5. POWER ARCHITECTURE (the keystone)

5.1 Three Power Tiers (matched to energy source)

rev 3.3: numbers re-baselined to the OptiPlex (~80W) brain, not the shelved Z440 (~350W). The old 350–400W "driving" figure was a Z440 number — gone.

5.2 The Numbers (why it tiers)

8-camera record (~58W):  ~116 Ah/day @ 12V → battery dies in hours
2-camera tier (~15W):    ~30 Ah/day  → 400W solar covers it ✅
ESP32-only idle (~5W):   ~10 Ah/day  → trivial

400W solar realistic PNW input:
├── Good summer day: ~1,500-2,000 Wh
├── Average: ~800-1,200 Wh
└── Poor/winter: ~300-600 Wh

⚠️ Winter-margin reality (rev 3.3): the 15W parked tier needs 15W × 24h ≈ 360 Wh/day — that's the floor of the poor/winter range, so on a bad PNW/BC week solar does NOT sustain even the 2-cam tier and it leans on battery reserve. Honest statement: solar sustains the long-park tier in ~3 seasons; deep winter draws the bank down. Default winter parked state should be ESP32-only (~2–5W ≈ 50–120 Wh/day), which solar covers year-round. This matters more after the a future move (worse latitude/cloud).

5.3 Power Flow (rev 3.8 — full layout in §5.9)

STARTER SIDE                          HOUSE SIDE (all − to the BATTERY-NEG BUS BAR, §5.9)
[start battery + 2nd lead-acid         [HOUSE 100Ah hood] ──── carputer loads
 (winch buffer)]                          │  ├ 12V→19V converter → OptiPlex
   │ alternator                           │  ├ Pi · ESP32 · cameras · radio
   └─[VEVOR 40A DC-DC]──(ign/V-sense)──────┘  └ inverter (− direct to bus bar)
                                          │
400W Solar → EPEVER MPPT (in bed) ───pos-only lead───┘ (neg → bed bus bar/chassis)
                                          │
                          [ESP32 contactor + pre-charge]── BED 600Ah (removable, Anderson)
GROUND: battery-neg bus bar = star point; ONE heavy bond → frame (CAN/sensor reference)

5.4 Engine-Off Logic + Sleep-State Ladder (rev 3.5)

Engine off → ESP32 isolates the OptiPlex converter to the house battery (starting battery never drained), accessories on a 15-min timer. The OptiPlex then walks an ACPI sleep ladder, shallow→deep, run by the Pi wake-coordinator (clock + door/ignition events + battery telemetry), executed on the OptiPlex via systemctl suspend/hibernate + RTC/BIOS-Auto-On self-wake.

ACPI states (5080 Micro estimates — ⚠️ bench-measure, gates §13):

(S4 and S5 draw the same with WOL-standby alive — S5's value is a CLEAN boot + robustness on long parks, not power.)

The ladder (the owner's tuning):

0–7 min parked   → S0 (stay on — finish housekeeping, instant return)
7 min – 2 hr     → S3 suspend (~1-3W, wakes 2-5s → ready before the door)
2 hr+            → S4 hibernate (~0.5-2W, wakes ~20s)
nightly 03:00    → COLD BOOT (BIOS "Auto On Time" / rtcwake) → maintenance window → back to rung
24hr no USER session  OR  battery low
                 → ESP32 HARD-CUTS the OptiPlex converter = TRUE 0W, stays off till you return

Nightly 03:00 maintenance window (cool + idle): pull the Pi→5080 event-log backlog · back up 5080→NAS/Hetzner/home · roll the day's events into RAG/memory · OS/ESPHome/app updates over Tailscale · health heartbeat (a missed 3am wake = something's wrong). Then drop back to its rung.

• ⚠️ The 3am maintenance wake does NOT count as a "user S0 session" — otherwise it resets the 24hr storage timer every night and the truck never reaches deep storage. Only ignition / you-returned wakes reset that timer.
• Battery-gated: skip the 3am wake if the house bank is low (don't spend ~3-9Wh on housekeeping when short — matters on thin winter solar). LVD is the hard backstop.

