Project Diary - Day 33 Completion

Day 33 opened with the planned first-pass reread of Parts 01–06. The reread surfaced something useful: both pipeline breakdowns proposed as "missing depth" earlier (the Training Pipeline and Inference Pipeline) were already built into Parts 3 and 5, verbatim. The series didn't need expansion — it needed better internal linking. That redirected the day toward the Six Layers stack diagram on Part 1 and a new glossary/definitions concept.

Completed This Session

Reread — Parts 01–06

Confirmed all six parts already match the depth standard Craig cited (the Training Pipeline example turned out to already exist in Part 3)
Part 6 confirmed self-contained, no action needed
Breadcrumb trail confirmed correct and consistent across all 7 parts + index

Part 1 — Six Layers Linked

Each row in the Six Layers stack diagram on 01-ai-stack-overview.html now links directly to its corresponding part (Application→7, Interface→6, Inference→5, Model→4, Training→3, Hardware→2)
Added hover state (highlight + arrow) so the rows read as clickable
Fixed striping/border CSS to key off the new link wrapper instead of the inner div, after wrapping broke the original nth-child selectors
Added target="_blank" rel="noopener" to all six links per Craig's direction — overview stays open while each layer opens in a new tab

Glossary/Definitions Concept — Prototype Built

Standalone click-to-reveal glossary prototype built and validated as a concept (modal dialog, term chips, keyboard/backdrop close)
Two working entries populated: GPU (What It Is / How It's Built / analogy / Why AI Uses It) and Self-Attention, using Craig's own fox-and-rabbit Winograd Schema example as the worked case
Established the source-of-truth rule: each glossary term maps to one canonical series page, even when the term is mentioned on several

Site Reference Update

Verified current Anthropic support/feedback contact channels via web search; corrected a personal bookmark page that had placeholder addresses

Carry Forward — Day 34+

Part 5 — Inference walkthrough — a concrete worked example (e.g. "What's the capital of France?") traced through all 8 existing pipeline steps, same treatment as the Self-Attention glossary entry
Part 7 — target links on the Complete Stack recap — mirror the Part 1 treatment; 5 of 6 rows map cleanly to Parts 2–6, but the Application Layer row points to Part 7 itself — decide whether to leave that one row unlinked or point it elsewhere
Glossary build-out — remaining terms: TPU, VRAM, Interconnect, Cluster, Tokenization, Transformer, Parameter, RLHF, Context Window, Temperature; also resolve modal-vs-inline and fixed-vs-flexible section design questions from the prototype
Outbound Reference Material links — the original target="_blank" question for TMA/deep-dive/learning links across all 7 parts is still open (distinct from the internal series links, now done)

Project Diary — Day 33 Completion