← Workshop home

Every file and link needed to prepare for and run the workshop, in one place. Participants get a trimmed version at รวมไฟล์และลิงก์.

Links are absolute so this page works the same whether you read it in the repository or as the published page at arnans.github.io/iot-training/instructor-resources.html. Rows marked (repo) need access to the private iot-training repository.

1. Board firmware — twelve images

One image per board type, per zone colour, plus a second sensor image for Session 4. The zone colour is baked in: a red-zone image only joins the red SSID. Sensor and relay images built and verified 12 Sep 2026; the ESP-NOW sensor image and the receiver image are new on 20 Sep 2026 and need a build and a bench pass (TEST-PLAN test M) before publishing. The base sensor image is unchanged in behaviour (it only gained internal ids) and need not be re-flashed.

Receiver devkits: buy one model. The log banner only appears on the USB port the image's logger writes to; the receiver image is built for native-USB boards (Super Mini, XIAO C3) by default and needs --receiver-console uart0 for boards with a CP2102/CH340 chip (DevKitM-1). Sort the receivers by zone and colour-code them like the other boards; they are handed out at the start of Session 4, one per team, and returned with the kit.

To flash a board: plug it into a laptop over USB-C, open https://web.esphome.io in Chrome or Edge → Connect → pick the serial port → Install → choose the .bin for that board and zone. Needs internet.

The colour on the board must match the colour on the kit and on the team card. A mismatched board joins the wrong WiFi and the team's Home Assistant never sees it. The board says which it is — its name reads gogo-iot-red-a1b2c3.

Spare boards flashed during the workshop must be flashed to that table's zone colour.

2. Home Assistant virtual machine

The same image participants install. Put it on the instructor kit and on every rescue laptop.

The two files are not interchangeable and cannot be imported on the wrong host: the ARM64 OVF declares <Platform architecture="ARM">, so VirtualBox on an Intel host rejects the file instead of importing something that will not boot. Apple Silicon needs VirtualBox 7.2+; 7.1 and older cannot run the ARM64 guest. Build notes and the exact settings of both images: hass/README.md (repo).

Import steps, settings and the login are in the participant guide: เตรียมแล็ปท็อป. The two things that go wrong: Adapter 1 must be Bridged and set to the laptop's own WiFi card, and the VM is reached at http://<its IP> — never homeassistant.local, because every team's VM answers to that name. On ARM64 the console's IP line reads enp0s8, not enp0s3.

Login: TODO / TODO. Not on the team card and not in the install guide — project it at Activity 3.1, when teams first log in. Until then participants only check that the Login page loads (install guide step 3, and the pre-start check in Session 1).

3. Worksheets and printed material

The renders/ pages are self-contained: open one once and it keeps working with no internet.

4. Preparing to teach

Instructor kit — a spare laptop mirroring a team kit, projected. It also rescues a team whose laptop dies. Three things teams don't have: the Auto / Off / Manual pump control box for the Session 2 debrief, a Home Assistant Cloud account for the Session 3 remote-access demo, and a saved morning history graph for teams whose own graph holds too little Session 2 data.

Kit prep, the two that fail silently: flip the float cylinder on its rod so it closes when wet (otherwise a cut wire reads "water present" and a pump can run dry), and colour-code board, kit and team card as one matching set.

5. Tools and external links

6. Running the activities

Each activity in the worksheet opens with a checklist for the group, ใบงาน: one item per sub-section, holding the question and its answer boxes, with a link down to the steps. This section follows the same items, so a TA looking at a team's screen can find the matching entry here. Each item gives Done when (what a finished answer looks like) and If stuck (what to check first). Minutes are the design estimate, not a rehearsed timing.

How to run it. Project the checklist and talk through it one item at a time; teams work down it while you talk. The prose under each item is for the team that falls behind or wants the reason. Keep the progress board open on a TA phone: every tick a team makes lands there within seconds, so it shows who is behind before they raise a hand. A team ticks either the item in the checklist or the ทำข้อนี้เสร็จแล้ว box at the foot of the sub-section; both are the same tick. The three "we understood this" boxes in 8.3 are not counted.

