feat: claude review of <= 66

52-vfd-pump-predictive-maintenance/README.md

@@ -12,7 +12,7 @@ This project is a [downtime prevention](https://blues.com/downtime-prevention/)
 
 ## 1. Project Overview
 
-**The problem.** Industrial centrifugal pumps almost universally run behind a VFD. Modern drives from ABB, Yaskawa, Danfoss, and Schneider all expose the pump's operating telemetry through Modbus holding registers — motor current, output frequency, output torque, drive temperature, runtime hours, active fault code, and almost nobody reads them. The VFD is sitting in the cabinet doing the work; the data is right there. What's missing is the network path off the plant floor.
+**The problem.** Industrial centrifugal pumps almost universally run behind a VFD. Modern drives from ABB, Yaskawa, Danfoss, and Schneider all expose the pump's operating telemetry through Modbus holding registers — motor current, output frequency, output torque, drive temperature, runtime hours, active fault code. Almost nobody reads them. The VFD is sitting in the cabinet doing the work; the data is right there. What's missing is the network path off the plant floor.
 
 A failing pump rarely just stops. It signals first: motor current can shift at constant frequency as bearing drag, fouling, valve position, fluid viscosity, or impeller condition change the load on the drive. Transient electrical faults can cluster on a stressed contactor or in a cavitating hydraulic regime. Actual runtime can drift above expected duty cycle as a downstream valve fouls and the pump runs longer to do the same work. *None of these telemetry shifts are diagnostic by themselves*, but they are leading indicators a maintenance team would happily act on a week early. This project is the device that catches those anomalies and routes them out, before a tank runs dry on a Saturday night. See the [Signal limitations](#12-limitations-and-next-steps) note for what this telemetry can and cannot conclude.
 
@@ -165,7 +165,7 @@ Dependencies:
 
 Six holding registers are pulled in a single Modbus transaction (`requestFrom` with `HOLDING_REGISTERS`, quantity 6, starting address taken from `reg_freq`). Reading all six in one transaction is roughly 8× cheaper in time and bus utilization than six individual reads. The firmware assumes the six registers are contiguous; on drives where they aren't, override `reg_*` variables individually and the firmware falls back to per-register reads.
 
-Polling cadence is `sample_minutes` (default 1 minutes). For each sample, `accumulate` updates the rolling hourly windows for current, frequency, torque, and drive temperature — separately tracking samples taken while the pump is *running* (frequency > 1 Hz) versus *stopped*, since a hot drive at 0 Hz is a different signal than a hot drive under load.
+Polling cadence is `sample_minutes` (default 1 minute). For each sample, `accumulate` updates the rolling hourly windows for current, frequency, torque, and drive temperature — separately tracking samples taken while the pump is *running* (frequency > 1 Hz) versus *stopped*, since a hot drive at 0 Hz is a different signal than a hot drive under load.
 
 ### Event payload design
 
@@ -226,7 +226,7 @@ The OPTA + expansion is line-powered (24 VDC), so MCU sleep is not the goal —
 ### Retry and error handling
 
 - The first `hub.set` uses `notecard.sendRequestWithRetry()` with a 5-second window — there is a known cold-boot race condition where the host comes up before the Notecard is ready to receive I²C transactions.
-- Modbus reads are retried up to 3× per cycle on `lastError() != 0`. If all three fail, the firmware skips that sample (no NaN ever appears in any payload, JSON has no valid `NaN` literal and templated Notes validate field types) and emits a separate `modbus_unreachable` event Note. The event is rate-limited to once per hour to avoid alarm fatigue when the drive itself is powered off for service.
+- Modbus reads are retried up to 3× per cycle on `lastError() != 0`. If all three fail, the firmware skips that sample (no NaN ever appears in any payload: JSON has no valid `NaN` literal, and templated Notes validate field types) and emits a separate `modbus_unreachable` event Note. The event is rate-limited to once per hour to avoid alarm fatigue when the drive itself is powered off for service.
 - Notecard requests use `notecard.requestAndResponse()` and check both `NULL` return and the `err` field on the response object before trusting the data.
 
