← Brief
Open SVG
Download PNG
How evidence becomes linked data
An unchanged PDF is parsed into source cells with coordinates. A JSON claim references the cell. A review references the claim. A verified fact retains both. Multiple report answers reference that fact, forming a traceable graph.
EVIDENCE / DATA SHAPES / TRACEABLE RELATIONSHIPS
How evidence becomes linked data
The same electricity-bill example, shown as small records connected by IDs.
PDF
Original document
March bill · original bytes retained
{
"id":
"DOC-17",
"file":
"march-bill.pdf",
"version":
1,
"sha256":
"<content hash>" }
01
Parsing layer
Read text and tables.
Keep the exact source location.
{
"id":
"SPAN-42",
"document_id":
"DOC-17",
"page":
2,
"table":
1,
"row":
8,
"column":
3,
"text":
"4,238 kWh",
"bbox":
[410, 520, 550, 545],
"coordinate_system":
"page points, top-left" }
Page + cell + bounding box make highlighting possible.
02
Extraction layer
Turn the source text into a claim.
Point back to the parsed evidence.
{
"id":
"CLAIM-88",
"evidence_id":
"SPAN-42",
"entity":
"Acme Ltd",
"metric":
"electricity",
"value":
4238,
"unit":
"kWh",
"period":
"2026-03",
"status":
"needs_review" }
A proposed claim, not yet accepted as a verified fact.
03
Source review
Show the source beside the claim.
Record the human decision.
{
"id":
"REVIEW-09",
"claim_id":
"CLAIM-88",
"decision":
"confirmed",
"reviewer_id":
"USER-07",
"checked":
["value", "entity", "period"],
"reviewed_at":
"2026-04-04T10:30:00Z" }
Rejected or corrected claims retain their review history.
04
Knowledge layer
Normalise the accepted value.
Reuse it with its evidence intact.
{
"id":
"FACT-31",
"entity":
"Acme Ltd",
"metric":
"electricity",
"period":
"2026-03",
"value":
4.238,
"unit":
"MWh",
"status":
"verified",
"claim_ids":
["CLAIM-88"],
"review_id":
"REVIEW-09",
"conversion":
"kWh / 1000" }
The fact preserves meaning, unit conversion and approval.
05 / GRAPH CONCEPTS
The IDs form a graph you can follow.
Each answer references a fact; the fact leads back through claims to the original file.
Source document
DOC-17
Evidence span
SPAN-42
Extracted claim
CLAIM-88
Verified fact
FACT-31
Customer questionnaire
{
"report_id":
"CUSTOMER-2026",
"answer_id":
"ENERGY-MARCH",
"fact_ids":
["FACT-31"] }
Sustainability report
{
"report_id":
"VSME-2026",
"answer_id":
"ENERGY-MARCH",
"fact_ids":
["FACT-31"] }
ILLUSTRATIVE SCHEMA · Coordinates and IDs are examples. Annual totals need all monthly inputs.