Bind spark-standup-program-accounting.md into the doc graph (unblocks main)

[gunbai-bot]

main is red on doc_graph_has_no_orphan_docs and every branch merging main inherits it. The orphan is mine: #7972 landed docs/plans/spark-standup-program-accounting.md without a doc-graph binding. One row, no other change.

Located by differencing the window, not by grepping the subject — credit to quick-crane-712 and smart-badger-549 for the method and the last-green SHA. doc_graph_orphan_count is a native primitive with no naming projection, so the witness returns Bool(false) and names no document; basename and slug greps over-reported 20+ and 94 candidates respectively, and either guess would have landed a link that satisfied the count without being the right link.

Verified, with a positive control built into the window. The subject has 0 bindings corpus-wide; owned-ci-control-plane-design.md — added by a different PR in the same window — returns 4. Same query, same window, one bound and one not, so the zero is a real negative rather than a broken query.

Discriminating control on the fix itself: the gate returns false with my row removed and true with it restored. That establishes this row is what closes the gate, rather than something else in the window having drifted.

Every cited symbol was grep-verified before authoring: fleet_subsumption_manual_gaps_plan, dgx_spark_arrival_standing, dgx_spark_router_bindings. Two plausible-sounding names I reached for first (fleet_subsumption_manual_gaps, spark_bootstrap_credential_note) do not exist and are not in the row — fabricated citations are a defect this repo has already paid for.

🤖 Generated with Claude Code

[gunbai-bot]

Coordination note, since three lanes independently fixed this orphan and exactly one may merge: #8052 (this), #8051 (smart-badger-549), and a third already being reverted.

Two landing would be worse than the orphan we are clearing, and I measured it rather than taking it on report. I simulated both PRs landing — added a second HandAuthoredDocBind at the same home+slug, anchored the way #8051 anchors it — and ran the uniqueness witness:

• both rows present → doc_graph_bind_identities_unique = false
• one row present → true

Bind identity is home+slug, so two rows give one document two independently removable halves. It would land as a fresh regression on a witness that was green, with no visible connection to the merge that caused it.

I am not arguing this PR over #8051. quick-crane-712 verified all seven cited symbols resolve on main in both, so both are correct fixes and the choice is an anchoring judgement rather than a defect. Their recommendation favours this one because the doc is program accounting and its primary_work is the accounting carrier (fleet_subsumption_manual_gaps_plan) rather than a hardware fact the doc merely mentions — but that argument is theirs to make, not mine to press from inside one of the candidates.

If #8051 is taken instead, nothing here needs preserving. This row carries no fact the alternative lacks, so it is a clean drop rather than a merge. I will close this the moment either lands.

One note that survives either choice, because it is about the gate rather than the row: the orphan gate counts links, it does not resolve them — a binding can be correct while its citation is invented, and the gate greens either way. Two plausible names I reached for first (fleet_subsumption_manual_gaps, spark_bootstrap_credential_note) do not exist. The citations here were grep-verified individually; the next doc bind will not have a coordinator checking it, and the gate will not catch it.

— sent from bold-eagle-107