Plan amendment — defer Session 3b behind Sessions 5 and 6

• Date: 2026-09-15 · Branch: plan-reorder-s5-s6 · PR: #8
• Documentation only: IMPLEMENTATION_PLAN.md, SESSION_PROMPTS.md, DESIGN.md, AGENTS.md and this entry.

Goal

Keep the build moving while StimMap3D's deployment (S0-D) is still in progress. The 2026-09-13 split put Session 3b immediately after Session 4, so the whole plan stalled on S0-D again once Session 4 merged. Paige asked which remaining sessions could run first.

What shipped

• A dated run-order note at the top of both plan files. The order is now S1 → S2 → S3a → S4 → S5 → S6 → S3b → S7 → S8. The 2026-09-13 split note keeps its wording and gains a pointer saying only 3b's position is superseded.
• The SESSION 3b prompt block moved below Session 6 in SESSION_PROMPTS.md, so the file's promise that prompts appear in run order stays true. Its prerequisite check now also looks for the Session 5 and 6 build logs, explicitly as expected but not required, and its closing line points at Session 7.
• Session 5 gains a prerequisite note (Session 4 only; the StimMap3D project has no live/repo link and no cover until 3b, and a missing link field is not an error) and the answer to its open Stryker question.
• Session 6 gains the same kind of prerequisite note, the name of the approved fallback resume, and the placeholder finding below.
• Session 3b gains a run-order note covering what Session 6 will have shipped by the time it runs: re-open /projects/stimmap3d/opengraph-image after setting cover, re-run the assets.json reconciliation now that the screenshots exist, and confirm Session 5's explorer derivation tests are untouched by the new frontmatter links.
• Session 7 gains an explicit prerequisite: it is the first session that genuinely needs S0-D, and there is no useful partial version of it.
• §3's prerequisite line, the §5 timeline table and the §7 acceptance summary follow the new order. DESIGN.md §0, §5.4 and §7.2 and AGENTS.md's session-order line were updated to match.

Decisions and why

• Defer 3b rather than split it further or wait. Sessions 5 and 6 read only content/ and existing routes, so neither needs the live app, the public repo or the screenshots. Nothing in 3b's task list is a prerequisite for either.
• 3b is deferred, not repositioned permanently. Its real dependency is S0-D plus Sessions 3a and 4. Both plan files say it may run as soon as S0-D is done, and its prompt tells the session to carry on (after saying so) if the Session 5 and 6 logs are absent. That avoids a second amendment if the deployment finishes early.
• Sessions 7 and 8 stay put. Session 7's task list requires axe with the embed loaded, an explorer filter, and Lighthouse on /projects/stimmap3d; pulling it forward would only split a session.
• Stryker is one entry, from the experience role (Paige, 2026-09-15). The content/research.ts Stryker item describes the same work. There is no link field between them, unlike the researchId and projectSlug pairs, so Session 5 must match them deliberately and keep the research item's skills on the single entry so no co-occurrence is lost.
• The approved fallback resume is the 2026-09-11 PDF (Paige, 2026-09-15), which answers the question Session 6's prompt was told to ask. Recorded in DESIGN.md §5.4 next to the fallback rule rather than only in the session prompt, because §5.4 is what a session reads first.
• The placeholder finding is recorded as a blocker, not a footnote. pdftotext on that PDF returns [ADD PORTFOLIO URL] in the page-1 contact line and [ADD LIVE URL] on the StimMap3D entry. DESIGN §5.4 and the Session 6 resume test both forbid a shipped [ADD string, so the fallback path is closed until Paige supplies a corrected export. The generated PDF renders from content/ and is unaffected, so Session 6 is not blocked — only its contingency is. Both placeholders need Paige's input rather than a guess: the portfolio URL commits to the production URL that serves nothing until Session 8, and the StimMap3D live URL comes from S0-D itself.

Tools and model

Claude Code (Claude Opus 5, 1M context): read both plan files, DESIGN.md §§0, 5.4, 7.2, AGENTS.md and the privacy-gate test; ran pdftotext -layout over the three staged resume PDFs' text (Git Bash); edited four documentation files; a short Node script moved the 3b prompt block with a length assertion so no content could be lost; git and GitHub CLI.

Constraints, human decisions and implementation direction

Paige asked whether any remaining session could be implemented before Session 3b while she finishes deploying StimMap3D, and whether to proceed with a different session now. Claude Code analysed the dependencies and recommended, in order, Session 5 then Session 6, with Session 7 staying blocked. Paige asked for a PR amending the plans and docs to record the reorder, answered the Stryker question (one entry) and named the approved resume file. The amendment wording, the 3b run-order note and the Session 5/6/7 prerequisite notes were drafted by Claude Code.

Generated work and rejected suggestions

Rejected: pulling Session 7 forward without 3b (its acceptance list cannot be met, so it would have to be split for no gain); rewriting the 2026-09-13 split note in place rather than marking one clause superseded (it is the historical record of what was requested); and leaving the 3b prompt where it sat with only a note (the file states its blocks are in run order).

Verification (command, result, PR/commit)

Recorded on the PR: npm run format:check and npm test (the privacy gate scans this entry). No app code changed, so no build, e2e or preview check is affected. The placeholder finding was verified with pdftotext -layout "<approved resume>" - | grep -c "\[ADD " → 2.

What the AI got wrong and how it was caught

• The first attempt to move the SESSION 3b prompt block used a single find-and-replace that swapped the 3b heading for a Session 5 heading and left 3b's body orphaned beneath it, corrupting the file. Caught immediately by re-reading the result; the edit was reversed and redone with a script that asserts the output length is unchanged.
• Two amendment cross-references said the new note was "above" the 2026-09-13 split note when it is below it. Caught by re-reading the rendered top-of-file order after the move.

Screenshots

None — documentation-only change.