeats, outcomes, and log links. The initial view prioritizes the diff, with the complete batch history on the later **运行记录** tab; **运行记录** lists all runs compactly. Clicking anywhere on a batch row opens a narrow, full-height right drawer with Messages, checks, and raw JSON views. A selector switches between the runs in that batch. Its timeline includes recorded inputs, tool calls/results, screenshot thumbnails that open at full size, and the new model output. Opening a run positions the timeline at its highlighted new output; a locate button returns to it. The timeline adapts Bifrost's Apache-2.0 components; the pinned source, modifications, and license ship with the web assets. External log links remain available for provenance. The GUI uses Voyage components with GitHub as the default theme. The top-right standard VoyageToolbar provides light/dark and theme selection. Its feedback and language placeholders remain disabled until those product capabilities are configured. All styles and the toolbar are bundled locally for offline use; `npm run build:ui` synchronizes the pinned assets and `npm run check:ui` verifies them.
+The sidebar starts collapsed and lists saved experiment sessions by title and full session ID, with search and replay counts. Each session retains all replay batches, scenarios, repeats, outcomes, and log links. The initial view prioritizes the diff, with the complete batch history on the later **运行记录** tab; **运行记录** lists all runs compactly. Clicking anywhere on a batch row opens a narrow, full-height right drawer with Messages, checks, and raw JSON views. Fixed index bookmarks switch between the runs in that batch without moving the stack; selection highlights the active bookmark, and hover backgrounds stay opaque. Its timeline includes recorded inputs, tool calls/results, screenshot thumbnails that open at full size, and the new model output. Opening a run positions the timeline at its highlighted new output; a locate button returns to it. The timeline adapts Bifrost's Apache-2.0 components; the pinned source, modifications, and license ship with the web assets. External log links remain available for provenance. The GUI uses Voyage components with GitHub as the default theme. The top-right standard VoyageToolbar provides light/dark and theme selection. Its feedback and language placeholders remain disabled until those product capabilities are configured. All styles and the toolbar are bundled locally for offline use; `npm run build:ui` synchronizes the pinned assets and `npm run check:ui` verifies them.
 
 Replay invokes a single model request. Returned tool calls are inspected as data and never executed. A model response does not prove that an external action occurred.
 

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/README.md

@@ -34,6 +34,8 @@ ablation --workspace ./experiments import request.json --format request
 ablation --workspace ./experiments gui
 ```
 
+The GUI starts at `http://127.0.0.1:8766/ablation/`. It tries subsequent ports only when the default port is occupied. `--port PORT` binds an explicit port without fallback; `--port 0` requests a random port.
+
 The example addresses are placeholders. Replace them with your own endpoints and supply credentials through the named environment variables. Configuration stores variable names, not secret values. Source retrieval and model inference use separate endpoint and credential settings.
 
 Choose an explicit import format:
@@ -99,7 +101,7 @@ The initial view fixes the imported original request on the left and shows the m
 
 The summary counts historical runs for the selected versions and scenario, so a later passing replay does not hide earlier failures. The response diff shows the latest completed outputs; inspect individual runs for earlier evidence.
 
-The sidebar starts collapsed and lists saved experiment sessions by title and full session ID, with search and replay counts. Each session retains all replay batches, scenarios, repeats, outcomes, and log links. The initial view prioritizes the diff, with the complete batch history on the later **运行记录** tab; **运行记录** lists all runs compactly. Clicking anywhere on a batch row opens a narrow, full-height right drawer with Messages, checks, and raw JSON views. A selector switches between the runs in that batch. Its timeline includes recorded inputs, tool calls/results, screenshot thumbnails that open at full size, and the new model output. Opening a run positions the timeline at its highlighted new output; a locate button returns to it. The timeline adapts Bifrost's Apache-2.0 components; the pinned source, modifications, and license ship with the web assets. External log links remain available for provenance. The GUI uses Voyage components with GitHub as the default theme. The top-right standard VoyageToolbar provides light/dark and theme selection. Its feedback and language placeholders remain disabled until those product capabilities are configured. All styles and the toolbar are bundled locally for offline use; `npm run build:ui` synchronizes the pinned assets and `npm run check:ui` verifies them.
+The sidebar starts collapsed and lists saved experiment sessions by title and full session ID, with search and replay counts. Each session retains all replay batches, scenarios, repeats, outcomes, and log links. The initial view prioritizes the diff, with the complete batch history on the later **运行记录** tab; **运行记录** lists all runs compactly. Clicking anywhere on a batch row opens a narrow, full-height right drawer with Messages, checks, and raw JSON views. Fixed index bookmarks switch between the runs in that batch without moving the stack; selection highlights the active bookmark, and hover backgrounds stay opaque. Its timeline includes recorded inputs, tool calls/results, screenshot thumbnails that open at full size, and the new model output. Opening a run positions the timeline at its highlighted new output; a locate button returns to it. The timeline adapts Bifrost's Apache-2.0 components; the pinned source, modifications, and license ship with the web assets. External log links remain available for provenance. The GUI uses Voyage components with GitHub as the default theme. The top-right standard VoyageToolbar provides light/dark and theme selection. Its feedback and language placeholders remain disabled until those product capabilities are configured. All styles and the toolbar are bundled locally for offline use; `npm run build:ui` synchronizes the pinned assets and `npm run check:ui` verifies them.
 
 Replay invokes a single model request. Returned tool calls are inspected as data and never executed. A model response does not prove that an external action occurred.
 

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/README.zh-CN.md

@@ -31,6 +31,8 @@ ablation demo --open
 
 完整配置、命令与示例见 [English README](README.md)。
 
+GUI 默认地址为 `http://127.0.0.1:8766/ablation/`，仅在端口被占用时依次尝试后续端口。`--port PORT` 指定端口且不自动回退；`--port 0` 显式使用随机端口。
+
 ## 接入数据源
 
 请求文件可以直接导入；HTTP 记录可使用 `request.body` 和 `response.body` 保存 JSON 正文。日志默认读取 `params` 中的请求字段，并用顶层 `model`、`input`、`messages`、`tools` 覆盖同名字段。任意嵌套结构可配置 `--source-request-pointer` 选出完整请求，或通过 `--source-fields-json` 将顶层请求字段映射到原始记录中的 JSON Pointer。`--source-response-pointer` 用于历史响应；没有响应时省略。映射路径不存在会报错。

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/docs/implementation-contract.md

@@ -43,6 +43,8 @@ Scenario origin is `source`, `real`, `derived`, or `synthetic`; the built-in dem
 
 Loopback only, validate Host/Origin, require `X-Ablation-Client: local` on mutations. Errors `{error:{code,message}}` with non-2xx. No API reads arbitrary paths. Files imported through GUI are uploaded JSON content. Recognized credential fields and configured secret values are redacted. Source URLs reject credential query fields; patches that introduce credentials into the effective request are rejected. Prompt and response text still require review before sharing.
 
+The GUI lives at `/ablation/`; the routes below use that prefix (for example, `/ablation/api/experiments`). Legacy unprefixed API routes remain compatible. Default GUI startup binds port 8766, trying subsequent ports only on address-in-use errors. An explicit `--port` does not fall back.
