Robust Reasoning Processes

The field in one paragraph

Robust reasoning processes (RRP) is the study and engineering of procedures that turn effort and evidence into trustworthy conclusions — and keep doing so when interested parties try to corrupt them. Its objects are processes, not people or models: peer review, prediction markets, audits, expert panels, LLM judge pipelines, debate protocols. Its central activity is measurement: what does a process deliver, what does it cost, and what does it cost to corrupt? Civilization runs on reasoning processes that were never benchmarked; AI has just made them studyable — an LLM-based process can be re-run, perturbed, versioned, and attacked on purpose — and made the problem urgent, because the cost of producing plausible reasoning has collapsed while the cost of verifying it has not.

The map

Every process gets two headline numbers: the cost of extracting validated information from it, and the cost an adversary pays to bend its output. Plotting the field’s Process Catalogue on that plane is our version of a periodic table — and the most important feature is the quadrant that is still empty.

Positions are version-0 qualitative judgments from the Process Catalogue, drawn to be argued with. Replacing them with measured coordinates is one of the field's central projects.

The three layers

• The Judge — the consumer of reasoning, who cannot check the work object-level. What they need: a utility function over information, a way to price information against it, and the process-conditional weights that say how much each source should move them.
• The Processes — the procedures themselves: their anatomy, what their verdicts ground out in, their measured properties, and the attack-and-hardening cycle that makes them robust.
• The Environment — the game the processes live inside: producers, resolvers, accreditors, adversaries; the identity and track-record infrastructure every incentive scheme silently assumes; and the market, legal, and cultural constraints.

The book

#	Chapter	Status
1	Robust Reasoning Processes	draft
2	Cruxes	draft
I	The Judge
3	The Core Model	draft
4	Epistemic Impact Analysis	draft
5	Constructing Utility Functions	draft
6	What Can You Use from an Untrustworthy Source?	draft
—	Interlude: The Funding Effect	draft
II	The Processes
7	The Process Catalogue	draft (v0)
8	What Grounds an Oversight Protocol?	draft
9	Consistency Evaluations	draft
10	What Is a Strong Reasoner?	draft
III	Hardening
11	Hardening: Overview	draft
12	Calibration	draft
13	Verifiability Asymmetry	draft
14	Independence & Decorrelation	draft
15	Invariance & Low Sensitivity	draft
16	Incentive-Compatibility	draft
17	Deterrence	draft
18	Certification and Gyms	planned
IV	The Environment
19	Identity and Track-Record Infrastructure	planned
20	The Market for Truth	planned
21	Law, Liability, and Private Deliberation	planned
22	Culture, Consent, and Adoption	planned
V	Applications
23	The Reliability Ladder	draft
24	Overseeing Automated Research	draft
25	LLM Epistemics in Production	draft
26	Open Problems	aggregated

A Glossary & Notation page indexes the field’s terms and symbols.

How this book reasons about itself

Three norms, applied throughout:

1. Formalisms are graded. Every equation carries an explicit grade — [exact] (correct by construction), [standard shape] (precedent in an adjacent rigorous field), or [heuristic] (a deliberately simple decomposition, not a law). A field about robust reasoning should grade its own reasoning first. See The Core Model.
2. Failures are published. The deployment write-ups in LLM Epistemics in Production report false-positive rates, overconfidence, and a miscounted competition alongside the wins. Documented corruption and documented failure are the best data the field has.
3. Positions are dated and versioned. Every page carries a status line; everything here is a working draft of a field that does not exist yet, stated so it can be argued with and eventually replaced by measurement.

Status

This wiki is in an early, exploratory stage. Pages are working notes, not settled positions. It is part of the CAIRN project by QURI.