2026-09-13, PR #5

Plan amendment — split Session 3 into 3a and 3b

One entry of the build log, as committed in docs/build-log/plan-split-session-3-2026-09-13.md. All entries

Date: 2026-09-13 · Branch: plan-split-session-3 · PR: #5
Documentation only: IMPLEMENTATION_PLAN.md, SESSION_PROMPTS.md, DESIGN.md, AGENTS.md and this entry.

(redacted for publication, 2026-10-02)

Goal

Let work continue while StimMap3D's deployment (S0-D) is still in progress. Session 3 required S0-D before it could start, and Session 4 builds on Session 3's MDX pipeline, so both sessions were blocked on the deployment.

What shipped

A dated split note at the top of both plan files. The run order is now S1 → S2 → S3a → S4 → S3b → S5 → S6 → S7 → S8.
Session 3a (s3a-projects-mdx): the MDX pipeline, the extended ProjectSchema, folding content/projects.ts into MDX frontmatter, every project card, the /projects index with its URL-synced filter, a reusable click-to-load EmbedFrame, and the StimMap3D page body with its disclaimer. It has no S0-D prerequisite. The 2026-09-12 Session 3 amendment now applies to 3a.
Session 4 now requires 3a to be merged, reads 3a's build log, reuses EmbedFrame for the valedictorian video and leaves the StimMap3D page alone.
Session 3b (s3b-stimmap3d-embed), placed after Session 4 in both files: live and repo links, a response-header check before framing, the embed on the StimMap3D page, the screenshots with public-commit provenance, cover and poster, re-checking 3a's quotes and metrics against that commit, the session-pack link, Playwright embed checks and the Lighthouse score. It requires S0-D, a public repo, and 3a and 4 merged.
S0-D, the timeline table, the acceptance summary and the session-time estimate now refer to 3a/3b.

Decisions and why

Split Session 3, not Session 4. Only a small part of Session 3 depends on the live app or public repo. Splitting Session 4 instead would have left its capstone and thesis pages, the heaviest remaining work, waiting on the deployment.
StimMap3D screenshots move to 3b. Repository provenance records a full commit SHA, and StimMap3D's commits may still change before the repo goes public, which would invalidate it.
EmbedFrame is built generically in 3a so Session 4's click-to-load video reuses it instead of adding a second component, and 3b only mounts it with the live URL.
No placeholder URL in content/. 3a leaves the embed and links out rather than committing a fact that is not yet true.
3a records the StimMap3D commit it quoted from so 3b can re-check quotes and metrics (such as the test count) against the public revision.
DESIGN.md and AGENTS.md were updated in a follow-up commit. The first commit touched only the two plan files, as Paige scoped it. Paige then asked for edits to DESIGN.md and AGENTS.md only if Claude Code recommended them. It did: AGENTS.md is loaded into every agent's context, and DESIGN.md §0 said StimMap3D must be public "before Session 3", which could stop a 3a session. Five references now say 3a or 3b (S0-D prerequisite, the projects.ts fold, the StimMap3D screenshot validation, and the session count in AGENTS.md). The split note's reading rule stays as a fallback.

Tools and model

Claude Code (Claude Opus 5): read the plan files, both session prompts, the current project loader and routes, and the private source inventory's StimMap3D section; edited the two plan files; git and GitHub CLI.

Constraints, human decisions and implementation direction

Paige asked whether Session 4 could run before Session 3 while S0-D is in progress. Claude Code recommended splitting Session 3 (3a → 4 → 3b) over splitting Session 4, waiting, or running Session 4 as written. Paige chose that option and asked for the change note in IMPLEMENTATION_PLAN.md and SESSION_PROMPTS.md, with a split Session 3 prompt, the new order and new branch and log names, in a new PR. The task lists, prerequisites and acceptance wording for 3a and 3b were drafted by Claude Code.

Generated work and rejected suggestions

Rejected options: splitting Session 4 (moves only the smaller Research and Leadership work forward), waiting for S0-D (stalls the build), and running Session 4 as written (it would improvise an MDX pipeline outside the approved Session 3 amendment and leave MDX-body metrics outside the claims gate).