+
 - GET `/api/experiments` -> `{experiments:[...]}`; GET `/api/experiments/<id>` -> experiment.
 - POST `/api/import` source + title -> experiment; POST `/api/demo` -> synthetic example experiment.
 - GET `/api/experiments/<id>/diff?base=...&candidate=...&scenario=...&artifact=...` -> compare.

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/src/ablation_workbench/__init__.py

@@ -1,3 +1,3 @@
 """Local prompt experiment workbench."""
 
-__version__ = "0.3.0"
+__version__ = "0.3.2"

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/src/ablation_workbench/cli.py

@@ -92,7 +92,7 @@ def parser():
     demo.add_argument("--open", action="store_true")
     gui = sub.add_parser("gui", help="Serve the local review workspace")
     gui.add_argument("experiment", nargs="?")
-    gui.add_argument("--port", type=int, default=0)
+    gui.add_argument("--port", type=int, help="Bind this port (default: 8766, then the next free port; 0: random)")
     gui.add_argument("--no-open", action="store_true")
     candidate = sub.add_parser("candidate", help="Create an immutable candidate")
     candidate.add_argument("experiment"); candidate.add_argument("--label", required=True)

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/src/ablation_workbench/server.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 from datetime import datetime, timezone
+import errno
 from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 import json
 import mimetypes
@@ -24,6 +25,14 @@ from .runner import AgentRunner, workspace_lock
 ACTIVE = {"queued", "running", "cancelling"}
 MAX_BODY = 32 * 1024 * 1024
 WEB = Path(__file__).parent / "web"
+DEFAULT_PORT = 8766
+BASE_PATH = "/ablation"
+
+
+def route_path(path):
+    if path == BASE_PATH or path.startswith(BASE_PATH + "/"):
+        return path[len(BASE_PATH):] or "/"
+    return path
 
 
 def now():
@@ -218,7 +227,11 @@ def make_server(workbench, port=0):
             try:
                 if not self.valid_local(): self.error(ValueError("Local origin required"), 403); return
                 parsed = urlsplit(self.path)
-                path = parsed.path
+                if parsed.path in ("/", "/index.html", BASE_PATH):
+                    target = BASE_PATH + "/" + ("?" + parsed.query if parsed.query else "")
+                    self.send_bytes(b"", 302, extra={"Location": target})
+                    return
+                path = route_path(parsed.path)
                 if path == "/api/events":
                     self.send_events()
                     return
@@ -258,7 +271,7 @@ def make_server(workbench, port=0):
         def do_POST(self):
             try:
                 if not self.valid_local(mutation=True): self.error(ValueError("Local origin and client header required"), 403); return
-                path = urlsplit(self.path).path
+                path = route_path(urlsplit(self.path).path)
                 body = self.read_body()
                 if path == "/api/import":
                     source = body.get("source", {})
@@ -297,7 +310,16 @@ def make_server(workbench, port=0):
                 self.send_json(result)
             except (ValueError, OSError, RuntimeError, KeyError, TypeError) as exc: self.error(exc)
 
-    server = ThreadingHTTPServer(("127.0.0.1", port), Handler)
+    # Bind directly so another process cannot claim the port between a probe and bind.
+    selected_port = DEFAULT_PORT if port is None else port
+    while True:
+        try:
+            server = ThreadingHTTPServer(("127.0.0.1", selected_port), Handler)
+            break
+        except OSError as exc:
+            if port is not None or exc.errno != errno.EADDRINUSE or selected_port >= 65535:
+                raise
+            selected_port += 1
     server.daemon_threads = True
     server.agent_runner = agent
     server.replay_jobs = jobs
@@ -319,10 +341,10 @@ def close_jobs(server):
         time.sleep(0.1)
 
 
-def serve(workbench, experiment_id=None, port=0, open_browser=True):
+def serve(workbench, experiment_id=None, port=None, open_browser=True):
     if experiment_id: workbench.get(experiment_id)
     server = make_server(workbench, port)
-    url = f"http://127.0.0.1:{server.server_port}/" + (f"?experiment={experiment_id}" if experiment_id else "")
+    url = f"http://127.0.0.1:{server.server_port}{BASE_PATH}/" + (f"?experiment={experiment_id}" if experiment_id else "")
     print(json.dumps({"url": url, "workspace": str(workbench.workspace)}), flush=True)
     if open_browser: webbrowser.open(url)
     from .updates import start_update_checker

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/src/ablation_workbench/web/app.js

@@ -15,7 +15,7 @@
   const drawerMemory=new Map();
   const drawerMemoryKey=() => `${state.experiment?.id}|${drawerRunId}`;
   let drawerTargetTool;
-  let drawerBatchKey=null,drawerTransition=0,drawerAnimations=[];
+  let drawerBatchKey=null,drawerAnimations=[];
 
   function readStorage(key) { try { return JSON.parse(localStorage.getItem(key)); } catch { return null; } }
   function persist() {
@@ -35,7 +35,7 @@
   async function api(path, body, method = body === undefined ? 'GET' : 'POST') {
     const options = {method, headers:{Accept:'application/json'}};
     if (body !== undefined) { options.headers['Content-Type']='application/json'; options.headers['X-Ablation-Client']='local'; options.body=JSON.stringify(body); }
-    const response = await fetch(path, options);
+    const response = await fetch('/ablation' + path, options);
     let value;
     try { value = await response.json(); } catch { throw new Error(`服务返回了无法读取的内容（HTTP ${response.status}）`); }
     if (!response.ok) throw new Error(value.error?.message || value.error || `请求失败（HTTP ${response.status}）`);
@@ -203,7 +203,7 @@
   function experimentDetails() {
     const e=state.experiment,source=e.source || {};
     const synthetic=e.synthetic || source.kind==='demo' || source.fidelity==='synthetic';
