A probe aimed at decentralized AI.

Tracked production systems across ZK proving, distributed training, FHE, secure MPC, and ML‑network consensus, each traced back to the paper it's actually built on, with every claim marked by how directly it's sourced. Not a news feed. A chain of custody.

–

SYSTEMS TRACKED

–

SEED PAPERS

–

FULLY GROUNDED

–

LAST CHECKED

WHY THIS EXISTS

What's changed since I last checked, and does it actually touch anything real?

This tracker follows a deliberately small list of systems that already ship in production, and the foundational research each one is demonstrably built on: not everything "important," only what a tracked system's own docs or whitepaper actually cites. When new research lands, it gets checked against each system's property matrix; a note only gets written if something in that matrix actually moves.

— pull, not push. No cron. You trigger a sweep when you want one.

CASE FILES

Tracked systems

Open a file to see its architecture, its sourced property matrix, and the citation chain down to the seed paper it's built on.

loading index.json…

SEED PAPERS

Foundations

One paper per field, chosen because a tracked system's own docs demonstrably cite it, not because it's subjectively "important."

loading index.json…

WATCH LOOP

Signal log

A daily check against a fixed, narrow set of sources (IACR ePrint, relevant arXiv categories, and tracked systems' own repos) — not the entire literature.

checkpoint — watch/checkpoint.json

loading index.json…

What a run does

Pull new items since the last checkpoint from the fixed source list — IACR ePrint, arXiv (cs.CR, cs.LG, cs.DC), and tracked systems' own repos — not the entire literature.
Filter for relevance against the tracked systems/fields.
For anything touching a tracked system — a paper, a release, a merged PR that looks significant: read it fully, check it against that system's property matrix. A note is written only if the matrix actually changes.
For anything field‑relevant but untied to a tracked system: filed as "related, unverified relevance" (a visibly lower tier, no full write‑up).
Checkpoint updates. Short report: N new, M matrix‑impacting, rest filed.

INSTRUMENT NOTES

Calibration

How to read every stamp on this instrument before you trust what it's showing you.

AI‑generated content notice: the research, analysis, and "does this affect system X" judgments here are produced primarily by an AI (Claude), with a human maintainer directing scope and spot‑checking output. This is not peer‑reviewed material and carries no claim of novel authorship over the analysis. Treat every claim as a starting point for your own verification; see each entry's confidence tier.

Signal‑integrity tiers

GROUNDED

Directly sourced: a project's own README, docs, or paper states the claim explicitly. The strongest return this instrument gives.

INFERRED

Reasoned from general sourcing, cross‑referenced secondary coverage, or convergence across sister sources: plausible, not a direct quote. Recheck before relying on it.

OPEN QUESTION

Flagged explicitly as unresolved: a withdrawn paper, a naming gap, an unverifiable claim, carried forward rather than quietly dropped.

Structure

/systems: one file per tracked production system: what it does, what it's built on, known limitations, and a property matrix checked against incoming research.
/foundations: one foundational paper per field.
/watch: checkpoint + dated log, one entry per triggered run.
index.json: generated mirror of the above, regenerated by scripts/build_index.py. This page reads from it directly: edit a markdown file, regenerate, push, and the live page reflects it.

Part of misbar.tech: prove what AI can do.