LRRK Research

The vocabulary is the research.

The hard problem is not detection. It is saying exactly what is known, how it came to be known, and what remains unknown — in language that survives review years later.

The programme has one object: a Security Story that a reviewer can audit sentence by sentence. Everything below — the method, the vocabulary, the papers, the unresolved problems — exists to make that object accountable rather than persuasive.

§ 01

Method

The full reference: assertion taxonomy, evidence outcomes, confidence bands and capability truth labels used everywhere in the archive.

Fact: An observation with an intact custody chain. Cited by reference, never paraphrased into certainty.
Inference: A reading that follows from facts under stated assumptions. It can be withdrawn without deleting the facts.
Judgment: An analyst's disposition on a contested question, signed and attributable to a person.
Recommendation: An action proposed under named limits. Never presented as a finding.
Preserved unknown: Something the authorization did not permit us to examine. Recorded with the same weight as a finding.

§ Method · reference

How an LRRK Security Story is built

Facts, inferences, judgments, recommendations and preserved unknowns — each with its own provenance.

§ Frameworks · reference

Cybersecurity assurance frameworks, and what a Security Story adds

How the model sits alongside NIST CSF, IEC 62443, SOC 2 and ISO/IEC 27001 — and why scores give way to stories.

§ 02

Library

Published papers and working documents. Each release is a fixed artifact — downloads are counted, the text is not revised in place.

Kind

SearchSort

Semantic Airlock with Proof-Carrying Curiosity

A proposed KAT-bound research architecture for systems that sense, move, and act

Curiosity was never the dangerous capability; allowing a plausible question to become its own authorization was. This white paper develops an LRRK-specific formulation: hypotheses should be abundant, plural and contestable, while access and action stay scarce, purpose-bound, temporary, attributable and enforced outside…

Prepared for: LRRK LLC
Series: LRRK White Paper 02
Version: 0.9 — August 2026
Status: Proposed architecture · not an adopted standard
Scope: Information-flow governance, authorization, research method

semantic airlock
proof-carrying curiosity
kinematic assurance topology
capability leases
information-flow control
provenance ledger
severe testing
human disposition

Downloads · counting…

Download PDF ↓

§ Research paper · 2026

DOCX · 53 KB

Select

Restricted, Still Available

U.S. drone policy at the intersection of market access, industrial strategy, and product-security assurance

Restriction and availability are not a paradox but a conjuncture: prospective FCC equipment authorization, entity-specific procurement rules, defense supply-chain requirements, durable inventories and fragmented retail run on different clocks. An August 2026 audit of 44 commercially relevant manufacturers found a clear…

Prepared for: LRRK LLC
Date: August 12, 2026
Method: Regulatory analysis + retail availability audit
Scope: Federal UAS market access, procurement, retail, assurance

conjunctural analysis
FCC Covered List
American Security Drone Act
retail persistence
UAS cybersecurity
product-security passport

Downloads · counting…

Download DOCX ↓

§ 03

Vocabulary

The controlled terms used across every Story. They are deliberately few, and none of them is a score.

§ Assertion types

fact: Observed, custody intact.
inference: Derived under stated assumptions.
judgment: Analyst disposition on a contested reading.
recommendation: Proposed action, bounded by limits.
unknown: Not examined; scope excluded it.

§ Confidence bands

high: Multiple independent lines, no live contradiction.
moderate: One strong line, or several weak agreeing lines.
low: A single artifact, or unresolved contradiction.
not rated: No evidence has entered the archive yet.

§ Evidence outcomes

established: Condition reproduced under documented method.
not established: Attempted, not reproduced.
rejected: Actively disproved; the rejection is preserved.
inconclusive: Method exhausted without disposition.
not evaluated: Outside the authorized scope.

§ Currentness

current: Reading holds as of the stated date.
challenged: A contradicting record is open.
stale: Time has passed without re-examination.
superseded: A later signed version replaces it.

§ 04

Open questions

Four problems the archive has not solved. Each records our current stance and what would change it.

Revisable interpretation

A reading must be able to change without erasing the reading it replaced, or the reason it changed.

Current stance: Versions are append-only; every supersession carries a reviewer and a written reason.
Would change our mind: Evidence that readers reliably mistake a superseded version for the current one.

Provenance at scale

Evidence must be cited by reference across thousands of subjects while protected material stays out of the narrative.

Current stance: Narratives cite hashes and custody records; they cannot inline restricted material.
Would change our mind: A citation scheme that stays verifiable when the underlying artifact is lawfully destroyed.

Honest uncertainty

A low-confidence sentence must never be absorbed into a confident summary.

Current stance: Confidence is carried per assertion, not per document, and summaries inherit the weakest band.
Would change our mind: Reader studies showing per-assertion bands are ignored in practice.

Currentness decay

A story must become stale on a defensible schedule, and a stale story must still be readable.

Current stance: Decay is a function of subject volatility, not a fixed interval.
Would change our mind: A volatility signal accurate enough to set intervals per product line.

Synthetic demo — no publications are represented here.

§ 05

Citation & reuse

Papers in the library are fixed artifacts. Cite the release you read, including its date, because a later release may reach a different reading.

Suggested form

LRRK LLC. Title of paper. LRRK Research Library, release date. Retrieved from lrrk.ai/research.

Quote freely with attribution. Do not present a figure from one release as the current position without checking the library for a later release of the same title.