-    openModal('experiment-details','实验详情',`<dl class="config-list"><dt>数据来源</dt><dd>${synthetic?'合成示例':esc(source.kind || '未标注')}<br><span class="muted">${esc(source.ref || '')}</span></dd><dt>来源完整度</dt><dd>${esc(source.fidelity || '未标注')}</dd><dt>实验编号</dt><dd><code>${esc(e.id)}</code></dd><dt>记录版本</dt><dd>r${esc(e.revision)}</dd><dt>创建时间</dt><dd>${esc(formatTime(e.created_at))}</dd><dt>原始请求指纹</dt><dd><code>${esc(e.baseline?.hash || '未提供')}</code></dd></dl>${source.notes?.length?`<details class="preview-details"><summary>来源说明</summary><p class="dialog-explanation">${esc(Array.isArray(source.notes)?source.notes.join(' '):source.notes)}</p></details>`:''}<div class="form-actions"><a class="vg-btn btn" id="export-link" href="${endpoint('/export')}" download>导出实验</a><button class="vg-btn btn" data-action="close-modal">关闭</button></div>`);
+    openModal('experiment-details','实验详情',`<dl class="config-list"><dt>数据来源</dt><dd>${synthetic?'合成示例':esc(source.kind || '未标注')}<br><span class="muted">${esc(source.ref || '')}</span></dd><dt>来源完整度</dt><dd>${esc(source.fidelity || '未标注')}</dd><dt>实验编号</dt><dd><code>${esc(e.id)}</code></dd><dt>记录版本</dt><dd>r${esc(e.revision)}</dd><dt>创建时间</dt><dd>${esc(formatTime(e.created_at))}</dd><dt>原始请求指纹</dt><dd><code>${esc(e.baseline?.hash || '未提供')}</code></dd></dl>${source.notes?.length?`<details class="preview-details"><summary>来源说明</summary><p class="dialog-explanation">${esc(Array.isArray(source.notes)?source.notes.join(' '):source.notes)}</p></details>`:''}<div class="form-actions"><a class="vg-btn btn" id="export-link" href="/ablation${endpoint('/export')}" download>导出实验</a><button class="vg-btn btn" data-action="close-modal">关闭</button></div>`);
   }
   function switchTab(tab) {
     state.tab=tab;
@@ -566,7 +566,7 @@
   }
   function startUpdateEvents() {
     if(updateEvents || !window.EventSource)return;
-    const source=new EventSource('/api/events');
+    const source=new EventSource('/ablation/api/events');
     updateEvents=source;
     source.onopen=()=>{
       if(updateEvents!==source)return;
@@ -628,7 +628,6 @@
     if(focus)(open?$('[aria-current="true"]',picker):$('#run-drawer-all'))?.focus({preventScroll:true});
   }
   function stopDrawerMotion() {
-    drawerTransition++;
     drawerAnimations.forEach(animation=>animation.cancel());drawerAnimations=[];
     $$('.run-drawer-outgoing,.run-drawer-backplate',$('#run-drawer')).forEach(sheet=>sheet.remove());
     $$('[data-motion-hidden]',$('#run-drawer')).forEach(label=>label.removeAttribute('data-motion-hidden'));
@@ -637,49 +636,10 @@
   function liveBookmarks() {
     return $$('.run-drawer-spine',$('#run-drawer')).filter(button=>!button.closest('.run-drawer-outgoing,.run-drawer-backplate'));
   }
-  function cloneOutgoingDrawer() {
-    const panel=$('#run-drawer-panel'),viewport=$('#run-drawer-viewport');
-    const snapshot=source=>{
-      const clone=source.cloneNode(true),sources=[source,...source.querySelectorAll('*')],copies=[clone,...clone.querySelectorAll('*')];
-      const positions=sources.map((element,index)=>element.scrollTop||element.scrollLeft?[index,element.scrollTop,element.scrollLeft]:null).filter(Boolean);
-      copies.forEach(element=>{element.removeAttribute('id');element.removeAttribute('autofocus');element.removeAttribute('aria-controls');element.removeAttribute('aria-labelledby');});
-      return {clone,restore:()=>positions.forEach(([index,top,left])=>{copies[index].scrollTop=top;copies[index].scrollLeft=left;})};
-    };
-    const page=snapshot(panel),body=snapshot(viewport),clone=page.clone,rect=panel.getBoundingClientRect();
-    clone.classList.add('run-drawer-outgoing');clone.inert=true;clone.setAttribute('aria-hidden','true');
-    clone.dataset.outgoingRun=drawerRunId;
-    clone.style.width=`${rect.width}px`;clone.style.left=`${rect.left}px`;clone.style.top=`${rect.top}px`;
-    $('.run-drawer-surface',clone).style.width=`${rect.width}px`;
-    const backplate=document.createElement('div');backplate.className='run-drawer-backplate';backplate.inert=true;backplate.setAttribute('aria-hidden','true');
-    backplate.style.width=`${rect.width}px`;backplate.append(body.clone);
-    $('.run-drawer-surface',backplate).style.width=`${rect.width}px`;
-    $('#run-drawer').append(backplate,clone);page.restore();body.restore();
-    return clone;
-  }
-  function animateRunDrawer(origin,outgoing,opening) {
-    const panel=$('#run-drawer-panel'),surface=$('#run-drawer-surface'),rect=panel.getBoundingClientRect();
-    if(matchMedia('(prefers-reduced-motion: reduce)').matches){outgoing?.remove();return;}
-    const token=drawerTransition,duration=320,easing='cubic-bezier(.22,.9,.22,1)';
-    const active=$('.run-drawer-spine',panel),activeRect=active.getBoundingClientRect();
-    const labelTop=activeRect.top-rect.top,step=activeRect.width;
-    surface.style.width=`${rect.width}px`;
-    if(opening){
-      drawerAnimations.push(panel.animate([{transform:`translateX(${Math.min(96,rect.width)}px)`,opacity:.6},{transform:'translateX(0)',opacity:1}],{duration:280,easing}));
-    }else{
-      drawerAnimations.push(panel.animate([{left:`${origin?origin.right:rect.left}px`,top:`${origin?origin.top-labelTop:rect.top}px`,width:'0px'},{left:`${rect.left}px`,top:`${rect.top}px`,width:`${rect.width}px`}],{duration,easing}));
-      drawerAnimations.push(surface.animate([{opacity:.8},{opacity:1}],{duration:120,delay:60}));
-      if(outgoing){
-        const old=outgoing.getBoundingClientRect(),oldLabel=$('.run-drawer-spine',outgoing).getBoundingClientRect();
-        const previous=liveBookmarks().find(button=>button.dataset.runDetail===outgoing.dataset.outgoingRun),target=previous?.getBoundingClientRect();
-        if(previous)previous.dataset.motionHidden='true';
-        drawerAnimations.push(outgoing.animate([{left:`${old.left}px`,top:`${old.top}px`,width:`${old.width}px`,opacity:1},{left:`${target?target.right:rect.left}px`,top:`${target?target.top-(oldLabel.top-old.top):0}px`,width:'0px',opacity:.9}],{duration,easing,fill:'forwards'}));
-      }
-    }
-    Promise.allSettled(drawerAnimations.map(animation=>animation.finished)).then(()=>{
-      if(token!==drawerTransition)return;
-      surface.style.width='';outgoing?.remove();$('.run-drawer-backplate',$('#run-drawer'))?.remove();
-      $$('[data-motion-hidden]',$('#run-drawer')).forEach(label=>label.removeAttribute('data-motion-hidden'));drawerAnimations=[];
-    });
+  function animateRunDrawer() {
+    if(matchMedia('(prefers-reduced-motion: reduce)').matches)return;
+    const panel=$('#run-drawer-panel'),width=panel.getBoundingClientRect().width;
+    drawerAnimations.push(panel.animate([{transform:`translateX(${Math.min(96,width)}px)`},{transform:'translateX(0)'}],{duration:280,easing:'cubic-bezier(.22,.9,.22,1)'}));
