Wiki–LLM research pipeline
An audit is only as trustworthy as the process behind it. This is the machine: public sources in on the left, evidence-cited findings out on the right, every claim traceable to where it came from.
From sources to evidence to findings
- Site
- Docs
- GitHub
- Discord
- X / posts
- Search
- Screenshots
-
Collect
Every public surface, captured and dated.
Produces: dated captures -
Normalize
Screenshots and pages become clean, searchable markdown.
Produces: the corpus -
Extract
Pull the claims, contradictions, and open questions out of the noise.
Produces: claim list -
Evidence
Tie each claim to its source and a verdict — verified, contested, unbacked.
Produces: evidence matrix -
Synthesize
The patterns across claims become findings, ranked by risk.
Produces: findings, ranked -
Translate
Findings become the maps, the report, the fix list.
Produces: the deliverables
- Evidence-cited findings ledger
- Trust & confusion maps
How to read it
Left to right, raw public material becomes structured evidence. Each stage produces one artifact the next stage builds on. Nothing reaches a finding without a source attached.
The real Obsidian + LLM vault workflow: sources normalized to markdown, claims held in an evidence matrix. Static schematic of the live system.
Why it matters
The desk's whole promise is "we cite the public record." A promise like that has to be a system, not a habit. This is where the citation comes from — every claim in an audit traces back through this pipeline to a dated capture.
See the method