Feynman: Inspectable Research

Overview

Feynman wraps the Pi agent runtime with research-specific workflows, source routes, decision records, and a local workbench. Across the series, a conversational question becomes a planned run: the system chooses direct or bounded delegated search, distinguishes legal access from fetched text, turns findings into artifacts, and records provenance, status, verification, and next actions. PaperRank makes reading order inspectable through separate signals, local citation context, sensitivity, and follow-up evidence; the workbench preserves lineage across sessions and versions. These mechanisms make gaps and human judgment visible rather than proving scientific correctness. The reviewed 0.3.49 repository snapshot supports a design assessment, not independent evidence of accuracy, speed, adoption, or research quality.

Learning outcomes

Feynman’s core output is an inspectable research run, not only a polished answer.
Pi provides runtime machinery; Feynman adds research workflows, source routes, guardrails, and artifacts.
PaperRank proposes a read-first order; its signals do not validate a paper’s scientific claims.
The workbench preserves lineage and partial states, but the snapshot does not prove better research outcomes.

Keywords

FeynmanresearchevidencePiPaperRankverificationprovenanceworkbenchlineagereproduction

Chapters

Feynman’s Inspectable Loop

Why artifacts, states, and next actions matter

9 min · Why artifacts, states, and next actions matter

9 min

Feynman and Pi: Runtime Boundary

How the wrapper separates readiness from research sufficiency

9 min · How the wrapper separates readiness from research sufficiency

9 min

Feynman: Source-Grounded Runs

How scope, source access, and verification shape a run

9 min · How scope, source access, and verification shape a run

9 min

Feynman’s Research States

Why artifacts, status, and proof must stay separate

9 min · Why artifacts, status, and proof must stay separate

9 min

Feynman PaperRank: Read-First Triage

How signals set a reading order

9 min · How signals set a reading order

9 min

Feynman Science Workbench

How plans, versions, claims, and checks keep lineage

9 min · How plans, versions, claims, and checks keep lineage

9 min

Feynman: Research Failure Modes

How gaps, privacy, and scope stay visible

11 min · How gaps, privacy, and scope stay visible

11 min

Feynman: Research Judgment

Where inspectability helps—and human review remains essential

10 min · Where inspectability helps—and human review remains essential

10 min

Preview every chapter · Unlock full text, audio, and video from chapter 4

Plans

Ch.1-3 FreeCh.4-8 Premium

FAQ

What makes Feynman’s research output inspectable?

Feynman makes research output inspectable by linking sources, actions, artifacts, statuses, checks, and next actions. That structure exposes uncertainty without making the result automatically correct.

How does Feynman turn a question into a source-grounded run?

Feynman defines the evidence threshold, chooses direct work or bounded delegation, gathers through source-specific routes, then synthesizes and verifies the result. Access and provider limits remain explicit.

What does PaperRank actually tell a researcher?

PaperRank suggests which candidate paper to read first within a bounded set and explains that order through separate signals and evidence. It is triage, not scientific validation.

When is Feynman’s science workbench worth using?

Feynman’s workbench is worth using when successive sessions, executions, revisions, claims, and checks need to remain connected in a local research context. A narrow, short-lived question may need only the CLI.

Does Feynman guarantee scientific correctness?

No. Feynman does not guarantee scientific correctness; the repository supports inspectable mechanisms, while sources, models, providers, executions, and human judgment determine whether a conclusion is sound.