   }
   function renderRunStack(batch,run) {
     const stack=$('#run-drawer-stack'),key=`${state.experiment.id}|${run.batch_id || run.id}`;
@@ -689,21 +649,20 @@
     const width=$('#run-drawer').clientWidth,n=batch.length,preferredEdge=width>=1000?(n<=6?36:28):(n<=6?28:20);
     const minimumPanel=width>=1000?560:width>=480?360:220;
     const edge=Math.max(0,Math.min(preferredEdge,Math.max(14,(width-minimumPanel)/n),(width-Math.min(160,width))/n));
-    const panelWidth=Math.min(1040,width-n*edge),backOrder=batch.filter(item=>item.id!==run.id).map(item=>item.id).reverse();
+    const panelWidth=Math.min(1040,width-n*edge);
     $('#run-drawer').style.setProperty('--run-panel-width',`${panelWidth}px`);
     $('#run-drawer').style.setProperty('--run-bookmark-step',`${edge}px`);
     $('#run-drawer').style.setProperty('--run-index-size',edge<18?'10px':'11px');
-    $('#run-drawer-panel').style.setProperty('--bookmark-depth',1);
     const statusText=item=>activeStatus(item.status)?'运行中':item.error?'执行错误':item.passed===false?'未通过':item.passed===true?'通过':'未评分';
     for(const [index,item] of batch.entries()){
       const name=item.candidate==='baseline'?'原提示词':label(state.experiment.candidates,item.candidate),context=`${name} · ${label(state.experiment.scenarios,item.scenario)} · 第 ${item.repeat ?? 1} 次 · ${statusText(item)}`;
       let sheet=$$('[data-stack-run]',stack).find(element=>element.dataset.stackRun===item.id);
       if(!sheet){sheet=document.createElement('section');sheet.className='run-drawer-sheet';sheet.dataset.stackRun=item.id;sheet.setAttribute('role','presentation');stack.append(sheet);}
-      const selected=item.id===run.id,depth=selected?1:backOrder.indexOf(item.id)+2;
+      const selected=item.id===run.id,depth=batch.length-index;
       let button=liveBookmarks().find(button=>button.dataset.runDetail===item.id);
       if(!button){button=document.createElement('button');button.className='run-drawer-spine';button.dataset.runDetail=item.id;button.innerHTML=`<span class="spine-number">${index+1}</span><span class="spine-status-dot" aria-hidden="true"></span>`;}
-      const owner=selected?$('#run-drawer-panel'):sheet;if(button.parentElement!==owner)owner.prepend(button);
-      sheet.hidden=selected;sheet.inert=selected;sheet.dataset.visible=String(!selected);sheet.style.zIndex=batch.length-depth+1;sheet.style.setProperty('--bookmark-depth',depth);
+      if(button.parentElement!==sheet)sheet.prepend(button);
+      sheet.dataset.visible='true';sheet.style.zIndex=batch.length-depth+1;sheet.style.setProperty('--bookmark-depth',depth);
       button.id=`run-drawer-bookmark-${item.id}`;button.dataset.bookmarkIndex=index;button.disabled=false;button.title=context;button.setAttribute('aria-label',`回放 ${index+1}：${context}`);button.removeAttribute('role');button.removeAttribute('aria-selected');button.setAttribute('aria-current',selected?'page':'false');button.setAttribute('aria-pressed',String(selected));button.tabIndex=selected?0:-1;
       if(selected)$('#run-drawer-surface').setAttribute('aria-labelledby',button.id);
       $('.spine-status-dot',button).className=`spine-status-dot ${item.passed===false||item.error?'error':item.passed===true?'success':'warning'}`;
@@ -754,9 +713,7 @@
     if(!state.experiment?.runs.some(run=>run.id===runId))return;
     const changed=drawerRunId!==runId,opening=!$('#run-drawer').open,pickerWasOpen=!$('#run-drawer-picker').hidden;
     stopDrawerMotion();
-    const bookmark=liveBookmarks().find(button=>button.dataset.runDetail===runId),origin=bookmark?.getBoundingClientRect();
     saveDrawerPosition();
-    const outgoing=!opening&&changed&&!matchMedia('(prefers-reduced-motion: reduce)').matches?cloneOutgoingDrawer():null;
     setRunPicker(false);
     if(changed) {
       globalThis.AblationMessages.clear();$('#run-drawer-messages').replaceChildren();
@@ -772,7 +729,7 @@
       $(`#run-drawer-surface .run-drawer-tabs [data-run-view="${name}"]`).setAttribute('aria-selected',String(name===drawerView));
     }
     renderRunDrawer();restoreDrawerPosition();
-    if(changed||opening)animateRunDrawer(origin,outgoing,opening);
+    if(opening)animateRunDrawer();
     if(!opening)(pickerWasOpen?$('#run-drawer-all'):liveBookmarks().find(button=>button.dataset.runDetail===runId))?.focus({preventScroll:true});
     const run=state.experiment.runs.find(item=>item.id===runId);
     $$('.run-batch-open').forEach(button=>button.setAttribute('aria-current',String(button.closest('[data-batch-group]').dataset.batchGroup===(run.batch_id || run.id))));

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/src/ablation_workbench/web/index.html

@@ -5,20 +5,20 @@
   <meta name="viewport" content="width=device-width, initial-scale=1">
   <meta name="color-scheme" content="light dark">
   <title>Ablation · Prompt workbench</title>
-  <link rel="icon" href="/favicon.svg" type="image/svg+xml" sizes="any">
-  <script src="/theme.js"></script>
-  <link rel="stylesheet" href="/voyage.css">
-  <link rel="stylesheet" href="/styles.css">
-  <link rel="stylesheet" href="/messages.css">
-  <script src="/messages.js" defer></script>
-  <script src="/run-results.js" defer></script>
-  <script src="/app.js" defer></script>
-  <script src="/voyage-toolbar.js" defer></script>
+  <link rel="icon" href="/ablation/favicon.svg" type="image/svg+xml" sizes="any">
+  <script src="/ablation/theme.js"></script>
+  <link rel="stylesheet" href="/ablation/voyage.css">
+  <link rel="stylesheet" href="/ablation/styles.css">
+  <link rel="stylesheet" href="/ablation/messages.css">
+  <script src="/ablation/messages.js" defer></script>
+  <script src="/ablation/run-results.js" defer></script>
+  <script src="/ablation/app.js" defer></script>
+  <script src="/ablation/voyage-toolbar.js" defer></script>
 </head>
 <body class="focused-workbench vg-app vg-fullscreen">
   <a class="skip-link" href="#main">跳到实验内容</a>
   <header class="global-header vg-header">
