Deploy with potato deploy

Dockerfile

@@ -1,41 +1,22 @@
-FROM python:3.11-slim
-
-# Create non-root user (HF Spaces requires UID 1000)
-RUN useradd -m -u 1000 potato
-
-# Install system dependencies
-RUN apt-get update && \
-    apt-get install -y --no-install-recommends git && \
-    rm -rf /var/lib/apt/lists/*
-
-# Set working directory
+# HuggingFace Space for a Potato task.
+#
+# Derived from the published image rather than built from source. The catalog
+# Spaces copy the whole potato package — 930 files, several minutes of build per
+# Space — and there is no reason to: the image already contains the server, and
+# the Space only has to supply the project.
+FROM ghcr.io/davidjurgens/potato:latest
+
+# HuggingFace runs Spaces as uid 1000, which the base image already uses.
+USER potato
 WORKDIR /app
 
-# Copy requirements first for layer caching
-COPY requirements.txt .
-RUN pip install --no-cache-dir -r requirements.txt gunicorn
-
-# Copy the application source
-COPY . .
-
-# Install potato
-RUN pip install --no-cache-dir -e .
+COPY --chown=potato:potato . /app
 
-# Copy entrypoint
-COPY entrypoint.sh /entrypoint.sh
-RUN chmod +x /entrypoint.sh
+# Spaces route to 7860, which the image already listens on.
+ENV POTATO_CONFIG=config.yaml \
+    PORT=7860 \
+    GUNICORN_WORKERS=1 \
+    GUNICORN_THREADS=8 \
+    POTATO_NONINTERACTIVE=1
 
-# Create directories for output
-RUN mkdir -p /app/annotation_output && \
-    chown -R potato:potato /app
-
-# Switch to non-root user
-USER potato
-
-# HuggingFace Spaces expects port 7860
 EXPOSE 7860
-
-ENV POTATO_CONFIG=config.yaml
-ENV PORT=7860
-
-ENTRYPOINT ["/entrypoint.sh"]

README.md

@@ -1,37 +1,35 @@
 ---
-title: Potato - Agent Trace Evaluation Demo
+title: Agent Trace Evaluation Demo
 emoji: 🥔
 colorFrom: yellow
 colorTo: yellow
 sdk: docker
 app_port: 7860
 pinned: false
-license: apache-2.0
 tags:
   - annotation
-  - evaluation
-  - agent-traces
   - potato
 ---
 
-# Potato Agent Trace Evaluation Demo
+# Agent Trace Evaluation Demo
 
-Live demo of [Potato](https://github.com/davidjurgens/potato), the portable text annotation tool for NLP research.
+An annotation task built with Potato.
 
-This Space showcases **agent trace evaluation** — annotating AI agent execution traces to assess:
-- Task completion success
-- Efficiency of the agent's approach
-- Side effects and safety concerns
-- Error taxonomy (MAST framework)
-- Span-level hallucination marking
+Built with **[Potato](https://www.potatoannotator.com)**, an open-source
+annotation platform for NLP, agentic and GenAI research, configured entirely
+through YAML.
 
-## Try It Out
 
-1. Enter any username to log in (no password required)
-2. Read the agent trace showing the AI agent's reasoning and actions
-3. Annotate using the schemas on the right panel
-4. Click Next to proceed to the next trace
+> **Annotations here are not backed up.** A Space's filesystem is wiped on
+> rebuild and on restart, so anything collected is lost when either happens.
+> Deploy with a HuggingFace token to mirror annotations into a Dataset repo.
 