The two questions from the Activity 4 pilot, s2.time.trace and s2.time.why, and the Session 1 lighting keys (s1.light.on, s1.rule.dark.*, s1.rule.bright.*, s1.auto.pass, s1.timer.at, s1.timer.ok) are on the record sheet like every other answer. Session 1 no longer records the supply voltage, the wired-right-first-time self-report, the control-box position or the manual-round counts (Roger, 15 Sep 2026).

Session 1 · Activity 1 — Sense (35 min)

Teams leave with the sensor board's name and IP, a first reading of every sensor from the board's own web page, and a sense that the float is a yes/no sensor. No Home Assistant yet.

1.1 Plug in the float switch

Done when the float's Grove lead is in port 1, the board is on the laptop's USB, and the float is floating in the glass.

If stuck A float in any other port never changes state; only port 1 is configured.

1.2 Read the board's name and IP

Done when a name like gogo-iot-red-a1b2c3 is written down whose colour matches the team card, with an IP under it. The IP line appears 10–20 seconds after the name.

If stuck The log is blank or the name has scrolled past: press Reset device in the web page. Unplugging the cable or pressing the board's button drops the browser's connection. Firefox cannot connect, so use Chrome or Edge. A name but no IP within a minute: the board cannot join WiFi. Check the colour, and swap the board if it is wrong.

1.3 Open the board's web page

Done when http://<IP> shows the list of entities, updating without a refresh.

If stuck The laptop is not on farm-laptop-<colour> for the board's colour. The page lives on the board, not on the internet.

1.4 What each sensor measures

Done when each row has a room value and a test value that visibly differs from it: humidity rises when breathed on, light drops when covered, vibration rises from a few mg at rest when the table is knocked. The float reads ON floating and OFF lifted. The light and humidity values change within about one to two seconds, because the board reads them once a second. Session 2 uses the room and covered light numbers to set its thresholds.

If stuck Humidity barely moves: breathe slowly for the full five seconds, close to the board. The float never changes: the lead is not in port 1.

End-of-activity question

A good answer says humidity decides watering (standing in for soil moisture on a real farm), and the float tells whether the reservoir still has enough water for the pump. It only says yes or no, not how much.

Session 1 · Activity 2 — Act (40 min)

Teams leave with the 12 V lamp circuit wired by their own hands, the relay board powered from the charge controller's USB, and the LED module lighting on command. No water in Session 1 beyond the float's glass; the pump is first wired in Session 2 (4.6).

2.1 Know the relay board

Done when every member can point at COM and NO (used) and at 7-60 Input / GND (never used) on the real board.

2.1 Read name and IP, click the relay

Done when a gogo-relay-<colour>-… name and IP are written down. The relay clicks from the web page switch and from the board's own button. The USB lead is then unplugged from the laptop, but left in the board.

If stuck Same as 1.2. The LED blinking means no WiFi; a steady LED means connected.

2.2 Charge controller parts and the relay circuit

Done when the team can find the middle pair (battery symbol, where the adapter goes), the right pair (lamp symbol, the 12 V output) and the two USB ports, and can say from the diagram why the +12 V lead goes through COM and NO while the negative lead goes straight to the lamp.

2.2 Wire it

Done when all six steps are done and a second person has checked every screw. No bare ends touch, no wire is in NC or 7-60 Input / GND, and the adapter is still unplugged.

If stuck Check for a wire on NC instead of NO (the LED is then lit whenever the relay is Off), and for the load pair and the battery pair swapped. The LED's red lead belongs on NO and the black lead on the negative row; a reversed LED stays dark but is not damaged.

2.2 Power on and light the lamp

Done when the charge controller screen shows about 12 V. The relay board lights and rejoins WiFi. The LED lights on On and darkens on Off, and its lenses point away from the GoGo-IoT board, at least 30 cm off. A lamp shining onto the light sensor makes the Activity 3 rules fight themselves.

If stuck Screen dark: unplug first, then check the adapter is on the middle pair, red to plus. Screen on but the relay board is dark: reseat its USB lead or try the other port. Relay clicks but the LED stays dark: swap the LED's two leads. The page will not open after a minute: move the relay board's USB lead to the laptop briefly to read a new IP (the relay restarts Off), then move it back.

The Auto / Off / Manual control-box talk has no worksheet item any more (removed 15 Sep 2026). If you give it, it is two minutes before the break with nothing to fill in; Session 2's debrief demo carries the idea.

