Paper / graph noteResearch map

Evidence before answers

A fictional paper map showing how a research question becomes a testable, evidence-grounded workflow.

Synthetic design example. Not a real paper; no empirical results are claimed. All section and page references are illustrative.

Evidence connectionInferred direction
01

Research questions

What needs answering?

q-trust

Can a verdict be checked?

A useful answer should reveal which evidence supports it, not merely sound convincing.

Example §1 Motivation · p. 1
Evidence & details
  • This is a fictional research question used to demonstrate the map format.
q-transfer

What changes under drift?

Ask whether the same evidence workflow remains useful when language, topics, and source material change.

Example §1 Questions · p. 1
02

Related work

Where existing work stops

r-prediction

Prediction is not evidence

The toy comparison distinguishes a model's answer from a traceable justification for that answer.

Example §2 Context · p. 2
r-retrieval

Retrieval supplies context

Finding relevant passages creates candidate support, but relevance alone does not establish the conclusion.

Example §2 Context · p. 2
03

Method

How the paper responds

m-retrieve

01 / Retrieve candidates

Turn the question into a small set of candidate passages while preserving their source locations.

Example §3.1 Retrieval · p. 3
Evidence & details
  • Input: a question and a permitted document collection.
  • Output: candidate passages with stable source references.
m-check

02 / Check support

Separate supporting passages, conflicting evidence, and missing information before producing a verdict.

Example §3.2 Verification · p. 3
Evidence & details
  • Retain uncertainty instead of treating absent evidence as a negative finding.
m-answer

03 / Explain the decision

Return a concise answer with its evidence trail and clearly bounded uncertainty.

Example §3.3 Output · p. 4
04

Experiments

What the evidence supports

e-compare

Compare like with like

Illustrative protocol: hold questions fixed and compare the evidence workflow with a retrieval-free baseline.

Example §4 Evaluation plan · p. 5
Evidence & details
  • No real scores are supplied. A real paper map must preserve the authors' metrics, units, splits, and uncertainty.
e-ablate

Remove the support check

Illustrative ablation: remove verification to ask whether it changes citation faithfulness rather than just fluency.

Example §4 Ablation plan · p. 5
05

Future work

What remains open

f-driftInferred

Test changing sources

A proposed next step is to evaluate the workflow when the evidence collection changes over time.

No source location supplied
Evidence & details
  • This is an explicitly inferred proposal, not an author-stated conclusion.
f-conflictInferred

Keep disagreement visible

Explore how conflicting sources should appear in an explanation instead of being silently averaged away.

No source location supplied

Follow the connections. Expand a node for evidence and reading notes. Dashed nodes are interpretations, not reported findings. Hover or focus a node to trace its connections.

Connection index