 ### Key code snippets

53-municipal-wastewater-lift-station-monitor/README.md

@@ -14,11 +14,11 @@ This project is a [downtime prevention](https://blues.com/downtime-prevention/)
 
 **The problem.** A **lift station** (also called a pump station) is a below-grade concrete vault or roadside cabinet that collects raw sewage from the surrounding gravity sewer system and pumps it uphill toward the treatment plant. Every municipality has dozens of them, often scattered across low-lying neighborhoods, industrial zones, and rural road shoulders — most with no onsite staff and no way to know what's happening inside until a citizen calls to report a smell or, worse, a spill.
 
-When a lift station fails, the **wet well** — the collection basin that feeds the pumps — fills up and overflows. The result is a **SSO**: a sanitary sewer overflow. SSOs draw immediate regulatory attention; they trigger EPA reporting obligations, risk consent decree violations, and require expensive emergency cleanups. A station that fails on a Friday evening and isn't discovered until Monday morning is a public health event, a PR crisis, and a significant unplanned expense all at once. The failure modes are rarely dramatic: a pump fails to start because its float control sticks, a discharge check valve fails and allows backflow that clogged the impeller, or a wet-well float switch trips but no one receives the alarm because the SCADA dial-up modem lost its phone line. Each of these is detectable minutes after it starts — if someone is watching.
+When a lift station fails, the **wet well** — the collection basin that feeds the pumps — fills up and overflows. The result is an **SSO**: a sanitary sewer overflow. SSOs draw immediate regulatory attention; they trigger EPA reporting obligations, risk consent decree violations, and require expensive emergency cleanups. A station that fails on a Friday evening and isn't discovered until Monday morning is a public health event, a PR crisis, and a significant unplanned expense all at once. The failure modes are rarely dramatic: a pump fails to start because its float control sticks, a discharge check valve fails and allows backflow that clogs the impeller, or a wet-well float switch trips but no one receives the alarm because the SCADA dial-up modem lost its phone line. Each of these is detectable minutes after it starts — if someone is watching.
 
 This project is that watcher. It straps to the inside of the station, samples the wet-well level every 60 seconds, measures current draw on each pump, and monitors the high-water float switch — four sensing points across three sensor types. Onboard edge logic on the STM32-based host MCU evaluates three fault rules every 60 seconds and routes alerts to the [Blues Notehub](https://blues.com/notehub/) cloud service the instant any rule trips — over cellular where a tower is in reach, and over the Skylo satellite network where one isn't. The on-call crew gets paged before the wet well overflows, not after.
 
-**Why Notecard for Skylo.** The wireless-first architecture here isn't a convenience — it's a necessity. Lift stations sit in concrete vaults underground, often with no AC power outlet in the vault itself (power runs to the pump control panel, not a wall socket). They're geographically distributed across a municipality in a pattern that matches the sewer network, not the municipal network — there's no fiber running to a roadside pump cabinet, and there's no corporate WiFi AP that can reach through a concrete lid to a sensor inside. Utility supervisors would need to deploy and maintain a WiFi access point at every single station to achieve what one Notecard covers automatically.
+**Why Notecard for Skylo.** The wireless-first architecture here isn't a convenience — it's a necessity. Lift stations sit in concrete vaults underground, often with no AC power outlet in the vault itself (power runs to the pump control panel, not a wall socket). They're geographically distributed across a municipality in a pattern that matches the sewer network, not the municipal network: there's no fiber running to a roadside pump cabinet, and there's no corporate WiFi AP that can reach through a concrete lid to a sensor inside. Utility supervisors would need to deploy and maintain a WiFi access point at every single station to achieve what one Notecard covers automatically.
 
 [Notecard for Skylo](https://shop.blues.com/products/notecard-for-skylo?utm_source=dev-blues&utm_medium=web&utm_campaign=store-link) (NOTE-NBGLWX) carries three radios on one M.2 module — cellular (LTE-M / NB-IoT / GPRS), WiFi, and satellite over the [Skylo](https://www.skylo.tech/) non-terrestrial network (NTN) — and selects among them automatically. The firmware sets a single [`card.transport`](https://dev.blues.io/api-reference/notecard-api/card-requests/#card-transport) preference of `wifi-cell-ntn`: prefer WiFi where an accessible AP happens to be in range (rare for a sealed vault), fall back to cellular (the de-facto primary at the vast majority of municipal infrastructure), and fall back again to Skylo satellite at stations on the fringe of the service territory, beyond the reach of any cellular carrier. Failover is handled inside the Notecard; the host firmware never branches on which network is live. That collapses what used to be a two-SKU decision — a cellular Notecard for in-coverage stations, a satellite device for the rest — into a single part number, a single antenna kit, and a single firmware image that deploys unchanged across the entire fleet. There is nothing to swap when a station turns out to have weaker coverage than the survey suggested: the same board that runs on cellular downtown automatically reaches the Skylo network at the rural lift station.
 