-    <div class="header-start"><button class="vg-iconbtn history-toggle" data-action="toggle-history" aria-controls="session-history" aria-expanded="false" aria-label="展开会话历史" title="展开会话历史"><svg viewBox="0 0 20 20" width="18" height="18" fill="none" stroke="currentColor" stroke-width="1.4" aria-hidden="true"><rect x="2.5" y="3" width="15" height="14" rx="2"/><path d="M7 3v14M4.5 6h.5m-.5 3h.5m-.5 3h.5"/></svg></button><a href="/" class="brand" aria-label="Ablation 首页" title="Ablation · 提示词原版与变体对比"><svg class="brand-mark" viewBox="0 0 32 32" width="29" height="29" fill="none" stroke="currentColor" stroke-width="1.7" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><rect x="2" y="5" width="11" height="22" rx="3" opacity=".65"/><path d="M5.5 11h4m-4 5h4m-4 5h4" opacity=".65"/><rect class="brand-variant" x="19" y="5" width="11" height="22" rx="3"/><path class="brand-variant" d="M22.5 11h4m-4 10h4"/><path d="M15 16h2"/></svg><strong>ablation</strong></a></div>
+    <div class="header-start"><button class="vg-iconbtn history-toggle" data-action="toggle-history" aria-controls="session-history" aria-expanded="false" aria-label="展开会话历史" title="展开会话历史"><svg viewBox="0 0 20 20" width="18" height="18" fill="none" stroke="currentColor" stroke-width="1.4" aria-hidden="true"><rect x="2.5" y="3" width="15" height="14" rx="2"/><path d="M7 3v14M4.5 6h.5m-.5 3h.5m-.5 3h.5"/></svg></button><a href="/ablation/" class="brand" aria-label="Ablation 首页" title="Ablation · 提示词原版与变体对比"><svg class="brand-mark" viewBox="0 0 32 32" width="29" height="29" fill="none" stroke="currentColor" stroke-width="1.7" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><rect x="2" y="5" width="11" height="22" rx="3" opacity=".65"/><path d="M5.5 11h4m-4 5h4m-4 5h4" opacity=".65"/><rect class="brand-variant" x="19" y="5" width="11" height="22" rx="3"/><path class="brand-variant" d="M22.5 11h4m-4 10h4"/><path d="M15 16h2"/></svg><strong>ablation</strong></a></div>
     <div class="header-actions"><button class="vg-btn btn" data-action="import">导入数据</button><button class="vg-btn btn quiet" id="update-indicator" data-action="updates" hidden>有可用更新</button><div id="voyage-toolbar" aria-label="外观与工具"></div><button class="vg-iconbtn" data-action="settings" aria-label="设置与版本" title="设置与版本"><svg viewBox="0 0 20 20" width="17" height="17" aria-hidden="true"><path d="M3 5h14M3 10h14M3 15h14M7 3v4m6 1v4m-7 1v4" stroke="currentColor" stroke-width="1.5" fill="none"/></svg></button></div>
   </header>
   <div id="upgrade-banner" class="upgrade-banner" role="status" aria-live="polite" hidden><span>Ablation 已升级到 <strong id="upgrade-version"></strong>，刷新页面以加载新版界面。</span><button class="vg-btn btn" data-action="reload-ui">刷新页面</button></div>

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/src/ablation_workbench/web/styles.css

@@ -227,11 +227,11 @@
 .run-drawer-sheet[data-visible=false]{visibility:hidden;opacity:0;pointer-events:none}
 .run-drawer-spine{position:absolute;left:calc(-1*var(--bookmark-depth,1)*var(--run-bookmark-step));top:calc(12px + (var(--bookmark-depth,1) - 1)*3px);display:flex;flex-direction:column;align-items:center;justify-content:flex-start;gap:8px;width:var(--run-bookmark-step);height:120px;min-width:0;border:1px solid var(--border);border-right:0;border-radius:6px 0 0 6px;background:var(--bg);color:var(--muted);padding:18px 0 0;text-align:center;box-shadow:-3px 2px 8px #0002;pointer-events:auto;overflow:visible}
 .run-drawer-spine[data-motion-hidden=true]{visibility:hidden}
-.run-drawer-spine:after{content:'';position:absolute;left:100%;top:-1px;bottom:-1px;width:calc((var(--bookmark-depth,1) - 1)*var(--run-bookmark-step) + 1px);border-block:1px solid var(--border);background:inherit;pointer-events:none}
-.run-drawer-spine:hover{background:var(--blue-bg)}
-.run-drawer-spine[aria-current=true],.run-drawer-spine[aria-pressed=true]{border-left-color:var(--acc);background:var(--blue-bg);color:var(--fg)}
+.run-drawer-spine:after{content:'';position:absolute;left:100%;top:-1px;bottom:-1px;width:calc((var(--bookmark-depth,1) - 1)*var(--run-bookmark-step) + 1px);border-block:1px solid var(--border);background:var(--bg);pointer-events:none}
+.run-drawer-spine:hover{background:color-mix(in srgb,var(--acc) 8%,var(--bg));opacity:1}
+.run-drawer-spine[aria-current=true],.run-drawer-spine[aria-pressed=true]{border-left-color:var(--acc);background:color-mix(in srgb,var(--acc) 14%,var(--bg));color:var(--fg)}
 .run-drawer-spine[aria-current=true] .spine-number,.run-drawer-spine[aria-pressed=true] .spine-number{color:var(--fg)}
-.run-drawer-spine:focus-visible{outline-offset:-3px;background:var(--blue-bg)}
+.run-drawer-spine:focus-visible{outline-offset:-3px;background:color-mix(in srgb,var(--acc) 14%,var(--bg))}
 .spine-number{font:var(--run-index-size,11px)/1.5 var(--mono);font-variant-numeric:tabular-nums;color:var(--muted);flex:none;white-space:nowrap}
 .spine-status-dot{display:block;width:6px;height:6px;border-radius:50%;background:var(--muted);flex:none}
 .spine-status-dot.success{background:var(--green)}

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/tests/browser_smoke.cjs

@@ -27,7 +27,7 @@ async function main(){
   child=spawn(python,['-m','ablation_workbench','--workspace',workspace,'gui','--no-open'],{cwd:process.env.ABLATION_USE_INSTALLED?workspace:root,env:appEnv,stdio:['ignore','pipe','pipe']});
   let output='',errors='';child.stdout.on('data',b=>output+=b);child.stderr.on('data',b=>errors+=b);
   const url=await until(()=>{if(child.exitCode!==null)throw Error(errors);try{return JSON.parse(output.trim()).url;}catch{return null;}});
-  const api=async(route,body)=>{const response=await fetch(new URL(route,url),{method:body===undefined?'GET':'POST',headers:{'Content-Type':'application/json','X-Ablation-Client':'local'},body:body===undefined?undefined:JSON.stringify(body)});const data=await response.json();assert(response.ok,JSON.stringify(data));return data;};
+  const api=async(route,body)=>{const response=await fetch(new URL('/ablation'+route,url),{method:body===undefined?'GET':'POST',headers:{'Content-Type':'application/json','X-Ablation-Client':'local'},body:body===undefined?undefined:JSON.stringify(body)});const data=await response.json();assert(response.ok,JSON.stringify(data));return data;};
   browser=await chromium.launch({headless:true});
   const context=await browser.newContext({viewport:{width:1440,height:1000},permissions:['clipboard-read','clipboard-write']});
   const page=await context.newPage(),pageErrors=[];page.on('pageerror',e=>pageErrors.push(e.message));
@@ -204,8 +204,8 @@ async function main(){
   await page.locator('.batch-copy strong').first().click();