Deep storage (24hr idle or low battery): the always-on ESP32 (on its own 12V) hard-cuts the converter feed → the OptiPlex is genuinely off at 0W (better than S5's ~0.5-2W standby), no nightly wakes. Return = the ESP32 re-powers the converter on ignition OR keyring beacon OR dash button. (WOL is gone once power is cut — but you're physically present, so the ESP32 re-power path covers it.)

• Reverse cam NEVER on the OptiPlex path — GreenYi → radio is instant; the OptiPlex's ~10-25s resume is irrelevant to backing up.

5.5 12V vs 48V Decision — ✅ SETTLED: 12V

Resolved (rev 3). The deciding check below is answered: the EPEVER 30A TracerAN is 12V/24V-only — no 48V (confirmed). Per the rule below, that forces 12V (matches the truck, EPEVER works, loads native). System voltage is 12V, settled — the 48V golf-cart bank is unused for this build. The analysis below is retained for the record; it is no longer an open decision, and this is NOT a pre-spend blocker (the stale §13 entry has been corrected).

You own both a 12V 100Ah (1,280 Wh) and 48V 100Ah (4,800 Wh) LiFePO4.

48V advantages: 4× the energy (solves camera wall — ~2.7 days @ 58W),
                1/4 the current, thinner/cooler wiring, more efficient inverter
48V costs: truck is 12V-native → need 48V→12V converter for 12V loads,
           12V→48V DC-DC charger (rare/pricey) for alternator charging,
           and a 48V-capable MPPT (EPEVER 30A may be 12V/24V only)

DECIDING FACTOR: Does the EPEVER 30A MPPT support 48V?
├── YES → 48V attractive (solar already handled, huge capacity)
└── NO  → staying 12V is simpler (matches truck, EPEVER works, loads native)  ← ✅ THIS BRANCH

ANSWER: EPEVER 30A TracerAN = 12V/24V-only (no 48V), CONFIRMED → 12V chosen.

PRAGMATIC PATHS:
A) Stay 12V: simplest, 1,280Wh, fine for driving + 2-cam parked tier  ← ✅ SELECTED
B) Go 48V: 4,800Wh end state, but +48V MPPT +48V→12V conv +12V→48V charger  (not pursued)

~~TODO: confirm EPEVER 30A model's 48V capability~~ → DONE: 12V/24V-only confirmed. Decision closed.

5.6 Output & DI Maps (rev 3.6 — DO triggers → Bosch panel)

Architecture: the controller's 8 digital outputs each TRIGGER one socketed Bosch 30-40A relay on the external fuse/relay panel (§6). The panel relays carry the load; the DO just pulls the coil. So there's no "10A onboard relay" limit — the Bosch relays handle 30-40A, sized per load, and swap out individually. Tiny loads/enables (≤500mA: OptiPlex pulse, converter enable, compressor signal) can be driven by a DO directly, no Bosch relay.

Outputs (rev 3.17 map — matches the full-wiring sheets; safety on the board, lights on the RS485 module):

• O1 — OptiPlex converter power → Electronics-Salon MD-D262T/12V passive latching DPDT (NOT in the Bosch panel; TE/Schrack RT424F12, zero standby. V+/S/R active-low → 2 DOs; Switch X = converter 12V, Switch Y = read-back to DI7. Holds through reboot/OTA, true 0W cut — §6.3)
• O2 — accessories 15-min timer cut → Bosch relay
• O3 — OptiPlex power-btn pulse (DO DIRECT; cold-boot / hard reboot)
• O4 — camera tier (Switch B) power → Bosch relay (cut when asleep)
• O5 — bed contactor coil → Bosch relay (§5.9 combiner; sheet 6)
• O6 — pre-charge relay coil (DO direct — Bosch coil <500mA; sheet 6)
• O7-O8 — spare (compressor DROPPED from the build — rev 3.17; runs standalone outside the carputer)
• Lighting DOs live on the RS485 Modbus module (§6.1/§6.2): M-DO1 rock · M-DO2 KC-bar trigger · M-DO3 underbody · M-DO4 pods · M-DO5 amber/marker · M-DO6 master isolation

Digital Inputs (8): 1 door (dome-light wire) · 2 ignition (12V-switched) · 3 reverse · 4 dash button · 5 cab-temp/sensor · 6 combiner override (+ ADC2 bed-V sense feeds the combiner logic, sheet 6) · 7 O1 read-back (the latching DPDT relay's pole-2 aux contact = true on/off state, keeps O1 state-synced through reboots, §6.3) · 8 spare

Wake/state via physical DI wires, not CAN — more reliable, and it's why no always-on CAN node is needed (§8).

