# Why Self-Distillation Helps and Hurts: Denoising vs. Signal Forgetting

ORID `VnA5q5jXVz` · **claim-faithful** repair 2026-07-27T19:08:48.247863+00:00

Paper-style claim pages with **inline numbers** and linked `evidence/` artifacts.

## Claims

- [Claim 1: 01-deterministic-equivalent-recursions-prediction-r](./01-deterministic-equivalent-recursions-prediction-r/) — VERIFIED (2/2)
- [Claim 2: 02-test-risk-function-self-training-iteration](./02-test-risk-function-self-training-iteration/) — VERIFIED (2/2)
- [Claim 3: 03-spiked-covariance-models-self-training-iteration](./03-spiked-covariance-models-self-training-iteration/) — VERIFIED (2/2)
- [Claim 4: 04-iterated-generalized-cross-validation-igcv-estim](./04-iterated-generalized-cross-validation-igcv-estim/) — VERIFIED (2/2)
- [Claim 5: 05-theoretical-denoising-forgetting-risk-tradeoff-d](./05-theoretical-denoising-forgetting-risk-tradeoff-d/) — VERIFIED (2/2)
- [Conclusion](./conclusion/)

## Artifacts

- [`evidence/claim_1.json`](../evidence/claim_1.json)
- [`evidence/claim_2.json`](../evidence/claim_2.json)
- [`evidence/claim_3.json`](../evidence/claim_3.json)
- [`evidence/claim_4.json`](../evidence/claim_4.json)
- [`evidence/claim_5.json`](../evidence/claim_5.json)