docs: trinity landed on the 14th, carrying Windows and a return window

[Gerrrt]

What changed

The ProDesk was reported bought and delivered. It already was — one order, 2026-09-08, and hardware.md has carried it as trinity since it arrived. ADR-0034 decided one box does both jobs and the roadmap says a second is deferred, not ordered, so nothing here is a new purchase and no row on the buy table moves.

What the arrival paperwork carries that the documents did not:

• The date was a day late. The carrier's notice puts the drop-off at 15:05 local on 2026-09-14; the commit that recorded it was written that evening and dated by UTC, which had already turned over. Corrected in four files. The opening date is a separate reading, taken on the 15th, and stands.
• It arrived carrying Windows 11 Pro, which the rehearsal wipes — recorded with what it does not mean: the licence is OEM, it dies with the install, and it does not touch the buy table's two Windows 11 Pro keys, which are ADR-0029's lab-domain endpoints and a different machine.
• The warranty runs to 2027-09-08. ADR-0034 accepts that the tier is down until a replacement arrives; inside that year a dead trinity has a claim behind it.
• The seller-return window closes 2026-10-08, and this one bites. The spec, serial and NIC count are still the listing's word rather than the machine's, and the step that would settle them — installing pfSense over the Windows partition — is also the step that ends the return. So that reading gets a date. It is now the only item on the rehearsal's blocker list with a deadline, and pointedly not one of the two things that list is blocked on: the machine can be proved before the I226 card or the installer stick land, and has to be.

Two smaller things. The TS150, boot SSD and bracket were recorded under the same UTC slip and are not audited here — the roadmap now says so rather than implying four dates were checked when one was (#413's to settle). And the two SM863a SSDs were delivered 2026-09-11 while the roadmap still had them moving; moved, with the caveat that delivery is recorded against the order and both drives being in it is #418's to confirm at the bays.

Why

#92. The purchase half is done; the rehearsal is what remains, and it now has one prerequisite that is not waiting on a package.

Blast radius

Documentation only. No host, VLAN or service is affected.

• No change to network segmentation or firewall rules
• No new port published to a VLAN that could not already reach the service
• No credential added outside secrets/*.sops.yaml

Verification

python3 scripts/check_docs.py → docs OK — 82 Prometheus + 18 Loki rules, 7 dashboards, 141 panels, 10 assertions
./scripts/lint.sh --require-all → lint passed

Both re-run after rebasing onto fc4490e. The buy table and README's "two items now, one later" are untouched, as are the ADR and runbook counts.

• make validate passes
• Deployed to the lab and confirmed working — n/a, documentation only
• Docs updated

Left alone deliberately

An HP 2.5" drive caddy bought 2026-09-12 appears nowhere in the repo, against the roadmap's rule that a PR implying a purchase edits that section in the same commit. Recording it means saying which machine it is for — most likely Saruman's SFF bays under #418 — and that is a guess this PR is the wrong place to make. Tracked separately.

Refs #92

🤖 Generated with Claude Code