oracle's laptop cell reads 72 % and its alert is silenced to 2026-10-08: buy and fit it, second in line after prometheus

[Gerrrt]

What is true

oracle's laptop cell reads 72 % of design capacity — worse than the cell prometheus had when #454 was opened — and HostBatteryHealthLow for it is silenced until 2026-10-08. #454 is scoped to prometheus by title and body; docs/runbooks/replace-the-laptop-cell.md names oracle as "next in line" and "unbought" and nothing tracks it. When the silence expires, the alert comes back with no issue behind it.

oracle is second, not first, for the reason #454 gives: prometheus dying is the estate going blind; oracle dying is the backup archive and the wiki's drift check going quiet. Second is still a list.

Order of work

• Identify the cell: oracle is the Dell; hardware.md has the model, the runbook's §1 has the reading procedure. Record the part number in hardware.md before buying. Done 2026-09-19 in docs(oracle): the cell is a Dell M5Y1K, recorded before it is bought, and the runbook says what differs on the Dell (#531) #540: the host reports DELL VN3N047, which is Dell part VN3N0 — the Inspiron 15-3565's latched 4-cell 40 Wh 14.8 V pack, primary part number M5Y1K (equivalents WKRJ2, HD4J0, 991XP, GXVJ3). A bare battery, not a kit: it is a slide latch, not glue.
• Buy it. roadmap.md's consumable exception covers it — same clause, same paragraph, the Never line does not apply. Bought 2026-09-19, eBay 357495025211, sold as genuine Dell M5Y1K 40 Wh — the listing's claim until the pack is in hand. Recorded in hardware.md and off the roadmap's buy table.
• Before the fit: A battery disconnect resets the RTC, and the stack writes backdated samples that nothing notices #519's question — whether this Dell's RTC survives a battery disconnect — is worth answering by a rule or a decision first, because prometheus's did not and the stack wrote backdated samples for three hours. The hardware half is answered in docs(oracle): the cell is a Dell M5Y1K, recorded before it is bought, and the runbook says what differs on the Dell (#531) #540: the 3565's coin cell is a separate CR2032 behind the keyboard and palmrest, which a latch swap never reaches, and the kernel drives the RTC as rtc_cmos — so it is expected to survive, and the runbook now has the journal check to run at the fit. The rule-or-decision half stays with A battery disconnect resets the RTC, and the stack writes backdated samples that nothing notices #519.
• Fit per replace-the-laptop-cell.md, including step 8, the mains pull. Extend the runbook where the Dell differs.
• Delete the silence rather than let it expire.

Done when

charge_full at or near design at a low cycle count, the mains pull measured, the silence gone, and hardware.md carrying both cells as parts.

Refs #454, #519.