Wire sizing (rev 3.17 — SIMPLIFIED TO THREE PURCHASED SIZES): DO→coil trigger = owned solid Cat5 / 20-22 AWG (one Cat5 = 8 triggers; secure for vibration, §6). Every relay→load run = 16 AWG, fused ≤10A (the fuse protects the wire; all remaining loads are ≤10A now the compressor is out and the KC bar's 12 AWG ships in its kit harness). 6 AWG = DC-DC in/out, MPPT, computer-box feed. 2 AWG = trunk ±, battery leads, frame bond (covers the old 4 AWG jobs). Rule: any future load >10A gets its own dedicated thicker run — never up-fuse a 16 AWG wire.

5.7 Lighting — high-output LED, on with ignition off (rev 3.4)

Goal: multiple high-output LED bars (e.g. KC FLEX ERA) controllable from voice/app/switch, on with the engine off, fed from the house bank, protected by the ESP32. (Exact quantity/sizes still pending — needed to finalize the wire/fuse/relay map.)

Lighting splits by cluster (rev 3.32, Option B §6.5):

• FRONT board → KC front-bumper light rack (forward-facing). On-road interlock applies (inhibit above ~5 mph on-road, #4 below) + the diode-OR'd manual KC switch fail-safe (#2 below). Plus any front rock/underbody.
• BED board → tent/rack CAMP lights (the 3+ sets, rearward/hangout). No on-road interlock (camp lights, run anytime); local tailgate switch control.

Reference load — KC FLEX ERA 20" kit (SKU 0292 = two 10" bars): 216W white / 18A, 24W amber / 2A, 9-36V. A 50" Flex Era is 45A — that one needs a proper contactor, not the harness relay.

1. Fed from the HOUSE bank, gated by the always-on Waveshare — never the starting battery. "On with ignition off" works because the feed is the always-on 12V house bus, switched by a relay the (always-on) ESP32 controls. This is why the Waveshare is always-on.

2. The ESP32 triggers, it doesn't carry the current. 18A would smoke the 10A relays. But the KC kit already includes a harness relay + illuminated 3-position switch, so no separate Bosch relay is needed:

House 12V (always-on, fused ~25A/kit) → KC harness relay (carries 18A) → light bar
                                              ↑ trigger (mA)
                          ┌───────────────────┴───────────────────┐
                   KC dash switch (manual, always works)    Waveshare relay output (voice/app/auto)
                   ──── diode-OR'd: either source turns it on ────

Keep the manual KC switch as a fail-safe (lights work even if the computer is dead); the Waveshare trigger adds voice/app/automated control.

3. The real reason to route through the ESP32 = protection. Runtime on the FEENCE 100Ah (1,280Wh):

White = work-session lighting, not all-night. The ESP32 enforces: LVD cutoff (kill lights if the bank drops below threshold), auto-off timer (don't drain the bank because you forgot), and a runtime estimate from the WonVon shunt on the Flask/PWA dash.

4. On-road interlock (FRONT board, KC bumper rack). Forward-facing high-output bars are off-road-only when lit on public roads → software gate: the front board inhibits the KC bumper rack above ~5 mph on-road — speed comes from the OptiPlex's CAN over MQTT (you're driving = OptiPlex awake), or fall back to a vehicle-speed DI; rock/underbody/amber-marker and the bed camp lights stay freely available (camp lights are rearward — no interlock).

5. Wiring (per kit, once quantity is set): ~12 AWG, own ~20-25A fuse off a distribution block on the house bus, one relay-trigger circuit each. §5.6 already reserves R5 rock lights / R6 light bar / R7 underbody.

5.8 Battery Purchase — buy SOON (2026 rising market) (rev 3.37)

Timing: buy the capacity-add battery sooner, not later. 2026 is a rising market on both ends — battery-grade lithium carbonate ~doubled (to ~$26k/ton, highest in 2+ yrs), and the US Section 301 tariff on Chinese non-EV Li batteries jumped 7.5% → 25% (Jan 2026), with combined effective LFP-cell rates heading toward ~82% under already-scheduled increases. The "wait for it to get cheaper" era is over; waiting most likely costs more.

