Day 38 fell on Saturday July 18, 2026 — three days after Day 37 on July 15. The Hot item carried since Day 30 — TMA breadcrumbs — was closed out in full, along with an unplanned but necessary filename cleanup that surfaced along the way.
Session opened with the AI.tar restart package and a review of the Day 38 2DO. TMA breadcrumb work was tackled file-by-file, with each file's background color checked before applying breadcrumb CSS to avoid low-contrast text on the handful of pages with dark or gradient body backgrounds.
All 19 TMA files now carry breadcrumb navigation. Three (TMA-index.html,
AI-index.html, LLM-index.html) already had breadcrumbs from
a prior session. The remaining 16 content pages were completed today:
One page (AI-Types-Detail.html) required custom breadcrumb contrast
colors (#FFF text / #00FF00 links) due to its dark
gradient body background — standard #666/#006600
would have been unreadable. All other pages used the standard breadcrumb CSS.
While reviewing breadcrumb labels, confirmed that LLM-Capabilities.html
and LLM-Capabilities_MORE.html were misnamed — both pages are
actually about LLM architecture/components, not capabilities. Renamed to:
LLM-Capabilities.html → LLM-Architecture-Overview.htmlLLM-Capabilities_MORE.html → LLM-Architecture-Detailed.htmlThe rename broke 10 links site-wide (5 in ai-stack/, 2 in
deep-dive-index.html, 1 in learning/AI-architecture-stack.html,
2 in LLM-index.html). All 10 fixed with a two-pass sed
across the affected files. PROJECT-STRUCTURE.html still has the old
names in its cosmetic directory-tree listing — parked, not a functional link.
LLM-Tokenization.html, which briefly failed with a
CSS parse error mid-session and was resolved)check_tree -a from /AI — 0 orphans, 0 broken links, site-wide~ backup file found and removedtar czf AI.tar AI/The TMA breadcrumb work was more varied than the earlier deep-dive/learning/resources
sprint — several of these 16 pages carry their own imported CSS themes
(:root variable palettes, gradient headers) rather than the site's
house style, which is exactly why the breadcrumb work had to be checked
page-by-page for contrast rather than applied as a blind find-and-replace.
The rename cleanup was a good catch that came out of simply reading the content while building breadcrumb labels — a reminder that file names drift from content over time, and that periodic spot-checks during unrelated work (here, breadcrumbs) surface issues that a pure link/orphan audit wouldn't catch.
cmh — AI Education Project Day 38 — July 18, 2026
Next session: Day 39 — see 2DO for open items