Gaia Registry Meta Audit Series

A six-link, gate-checked curation pass expanded the starless layer of the registry by thirteen generic references. The pass also surfaced a taxonomy correction: feed-monitoring, previously held as a Unique Skill, was reclassified to Basic once curation established a fusion path forward into a new Extra. This report records the full set of additions and focuses on the reasoning behind the feed-monitoring decision.

Executive Summary

This pass was run as a prompt-chaining workflow: six sequential links, each handled by a fresh sub-agent, with a programmatic gate between every link. Evidence was gathered against arXiv, public repositories, and community sources; all thirty-nine source URLs were confirmed resolvable before the design link began. Every addition is a starless reference — a generic taxonomy node that carries no stars of its own and holds the shared, capability-level evidence pool its future named children inherit.

The headline taxonomy event of the pass was the reclassification of feed-monitoring from the Unique tier to Basic, detailed in §3.

Additions

Thirteen starless references were added. Four are Basic Skills; nine are Extra Skills that draw on existing prerequisites already present in the graph.

| Reference | Tier | Prerequisites | |-----------|------|---------------| | time-series-forecasting | Basic | — | | synthetic-data-generation | Basic | — | | probabilistic-programming | Basic | — | | cultural-localization | Basic | — | | financial-modeling | Extra | statistical-analysis, data-analysis, generate-sql | | supply-chain-optimization | Extra | statistical-analysis, data-analysis, detect-anomaly | | causal-inference | Extra | statistical-analysis, hypothesis-generate, logical-inference | | regulatory-compliance-mapping | Extra | knowledge-graph-build, extract-entities, security-audit | | threat-intelligence-synthesis | Extra | feed-monitoring, knowledge-graph-build, security-audit | | ontology-alignment | Extra | knowledge-graph-build, schema-design, extract-entities | | explainability-audit | Extra | evaluate-output, statistical-analysis, data-visualize | | edge-optimization | Extra | fine-tune, evaluate-output, performance-tuning | | adversarial-robustness-testing | Extra | agent-eval, prompt-injection-defense, guardrails |

One candidate, multi-modal-fusion, was rejected at human review as redundant with the existing multimodal-reasoning reference. A second candidate, probabilistic-programming, was retained as distinct from statistical-analysis: it carries the Bayesian and uncertainty-quantification evidence pool (posteriordb, PyMC, Stan), keeping that material on the reference where it fits best.

The feed-monitoring Reclassification

The defining property of a Unique Skill is the absence of a fusion path forward — it is a Basic that reached elite depth and has nowhere left to fuse. feed-monitoring had been held as a Unique on exactly that basis.

This pass changed that fact about the graph. The new Extra threat-intelligence-synthesis builds on feed-monitoring as a prerequisite: continuous monitoring of security and advisory feeds is the upstream input that threat-intelligence synthesis draws on, alongside knowledge-graph-build and security-audit. The moment that prerequisite edge exists, feed-monitoring has a fusion path forward — and a reference with a fusion path forward is, by definition, not a Unique.

The validation gate enforced this automatically. The first validation run after the additions failed with a single error:

> Unique skill 'feed-monitoring' is referenced as a prerequisite by another skill — unique skills must be graph-isolated.

Rather than route the new Extra around feed-monitoring, the correct fix was to update the stale fact: feed-monitoring was reclassified from Unique to Basic via gaia dev reclassify feed-monitoring basic, then wired as a prerequisite of threat-intelligence-synthesis. Validation then passed cleanly. The Unique count in the graph moved from three to two; the Basic count rose accordingly.

The lesson is that Unique status is not permanent — it is a statement that no fusion path is known yet. When curation discovers such a path, the honest action is to demote the reference and let it serve as a prerequisite, not to preserve its isolation artificially.

Method and Gates

Each link advanced only after a programmatic check passed:

• Scope — enumerated the existing references and proposed research targets.
• Research — gathered evidence; all source URLs verified resolvable.
• Design — mapped candidates to tiers and prerequisites, forming an acyclic graph.
• Human review — accept / reject / reclassify decisions, including the rejection of multi-modal-fusion.
• Mutation — all changes applied through gaia dev commands; no hand-edited node files.
• Ship — regenerated docs and confirmed validation, manifest version lockstep, and a clean build.

All registry mutations were performed programmatically in keeping with the Programmatic-First policy.