Documentation Impact Assessment — PRD Alignment
This note captures the ripple effects of adopting the three new MegaContext PRDs and migrating toward a nanochat-based implementation.
High-priority updates
- Legacy POC plan (archived) — superseded by the PRD stack; mark legacy phases (Lightning notebooks, staged focus experiments) as archival or rewrite as historical context.
- POC Implementation — rewrite to describe nanochat-oriented modules, referencing MegaContext End-to-End Training, MegaPrediction Training, and Cognitive-Core Training as the new milestones.
- Legacy “Alternating Optimization” note — scrub references and point everything at MegaContext End-to-End Training instead.
- Ops / Telemetry — align metrics and runbooks with the nanochat migration (ΔNLL@H, gist losses, composite MC telemetry).
- Base Runtime — update runtime instructions to cover nanochat CLI usage and MegaPrediction inference hooks.
Secondary adjustments
- Plans index — ensure it links to the new PRD index and clarifies that previous milestone plans are historical.
- POC Scope — annotate limitations that no longer apply (Lightning reliance, two-level gist assumption) and highlight the new baseline expectations.
- Architecture Details / System Properties — verify language surrounding alternating training, two-context loops, and focus allocation matches the PRD terminology.
- Telemetry notebooks — consider migrating key guidance into markdown to reduce dependence on legacy notebooks.
Archival candidates
- Legacy notebook references inside GistNet Training, LensNet Training, and related component notes.
- Detailed walkthroughs of the alternating training schedule (retain only as historical appendix if needed).
New documentation to add
- Nanochat Integration Guide (developer onboarding, environment setup, CI notes).
- Migration Status tracker referencing Migration Plan - Nanochat Integration for phase-level progress.
- PRD Progress Tracker (checkpoints, open questions) to keep the PRDs living documents.