feat: document agent calibration, structure anchoring, and probe metrics - #257
Merged
Conversation
- Added support for tracking invisible text length in document agent features. - Updated the `probe_page_features` and related functions to include metrics for both visible and invisible text. - Improved documentation for clarity on text rendering modes and their impact on text metrics. This change enhances the accuracy of text extraction and analysis in document processing.
- Deleted the `inspect.pages` tool from the document agent, streamlining the toolset. - Updated prompts and decision-making logic to reflect the removal of the inspect action. - Adjusted the planner and executor to handle legacy actions appropriately, ensuring compatibility with the new structure. - Enhanced the sampling logic in the planner to include optional random page selection when extrema are not available. This change simplifies the document profiling process and improves overall clarity in tool usage.
- Replaced the offset calibration method in `propose_shard_plan` with a new implementation from `structure_anchoring`. - Consolidated TOC node extraction and hierarchy handling by integrating new functions for anchoring and offset calibration. - Removed obsolete pruning logic for out-of-scope nodes, streamlining the skeleton extraction process. - Enhanced the handling of TOC ranges and improved the overall clarity of the document agent's structure. These changes improve the maintainability and efficiency of the document agent's processing logic.
- Updated budget parameters in `.env.example` files for both API and worker to support increased visual and calibration budgets. - Introduced a new calibration stage in the `BudgetStage` and updated the `BudgetTracker` to accommodate the new calibration budget logic. - Enhanced the `calibration` phase in various modules, including `loop.py`, `procedure.py`, and `service.py`, to improve the handling of offsets and visual stages. - Refactored the `inspect_pages` tool to clarify budget limits and ensure proper handling of visual stages during inspections. These changes improve the efficiency and accuracy of the calibration process within the document agent, allowing for better handling of visual budgets and offsets.
Co-authored-by: Cursor <cursoragent@cursor.com>
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
Comment on lines
+17
to
+20
| from app.services.document_agent.agents.calibration.procedure import ( | ||
| build_calibration_payload, | ||
| finalize_calibration_result, | ||
| ) |
Comment on lines
+28
to
+31
| from app.services.document_agent.structure.structure_anchoring import ( | ||
| deserialize_skeleton_anchor, | ||
| serialize_skeleton_anchor, | ||
| ) |
Comment on lines
+34
to
+40
| from app.services.document_agent.structure.structure_anchoring import ( | ||
| SkeletonAnchor, | ||
| locate_null_page_parent_overrides, | ||
| offset_guided_anchoring, | ||
| prune_unanchored_toc_leaves, | ||
| serialize_skeleton_anchor, | ||
| ) |
Comment on lines
+301
to
+303
| from app.services.document_agent.structure.structure_anchoring import ( | ||
| bulk_offset_matches, | ||
| ) |
|
|
||
| from loguru import logger | ||
|
|
||
| from app.services.document_agent.agents.calibration.loop import run_calibration_phase1 |
Comment on lines
+813
to
+816
| from app.services.document_agent.agents.calibration.procedure import ( | ||
| finalize_calibration_result, | ||
| flat_toc_entries, | ||
| ) |
Comment on lines
+817
to
+819
| from app.services.document_agent.agents.calibration.service import ( | ||
| calibrate_offset, | ||
| ) |
Drop arithmetic offset fallback and legacy page_locate budget so shard planning only uses VLM-calibrated offsets, and reuse one collapse helper across calibration and page_memory. Co-authored-by: Cursor <cursoragent@cursor.com>
Avoid early Timestamp/NaT stringification in postprocess_tb so sparse datetime columns still render clean HTML without NaT leakage. Co-authored-by: Cursor <cursoragent@cursor.com>
Patch verify on the live anchoring globals and drop brittle SkeletonAnchor isinstance checks so contract autouse rebinds cannot miss the mock. Co-authored-by: Cursor <cursoragent@cursor.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
mainand sync document-agent work: text probe metrics, removeinspect.pagesfrom the agent tool surface, structure anchoring/TOC handling, and calibration stage/budgeting.Test plan
Made with Cursor