MegaContext PRD Index

All entries are plan-of-record (POR). Update the linked PRDs + PRD Progress Tracker whenever requirements change or implementation work lands.

Quick reference for the current MegaContext product requirement documents:

1. MegaContext End-to-End Training — Unified training plan covering horizon evaluation, ΔNLL supervision, and gist-level losses for the full MegaContext loop.
2. MegaAttention Training — Extending sparse attention with hierarchical attention via GistNet
3. MegaPrediction Training — Roadmap for the multi-LOD decoding head that forecasts gist vectors alongside tokens to enable gist-first inference paths.
4. Cognitive-Core Training — Strategy for the cognitive-core model that externalizes knowledge into composable MegaContexts and evaluates composite focus behavior.
5. Hierarchical KV Caching Strategy - Strategy for efficiently updating KV caches as the Focus Allocator performs edits.
6. PRD Progress Tracker — Status board summarizing progress, blockers, and next steps for each PRD.