   await page.locator('.run-drawer-tabs [data-run-view="messages"]').click();
   await page.locator('#run-drawer .bifrost-messages').waitFor();
-  assert.equal(await page.locator('#run-drawer-stack .run-drawer-sheet[data-visible="true"]').count(),1);
-  assert.equal(await page.locator('#run-drawer-panel > .run-drawer-spine').count(),1);
+  assert.equal(await page.locator('#run-drawer-stack .run-drawer-sheet[data-visible="true"]').count(),2);
+  assert.equal(await page.locator('#run-drawer-panel > .run-drawer-spine').count(),0);
   const baselineSpine=page.locator(`#run-drawer-stack .run-drawer-sheet[data-visible="true"] [data-run-detail="${exp.runs[0].id}"]`);
   if(await page.locator('#run-drawer-panel').getAttribute('data-run-id')!==exp.runs[0].id)await baselineSpine.locator('.spine-number').click();
   await page.locator('.run-drawer-tabs [data-run-view="messages"]').click();

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/tests/browser_updates.cjs

@@ -57,13 +57,13 @@ async function stop(service){send(service,{type:'stop'});await until(()=>service
 async function main(){
   let service=await start();
   const initial=service.version,parts=initial.split('.').map(Number);parts[2]+=1;
-  const next=parts.join('.'),url=`http://127.0.0.1:${service.port}/`;
+  const next=parts.join('.'),url=`http://127.0.0.1:${service.port}/ablation/`;
   browser=await chromium.launch({headless:true});
   const page=await browser.newPage({viewport:{width:1280,height:900}}),errors=[];
   page.on('pageerror',error=>errors.push(error.message));
   let updates=0,jobs=0,connections=0,versions=0;
-  page.on('request',request=>{const p=new URL(request.url()).pathname;if(p==='/api/update')updates++;if(p==='/api/jobs')jobs++;});
-  page.on('response',response=>{const p=new URL(response.url()).pathname;if(p==='/api/events'&&response.ok())connections++;if(p==='/api/version'&&response.ok())versions++;});
+  page.on('request',request=>{const p=new URL(request.url()).pathname;if(p==='/ablation/api/update')updates++;if(p==='/ablation/api/jobs')jobs++;});
+  page.on('response',response=>{const p=new URL(response.url()).pathname;if(p==='/ablation/api/events'&&response.ok())connections++;if(p==='/ablation/api/version'&&response.ok())versions++;});
   await page.clock.install();
   await page.goto(url);
   await page.locator('#empty-state').waitFor({state:'visible'});

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/tests/browser_workbench.cjs

@@ -74,7 +74,7 @@ async function assertPhysicalBookmarks(page,count){
   assert.equal(geometry.bookmarks.length,count,`All ${count} page bookmarks remain visible`);
   assert.equal(geometry.bookmarks.filter(item=>item.current==='page').length,1,'Exactly one ordinary bookmark is current');
   const current=geometry.bookmarks.find(item=>item.current==='page');
-  assert(current.main&&current.id===geometry.active&&current.pressed==='true','Current bookmark is attached directly to the live page');
+  assert(!current.main&&current.id===geometry.active&&current.pressed==='true','Current bookmark stays in its fixed stack position');
   assert.notEqual(geometry.stackRole,'tablist','Bookmarks are attached page controls rather than a separate navbar');
   assert(Math.abs(geometry.panel.top)<1&&geometry.panel.height>=geometry.viewport.height-1,'Live page remains top-aligned and full height');
   assert(geometry.panel.left>=0&&geometry.panel.right<=geometry.viewport.width+1,'Live page fits the right side of the viewport');
@@ -82,7 +82,7 @@ async function assertPhysicalBookmarks(page,count){
   assert(['hidden','clip'].includes(geometry.clip.overflow),'Page viewport clips content during physical expansion');
   assert(geometry.bookmarks.every(item=>item.owner===item.id&&item.role!=='tab'),'Each ordinary button belongs to its corresponding page');
   assert(geometry.bookmarks.every(item=>Math.abs(item.frame.left-geometry.panel.left)<2&&Math.abs(item.frame.top)<1&&Math.abs(item.frame.width-geometry.panel.width)<2&&item.frame.height>=geometry.viewport.height-1),'Background pages share the live page rectangle; there are no detached full-height peer rails');
-  assert(geometry.bookmarks.every(item=>item.label.height>=100&&item.label.height<=140&&item.label.top>=0&&item.label.bottom<=180&&item.label.right<=geometry.panel.left+2&&item.horizontal&&item.status&&item.reachable&&item.number.left>=-1&&item.number.right<=geometry.viewport.width+1),'Every attached bookmark/status protrudes only in the top-left and remains directly clickable');
+  assert(geometry.bookmarks.every(item=>item.label.height>=100&&item.label.height<=140&&item.label.top>=0&&item.label.bottom<=Math.min(geometry.viewport.height,132+(count-1)*3)&&item.label.right<=geometry.panel.left+2&&item.horizontal&&item.status&&item.reachable&&item.number.left>=-1&&item.number.right<=geometry.viewport.width+1),'Every attached bookmark/status protrudes only in the top-left and remains directly clickable');
   assert(geometry.bookmarks.every(item=>item.neck.content!=='none'&&item.neck.width>=geometry.panel.left-item.label.right-2),'Bookmark necks connect each index sticker to its page');
   return geometry;
 }
@@ -135,6 +135,7 @@ async function sampleMotion(page,selector){
 async function main(){
   server=http.createServer((req,res)=>{
     const url=new URL(req.url,'http://127.0.0.1');
+    url.pathname=url.pathname.replace(/^\/ablation(?=\/|$)/,'')||'/';
     if(req.method!=='GET'){mutations.push(req.url);res.writeHead(405);res.end();return;}
     if(url.pathname==='/api/events'){res.writeHead(200,{'Content-Type':'text/event-stream'});res.write('retry: 1000\n\n');return;}
     const json=data=>{res.writeHead(200,{'Content-Type':'application/json','Cache-Control':'no-store'});res.end(JSON.stringify(data));};
@@ -153,7 +154,7 @@ async function main(){
   });
   server.on('connection',socket=>{sockets.add(socket);socket.on('close',()=>sockets.delete(socket));});
   await new Promise(resolve=>server.listen(0,'127.0.0.1',resolve));
-  const base=`http://127.0.0.1:${server.address().port}/`;
+  const base=`http://127.0.0.1:${server.address().port}/ablation/`;
   browser=await chromium.launch({headless:true});
   const page=await browser.newPage({viewport:{width:1440,height:1000},reducedMotion:'no-preference'});
   page.on('pageerror',error=>errors.push(error.message));
