Project Diary - Day 35 Completion

Day 35 opened with the Day 35 2DO and full site tar. Primary work covered three areas: Part 5 inference worked example, TMA directory indexes, and glossary cleanup. The Transformer deep dive page was scoped and added to the running 2DO for Day 36.

The running 2DO format introduced in Day 35 — with age tags showing when items were first added — is working well. Nothing gets dropped; parking lot items are clearly dated.

Completed This Session

Part 5 — Inference Worked Example

Added concrete worked example: “What is the capital of France?” traced through all 8 inference pipeline steps
Two-column table layout — abstract step alongside specific example
Token explanation callout at Step 2 — what a token is, why it matters
Ambiguity resolution callout at Step 4 — why “capital” maps to a city not money; self-attention and context explained with a real case
Installed as 05-inference.html — vnu clean

TMA — Index Pages

TMA/TMA-index.html — hub page linking to AI and LLM modules
TMA/AI/AI-index.html — 10 page listing with descriptions
TMA/LLM/LLM-index.html — 5 page listing with descriptions
All three symlinked to index.html — consistent with site pattern
vnu clean on all three
check_tree -a from /AI root — zero broken links

Glossary

Breadcrumb CSS and div added to glossary.html
index.html → glossary.html symlink confirmed in place
Working files cleaned up — prot-glossary.html and backup removed
vnu clean

Part 7 — Complete Stack Recap

Reviewed — 5 of 6 rows already linked to correct parts
Layer 6 (Application) intentionally left unlinked — on Part 7 itself; breadcrumbs will provide navigation when installed
Item closed

New Item Scoped — Transformer Deep Dive

A dedicated Transformer architecture page was scoped during the session. The 2017 Attention Is All You Need paper changed everything — the Transformer deserves its own page rather than a paragraph in Part 4.

Proposed scope: what problem it solved, the attention mechanism with a worked example, encoder/decoder structure, why the claim was right. Location: TMA/LLM/ as a peer to the existing LLM reference pages — now that TMA has proper index pages it's a first-class location.

check_tree Orphan Notes

Seven orphans reported — all non-HTML working files, not broken pages. Zero broken links. Items to address in Day 36:

glossary/NOTES.txt appears to be a directory — investigate
Add skip patterns to check_tree for .jpeg, .png, extensionless files in remember/ and addendum/
biomes/ images — decide if they need a linking page or exclusion

Carry Forward — Day 36

Transformer deep dive page — Day 36 primary candidate
AI-NEW.html entry point card for how-ai-works/
check_tree skip list — add exclusions for non-HTML files
Glossary build-out — remaining terms
TMA breadcrumbs — 18 files now (15 + 3 new indexes)
Site-wide consistent footer
collaboration/lessons-learned.html — Days 11–35 unrecorded

Project Diary — Day 35 Completion

Session Context