Strategy shift (rev 3.37 — SIZING UP for overland): the owner's target moved from "carputer-only" to ~7 kWh overland house power (fridge/camp/tent for days). That reopens the size question — but NOT toward the cheapest big "600Ah" sticker. The honest path to 7 kWh is either a reputable single 560Ah pack or two ~300Ah packs. The distributed two-pack option now wins on more than redundancy: at ~7 kWh a single pack is 55–60 kg — too heavy to lift out, which breaks the §5.9 removable-bed concept; two ~30 kg packs stay one-person liftable AND give redundant BMS.

⛔ Dumfume 600Ah — REJECTED (rev 3.37). Linked Jun 2026 at $705 / claimed 7680 Wh. Fails the weight-sanity rule below: 7680 Wh ÷ 48.9 kg = 157 Wh/kg, above the 120–150 band. Apples-to-apples, a genuine LiTime 560Ah (7168 Wh, slightly less capacity) weighs 55–60 kg (119–130 Wh/kg). Dumfume claims more Ah at 10–20% less weight → real capacity ≈ 470–500 Ah (~20–25% inflated). On honest capacity that's ~$116/kWh — no better than an established brand, with none of the support. (Dumfume's 314Ah tested clean on DIY Solar Forum; the 600Ah is a different story — eBay "AVOID THIS SELLER" + fake-review complaints.) Not a deal — an inflated label.

Candidates priced (Jun 2026):

Carputer-only baseline (superseded by the overland resize, kept for the record): Dumfume 314Ah $320 / 3.84 kWh / ~$83 · E-LekTech 440Ah $560 / 5.63 kWh / ~$99 · Goldenmate 400Ah $599 / 5.12 kWh / ~$117. Brand note: LiTime = AmpereTime = Power Queen (one owner); Redodo has the fewest reported QC issues of the budget brands.

Selection rules:

• $/kWh is the real comparison, not sticker price.
• Weight = capacity sanity check: large packs run ~120-150 Wh/kg; suspiciously light = inflated Ah. This rule just killed the Dumfume 600Ah (157 Wh/kg). Calibration: genuine LiTime 560Ah = 119–130 Wh/kg; LiTime 460Ah 8D = ~151 (tightly-packed = top of band). Anything above ~150 on a big pack = treat as inflated until a measured-capacity test proves otherwise.
• Vet reviews for measured-capacity tests + BMS reliability — unknown Amazon brands inflate Ah and ship weak BMSs. This is the single most important check.
• Sizing (rev 3.37): 3.84 kWh is plenty for the carputer alone (~38h awake at ~100W / weeks parked at ~5W + 400W solar). the owner is now sizing for overland → ~7 kWh (fridge/camp/tent for days) — that's the deliberate "serious overland house power" case, not over-buying. Note the 400W solar / EPEVER MPPT and the §5.9 combiner are unchanged; only the bank capacity grows.
• App-BMS is mostly redundant — you own the WonVon BT shunt and the carputer reads pack voltage already. And catching a drifting cell on a sealed pack is mostly non-actionable (warning, not repair — you replace the pack, not the cell). Not a reason to pick an unknown brand. (True cell-level serviceability = a DIY pack with bolted cells + JBD/Overkill smart BMS — a bigger project.)
• Mount COOL (cab/bed), not under-hood; no cheap LiFePO4 charges below 0°C — plan for winter.

Decision (rev 3.37): Dumfume 600Ah REJECTED. Target = ~7 kWh overland, recommendation two ~300–314Ah packs (LiTime or Redodo) — redundant BMS + each ~30 kg stays liftable, so the §5.9 removable-bed concept survives. Alternative: single LiTime 560Ah (~7.2 kWh, reputable) IF the owner accepts a 55–60 kg, effectively-fixed pack (then §5.9's "removable" premise needs a note). Final pick still PENDING a measured-capacity review on the chosen units (the one non-negotiable check) + the cold-charge plan (§5.10 bench-balance, no charging <0°C).

