Guide

From blank sector to publishable thesis.

Nonconsensus is not chat-first. It is evidence-first: frame the market, compound the corpus, generate a falsifiable memo, then go find what would kill it.

The loop

Frame a sector

Make the question narrow enough to be wrong: market, geography, horizon, lens, and your edge.

Source evidence

Auto-source the public web, paste private notes, or upload docs; every claim becomes corpus memory.

Generate

Thesis and discovery plans are built from an evidence snapshot, not chat vibes.

Publish or kill

Share the memo if it survives, or use the unknowns and kill criteria to invalidate it fast.

Framing a sector

Specific beats broad. A useful frame names the arena, the constraint, the time horizon, and the kind of winner you are looking for.

The transport bottleneck: machine data vs backhaul capacity in autonomous aerospace · US/global · 5yr
AI-era productivity platforms: who owns the workflow layer after agents mature? · US/global · 5yr
Robotics data layer: can proprietary embodied-world data become a defensible platform? · China/global · 3yr

Evidence

Auto-source

The underlying AI searches the web server-side, validates returned URLs, and stores cited evidence only.

Paste

Use transcripts, filings, expert-call notes, customer interviews, or any text you already trust.

Upload

PDF, DOCX, TXT, and MD become private document sources. The original file is not retained in v1.

Consensus / Δ / Unknown

Consensus is the market story. Δ is where the story is wrong, early, or mismeasured. Unknown is the gap that still gates conviction.

Anti-slop guarantee

Unsupported claims are downgraded to LOW with an unknown instead of being polished into fake certainty.

The memo

Differential insight

The one-line bet.

Consensus → Δ

What others believe, and where the memo departs.

Why-now

The timing argument.

Binding constraint

Technology, distribution, regulation, trust, capital, or workflow inertia.

Wedge

The entry point that can actually be tested.

72-hour MVP spec

The smallest buildable proof.

Fundability

Power-law venture outcome or honest cash business.

Biggest UNKNOWN

The thing that can still break the thesis.

Charts

Evidence-derived visuals appear when the corpus supports them; empty charts are skipped.

The method

Nonconsensus is the reference implementation of The Δ Protocol. A claim earns space only when it passes five gates:

Divergent

Name the consensus before claiming the Δ.

Evidenced

Tie the claim to sources or mark it unknown.

Material

Keep only deltas that change a decision.

Falsifiable

State what would break the thesis.

Timed

Put the claim on a horizon so it can resolve.

Read The Δ Protocol →

Publish & share

Publishing creates a public memo URL with the evidence count, share buttons for X/LinkedIn/WhatsApp, a 3:4 share card, and an OG preview. WeChat and Xiaohongshu are image-first: use the QR/share card path.

FAQ

Is my work private?

Yes by default. Public links only exist after you publish a memo. Private-document claims can inform a thesis, but private-doc quotes are withheld on public pages. We collect usage signals (like which features you use) to improve the product. Your research content is never used to train models unless you explicitly opt in; private documents never are.

Can I delete things?

Evidence rows can be removed from the Evidence tab. Reports are append-only so prior memo versions remain auditable.

What if generation fails?

Nothing is overwritten. Try again, use fewer or lighter sources, or refresh evidence if the underlying AI is temporarily overloaded.

Can someone fork a public memo?

The public page is read-only today. A reader can open Nonconsensus and build a counter-thesis in their own workspace.

Where does evidence come from?

Real web results, pasted text, and uploaded private documents. Auto-source URLs are validated before persistence.