docs(workflow): C001 trigger-issue draft (PR #381) mixes link bases — one pointer is dead, and the checker strips the fenced draft so it cannot catch it #382

Finding

PR #381 added a Stage 1 "Trigger Issue Draft" to doc/2026-10-19T06-09-02Z_research_research-intake-to-implementation/concepts/C001-transcript-preservation.md whose three Context Pointers use two different link bases, so no single place the draft can be read or copied from makes all three resolve — and the validation the PR cites cannot catch it, because the draft sits inside a fenced code block that the checker strips before link-checking.

Evidence

• PR: docs(workflow): prepare research angles and trigger issue for C001 #381 (merged, commit 25bc8067af63ad1aa969da302eb943934e96e983), still present on current main in the same file.
• Draft Context Pointers (file lines 75–77 on main):
  • [main.md](../main.md#c001) — resolves only relative to concepts/ (→ package main.md, which does exist, with an <a id="c001"> anchor).
  • [transcript.txt](../transcript.txt) — likewise resolves only relative to concepts/.
  • [concepts/C001-transcript-preservation.md](concepts/C001-transcript-preservation.md) — resolves only relative to the package root; relative to concepts/ it points at concepts/concepts/C001-transcript-preservation.md, which does not exist.
  • Verified by path resolution against current main: from concepts/ the first two exist and the third does not; from the package root the third exists and the first two escape to doc/main.md / doc/transcript.txt, which do not exist. Copied verbatim into a GitHub issue (its stated purpose — S1 drafts are the handoff for the research trigger), all three relative links resolve against the issue URL and break.
• The checker is blind to this by construction: scripts/research_workflow.py prose()/unfenced() strips fenced blocks, and links() runs on prose() output, so the ```markdown draft is exempt from check_link. Reproduced on a clean origin/main snapshot: python3 scripts/research_workflow.py check --repo-root . returns "ok": true while the broken pointer above is present — the same command PR docs(workflow): prepare research angles and trigger issue for C001 #381's body cites as its validation.

Why it matters

Stage 1's whole output is a self-contained, zero-context handoff (planning/workflow/stages.md: S1 = "trigger issue drafting"; C004: trigger issues are "zero-context handoff briefs for dedicated research workers"). A Stage 2 / handoff worker who follows the draft's own Concept Document pointer from the concept file's directory hits a nonexistent path, and one who pastes the draft into an issue gets three dead links. The workflow's link integrity guarantee — the thing check exists to enforce for every other link in the package — silently does not apply to the one artifact meant to leave the package.

Acceptance criteria

• The C001 trigger draft's Context Pointers all resolve from the draft's intended consumption point (e.g. make the Concept Document pointer C001-transcript-preservation.md / ./C001-transcript-preservation.md if the base is concepts/, or convert all three pointers to full repo-relative/blob URLs if the base is a GitHub issue) — one consistent base, verified by path resolution.
• scripts/research_workflow.py check (or an equivalent documented check) validates links inside fenced workflow:issue trigger drafts, or the concept doc records explicitly that draft links are unchecked and must be resolved when the issue is created.
• Re-running the check on the fixed tree passes, and deliberately breaking a pointer inside the draft makes the (extended) check fail.

Found by the pr-diff-review job reviewing PR #381 (merged) in developing-today/code. Cross-repo check: this research package and scripts/research_workflow.py exist only in developing-today/code, so this is a single-repo issue. Deduped: no open or closed issue mentions C001, the trigger draft, or research_workflow link checking.