End-of-activity question

A good answer says a 220 V pump needs a relay or contactor rated for the pump's AC current, including its start-up surge. It also needs 220 V-rated wiring, an enclosure and protection (breaker, RCD), and an electrician to wire it, not the team. Before the break every relay is Off and every lamp is dark.

Session 1 · Activity 3 — Logic (40 min)

Teams leave with both boards in their own Home Assistant, a light threshold they chose, and two automations, มืด-เปิดไฟ and สว่าง-ปิดไฟ on one shared threshold, passing 3 of 3. Fast teams also leave with a tested (and then disabled) ปิดไฟตามเวลา. Cut on 15 Sep 2026 to reduce reading during the live session: no history graph (Session 3 opens on it) and no "be the Logic yourself" round; 3.3 goes straight from threshold to rule sentences to the two automations.

3.1 Add both boards

Done when the representative's laptop is logged in with the credentials you project now, and both boards are added by IP (+ Add integration → ESPHome → Host = IP, port 6053 → Submit; if a list appears, choose Setup another instance of ESPHome), with the Device name in the Name and assign dialog matching the names written down (HA shows it title-cased). Renaming the devices is optional.

If stuck The Login page no longer loads: the VM's IP changed, so read it again from the VM window. Boards have mDNS disabled (it flooded the room's WiFi), so there are no discovered cards and .local names do not resolve — the IP is the only way. A mistyped IP that lands on a neighbour's board looks like success, but the team then watches the neighbour's sensors and its automation drives the neighbour's lamp. The entry already exists when Name and assign shows the Device name, so a wrong name means: close the dialog, ⋮ → Delete on that board's row, add again with the right IP. Cannot connect: the board's IP moved, so read it again at web.esphome.io (Reset device, IP ADDRESS line). For a board already added, use ⋮ → Reconfigure → new IP in Host; the rename and entities are kept. HA refuses another board's address by MAC.

3.2 Test the boards in Home Assistant

Done when the GoGo-IoT device page shows the same entities as the board's web page and the light value drops under a cupped hand, and the lamp toggles from the laptop and from a phone, with both screens agreeing.

If stuck The phone must be on the zone WiFi, not mobile data, and must use the IP from the VM screen, never homeassistant.local. If it still fails, skip this step and flag it to a TA; the rest of the activity does not need the phone.

3.3 Threshold, rule sentences, two automations

Done when the threshold sits roughly halfway between the open and covered readings taken now on the device page (80 and 7 → about 40), not copied from Activity 1 because the room may have changed. The two sentences read: dark, "if the light sensor is below the threshold, turn the lamp on"; bright, "if it is above, turn the lamp off". Then covering the board lights the lamp and uncovering darkens it, three rounds out of three, with nobody touching the switch. Both rules use the same threshold on purpose. The scenario is the flower-farm night light: the LED is the lamp, a cupped hand is dusk.

If stuck The value barely moves: the hand is not over the light sensor (right-hand group on the board). It never rises again: the lit LED is shining under the hand; turn it away. Covering does nothing: the reading was already below the threshold when the rule was saved. A crossing trigger needs the value to go above, then below again, so uncover, set the relay Off and cover again. The board sends a light value about once a second, so hold the cover for two seconds. Check that the trigger entity is ความสว่าง (light) and the action is the team's own relay. The lamp flickers while the hand moves slowly: that is expected with one shared threshold, tell them to cover fully and quickly; it is the opening lesson of Session 2, not a fault.

3.4 Optional: lamp off at a chosen time

Done when the lamp, switched On from the device page, goes off by itself at the minute the team chose, and the team has then switched the rule off. Left on, it fires every day at that time, including during Session 3 with the pump on the relay.

If stuck Nothing happens: the chosen time is already past, so pick a new future minute. Home Assistant runs on Asia/Bangkok time; read the laptop clock. The lamp went off early: สว่าง-ปิดไฟ fired because the sensor was covered then uncovered; start with the sensor open.

End-of-activity question

A good answer says the lamp switches on and off repeatedly, because one value is both the on and the off line, so every wobble across it fires a rule. Leave it there: Session 2 opens by measuring it and fixing it with a second threshold.

Session 2 · Activity 4 — automations and hysteresis (70 min)

Teams leave with four automations (five if they did the Session 1 timer): two watering rules on, the two garden-lamp rules and the timer off, and the pump wired for the first time. They also leave with their thresholds, two flicker counts and the written answer. Keys go into the team's JSON as s2.light.off, s2.chatter.*, s2.time.*, s2.pump_back_t, s2.hum.*, s2.mode.*, s2.q4; the on-threshold is Session 1's s1.light.on.

