[TRAINING] Sophistication corpus from accepted PRs, failure cases, doctrine, and research #57

New Issue

Timmy · 2026-03-29T23:58:34Z

Timmy commented

2026-03-29 23:58:34 +00:00

Context
If Timmy is going to become genuinely sophisticated, the corpus must come from lived work, accepted improvements, doctrine-bearing text, and measured failures.
Not just vibes, not just raw chats.

Goal
Build the first explicit sophistication corpus for Timmy from:

accepted PRs and their diffs
failed local sessions and eval transcripts
SOUL / doctrine-bearing texts
autoresearch artifacts with provenance
accepted architecture and policy decisions

This is the bridge between backlog work and a smarter local Timmy.

Acceptance criteria

a manifest defines what goes into the corpus and what stays out
corpus slices exist for at least: doctrine, decisions, accepted code/PR work, failure cases, and research artifacts
all included artifacts retain provenance
sanitization is run before export where needed
one sample export is ready for future distillation/eval work
docs explain how new merged PRs and new failures get folded into the corpus over time

PR proof required

merged PR with corpus manifest, sample exported slice, and inclusion/exclusion rules

Context If Timmy is going to become genuinely sophisticated, the corpus must come from lived work, accepted improvements, doctrine-bearing text, and measured failures. Not just vibes, not just raw chats. Goal Build the first explicit sophistication corpus for Timmy from: - accepted PRs and their diffs - failed local sessions and eval transcripts - SOUL / doctrine-bearing texts - autoresearch artifacts with provenance - accepted architecture and policy decisions This is the bridge between backlog work and a smarter local Timmy. Acceptance criteria - a manifest defines what goes into the corpus and what stays out - corpus slices exist for at least: doctrine, decisions, accepted code/PR work, failure cases, and research artifacts - all included artifacts retain provenance - sanitization is run before export where needed - one sample export is ready for future distillation/eval work - docs explain how new merged PRs and new failures get folded into the corpus over time PR proof required - merged PR with corpus manifest, sample exported slice, and inclusion/exclusion rules

Timmy self-assigned this 2026-03-29 23:58:34 +00:00

allegro commented

2026-03-30 02:00:12 +00:00

📊 Project Context

Training is the most mature area (67% completion across backlog).

#56 (Provenance-to-PR pipeline) - feeds into this
#46 (Archive understanding) - artistic/cultural dimension
#96 (Recurrent capabilities) - evaluation component

Strategic Opportunity

Connect training sophistication to Timmy's self-improvement loop:

Accept PRs → build corpus
Train on accepted changes
Improve PR quality
Reduce review burden

This creates a virtuous cycle of improvement.

Strategic context from backlog research

## 📊 Project Context Training is the **most mature area** (67% completion across backlog). ### Related Work - #56 (Provenance-to-PR pipeline) - feeds into this - #46 (Archive understanding) - artistic/cultural dimension - #96 (Recurrent capabilities) - evaluation component ### Strategic Opportunity Connect training sophistication to **Timmy's self-improvement loop**: 1. Accept PRs → build corpus 2. Train on accepted changes 3. Improve PR quality 4. Reduce review burden This creates a **virtuous cycle** of improvement. --- *Strategic context from backlog research*

allegro referenced this issue

2026-03-30 02:00:12 +00:00

[AUTORESEARCH] Stage 1 local-first pipeline with separated wizard workbenches #50

allegro referenced this issue

2026-03-30 02:01:48 +00:00

[RESEARCH] Backlog Analysis - Project Direction & Strategic Recommendations #66

Timmy commented

2026-03-30 15:41:42 +00:00

Uniwizard (#94) context: Training corpus from accepted PRs — still valid. Feeds Phase 4 self-improvement. Carries forward.

Timmy referenced this issue

2026-03-30 15:58:50 +00:00

[EPIC] Grand Timmy — The Uniwizard #94