Learning process
Finch's knowledge graph grows as it learns. Concepts that get referenced across many topics accumulate connections — the highest- connected entities form the spine of the substrate. Lower-degree entities are the surface that the next mastery cycle works on.
Average degree is the densification signal — the graph getting more connected over time, not just bigger. When this number trends up, concepts are consolidating rather than fragmenting.
// triples per Q&A pair, from substrate origin to today. higher ratio means more structured relations extracted per stored interaction — the densification signature.
// structural quality alongside the size metrics above. predicate vocabulary tracks the “snowflake” fight (canonical reuse vs. invented near-synonyms). leaf % measures how much of the graph is structurally dead-end. cross-domain entities measure transfer — concepts that span multiple topics.
// lifetime mastery-cycle outcomes per topic. attempts = cycles fired; passes = cycles that passed verification. per-topic variance reveals what the substrate finds easy vs hard; overall pass-rate trending up is the headline signal.
// mastery promotions per day, stacked by topic. each bar is one day; empty days are honest — the system isn't flipping something every day.
// scopes are the coarse-grained domains each triple was minted under. top 10 by count, updated live.
// the ten most-connected concepts in the graph, grouped by category. bars are normalized to the largest hub.
Connections form
+ GROWTH- ↳New learning adds new edges
- ↳Practice strengthens existing ones
- ↳Failures become structured relations
Concepts converge
− CONSOLIDATION- ↳Synonym concepts collapse into one
- ↳Weak, isolated concepts get pruned
- ↳The whole graph normalizes overnight