During the testing and hardening of 2026 (mostly October and November), many details of the system were refined. The pages describing the system give only the result; this page records what changed, for those interested in how the present design was reached.
Communications
- The message checksum was a simple XOR of the payload bytes until November 2026, when it was replaced by a CRC-8, which catches errors the XOR misses (such as two flipped bits or swapped characters).
- AR3's acknowledgments and its HB:0 message were sent as plain text until 4 November 2026. AR1 never recognised the plain-text acknowledgments, so every command was in effect sent four times; framing them stopped the repeats.
- The return from the backup link to RS-485 was made to wait for 60 seconds of steady RS-485, so that an intermittent wire could not make AR1 flap between the two links.
- The UNO R4's habit of holding back the last characters of a backup-link message was found in November 2026 testing, and AR1's receiving was adapted to cope with it.
- From 17 to 20 November 2026 the Fairhaven readings froze for three and a half days without any warning. A poor contact on the RS-485 A wire (at the switch that isolates the link) had broken the link, and AR1 did not move to the backup link because the comms path test control had been left forced to primary since 13 November: a Home Assistant restart had reset the control to auto on the dashboard without telling AR1. AR1 went on reporting the last readings and a healthy link. It was found when AR1 was restarted on site. Since then a forced path ends by itself after 30 minutes, the dashboard shows the setting AR1 is actually using, the comms alert is raised if a forced path goes silent, and AR1 reports the age of its data, with a No data from shed alert. AR2's self-restart now counts only messages from AR3, not mere WiFi connection, and AR2 and AR3 slow to hourly attempts instead of giving up after four.
- Message checking was first introduced to count bad messages only; rejection of bad messages (FRAME_REJECT) was tested on TestSys from 5 to 7 November 2026.
Control and modes
- Until 1 November 2026, when control returned from Autonomous mode to the dashboard, AR1 adopted AR3's last relay states. The system is still locked at that moment, so the pump stopped anyway while the dashboard went on showing it running; manual control now starts from a safe state with the pump off.
- Until 25 October 2026 the LOCK "kick" (repeatedly requesting pump off while locked) also applied in Autonomous mode, where it fought the autonomous logic, pulsing the pump off and on whenever the lock had been left on.
- Parts of the AR1 and AR3 code had the recirculation valve the wrong way round until 1 November 2026, so that recirculation cycles never reached the sprinklers.
- Cool-down detection compared readings five minutes apart until 1 November 2026, which could miss the fall when the outside and inside drops landed in different intervals; it now compares each with its peak.
- A restart during Autonomous mode used to lose five to ten minutes of autonomous control; since November 2026 each Arduino remembers its mode.
- Watchdogs were made effective on all three boards in October-November 2026 (the earlier AR2 and AR3 sketches requested a watchdog interval longer than the chip allows, so it never started). A physical remote reboot switch had been planned, but with working watchdogs it was not needed. AR3's start-up LCD check was added after writing to a missing display stalled AR3 into a restart loop.
Sensors and wiring
- A failed temperature sensor used to be reported with a believable default (30-32 °C); it is now sent as 0 with a failure flag.
- The tank gauge's first conversion (3.3 V = 3.2 m) read about 1.2 m too high near empty, so the dry, gel and water-saving thresholds could never have triggered; it was replaced by the water-meter calibration of 25 October 2026. In December 2025 the reading had been seen to climb above 4 m as the shed warmed from 20 to 35 °C. Glitch filtering was added on 3 November 2026, after ten days of Fairhaven readings (24 October - 3 November) showed brief 10-20 cm jumps.
- The DN50 flow meter's phantom readings were traced to noise on the Arduino's 5 V supply, and the meter was calibrated against the house water meter on 25 October 2026.
- Until 22 November 2026 the flow reading counted the pulses in a single tenth of a second, so at reduced flow it jumped between about 74 and 180 L/min; it is now averaged over 2 seconds. At the same time the fixed 147 L/min assumed when the flow meter fails was replaced by the last measured flow, because the reduced-flow valve can be adjusted.
- The oil and water pressure optocouplers were originally wired straight onto the shared switch wires. With the engine running, the SFS01 box's oil input was dragged down from 4.0 V to 2.1 V, so the box read "no oil pressure" and re-engaged the starter about every 20 seconds into the running engine (heard as a hiss). The diodes and pull-up resistors were fitted on 23-24 October 2026. At the same time two wiring faults on the opto board were found and corrected: the shared input common was floating, and OPTO 1's output ground had never been connected, which is probably why water pressure had never registered in Home Assistant.
- AR2's battery reading had been at full scale since May 2026. A correction factor of 0.558 had scaled a reading that was already at its maximum to look like 13.3 V, so it always showed 13.24 V and the battery rules always saw a healthy battery. On 21-22 November 2026 the battery reading moved to AR1 in the house, and AR3 now uses recirculation when it has no battery reading from AR1.
- While that divider was being fitted on 21 November 2026, a wiring slip put 12 V onto a ground wire. It shorted AR1's D0 and D1 pins (the RS-485 connection) and damaged the house Gravity module, which then corrupted about one message in twenty from AR3; both were replaced on 22 November. A loose solder joint on the Gravity module's power feed and a broken wire in its plug were repaired at the same time.
Dashboard
- On 1-2 November 2026: the Disable Shed Cooling switch stopped briefly energising the cooling relay at start-up and when switched; AR1 stopped running its own shed cooling cycle in Autonomous mode (which had made the Shed Cooling LED flash on every 2 minutes); the Time ON / Time OFF sliders became - and + buttons whose settings survive a Home Assistant restart; the sprinkler cycle's ON and OFF were corrected (they had been reversed); turning gel on now also selects reduced flow; and the System Current State dashboard was removed, its contents moved to Sprinkler Main Control and the Autonomous Mode State dashboard.
Remote recovery
- The Remote Recovery run sheet on the previous wiki was cut short during a bulk edit in May 2026; the full text was restored from its February 2026 version.
<< Development history: Autonomous mode | | Main.Contact >> |Table of Contents>