5.9 Power System Layout — Charging, Combiner, Grounding, Wiring (rev 3.8)

Charging (two separate, voltage-following sources)

• Solar → owned EPEVER 30A MPPT (kept). Alternator → VEVOR 40A alternator-only DC-DC (~$104, the 34-review line — NOT the combined-MPPT VEVOR, which does solar-OR-alternator, not additive).
• Why separate, not the combined unit: two independent chargers add up (~70A driving + sunny); the combined VEVOR can only run one input at a time (solar priority).
• DC-DC input taps a 2nd lead-acid on the STARTER side (a winch buffer — buffers the 400A+ winch pulls). The alternator charges that battery; the DC-DC's ignition/voltage-sense stops it draining the lead-acid when parked. ✅ valid (the DC-DC sees the same alternator voltage there).
• Set BOTH chargers to the same LiFePO4 profile (14.4V abs / 13.6V float) so they share cleanly. Optional longevity: set a ~90% daily target + occasional full charge for cell balancing; mount the big pack cool.
• VSR / dual-battery isolator: NOT used. The DC-DC does alternator-charging properly for LiFePO4 (right profile + current limit + isolation); a VSR is the inferior way and can't combine two house batteries (it drops the aux on discharge).

Combiner — removable bed battery (ESP32 contactor + pre-charge)

Combine the hood 100Ah + bed ~600Ah into one bank, automatically, fail-safe.

⚠️ rev 3.37 resize note: "440Ah" throughout §5.9 was the single-pack figure; the bed bank is now sized to ~7 kWh / ~600Ah (§5.8). The contactor (150–200A), trunk fuse (125A) and 4 AWG bed feed are sized to load current, not pack Ah, so they carry over unchanged. BUT the force-merge / separate-on-loss C-rate math below (and the "~90Ah usable / ~400Ah needed" figures) was computed for a single 440Ah pack — RECOMPUTE once the final config is locked: a single 560Ah ≈ scale the numbers; two ~300Ah packs = re-topologize (likely each pack its own SB175 + fuse, paralleled on the bed bus bar, with combiner logic per pack). Do this when §5.8's "PENDING measured-capacity review" resolves.

            ┌────────[ CONTACTOR (150-200A, continuous-duty, NO=default OPEN) ]────────┐
HOUSE BUS ──┤  ESP32 closes only when voltages match                                    ├── BED 600Ah
            └──[ Bosch relay (from panel) ]──[ ~5Ω 25-50W resistor ]───────────────────┘   (Anderson SB175)
                 PRE-CHARGE bypass: ESP32 closes this first → resistor limits inrush →
                 banks equalize → then close the contactor → open pre-charge.

• Switch the POSITIVE side (negatives are common — switching the negative wouldn't isolate). Contactor + pre-charge both in the positive.
• Coil driver: ESP32 DO → a Bosch relay → contactor coil (coil exceeds the DO's 500mA).
• Voltage sensing (how it knows ΔV) — rev 3.30: use the Modbus RTU DC Monitor (SKU 33931, ~$27). 4-ch, 16-bit, 0-36V direct over RS485 — front bank → ch1, bed stud A → ch2 (+2 spare). This closes the open ADC-pin question (read both banks over the same Modbus bus, no bare ADC pin, no DIY divider). ⚠️ Its current channels are only ±8A — useless for the 100A+ banks, so keep the WonVon shunt for main current; the DC Monitor is voltage-only here (its current is for <8A loads if ever wanted). Needs a common ground (the star ground satisfies this). Modbus latency is fine for the second-scale combiner sequence (§6.2). (Superseded the earlier DIYmall-0-25V-on-a-bare-ADC-pin plan — cleaner, higher-res, and no ADC-pin dependency.) The board's DI range is 5-36V, so the DI7 read-back (12V from the DPDT pole) and all 12V vehicle signals still work direct, no level-shift.
• Fail-safe: NO contactor defaults OPEN → ESP32 dead = bed battery isolated.
• Deeply-depleted bed battery: don't dump the house into it — ESP32 combines it only while a charge source is active (solar is always installed), so the source refills it. Close/equal voltages → straight pre-charge + combine.

