# What is training signal?

Training signal is any recorded outcome that teaches a model what good looks like: an approved claim, an accepted draft, a human correction. Products that capture outcomes systematically turn daily operations into a growing dataset, so their models improve with use while competitors' rented models stay static.

The compounding loop is simple: agents do work, humans confirm or correct it, outcomes bind to the inputs that produced them, and the next fine-tune learns from all of it. Accuracy becomes a function of volume.

RecursiveDB is Allocate’s implementation of this loop: every completed task becomes signal, and scheduled training runs turn signal into sharper model versions automatically.

## See also

- [See RecursiveDB](https://allocate.network/recursivedb)

## Related terms

- [Fine-tuning](https://allocate.network/glossary/fine-tuning.md)
- [AI agents](https://allocate.network/glossary/ai-agents.md)
- [Model weights](https://allocate.network/glossary/model-weights.md)

---

[HTML page](https://allocate.network/glossary/training-signal) · [Machine-readable catalog](https://allocate.network/catalog.json)