docs: reconcile the backlog entries stranded with the July drafts

[abrichr]

Follow-up to #45, #46, #47 and #48. Backlog bookkeeping only — no posts, no workflow changes.

Each of the three stranded July auto-draft branches carried near-miss candidates its scan had found. #48 merged, so its 2026-07-28 entries reached main. #46 and #47 were closed, so theirs never did. This adds them.

Also records the two declined drafts: the idea in each that would clear the substance bar, and the branch and closed PR where the prose still lives, so neither is quietly lost.

One entry is deliberately dropped. openadapt-flow #153 appears in both #46's scan and the ## Scan 2026-07-19 section already on main, so it is not repeated.

How the three were decided. By the repo's own gate, not by taste:

lint_post_substance.py --strict
  PASS  2026-07-28  safety-taxonomy branch    1 warning, 0 floor violations   -> merged (#48)
  FAIL  2026-07-19  one-command pairing       776 words (floor 850), no thesis -> closed (#46)
  FAIL  2026-07-20  .report_run_id TOCTOU     712 words (floor 850)            -> closed (#47)

The one that landed is draft: true. deploy.yml runs hugo --minify with no --buildDrafts, so nothing was published; flipping the flag is still a human call.

Watermark. Already on main at 2026-07-28T02:24:34Z via #48. Worth noting why #46 showed as CONFLICTING: its state.json carried 2026-07-19T07:19:05Z, which is earlier than main's 2026-07-19T14:02:39Z. Merging it would have rewound the scanner over ground it had already covered.

Pipeline state. All three branches are now accounted for, so the guard stops skipping. The remaining blocker is unchanged: ADMIN_TOKEN still has no pull_requests: write here, so the next authored post will push its branch and fail to open the PR. NEEDS_YOU.md B44/B10.

🤖 Generated with Claude Code