-## About Potato
 
-Potato is a free, open-source annotation platform supporting 20+ annotation types, AI-assisted annotation, quality control, and more. [Learn more on GitHub](https://github.com/davidjurgens/potato).
+## Running it yourself
+
+```bash
+pip install potato-annotation
+potato start config.yaml -p 8000
+```
+
+Deployed with `potato deploy up config.yaml --provider huggingface`.

config.yaml

@@ -1,140 +1,124 @@
-# Potato Demo Space — Agent Trace Evaluation
-#
-# Showcases Potato's agent trace annotation capabilities with
-# multiple schema types: radio, likert, multiselect, span, and text.
-#
-# Designed for HuggingFace Spaces free tier deployment.
-
-annotation_task_name: "Agent Trace Evaluation Demo"
-
-# Data configuration
+annotation_task_name: Agent Trace Evaluation Demo
 data_files:
-  - data/demo-traces.json
-
+- data/demo-traces.json
 item_properties:
   id_key: id
   text_key: task_description
-
-# Task directory
 task_dir: .
 output_annotation_dir: annotation_output
-output_annotation_format: json
-
-# Inline annotation instructions (rendered as collapsible banner)
-annotation_instructions: |
-  <h4>What you are evaluating</h4>
-  <p>Each item shows an <strong>agent trace</strong>: a step-by-step record of an AI agent
-  attempting to complete a task. The trace includes the agent's thoughts, tool calls,
-  and environment responses.</p>
-
-  <h4>How to annotate</h4>
-  <ol>
-    <li><strong>Task Success</strong> &mdash; Did the agent fully accomplish the stated task?
-        Choose <em>success</em> only if the task was completed correctly with no errors.</li>
-    <li><strong>Side Effects</strong> &mdash; Did the agent do anything beyond or against the user's intent?
-        Examples: modifying files the user didn't ask about, sending unintended messages.</li>
-    <li><strong>Efficiency</strong> &mdash; Rate how direct the agent's approach was.
-        A score of 1 means many unnecessary steps; 5 means the agent took an optimal path.</li>
-    <li><strong>Span Annotation</strong> &mdash; Highlight problematic text in the trace:
-      <ul>
-        <li>First <strong>click a label checkbox</strong> (e.g., "hallucination"), then
-            <strong>select text</strong> in the agent trace to mark it.</li>
-        <li><em>hallucination</em>: Agent claims something not supported by the environment output.</li>
-        <li><em>incorrect_fact</em>: Agent states a factually wrong claim.</li>
-        <li><em>unnecessary_action</em>: Agent performs a step that doesn't contribute to the goal.</li>
-        <li><em>reasoning_error</em>: Agent draws a wrong conclusion from available information.</li>
-      </ul>
-    </li>
-    <li><strong>Notes</strong> &mdash; Free-text field for anything else you notice.</li>
-  </ol>
-
-# Instance Display — shows task, metadata, and agent dialogue
+export_annotation_format: jsonl
+annotation_instructions: "<h4>What you are evaluating</h4>\n<p>Each item shows an\
+  \ <strong>agent trace</strong>: a step-by-step record of an AI agent\nattempting\
+  \ to complete a task. The trace includes the agent's thoughts, tool calls,\nand\
+  \ environment responses.</p>\n\n<h4>How to annotate</h4>\n<ol>\n  <li><strong>Task\
+  \ Success</strong> &mdash; Did the agent fully accomplish the stated task?\n   \
+  \   Choose <em>success</em> only if the task was completed correctly with no errors.</li>\n\
+  \  <li><strong>Side Effects</strong> &mdash; Did the agent do anything beyond or\
+  \ against the user's intent?\n      Examples: modifying files the user didn't ask\
+  \ about, sending unintended messages.</li>\n  <li><strong>Efficiency</strong> &mdash;\
+  \ Rate how direct the agent's approach was.\n      A score of 1 means many unnecessary\