At the start the lamp rules from Session 1 are live, so anyone shading the board lights the lamp. That is a free demo, not a fault. There is no recap activity any more (cut 16 Sep 2026): the session goes from the four checks straight into 5.1.

Before start (ก่อนเริ่ม) · ~3 min

Done when the four worksheet checks are ticked (they report to the dashboard as one activity, 15 Sep 2026): Home Assistant is up, the sensor board shows values, the relay board switches the LED, and the team has agreed to leave both boards and Home Assistant on for the half-day recording. Also confirm what the worksheet no longer asks: มืด-เปิดไฟ and สว่าง-ปิดไฟ are on with the same threshold, ปิดไฟตามเวลา (if it exists) is off, and covering the board lights the lamp. Today's VM IP is read off the screen; the record sheet still has a box for it.

If stuck The lamp will not light: same as Session 1's 3.4 (no crossing yet, or the room got darker than the threshold overnight, so raise the threshold or add light).

4.1 Lamp flickers near the threshold · ~5 min

Done when the team has a one-minute count with the off rule at the same value as the on rule. It should be clearly higher than the count they get in 5.2. There is no fixed expected number: TEST-PLAN test E still owes a bench count.

If stuck A count of 0–2 means the hand is not crossing the threshold. Have one person watch the value on the device page while another finds the height where it sits at the threshold, then waver there.

4.2 Separate the on and off thresholds · ~8 min

Done when the off threshold sits above the on-threshold and below the room reading, about a quarter of the range above (82 / 6 / on 40 → off about 60). Self-check passes on all three: ≤ 2 switches in a minute, cover → on, uncover → off.

If stuck The lamp never goes off: the off threshold is above the room reading. It still flickers: widen the gap, and turn the LED so it does not shine onto the board's light sensor.

4.3 Time window · 4.4 Lamp on when the time arrives · ~12 min tests, ~5 min fix

Done when

• Traces answer: the latest run stopped at the condition (the And if / Time box). The trigger fired but was blocked.
• Predictions (added 16 Sep 2026, window 18:30–07:30): dark after 18:30 → the lamp comes on (the crossing fires inside the window). Dark before 18:30 → it stays off: the crossing fired outside the window and was blocked, and nothing fires again at 18:30. The second answer is the reason 4.4 exists.
• After the fix: with the sensor covered before the start time, the lamp comes on at the start time (s2.time.fixed = ติด).
• Why re-check light in And if: the time trigger fires whether or not it is dark. Without the light condition the lamp would switch on at the start time in daylight.
• Tick: with the sensor kept covered, the lamp turns on at After and off at Before.

If stuck Traces shows no new run at all: nothing crossed. Uncover, set the relay Off, then cover again. The timed test does nothing: the chosen times are already past. Both After and the trigger's At must move to a future minute, and the same goes for Before and At on the off rule. Home Assistant runs on Asia/Bangkok time, so read the laptop clock.

4.5 Notification · ~4 min

Done when a message appears under the bell icon, bottom-left, when มืด-เปิดไฟ runs.

If stuck Covering does nothing now: the time window from 4.3 excludes the current time, so widen it. Send a notification via mobile_app is missing from the list: it only appears after a phone has logged in through the Companion app. Persistent notification is always available.

4.6 Wire the pump, run it dry · ~8 min

Done when both lamp rules and the timer are off and the relay is Off before any screw is touched; the LED's two leads are off the terminal block and the pump's leads sit on the same two points, + on the block terminal fed from NO, − on the negative terminal; the float sits in its glass reading Wet; the pump has run dry on a 10-second command, heard and felt, and stopped on Off; and the changeover time is written down. No hose, containers or tray: the pump runs dry all morning (16 Sep 2026), so keep every run short. The 30 s cutoff that used to backstop this is built last in Activity 5 since the 16 Sep reorder, so until then the only limit is the person watching the clock. Session 3 uses it to tell lamp commands from pump commands on the relay's history graph.

