NodeXLoop captures what agents attempt, where they fail, and what humans approve. Axiom verifies every decision. NodeX Trace converts that evidence into private corpora, evaluations, and specialized open-weight agents that improve with every approved run.
Attempts, tool calls, stalls, corrections, cost and outcomes — recorded at the loop level by NodeXLoop.
Axiom signs provenance, policy verdicts and approved uses. Nothing enters the corpus unsigned.
NodeX Trace builds evals, routing rules and tuned specialists from corrected decisions.
Version-bound agents return measurable production evidence — and the loop turns again.
Finds dead calls, stalled critics, wasteful retries, routing errors, context bloat, and missing stop conditions — then proposes a testable repair.
tracing run nxl-run-4471…
Leave with a private corpus, a governed specialist, and measured proof of what improved. One workflow, thirty days, measured before → after — and you own every trace.
Every row is signed by Axiom and labeled with its sharing tier. Dead calls, retries, and rejected branches are retained as hard negatives — failures become assets.
| Trace | Sequence | Axiom verdict | Outcome | Tier |
|---|
1 · The audit export grows a machine twin. Every NodeXLoop run already exports an audit report — provenance, cost, every governance decision, the full log, with estimated figures labelled as estimates. The same data as a machine-readable manifest is the corpus row: situation, attempted action, intervention, correction, outcome.
2 · Trace ingests manifests from wherever the run happened. The VS Code extension and the browser app are one source. Headless runs are the other — a CI box replaying a regression suite, or an edge device running a loop with no screen attached (a Jetson driving a robot loop can emit the same manifest it would show you in an editor).
3 · Nothing enters unsigned. Run manifests are sealed locally today — tamper-evident on the machine that made them, and they do not claim more than that. Server-side verification against Axiom at ingestion is the gate this product adds.
Agents are re-tested against versioned evidence and policies, so proof survives model changes. Open weights keep control local — no dependence on third-party model APIs.
Each specialist is sealed by an Axiom manifest and re-proven against versioned evidence before promotion. Proven agents can later be licensed through the Registry & Exchange. The three below are illustrative.