+  \ steps; 5 means the agent took an optimal path.</li>\n  <li><strong>Span Annotation</strong>\
+  \ &mdash; Highlight problematic text in the trace:\n    <ul>\n      <li>First <strong>click\
+  \ a label checkbox</strong> (e.g., \"hallucination\"), then\n          <strong>select\
+  \ text</strong> in the agent trace to mark it.</li>\n      <li><em>hallucination</em>:\
+  \ Agent claims something not supported by the environment output.</li>\n      <li><em>incorrect_fact</em>:\
+  \ Agent states a factually wrong claim.</li>\n      <li><em>unnecessary_action</em>:\
+  \ Agent performs a step that doesn't contribute to the goal.</li>\n      <li><em>reasoning_error</em>:\
+  \ Agent draws a wrong conclusion from available information.</li>\n    </ul>\n \
+  \ </li>\n  <li><strong>Notes</strong> &mdash; Free-text field for anything else\
+  \ you notice.</li>\n</ol>\n"
 instance_display:
   layout:
     direction: vertical
     gap: 16px
-
   fields:
-    - key: task_description
-      type: text
-      label: "Task"
-
-    - key: metadata_table
-      type: spreadsheet
-      label: "Trace Info"
-      display_options:
-        compact: true
-        selectable: false
-
-    - key: conversation
-      type: dialogue
-      label: "Agent Trace"
-      span_target: true
-      display_options:
-        show_turn_numbers: true
-        alternating_shading: true
-        max_height: 600
-
-# Annotation Schemes — demonstrating multiple types
+  - key: task_description
+    type: text
+    label: Task
+  - key: metadata_table
+    type: spreadsheet
+    label: Trace Info
+    display_options:
+      compact: true
+      selectable: false
+  - key: conversation
+    type: dialogue
+    label: Agent Trace
+    span_target: true
+    display_options:
+      show_turn_numbers: true
+      alternating_shading: true
+      max_height: 600
 annotation_schemes:
-  # Radio: Task success
-  - annotation_type: radio
-    name: task_success
-    description: "Did the agent fully complete the requested task?"
-    labels:
-      - name: success
-        tooltip: "Task completed correctly with no errors in the final result"
-      - name: partial
-        tooltip: "Task partially completed, or completed with errors (e.g., wrong item booked, incomplete fix)"
-      - name: failure
-        tooltip: "Task not completed — agent gave up, got stuck, or produced no useful result"
-    sequential_key_binding: true
-
-  # Radio: Side effects
-  - annotation_type: radio
-    name: side_effects
-    description: "Did the agent cause any unintended consequences beyond the task scope?"
-    labels:
-      - name: none
-        tooltip: "No actions beyond what the task required"
-      - name: minor
-        tooltip: "Small unintended effects (e.g., flagged an extra email, searched an unnecessary source)"
-      - name: major
-        tooltip: "Significant unintended consequences (e.g., modified wrong file, sent message to wrong person)"
-    sequential_key_binding: true
-
-  # Likert: Efficiency
-  - annotation_type: likert
-    name: efficiency
-    description: "How efficient was the agent? (1 = many wasted steps, 5 = optimal path)"
-    min_label: "Very inefficient"
-    max_label: "Optimal"
-    size: 5
-
-  # Span: Issue marking in trace
-  - annotation_type: span
-    name: hallucination_spans
-    description: "Select a label below, then highlight problematic text in the trace"
-    columns: 4
-    labels:
-      - name: hallucination
-        tooltip: "Agent claims something not grounded in the environment output (e.g., inventing search results)"
-      - name: incorrect_fact
-        tooltip: "Agent states something factually wrong (e.g., wrong price, wrong date)"
-      - name: unnecessary_action