Layout — bed sub-bus-bars + dual trunk (rev 3.9)

BED                                              FRONT
[bed 600Ah]──[ESP32 contactor+precharge]──┐      ┌── 100Ah ── computer fuse/relay box
[inverter] ───────────────────────────────┼ BED ─┤   FRONT +/− BUS BARS
[MPPT (solar)] ────────────────────────────┘ +/− │   (the STAR point; ONE chassis bond here)
                                  BUS BARS └─trunk +(fused BOTH ends)─┘
                                           └─trunk −(the bed↔front ground)─┘

• Bed gets its own +/− bus bars — bed battery (via contactor), inverter, MPPT all land there locally. One heavy + trunk and one heavy − trunk run to the front bus bars (the star point).
• Solves MPPT-no-battery: the MPPT always reaches the front 100Ah through the trunk; only the bed battery is removable (behind its contactor). Inverter fed locally by the bed battery (short heavy leads).
• Star, not daisy-chained: the inverter and the computer fuse-box each get their own cable to the + bus bar — never tap the computer off the inverter's cable. The low-ESR LiFePO4 at the bus bar clamps the voltage, so the inverter can't sag/noise the computer.

Grounding — single-point / star (one chassis bond)

• The − trunk is the dedicated bed↔front ground. ONE deliberate heavy bond, front − bus bar → frame — the only chassis tie (references CAN, the ignition/door/reverse DI wires, sensors, GreenYi to vehicle ground).
• Do NOT also chassis-bond the bed — two paths bed↔front (the trunk + the chassis) = a ground loop: circulating chassis currents → audio whine + CAN errors (the #1 noise cause). One bond, at the front, only.
• Never chassis-return the high-current stuff — all high-current negatives go to a bus bar with dedicated cable, never through the chassis. Do NOT chassis-ground the inverter.
• Wire = pure COPPER for all power runs. CCA (copper-clad aluminum) has ~40% more resistance + lower real ampacity + corrodes at terminals — if you must use it, upsize 1-2 gauges and fuse to its actual ampacity, not the copper-gauge number.

Connector (removable bed battery)

Anderson SB175 + handle accessory (the handle is the fix for "annoying to connect" — leverage). Or lever-action battery terminals (disconnect at the posts). Size 100A+ (inverter on the combined bank).

Plug-in behavior (rev 3.18 — sheet 6): plugging in is sparkless and automatic — the contactor is open, so the SB175 mates onto a dead circuit (only the ~110k ADC2 sense, ~0.1mA). The ESP32 sees pack voltage on ADC2 instantly, debounces ~5s, runs the pre-charge sequence → combined in ~15-25s when the banks rest within ~0.3V (the common case), no user action (announce via MQTT/PWA; policy = auto-combine when rules pass, or notify-and-confirm). Pre-charge physics (rev 3.19 correction): the resistor fixes the caps (τ≈50ms) and surface charge (seconds) — it can never close a real SoC gap (0.1-0.4A into 100-440Ah ≈ nothing). So the close criterion is acceptable inrush, not equal banks: ΔV ≤ 0.3V → close (~10-20A blip) · 0.3-0.7V + charge source → close = managed merge (brief 20-50A into the low bank, ~0.1C, decays to charger rate) · > 0.7V → stay isolated + alert, charge the bed first. Persistent ΔV after ~15s is information (a real SoC gap), not "not done yet."

FORCE-MERGE override (rev 3.20 — the "bed pack stored 6 months at 12.0V, leaving NOW" case): 12.0V resting ≈ 0-5% SoC → ΔV ~1.8V vs a charging front bank → the >0.7V band refuses by default, correctly. The override: allowed ONLY with ignition ON + DC-DC active + explicit user confirm (PWA gate / override switch). Profile: ~100-120A for the first 1-2 min (the SoC cliff lifts the bed's voltage fast), then 30-60A sustained with the DC-DC backfilling 40A — ~0.25C into the 440Ah, brief ~1C off the 100Ah. Refuse below ~11V (damaged / BMS-tripped → bench supply §5.10 only). What "keep it rare" actually means (rev 3.20a): the close event itself is cheap — contactors are make-rated for thousands of such cycles, the front bank's ~1C blip is in-spec, the DC-DC charging the bed while driving is the system's normal job. Keep force-merge rare for two other reasons: (1) the inrush is uncontrolled (protected by ratings headroom, not current regulation — a designed-routine deep merge would warrant a real current limiter), and (2) needing it often means the pack keeps sitting at 0-5% — and low-SoC storage is what actually harms LiFePO4 (calendar aging + BMS drain → over-discharge/brick risk). Store at 50-60% and every reinstall is a normal-band merge — routine forever, no override.