54-multi-site-walk-in-cooler-energy-setpoint-monitor/README.md

@@ -8,7 +8,7 @@ This reference application is intended to provide inspiration and help you get s
 
 </Note>
 
-This project is a cellular [energy savings](https://blues.com/energy-savings/) reference design that gives corporate operations teams a live view into the temperature, compressor runtime, and door behavior of every walk-in cooler in their fleet — without touching a single franchisee or operator's WiFi network.
+This project is a cellular [energy savings](https://blues.com/energy-savings/) reference design that gives corporate operations teams a live view into the temperature, compressor runtime, and door behavior of every walk-in cooler in their fleet, without touching a single franchisee or operator's WiFi network.
 
 ## 1. Project Overview
 
@@ -154,7 +154,7 @@ Pin-by-pin connections:
 - **SDA / SCL** → Notecard I²C (Notecarrier CX routes these to the M.2 slot internally; no external wiring needed).
 - **+VBAT** → Mojo `LOAD` output (bench use only); Mojo `BAT` input ← 5V DC output of the MeanWell IRM-10-5 (or equivalent 5V/2A AC/DC supply), which is wired to the AC service conductors in the NEMA 4X enclosure.
 
-**CT installation Note.** The SCT-013-030 is a split-core clamp; open the clamp, pass it over *one* hot leg of the compressor's circuit (not both, clamping both legs cancels the magnetic fields and you'll read zero). Route the TRRS lead back to the NEMA 4X enclosure through a cable gland. The compressor circuit is line-voltage; installation must be performed by a qualified electrician following applicable codes and lockout/tagout procedures. The CT itself is entirely non-contact and electrically isolated once clamped.
+**CT installation Note.** The SCT-013-030 is a split-core clamp; open the clamp, pass it over *one* hot leg of the compressor's circuit (not both; clamping both legs cancels the magnetic fields and you'll read zero). Route the TRRS lead back to the NEMA 4X enclosure through a cable gland. The compressor circuit is line-voltage; installation must be performed by a qualified electrician following applicable codes and lockout/tagout procedures. The CT itself is entirely non-contact and electrically isolated once clamped.
 
 **DS18B20 placement.** For accurate air temperature monitoring, position the stainless probe tip in open box air at mid-box height or along the return-air path, away from both the evaporator fan discharge and the door. This location gives the most representative reading of box air temperature, which is a useful proxy for stored-product conditions but is not a direct product core-temperature measurement. Avoid the evaporator fan discharge — it is the coldest zone in the box and reads several degrees below true representative air temperature; placing the probe there biases temperature-to-target deviation calculations low and will delay `temp_high` alerts. Avoid the door zone, which sees warm infiltration air on every open cycle. Keep the probe away from direct coil contact for the same reason.
 
@@ -425,4 +425,4 @@ Once the basic fleet visibility is established, the following extensions are the
 
 ## 12. Summary
 
-The corporate energy team that used to stall at the 50-store pilot now has a fleet-wide view: compressor apparent kWh per window, door-open seconds, temperature drift against a setpoint they configure from a browser. One cellular SKU, three sensors, and a field tech who never has to ask anyone for a WiFi password — and the outlier box running double the compressor-hours of its peers shows up on a dashboard instead of in next month's utility bill.
+The corporate energy team that used to stall at the 50-store pilot now has a fleet-wide view: compressor apparent kWh per window, door-open seconds, temperature drift against a setpoint they configure from a browser. One cellular SKU, three sensors, and a field tech who never has to ask anyone for a WiFi password. And the outlier box running double the compressor-hours of its peers shows up on a dashboard instead of in next month's utility bill.