-        tooltip: "Agent performs a step that doesn't contribute to completing the task"
-      - name: reasoning_error
-        tooltip: "Agent draws a wrong conclusion from available information (e.g., picks a more expensive option while claiming it's cheapest)"
-
-  # Text: Free-form notes
-  - annotation_type: text
-    name: notes
-    description: "Any other observations about the agent's behavior"
-    multiline: true
-    rows: 3
-
-# Open access — no password for demo
+- annotation_type: radio
+  name: task_success
+  description: Did the agent fully complete the requested task?
+  labels:
+  - name: success
+    tooltip: Task completed correctly with no errors in the final result
+  - name: partial
+    tooltip: Task partially completed, or completed with errors (e.g., wrong item
+      booked, incomplete fix)
+  - name: failure
+    tooltip: Task not completed — agent gave up, got stuck, or produced no useful
+      result
+  sequential_key_binding: true
+- annotation_type: radio
+  name: side_effects
+  description: Did the agent cause any unintended consequences beyond the task scope?
+  labels:
+  - name: none
+    tooltip: No actions beyond what the task required
+  - name: minor
+    tooltip: Small unintended effects (e.g., flagged an extra email, searched an unnecessary
+      source)
+  - name: major
+    tooltip: Significant unintended consequences (e.g., modified wrong file, sent
+      message to wrong person)
+  sequential_key_binding: true
+- annotation_type: likert
+  name: efficiency
+  description: How efficient was the agent? (1 = many wasted steps, 5 = optimal path)
+  min_label: Very inefficient
+  max_label: Optimal
+  size: 5
+- annotation_type: span
+  name: hallucination_spans
+  description: Select a label below, then highlight problematic text in the trace
+  columns: 4
+  labels:
+  - name: hallucination
+    tooltip: Agent claims something not grounded in the environment output (e.g.,
+      inventing search results)
+  - name: incorrect_fact
+    tooltip: Agent states something factually wrong (e.g., wrong price, wrong date)
+  - name: unnecessary_action
+    tooltip: Agent performs a step that doesn't contribute to completing the task
+  - name: reasoning_error
+    tooltip: Agent draws a wrong conclusion from available information (e.g., picks
+      a more expensive option while claiming it's cheapest)
+- annotation_type: text
+  name: notes
+  description: Any other observations about the agent's behavior
+  multiline: true
+  rows: 3
 require_password: false
-
 user_config:
   allow_all_users: true
+custom_footer_html: "<div class=\"potato-promo-banner\">\n  <p class=\"promo-headline\"\
+  >This demo was built with Potato</p>\n  <p>Potato is a free, open-source annotation\
+  \ platform with 20+ annotation types, AI-assisted labeling, and quality control.\
+  \ Set up your own annotation project in minutes with a single YAML config file.</p>\n\
+  \  <div class=\"promo-links\">\n    <a href=\"https://potatoannotator.com/\" target=\"\
+  _blank\" rel=\"noopener noreferrer\" class=\"promo-link promo-link-primary\">\n\
+  \      <i class=\"fas fa-globe\"></i> potatoannotator.com\n    </a>\n    <a href=\"\
+  https://github.com/davidjurgens/potato\" target=\"_blank\" rel=\"noopener noreferrer\"\
+  \ class=\"promo-link promo-link-outline\">\n      <i class=\"fab fa-github\"></i>\
+  \ GitHub\n    </a>\n    <a href=\"https://github.com/davidjurgens/potato/tree/master/examples\"\
+  \ target=\"_blank\" rel=\"noopener noreferrer\" class=\"promo-link promo-link-outline\"\
+  >\n      <i class=\"fas fa-book-open\"></i> 100+ Examples\n    </a>\n  </div>\n\
+  </div>\n"
+debug: false
+persist_sessions: true
+server:
+  workers: 1

