Research Vault Operations

Report

Polymath Research Atlas

Portfolio evidence-pipeline audit: transparent stage flags and metadata-backed bridge retrieval over a personal research vault — not a quality score.

Verdict Mixed

Published
Jul 2026
Verdict
Mixed
Verification
Provisional
Question
Where does the research portfolio have evidence gaps and useful project bridges?
Intent
Make portfolio quality inspectable through metadata-backed stage flags, not a vanity score.
Method
Index vault notes, compute six stage flags, and rank project bridges from shared records, tags, and links.
Next test
Freeze a privacy-safe atlas slice for public notebook export.

Polymath Research Atlas

Question

Where does the research portfolio have evidence gaps and useful project bridges?

Context

The atlas is Club B archive instrumentation for the Recursive Self Improvement Lab. It treats the research portfolio as an evidence pipeline, not a scoreboard. For each project it sets six independent binary flags — question, source, experiment, artifact, verification, publication — then ranks bridge candidates between project pairs using only shared tags, wikilinks, and co-mentioned records. No embeddings. No LLM similarity. No weighted quality scalar.

flowchart LR
  vault["Vault notes"]
  flags["Six stage flags"]
  bridges["Metadata bridges"]
  publish["Publication surface"]

  vault --> flags --> bridges
  flags --> publish

Findings

  • Sealed run (as_of 2026-07-23, schema vault-ops.polymath-research-atlas.v1): 385 vault markdown files indexed; snapshot digest 1386f637441d18ba…6f878c; 32 projects (11 active / 11 dormant / 10 paused); 64 bridge candidates.
  • Stage presence (projects with flag=1): question 3/32, source 6/32, experiment 4/32, artifact 3/32, verification 3/32, publication 3/32.
  • First-gap distribution: question 29, complete 1, source 1, artifact 1. Only parameter-golf clears every stage.
  • Verified and published in that sealed read: parameter-golf, wave-causal-circuit-search, nostradamous. Flags are independent — a project can publish while still showing an earlier first gap.
  • Top metadata bridges by support points: memory-store×vault-ops (39), memory-store×munshi (36), lorenz×sansara (35), munshi×sansara (34), phi9×vault-ops (34).
  • For RSI Lab readers: Club A projects that cleared verification/publication in this snapshot are exactly the sealed notebook surfaces; Club B often has method without a full evidence spine yet.

Method

Canonical vault notes are read-only input. Analysis lives in vault-ops/notebooks/polymath-research-atlas.py. Bridge support is

4⋅min⁡(shared records,5)+3⋅min⁡(shared tags,5)+2⋅min⁡(shared links,5)+direct links.4\cdot\min(\text{shared records},5) + 3\cdot\min(\text{shared tags},5) + 2\cdot\min(\text{shared links},5) + \text{direct links}.

Independent verification recomputes spot checks against live vault files before treating a claim as sealed. False negatives are expected when association metadata is missing.

Limits

  • The atlas is a visibility map, not a quality score and not an inheritance result.
  • False negatives are expected where metadata or links are missing.
  • Public notebook export waits on a privacy-safe frozen slice.
  • Stage flags do not rewrite sealed experiment verdicts on Club A reports.

Artifact

Live notebook: vault-ops/notebooks/polymath-research-atlas.py. Sealed analysis artifacts under vault-ops/artifacts/polymath-research-atlas/ (private; host paths stripped from this public writeup). No browser notebook export yet — the live vault dependency and uncommitted sources block a fail-closed site pin until a privacy-safe frozen slice exists.

  • research-ops
  • marimo
  • knowledge-graph
  • portfolio