SEPARATE-ON-CHARGE-LOSS rule (rev 3.21): at key-off / charge-source loss, if the bed is still below the flat band (rest < ~13.0V — an unfinished force-merge), the ESP32 OPENS the contactor. Otherwise paralleled banks must equalize: a ~95% front 100Ah bleeds 60-80Ah into a low bed overnight (15-25A tapering over ~10hr) until both sit ~20% — harmless to the batteries (charge moves, nothing is lost) but it silently empties the critical front reserve. Healthy banks (both flat-band, OCVs ≈ equal) have negligible cross-current → staying combined parked remains fine and intended. Re-merge is automatic when a charge source returns (morning solar counts). Reality check: the 100Ah can't "fill" a 440Ah (~90Ah usable vs ~400Ah needed) — the alternator does the filling; a 2hr drive ≈ +80-100Ah ≈ bed at 25-30% on arrival. Pre-checks: meter the SB175 pins (BMS may be in low-V disconnect = reads ~0V), and no charging below 0°C. Storage tip: store the bed pack at ~50-60% (rest ~13.2V) + quarterly top-up — and the 12.0V morning never happens. Chain order: battery + post → MRBF fuse → 275A rotary disconnect (red key, ~$12, at the pack) → SB175 → contactor stud A. The rotary gives a guaranteed-dead, key-out service lockout independent of software/contactor (welded-contactor or firmware-bug case). ⚠️ Unplug rule: SEPARATE FIRST (PWA / override switch / rotary OFF), then pull the SB175 — while combined, ADC2 reads bus voltage through the closed contactor, so the ESP32 cannot see a yank, and the connector would break load current (arc).

Wire / fuse / resistor table (copper; sizes + LOCATIONS)

Fuse rule: every positive is fused, near its source (at the bus bar, or at a battery's own + terminal). Negatives are NEVER fused — straight to the − bus bar. A battery-to-battery interconnect (the trunk) is fused at BOTH ends.

OptiPlex power notes

• Owned 12V→19V 10A converter — hardwire the converter's 19V output to the board's + and − DC-input pads INTERNALLY (open the 5080, two wires, omit the barrel jack for vibration). ⚠️ Verify polarity against the genuine Dell brick's tip with a meter before connecting. Add an inline fuse + reverse-polarity diode, and feed the converter's 12V input through the ESP32 power-mgmt relay (O1, so deep-storage can cut it). 19V is in tolerance (Dell spec 19.5V; trim to 19.5 if adjustable).
• Dell no-ID 800MHz throttle (real on OptiPlex Micros). Fix options, cheapest first:

1. Free in software (primary) — disable BD PROCHOT / set RAPL via MSR on Linux (what ThrottleStop does on Windows).
2. Harvest the genuine Dell adapter's ID chip — the owner has a real Dell PA-12 65W brick on hand (LA65NS2-01, output 19.5V / 3.34A / 65W). The center-pin 1-wire ID chip is parasitically powered → the 5080 reads it even with the brick unplugged from AC. Splice that center wire into the internal feed so the board sees a genuine Dell ID. Clean and free since the brick is owned.
3. ESP32 emulates the Dell 1-wire ID (backup, if no brick) — free, you own ESP32s.

• 65W ID caps the CPU envelope at 65W. That's enough for a snappy bursty local SLM (heavy work → Fireworks). To "go higher on watts" you'd need a 90W/130W Dell ID, not a higher voltage. Then raise the RAPL power limit — fine thermally because the SLM is bursty.

5.10 Bench supply & LiFePO4 cell balancing

• NICE-POWER 30V / 10A adjustable bench supply (~$60) — self-contained (no USB-PD trigger needed, unlike the DPS-150). Used to bench-balance the big LiFePO4 pack: set 14.4V boost and HOLD it (NOT float) so the BMS keeps all cells at the top long enough to passive-balance — leaving it a week is harmless. Do this on a new pack and occasionally after.
• ⚠️ The owned wheeled lead-acid charger is for the lead-acid starter + winch-buffer batteries ONLY. It has no lithium mode, and its desulfation/equalize pulses damage LiFePO4. Never put it on the lithium packs.