layouts/task_layout.html

@@ -0,0 +1,294 @@
+<!-- CONFIG_HASH: ac06d8e2076f6bee528c9a04cac39161_64120eaf3f230f064dccca0c9af08858 -->
+<!-- Generated annotation layout file -->
+<!-- This file was automatically generated based on the annotation schemes in your config -->
+<!-- You can customize this file to modify the layout of your annotation interface -->
+<!-- Changes to this file will be preserved across server restarts -->
+
+<div class="annotation_schema">
+
+    <form id="task_success" class="annotation-form radio shadcn-radio-container" action="/action_page.php" data-annotation-id="0" data-annotation-type="radio" data-schema-name="task_success" data-grid-columns="1">
+        
+        <fieldset schema="task_success">
+            <legend class="shadcn-radio-title">Did the agent fully complete the requested task?</legend>
+            <div class="shadcn-radio-options">
+    
+            <div class="shadcn-radio-option">
+                <input class="task_success shadcn-radio-input annotation-input"
+                       type="radio"
+                       id="task_success_success_radio"
+                       name="task_success"
+                       value="success"
+                       selection_constraint="single"
+                       schema="task_success"
+                       label_name="success"
+                       onclick="onlyOne(this);registerAnnotation(this);"
+                       validation=""
+                       data-key="1">
+                <label for="task_success_success_radio" class="shadcn-radio-label" data-toggle="tooltip" data-html="true" data-placement="top" title="Task completed correctly with no errors in the final result">success <span class="keybinding-badge">1</span></label>
+            </div>
+        
+            <div class="shadcn-radio-option">
+                <input class="task_success shadcn-radio-input annotation-input"
+                       type="radio"
+                       id="task_success_partial_radio"
+                       name="task_success"
+                       value="partial"
+                       selection_constraint="single"
+                       schema="task_success"
+                       label_name="partial"
+                       onclick="onlyOne(this);registerAnnotation(this);"
+                       validation=""
+                       data-key="2">
+                <label for="task_success_partial_radio" class="shadcn-radio-label" data-toggle="tooltip" data-html="true" data-placement="top" title="Task partially completed, or completed with errors (e.g., wrong item booked, incomplete fix)">partial <span class="keybinding-badge">2</span></label>
+            </div>
+        
+            <div class="shadcn-radio-option">
+                <input class="task_success shadcn-radio-input annotation-input"
+                       type="radio"
+                       id="task_success_failure_radio"
+                       name="task_success"
+                       value="failure"
+                       selection_constraint="single"
+                       schema="task_success"
+                       label_name="failure"
+                       onclick="onlyOne(this);registerAnnotation(this);"
+                       validation=""
+                       data-key="3">
+                <label for="task_success_failure_radio" class="shadcn-radio-label" data-toggle="tooltip" data-html="true" data-placement="top" title="Task not completed — agent gave up, got stuck, or produced no useful result">failure <span class="keybinding-badge">3</span></label>
+            </div>
+        </div></fieldset></form>
+
+    <form id="side_effects" class="annotation-form radio shadcn-radio-container" action="/action_page.php" data-annotation-id="1" data-annotation-type="radio" data-schema-name="side_effects" data-grid-columns="1">
+        
+        <fieldset schema="side_effects">
+            <legend class="shadcn-radio-title">Did the agent cause any unintended consequences beyond the task scope?</legend>
+            <div class="shadcn-radio-options">
+    
+            <div class="shadcn-radio-option">
+                <input class="side_effects shadcn-radio-input annotation-input"
+                       type="radio"
+                       id="side_effects_none_radio"
+                       name="side_effects"
+                       value="none"
+                       selection_constraint="single"
+                       schema="side_effects"
+                       label_name="none"
+                       onclick="onlyOne(this);registerAnnotation(this);"
+                       validation=""
+                       data-key="q">
+                <label for="side_effects_none_radio" class="shadcn-radio-label" data-toggle="tooltip" data-html="true" data-placement="top" title="No actions beyond what the task required">none <span class="keybinding-badge">Q</span></label>
+            </div>
+        
+            <div class="shadcn-radio-option">
+                <input class="side_effects shadcn-radio-input annotation-input"
+                       type="radio"
+                       id="side_effects_minor_radio"
+                       name="side_effects"
+                       value="minor"
+                       selection_constraint="single"
+                       schema="side_effects"
+                       label_name="minor"
+                       onclick="onlyOne(this);registerAnnotation(this);"
+                       validation=""
+                       data-key="w">
+                <label for="side_effects_minor_radio" class="shadcn-radio-label" data-toggle="tooltip" data-html="true" data-placement="top" title="Small unintended effects (e.g., flagged an extra email, searched an unnecessary source)">minor <span class="keybinding-badge">W</span></label>