@@ -208,31 +209,29 @@ async function main(){
   await assertPhysicalBookmarks(page,6);
   const edgeSelector='#run-drawer-stack .run-drawer-sheet[data-visible="true"] > .run-drawer-spine[data-run-detail]';
   const edgeRun=await page.locator(edgeSelector).first().getAttribute('data-run-detail');
+  const before=await assertPhysicalBookmarks(page,6);
   const switching=await sampleMotion(page,edgeSelector);
-  assert(switching.states.includes('running'),'Selecting a physical bookmark starts whole-page motion');
-  assert(switching.sameBookmark,'The chosen actual bookmark DOM node moves into the live page');
-  const [start,quarter,middle,end]=switching.samples;
-  assert(start.width<end.width-100&&quarter.width>start.width+20,'Incoming whole page expands from the chosen bookmark');
-  assert(switching.samples.every(sample=>Math.abs(sample.incoming.width-end.incoming.width)<2),'Page content keeps a fixed reading width while its viewport opens');
-  assert(Math.abs(start.bookmark.left-switching.origin.left)<2&&Math.abs(start.bookmark.top-switching.origin.top)<2,'Chosen bookmark starts at its original physical position');
-  assert(switching.samples.every(sample=>Math.abs(sample.bookmark.right-sample.left)<2&&Math.abs(sample.bookmark.top-sample.top-12)<2),'Incoming page and attached bookmark move together');
-  assert(start.outgoing&&quarter.outgoing,'Previous foreground remains visible while it retreats');
-  assert(quarter.outgoing.width<start.outgoing.width-20&&end.outgoing.width<5,'Previous whole page collapses into its attached index sticker');
-  assert(switching.samples.every(sample=>Math.abs(sample.outgoing.bookmark.right-sample.outgoing.left)<2&&Math.abs(sample.outgoing.bookmark.top-sample.outgoing.top-12)<2),'Outgoing page and its bookmark move together');
-  assert.deepEqual({inert:switching.backplateContent.inert,hidden:switching.backplateContent.hidden,ids:switching.backplateContent.ids},{inert:true,hidden:'true',ids:0});
-  assert(switching.backplateContent.content>30,'Stable backplate retains actual previous page content');
-  assert(switching.samples.every(sample=>sample.backplate&&sample.backplate.left<=end.left+1&&sample.backplate.right>=end.left+end.width-1&&sample.backplate.top<=1&&sample.backplate.height>=999&&sample.backplate.opaque),'Stable opaque content covers the right edge throughout the physical page transition');
+  assert.equal(switching.states.length,0,'Selecting a bookmark updates content without page motion');
+  assert.equal(switching.outgoingContent,null,'Switching needs no fading page clone');
+  const after=await assertPhysicalBookmarks(page,6);
+  const positions=geometry=>geometry.bookmarks.map(item=>({id:item.id,label:item.label})).sort((a,b)=>a.id.localeCompare(b.id));
+  assert.deepEqual(positions(after),positions(before),'All bookmark positions and sizes stay fixed after selection');
+  await page.locator(edgeSelector).first().hover();
+  const hover=await page.locator(edgeSelector).first().evaluate(button=>{
+    const style=getComputedStyle(button);const canvas=document.createElement('canvas');canvas.width=canvas.height=1;
+    const context=canvas.getContext('2d');context.fillStyle=style.backgroundColor;context.fillRect(0,0,1,1);
+    return {opacity:Number(style.opacity),alpha:context.getImageData(0,0,1,1).data[3]};
+  });
+  assert.deepEqual(hover,{opacity:1,alpha:255},'Hover background stays fully opaque');
   assert.equal(await page.locator('#run-drawer-panel').getAttribute('data-run-id'),edgeRun);
-  assert.equal(switching.keyboardUsesCurrentPanel,true,'Keyboard End selects real panel tab while clone exists');
-  assert.deepEqual({inert:switching.outgoingContent.inert,hidden:switching.outgoingContent.hidden,ids:switching.outgoingContent.ids},{inert:true,hidden:'true',ids:0});
   const rapid=await page.evaluate(()=>{
-    const next=()=>[...document.querySelectorAll('#run-drawer-stack .run-drawer-sheet[data-visible="true"] > .run-drawer-spine[data-run-detail]')].find(button=>getComputedStyle(button).visibility!=='hidden');
+    const next=()=>[...document.querySelectorAll('#run-drawer-stack .run-drawer-sheet[data-visible="true"] > .run-drawer-spine[data-run-detail]')].find(button=>getComputedStyle(button).visibility!=='hidden'&&button.getAttribute('aria-current')!=='page');
     next().click();const first=document.querySelectorAll('.run-drawer-outgoing').length;
     next().click();const second=document.querySelectorAll('.run-drawer-outgoing').length;
     const active=document.querySelector('#run-drawer-panel').dataset.runId;
     document.querySelector('[data-action="close-run-drawer"]').click();return {first,second,active};
   });
-  assert(rapid.first<=1&&rapid.second<=1,'Rapid switching replaces the previous outgoing clone');
+  assert(rapid.first===0&&rapid.second===0,'Rapid switching creates no animated clones');
   await page.locator('#run-drawer').waitFor({state:'hidden'});
   await until(()=>page.evaluate(()=>!document.querySelector('.run-drawer-outgoing,.run-drawer-backplate')&&![...document.querySelectorAll('[data-motion-hidden]')].some(element=>getComputedStyle(element).visibility==='hidden')&&!document.getAnimations().some(animation=>animation.playState==='running'&&animation.effect?.target?.closest('#run-drawer'))));
   await page.locator('.run-batch-open').first().click();await settleDrawer(page);
@@ -253,9 +252,8 @@ async function main(){
   await raw.evaluate(node=>node.scrollTop=240);
   const rawScroll=await raw.evaluate(node=>node.scrollTop);assert(rawScroll>0);
   const rawTransition=await sampleMotion(page,'#run-drawer-stack .run-drawer-sheet[data-visible="true"] [data-run-detail="run_fixture_2"]');
-  assert(Math.abs(rawTransition.outgoingContent.rawScrollTop-rawScroll)<2,'Outgoing raw clone retains the reading position');
-  assert.equal(rawTransition.keyboardUsesCurrentPanel,true);
-  assert.deepEqual({inert:rawTransition.outgoingContent.inert,hidden:rawTransition.outgoingContent.hidden,ids:rawTransition.outgoingContent.ids},{inert:true,hidden:'true',ids:0});
+  assert.equal(rawTransition.states.length,0,'Raw view switches immediately without page motion');
+  assert.equal(rawTransition.outgoingContent,null);
   await page.locator('.run-drawer-tabs [data-run-view="checks"]').click();
   // A missing tool and a non-tool assertion still lead to the saved model output.
