Research index

Dread research and publication discipline

Updated · Class: Reviewed claim · Review: Reviewed by Dread team guidance · Maintainer: Dread Research · Contact [email protected]

Dread separates public material into three classes: historical artifacts preserved as-is, reviewed claims with evidence attached, and project stories that may motivate but do not prove.

Content classes

The publication discipline is part of the credibility story for a security-adjacent lab. Technical pages should identify their class, review status, last-updated date, and maintainer.

Research publication pipeline Internal traces move through sanitization, human review, evidence classification, public HTML and discovery, then measurement. Failed review returns work to sanitization instead of publishing. PROCESS DESCRIPTION / CLAIM CONTROL INTERNALTRACESruns · notesraw evidence SANITIZEsecretsprivate hostsunsupported detail HUMANREVIEW YES CLASSIFYreviewed claimproject storyhistorical artifact PUBLIC LAYERserver-rendered HTMLimmutable sourcemachine discoveryexplicit boundaries MEASUREreadiness · coverage · promptsrank the next change NO / REVISE A failed gate loops back. Publication is not the default outcome.
Research moves from internal traces to public claims through sanitization, human review, evidence classification, and post-deploy measurement.

Research publication pipeline

  1. Internal traces begin as runs, notes, and raw evidence.
  2. Sanitization removes secrets, private hosts, and unsupported detail.
  3. Human review is the gate; failed review loops back to sanitization.
  4. Approved material is classified as reviewed claim, project story, or historical artifact.
  5. The public layer emits server-rendered HTML, immutable source links, machine discovery, and explicit boundaries.
  6. Measurement checks readiness, coverage, and prompt visibility to rank the next change.
  • Historical artifact: preserved record, not necessarily current claim
  • Reviewed claim: evidence-backed public statement
  • Project story: narrative context that motivates but does not prove

Research lanes

Current public lanes include local-first agent infrastructure, eval and artifact workflows, adversarial and observability research prototypes, LLM red-team methodology, prompt-baking/model-adaptation evaluation, OpenLens style/OPSD research scaffolding, and Janus-style representation-dependent retrieval experiments.

Silence rule

For unverified capabilities, the correct public state is silence, not a softened disclaimer. A short true page is better than a long impressive one.

Canonical links