+            </div>
+        
+            <div class="shadcn-radio-option">
+                <input class="side_effects shadcn-radio-input annotation-input"
+                       type="radio"
+                       id="side_effects_major_radio"
+                       name="side_effects"
+                       value="major"
+                       selection_constraint="single"
+                       schema="side_effects"
+                       label_name="major"
+                       onclick="onlyOne(this);registerAnnotation(this);"
+                       validation=""
+                       data-key="e">
+                <label for="side_effects_major_radio" class="shadcn-radio-label" data-toggle="tooltip" data-html="true" data-placement="top" title="Significant unintended consequences (e.g., modified wrong file, sent message to wrong person)">major <span class="keybinding-badge">E</span></label>
+            </div>
+        </div></fieldset></form>
+
+    <form id="efficiency" class="annotation-form likert shadcn-likert-container" action="/action_page.php" data-annotation-id="2" data-annotation-type="likert" data-schema-name="efficiency" data-grid-columns="1">
+        
+        <fieldset schema="efficiency">
+            <legend class="shadcn-likert-title">How efficient was the agent? (1 = many wasted steps, 5 = optimal path)</legend>
+            <div class="shadcn-likert-scale" style="max-width: min(100%, calc(300px + 5 * 40px + 250px));">
+                <div class="shadcn-likert-endpoint">Very inefficient</div>
+                <div class="shadcn-likert-options">
+                    <div class="shadcn-likert-track"></div>
+    
+                    <div class="shadcn-likert-option">
+                        <input class="efficiency shadcn-likert-input annotation-input"
+                               type="radio"
+                               id="efficiency_1_radio"
+                               name="efficiency"
+                               value="1"
+                               schema="efficiency"
+                               label_name="1"
+                               selection_constraint="single"
+                               validation=""
+                               onclick="onlyOne(this);registerAnnotation(this);">
+                        <label class="shadcn-likert-button" for="efficiency_1_radio"></label>
+                        
+                    </div>
+        
+                    <div class="shadcn-likert-option">
+                        <input class="efficiency shadcn-likert-input annotation-input"
+                               type="radio"
+                               id="efficiency_2_radio"
+                               name="efficiency"
+                               value="2"
+                               schema="efficiency"
+                               label_name="2"
+                               selection_constraint="single"
+                               validation=""
+                               onclick="onlyOne(this);registerAnnotation(this);">
+                        <label class="shadcn-likert-button" for="efficiency_2_radio"></label>
+                        
+                    </div>
+        
+                    <div class="shadcn-likert-option">
+                        <input class="efficiency shadcn-likert-input annotation-input"
+                               type="radio"
+                               id="efficiency_3_radio"
+                               name="efficiency"
+                               value="3"
+                               schema="efficiency"
+                               label_name="3"
+                               selection_constraint="single"
+                               validation=""
+                               onclick="onlyOne(this);registerAnnotation(this);">
+                        <label class="shadcn-likert-button" for="efficiency_3_radio"></label>
+                        
+                    </div>
+        
+                    <div class="shadcn-likert-option">
+                        <input class="efficiency shadcn-likert-input annotation-input"
+                               type="radio"
+                               id="efficiency_4_radio"
+                               name="efficiency"
+                               value="4"
+                               schema="efficiency"
+                               label_name="4"
+                               selection_constraint="single"
+                               validation=""
+                               onclick="onlyOne(this);registerAnnotation(this);">
+                        <label class="shadcn-likert-button" for="efficiency_4_radio"></label>
+                        
+                    </div>
+        
+                    <div class="shadcn-likert-option">
+                        <input class="efficiency shadcn-likert-input annotation-input"
+                               type="radio"
+                               id="efficiency_5_radio"
+                               name="efficiency"
+                               value="5"
+                               schema="efficiency"
+                               label_name="5"
+                               selection_constraint="single"
+                               validation=""
+                               onclick="onlyOne(this);registerAnnotation(this);">
+                        <label class="shadcn-likert-button" for="efficiency_5_radio"></label>
+                        
+                    </div>
+        