   for(const index of [1,2]){
@@ -317,16 +315,28 @@ async function main(){
     assert.equal(await page.locator('#run-drawer-panel').getAttribute('data-run-id'),'run_fixture_12');
     await page.keyboard.press('Escape');await page.locator('#run-drawer').waitFor({state:'hidden'});
   }
+  await page.setViewportSize({width:1920,height:1000});
+  for(let index=13;index<=30;index++)fixture.runs.push({...fixture.runs[5],id:`run_fixture_${index}`,repeat:index-3});
+  fixture.revision++;
+  await until(async()=>await page.locator('.history-batch').getAttribute('data-run-count')==='30');
+  await page.locator('.run-batch-open').first().click();await settleDrawer(page);
+  const thirty=await assertPhysicalBookmarks(page,30);
+  for(const runId of ['run_fixture_30','run_fixture_1','run_fixture_15']){
+    await selectReplay(page,runId);
+    const selected=await assertPhysicalBookmarks(page,30);
+    assert.deepEqual(positions(selected),positions(thirty),'Thirty replay bookmarks retain their geometry across selection');
+  }
   if(process.env.ABLATION_SCREENSHOTS){
     fs.mkdirSync(process.env.ABLATION_SCREENSHOTS,{recursive:true});
-    await page.setViewportSize({width:1440,height:1000});await page.locator('.run-batch-open').first().click();
+    await page.setViewportSize({width:1440,height:1000});
+    if(!await page.locator('#run-drawer').evaluate(node=>node.open))await page.locator('.run-batch-open').first().click();
     await page.screenshot({path:path.join(process.env.ABLATION_SCREENSHOTS,'replay-failure-details.png'),fullPage:true});
     await page.locator('.run-drawer-tabs [data-run-view="messages"]').click();
     await page.locator('.bf-message-scroll').waitFor();
     await page.screenshot({path:path.join(process.env.ABLATION_SCREENSHOTS,'stacked-replay-drawers.png'),fullPage:true});
   }
   assert.deepEqual(mutations,[]);assert.deepEqual(externalRequests,[]);assert.deepEqual(errors,[]);
-  console.log(JSON.stringify({latestPolling:true,historicalInspection:true,draftsPreserved:true,failureReasons:true,physicalPageBookmarks:true,normalMotion:true,transitionCloneSafe:true,stableContentCoverage:true,rapidSwitchClose:true,reducedMotion:true,allTwelveBookmarks:true,fullWidthContent:true,compact631:true,readingPosition:true,keyboardClose:true,mobile:true,pageErrors:errors}));
+  console.log(JSON.stringify({latestPolling:true,historicalInspection:true,draftsPreserved:true,failureReasons:true,physicalPageBookmarks:true,normalMotion:true,fixedBookmarkPositions:true,opaqueHover:true,rapidSwitchClose:true,reducedMotion:true,allTwelveBookmarks:true,allThirtyBookmarks:true,fullWidthContent:true,compact631:true,readingPosition:true,keyboardClose:true,mobile:true,pageErrors:errors}));
 }
 main().catch(error=>{console.error(error);process.exitCode=1;}).finally(async()=>{
   if(browser)await browser.close();

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/tests/test_cli.py

@@ -60,6 +60,13 @@ class CLITests(unittest.TestCase):
             self.assertNotIn("sk-fake-secret", error.getvalue())
             self.assertFalse(Path(directory, "config.json").exists())
 
+    def test_gui_port_defaults_and_explicit_override(self):
+        with tempfile.TemporaryDirectory() as directory, patch("ablation_workbench.server.serve") as serve:
+            for options, port in (([], None), (["--port", "9000"], 9000), (["--port", "0"], 0)):
+                status, _, error = self.invoke(directory, ["gui", "--no-open", *options])
+                self.assertEqual(status, 0, error)
+                self.assertEqual(serve.call_args.kwargs["port"], port)
+
     def test_generic_profile_mapping_flags_and_local_import(self):
         with tempfile.TemporaryDirectory() as directory:
             status, _, error = self.invoke(directory, ["config", "profile", "source", "--source-base-url", "https://logs.example", "--source-path", "/records/{ref}", "--source-request-pointer", "/payload", "--source-response-pointer", "/result", "--source-fields-json", '{"model":"/deployment"}', "--token-env", "EXAMPLE_SOURCE_TOKEN"])

{ablation_workbench-0.3.0 → ablation_workbench-0.3.2}/tests/test_server.py

@@ -1,5 +1,7 @@
 import json
 import os
+import errno
+import socket
 from pathlib import Path
 import subprocess
 import sys
@@ -46,6 +48,47 @@ class ServerTests(unittest.TestCase):
             with urlopen(req) as res: return res.status, json.load(res)
         except HTTPError as exc: return exc.code, json.load(exc)
 
+    def test_default_port_is_stable_and_falls_back_only_when_occupied(self):
+        with socket.socket() as occupied:
+            occupied.bind(("127.0.0.1", 0))
+            port = occupied.getsockname()[1]
+            with patch("ablation_workbench.server.DEFAULT_PORT", port):
+                fallback = make_server(self.workbench, port=None)
+                try:
+                    self.assertGreater(fallback.server_port, port)
+                finally:
+                    fallback.server_close()
+                with self.assertRaises(OSError) as error:
+                    make_server(self.workbench, port=port)
+                self.assertEqual(error.exception.errno, errno.EADDRINUSE)
+        with patch("ablation_workbench.server.DEFAULT_PORT", port):
+            for _ in range(2):
+                server = make_server(self.workbench, port=None)
+                self.assertEqual(server.server_port, port)
+                server.server_close()
+
+    def test_default_port_does_not_hide_other_bind_errors(self):
+        with patch("ablation_workbench.server.ThreadingHTTPServer", side_effect=OSError(errno.EACCES, "denied")) as bind:
+            with self.assertRaises(OSError):
+                make_server(self.workbench, port=None)
+            self.assertEqual(bind.call_count, 1)
+
+    def test_ablation_entry_assets_and_api(self):
+        for entry in ("/", "/ablation", "/ablation/"):
+            with urlopen(self.url + entry + "?experiment=saved") as response:
+                self.assertEqual(response.url, self.url + "/ablation/?experiment=saved")
+                html = response.read().decode()
+                self.assertIn('src="/ablation/app.js"', html)
+                self.assertIn('href="/ablation/"', html)
+        with urlopen(self.url + "/ablation/app.js") as response:
+            self.assertIn(b"fetch('/ablation' + path", response.read())
+        self.assertEqual(self.request("/ablation/api/version"), (200, {"version": __version__}))
+        status, experiment = self.request("/ablation/api/demo", {})
+        self.assertEqual(status, 200)
+        self.assertEqual(self.request("/ablation/api/experiments/" + experiment["id"])[0], 200)
+        self.assertEqual(self.request("/ablation/api/demo", {}, {"Origin": "https://other.example"})[0], 403)
+        self.assertEqual(self.request("/ablation-other/api/version")[0], 404)
+
     def test_gui_serves_bundled_theme_without_external_asset_requests(self):
         for asset, content_type in (("favicon.svg", "image/svg+xml"), ("voyage.css", "text/css"), ("theme.js", "javascript"), ("voyage-toolbar.js", "javascript"), ("messages.js", "javascript"), ("messages.css", "text/css"), ("run-results.js", "javascript"), ("bifrost-LICENSE.txt", "text/plain"), ("bifrost-NOTICE.txt", "text/plain")):
             with self.subTest(asset=asset), urlopen(self.url + "/" + asset) as response:
@@ -97,7 +140,7 @@ class ServerTests(unittest.TestCase):
             fetch.assert_not_called()
 
     def test_sse_sends_update_hint_and_closes_on_shutdown(self):
-        with urlopen(self.url + "/api/events", timeout=3) as response:
+        with urlopen(self.url + "/ablation/api/events", timeout=3) as response:
             self.assertIn("text/event-stream", response.headers["Content-Type"])
             self.assertEqual(response.readline(), b"retry: 1000\n")
             self.assertEqual(response.readline(), b"\n")