Skip to content

Tags: pavlov-net/gk

Tags

v0.1.1

Toggle v0.1.1's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature.
Overhaul temporal scoring with FSRS-inspired model (#11)

* refactor: overhaul temporal scoring with FSRS-inspired model

Replace ad-hoc temporal scoring with a principled model based on FSRS
(Free Spaced Repetition Scheduler) and Wozniak's Two Component Model.

Key changes:
- Remove access_count column entirely (redundant with stability)
- Write-only temporal bumps: reads are pure, writes strengthen knowledge
- Power-law forgetting curve: R = (1 + t/(S*d))^(-0.5)
- FSRS spacing effect: neglected knowledge grows more on write
- Multiplicative scoring with temporal floor (0.1)
- Simplify pruneStale to use retention * tierWeight directly
- Replace most/least accessed with most/least stable in health report

* fix: auto-migrate existing databases to drop access_count columns

Existing SQLite databases still have access_count columns from before
the temporal scoring overhaul. This adds an automatic migration in
initialize() that checks for and drops these columns on startup.

v0.1.0

Toggle v0.1.0's commit message
Bump version to 0.1.0