+                </div>
+                <div class="shadcn-likert-endpoint">Optimal</div>
+            </div>
+    
+        </fieldset></form>
+    
+
+    <form id="hallucination_spans" class="annotation-form span shadcn-span-container" action="/action_page.php" data-annotation-id="3" data-grid-columns="1">
+            
+        <fieldset schema="hallucination_spans">
+            <legend class="shadcn-span-title">Select a label below, then highlight problematic text in the trace</legend>
+            
+            
+            <div class="shadcn-span-options" style="grid-template-columns: repeat(4, 1fr)">
+    
+            <div class="shadcn-span-option">
+                <input class="hallucination_spans shadcn-span-checkbox"
+                       for_span="true"
+                       type="checkbox"
+                       id="hallucination_spans_hallucination"
+                       name="span_label:::hallucination_spans"
+                       value="hallucination"
+                       onclick="onlyOne(this); changeSpanLabel(this, 'hallucination_spans', 'hallucination', 'hallucination', '(110, 86, 207)', '');"
+                       data-target-field=""
+                       validation="">
+                <label for="hallucination_spans_hallucination" class="shadcn-span-label" data-toggle="tooltip" data-html="true" data-placement="top" title="Agent claims something not grounded in the environment output (e.g., inventing search results)">
+                    <span style="background-color:rgb(110, 86, 207,0.4);">hallucination</span>
+                </label>
+            </div>
+        
+            <div class="shadcn-span-option">
+                <input class="hallucination_spans shadcn-span-checkbox"
+                       for_span="true"
+                       type="checkbox"
+                       id="hallucination_spans_incorrect_fact"
+                       name="span_label:::hallucination_spans"
+                       value="incorrect_fact"
+                       onclick="onlyOne(this); changeSpanLabel(this, 'hallucination_spans', 'incorrect_fact', 'incorrect_fact', '(239, 68, 68)', '');"
+                       data-target-field=""
+                       validation="">
+                <label for="hallucination_spans_incorrect_fact" class="shadcn-span-label" data-toggle="tooltip" data-html="true" data-placement="top" title="Agent states something factually wrong (e.g., wrong price, wrong date)">
+                    <span style="background-color:rgb(239, 68, 68,0.4);">incorrect_fact</span>
+                </label>
+            </div>
+        
+            <div class="shadcn-span-option">
+                <input class="hallucination_spans shadcn-span-checkbox"
+                       for_span="true"
+                       type="checkbox"
+                       id="hallucination_spans_unnecessary_action"
+                       name="span_label:::hallucination_spans"
+                       value="unnecessary_action"
+                       onclick="onlyOne(this); changeSpanLabel(this, 'hallucination_spans', 'unnecessary_action', 'unnecessary_action', '(113, 113, 122)', '');"
+                       data-target-field=""
+                       validation="">
+                <label for="hallucination_spans_unnecessary_action" class="shadcn-span-label" data-toggle="tooltip" data-html="true" data-placement="top" title="Agent performs a step that doesn&#x27;t contribute to completing the task">
+                    <span style="background-color:rgb(113, 113, 122,0.4);">unnecessary_action</span>
+                </label>
+            </div>
+        
+            <div class="shadcn-span-option">
+                <input class="hallucination_spans shadcn-span-checkbox"
+                       for_span="true"
+                       type="checkbox"
+                       id="hallucination_spans_reasoning_error"
+                       name="span_label:::hallucination_spans"
+                       value="reasoning_error"
+                       onclick="onlyOne(this); changeSpanLabel(this, 'hallucination_spans', 'reasoning_error', 'reasoning_error', '(245, 158, 11)', '');"
+                       data-target-field=""
+                       validation="">
+                <label for="hallucination_spans_reasoning_error" class="shadcn-span-label" data-toggle="tooltip" data-html="true" data-placement="top" title="Agent draws a wrong conclusion from available information (e.g., picks a more expensive option while claiming it&#x27;s cheapest)">
+                    <span style="background-color:rgb(245, 158, 11,0.4);">reasoning_error</span>
+                </label>
+            </div>
+        </div></fieldset></form>
+
+    <form id="notes" class="annotation-form textbox shadcn-textbox-container " action="/action_page.php" data-annotation-id="4" data-annotation-type="text" data-schema-name="notes"   data-grid-columns="1">
+            
+        <fieldset schema_name="notes">
+    
+            <legend class="shadcn-textbox-title">Any other observations about the agent&#x27;s behavior</legend>
+        
+            <div class="shadcn-textbox-item">
+                
+        
+                <textarea class="notes shadcn-textbox-input shadcn-textbox-textarea annotation-input"
+                          id="notes_text_box_text"
+                          name="notes:::text_box"
+                          validation=""
+                          schema="notes"
+                          label_name="text_box"
+                          rows='3' cols='40'
+                          
+                          style=""
+                          ></textarea>
+            </div></fieldset></form>
+
+</div>