PDF Viewer Component for Sustainability Reports

[suung]

Background

We want to not only provide an end-to-end analysis but also support analysts in the process of reading through the report and verifying the answers.

Goals

Analysts can can see the report and navigate through it

Acceptance Criteria

Implementation

PDF Component in #21 depents on #13

We want to have the PDF component from main and then merge json schema component later

• Make a new branch from main
• Cherry-pick or move all relevant files from pdf-component into that new branch

PDF Viewer — BDD acceptance criteria

Feature: Verify analysis against the source PDF

• Scenario: Analyst opens the PDF viewer after analysis
  Given an analysis has completed for a report PDF with chunk results
  When the analyst expands the PDF viewer section
  Then the source PDF is shown next to the related chunks for that analysis

• Scenario: Analyst focuses on one question
  Given the PDF viewer is open with results for multiple questions
  When the analyst selects a single question
  Then only chunks for that question are listed and highlighted on the PDF

• Scenario: Analyst views all questions again
  Given the PDF viewer is filtered to one question
  When the analyst selects all questions
  Then chunks for every question in the analysis are available again

• Scenario: Analyst filters to evidence only
  Given retrieved and evidence chunks are shown for the current question scope
  When the analyst turns on evidence-only
  Then only chunks the model actually used as evidence remain visible and highlighted

• Scenario: Analyst turns evidence-only off
  Given evidence-only is on
  When the analyst turns evidence-only off
  Then retrieved chunks that were not cited as evidence are visible again

• Scenario: Analyst jumps from a chunk to the page
  Given a chunk is listed with a known page in the source PDF
  When the analyst selects that chunk
  Then the viewer navigates to that page and shows the matching highlight

• Scenario: Analyst inspects a highlight
  Given a chunk highlight is visible on the PDF page
  When the analyst hovers that highlight
  Then they can see the scores and order needed to judge the citation

• Scenario: Analyst moves between pages
  Given a multi-page PDF is loaded in the viewer
  When the analyst goes to the next or previous page
  Then the displayed page updates and the page indicator reflects the current position

• Scenario: No chunks match the current filters
  Given the current question and evidence filters match no chunks
  When the viewer updates
  Then the analyst is told there are no chunks to display instead of seeing a broken empty pane

• Scenario: PDF cannot be loaded
  Given the report PDF path or data is unavailable
  When the viewer tries to load the document
  Then the analyst sees a clear loading failure rather than a silent blank viewer

Wireframe

┌─ Streamlit: Analysis results ─────────────────────────────────────────┐ │ … answers / tables … │ │ │ │ ▾ 📄 PDF Viewer with Chunks │ │ ┌─ pdf-viewer-with-chunks ─────────────────────────────────────────┐ │ │ │ │ │ │ │ ┌─ SIDEBAR ─────────────┐ ┌─ VIEWER ─────────────────────────┐ │ │ │ │ │ Chunks by Question │ │ [Previous] Page 12 of 84 [Next] │ │ │ │ │ │ │ │ │ │ │ │ │ │ [ Question ▾ ] │ │ ┌────────────────────────────┐ │ │ │ │ │ │ All Questions │ │ │ │ │ │ │ │ │ │ Q1 How does org… │ │ │ PDF page (PDF.js) │ │ │ │ │ │ │ Q2 What targets… │ │ │ │ │ │ │ │ │ │ │ │ │ ▓▓▓ retrieved highlight │ │ │ │ │ │ │ ☐ Show evidence only │ │ │ ███ evidence highlight │ │ │ │ │ │ │ │ │ │ │ │ │ │ │ │ │ ┌─ chunk card ──────┐ │ │ │ (hover → tooltip) │ │ │ │ │ │ │ │ Chunk 3 [Evidence]│ │ │ │ sim / llm / chunk_order │ │ │ │ │ │ │ │ [Page 12] │ │ │ │ │ │ │ │ │ │ │ │ "…preview text…" │ │ │ └────────────────────────────┘ │ │ │ │ │ │ │ Sim: 0.850 LLM:… │ │ │ │ │ │ │ │ │ └────────────────────┘ │ │ │ │ │ │ │ │ ┌─ chunk card ──────┐ │ │ │ │ │ │ │ │ │ Chunk 7 [Page 14] │ │ │ │ │ │ │ │ │ │ (retrieved only) │ │ │ │ │ │ │ │ │ └────────────────────┘ │ │ │ │ │ │ │ └───────────────────────┘ └──────────────────────────────────┘ │ │ │ └──────────────────────────────────────────────────────────────────┘ │ └───────────────────────────────────────────────────────────────────────┘

[suung]

@mounacheikho-cmd as discussed, lets see what happens in

Scenario: User chose to only chunk a pdf
Given the pdf is only chunked and not mapped
When I go to the pdf screen
-> what should happen when it can't map chunks to questions
-> what should happen when it can't show evidence

Scenario: User mapped questions but didn't run the full analysis
Given there is no "is evidence" mapping
When I go to the pdf screen
It should still work
I should still be able to map the questions
And I should see clearly that i can't select "is evidence" because there is none

Scenario: Some questions are mapped fully, some are not maped
When I go to the pdf screen
I understand which questions are mapped or analysed and which note