Update plans with next steps and session completions

Add prioritized next steps: standards parsing (quick win), problem
extraction (core data asset), web templates (depends on problems).
Add research items for remark-math and Plurimath.js editing.
Document all completions from this session.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: jcasimir

plans/todos.md

@@ -4,8 +4,11 @@
 
 ## Data & Content
 
-- [ ] **Standards tagging** — Map 149 Grade 5 lessons to their CCSS standards. 110 standards (Gr 4-6) already imported. Could use AI to match lesson content against standard descriptions, or reference EngageNY's published curriculum maps.
-- [ ] **Concept dependency graph** — Build prerequisite edges between standards and lessons. E.g., "5.NF.1 (add fractions) requires 4.NF.3 (understand fraction equivalence)." This is the core differentiator.
+- [ ] **Standards parsing (quick win)** — Extract CCSS codes from lesson HTML (they're in `ny-list-focusstandards` CSS classes) and create StandardTagging records linking them to lessons/topics. 110 standards already imported for Gr 4-6. This is mechanical parsing, not AI — the standards are explicitly listed in the HTML.
+- [ ] **Problem extraction (core data asset)** — AI reads each problem set/exit ticket/homework HTML, identifies individual problems, extracts text/answer/type, creates Problem records. Each Problem links to its Expression records for structured math rendering. This is where we hit the formatting chaos and start normalizing. Multi-tier validation: automated structural checks → Fiverr human review. Design with validation status fields (extracted, auto_validated, human_validated).
+- [ ] **Web presentation templates** — Stop serving raw Aspose HTML. Build purpose-built templates for each content type (lesson plan, problem set, exit ticket, homework) using proper web CSS with MathML from Expression records. Depends on Problem extraction to be truly useful — otherwise we're just wrapping HTML blobs. Key insight: Aspose HTML is source data, not presentation layer.
+- [ ] **Standards tagging (AI-assisted)** — Beyond the explicit CCSS codes in the HTML, use AI to match lesson content against standard descriptions for deeper tagging. Builds on the mechanical standards parsing above.
+- [ ] **Concept dependency graph** — Build prerequisite edges between standards and lessons. E.g., "5.NF.1 (add fractions) requires 4.NF.3 (understand fraction equivalence)." This is the core differentiator. Depends on Problem extraction + Standards tagging.
 - [ ] **Inline homework/exit ticket solutions** — Currently links out to Google Drive PDFs. Fetch, parse, and render answer keys inline on lesson pages. 148 homework + 135 exit ticket solution URLs available.
 - [ ] **Parent newsletter recovery** — 34 parent newsletter URLs redirect to Google sign-in (broken). Find working URLs or recreate the content.
 - [ ] **Grades 4, 6-8 EngageNY content** — Convert remaining grade DOCXs through the Aspose pipeline. Source ZIPs available in concept-research project.
@@ -29,6 +32,11 @@
 - [ ] **Tech launch** — Show HN + Product Hunt in the same week. Drafts in `docs/publicity-plan.md`.
 - [ ] **Homeschool push** — Well-Trained Mind Forums, SEA Homeschoolers, Freedom Homeschooling. Time for July-August (curriculum planning season).
 
+## Research
+
+- [ ] **remark-math** — Evaluate https://github.com/remarkjs/remark-math for math rendering in web content. It's a remark/rehype plugin ecosystem for rendering LaTeX math syntax in Markdown/HTML via KaTeX or MathJax. Could be useful if we want to author/store math as LaTeX and render client-side, or for teacher-authored content. Compare with our current approach (MathML from Plurimath rendered natively by browsers). Key questions: do we need client-side rendering, or is server-side MathML sufficient? Does remark-math play well with Hotwire/Turbo?
+- [ ] **Plurimath.js for editing** — Plurimath has a JavaScript version that could enable WYSIWYG math editing in the browser. Teachers could create exercises with a dropdown to choose math format (LaTeX, visual editor, paste from Word). We'd store as MathML. Evaluate feasibility and integration with Stimulus controllers.
+
 ## Infrastructure
 
 - [ ] **pgvector embeddings** — Embed lesson content and concept descriptions for semantic search. PostgreSQL + pgvector already planned in the architecture.
@@ -48,3 +56,14 @@
 - [x] Broken link audit — 34 parent newsletter URLs identified and cleared
 - [x] Publicity plan drafted — 6 phases with draft copy for each channel (`docs/publicity-plan.md`)
 - [x] EMBARC extraction plan documented (`docs/embarc-extraction-plan.md`)
+- [x] Domain model designed (12 tables): Grade → ContentModule → Topic → Lesson with LessonPlan, ProblemSet, ExitTicket, Homework, Assessment, Problem, Standard, Expression
+- [x] HTML normalizer (Nokogiri-based) strips meaningless whitespace from Aspose output
+- [x] HTML parser splits lessons into components by detecting Name/Date section boundaries
+- [x] Round-trip validation: 149/149 lessons pass (normalize → split → reconstruct = original)
+- [x] Browse UI: Grade → Module → Topic → Lesson with Tailwind
+- [x] OMML → MathML pipeline: 13,660 expressions extracted, 13,659 converted (99.99%)
+- [x] Plurimath fork (jcasimir/plurimath) with fixes for accent/overbar expressions — PRs submitted upstream
+- [x] MathConverter wrapper handles project-specific edge cases (track changes, Wingdings symbols) with manifest tracking
+- [x] Expression model stores OMML, MathML, text representation, and conversion status
+- [x] PDF comparison tooling (pdftoppm + ImageMagick) for visual validation
+- [x] Key finding: Aspose HTML is source data, not presentation layer — web templates should be purpose-built