If stuck Relay clicks, pump silent: the + lead went to a terminal fed from NC, or the − lead is not on the negative terminal, or the controller's load pair is off (its load mode; tell the instructor). Pump runs with the relay Off: lead on NC. The lamp turned the relay on during the swap: a lamp rule was left enabled. Nothing watches the float yet, and the worksheet says so; the float rule is built in 6.1.

4.7 Watering rules · ~12 min

Done when the pump-on threshold sits a little above room humidity; off threshold 5–10 % above that. The pump does not start when the rule is saved. That is expected, because nothing crossed. Blowing on the board pushes humidity past the off threshold; when blowing stops and it falls back below the on threshold, the pump starts. How long that takes depends on the room; the worksheet stopped timing it on 16 Sep 2026 (Roger), so there is no number to record.

If stuck The pump never restarts after blowing: the on-threshold is below the room reading, so the value never crosses it on the way down.

4.8 Auto / Manual mode toggle · ~12 min

Done when a Toggle helper Pump Auto Mode exists, shows on a dashboard the team added (Overview (Legacy), any name, on the sidebar), and รดน้ำ - เปิดปั๊ม carries the condition Switch is on for it. With the toggle off, the blow-and-wait test from 4.7 no longer starts the pump while the relay's device page still runs it; with the toggle on, the pump starts again (s2.mode.off = ไม่เริ่ม, s2.mode.on = เริ่ม). Rebuilt by Roger on 16 Sep 2026 from "disable a rule, run the pump by hand": the toggle gates automatic starting only. Manual still allows the humidity stop rule and every protection rule to operate; switching the toggle off does not stop a running pump or isolate power. Switching it on while humidity is already below the threshold does not start the pump until a new crossing occurs. The Automations list ends as: รดน้ำ - เปิดปั๊ม, รดน้ำ - ปิดปั๊ม on; มืด-เปิดไฟ, สว่าง-ปิดไฟ and ปิดไฟตามเวลา off; Pump Auto Mode on.

If stuck The toggle is not on the new dashboard: check the team picked Overview (Legacy), which lists every entity, and scroll. The default Overview of current Home Assistant releases does not show helpers, which is the whole reason 4.8 adds the legacy dashboard. The condition search finds nothing: Create was not pressed, so the helper does not exist. Only รดน้ำ - เปิดปั๊ม gets the condition; the fail-safes built in Activity 5 must work in every mode, so they never check it. If a team also gated รดน้ำ - ปิดปั๊ม, remove that condition so its behavior matches the worksheet's definition of Manual.

Check this list at every table before the break. Activity 5's sabotage rounds assume exactly this state. A lamp rule left on switches the relay whenever the room light changes, and a timer rule left on switches it off at the same minute every day. Those are the likeliest causes of a pump "running by itself" or stopping unexplained in the failure lab.

End-of-activity question · ~3 min

A good answer gives the team's own on and off light thresholds. It says that one shared value makes the output flicker whenever the reading hovers at it, and that watering applies the same fix: the off threshold is set 5–10 % above the on threshold.

Session 2 · Activity 5 — Failure lab (50 min)

Every problem runs the same three steps: the paired team sabotages, our team fixes, the paired team sabotages again. While being sabotaged, the team watches only its Home Assistant screen. Required: all four problems seen, fixed and re-tested — nothing is optional since 16 Sep 2026, and the instructor decides live whether a slow table drops one. Four problems in 50 min is tighter than the old three-required shape, so watch the clock at problem 2 and keep 5.5 to a read-and-answer. Pairing is explained live, not in the worksheet.

Before every problem

Done when both watering rules are on, Pump Auto Mode is on, both lamp rules and the timer are off, the float sits in its glass reading Wet, both boards show values, and the two device pages are open in separate tabs. No float rule exists yet; problem 1 builds it.

If stuck A pump started by hand keeps running because the room never reaches the off threshold. That is expected; do not let teams change the humidity thresholds to make the pump run.

5.1 Problem 1 — water gone, pump still starts (Fail-safe 1)

Done when the sabotage (float lifted until Dry, then the pump commanded On from the board's button or the device page) runs the pump with nothing stopping it, because no rule looks at the float at all. After the fix, Fail-safe - น้ำหมด (two triggers, one condition, two actions) stops the pump within about a second, with a notification, whether started from the button or the device page, and lifting the float mid-run stops it too.

If stuck The first trigger is float → Dry (state, not numeric). The second trigger must be pump switch → On with the condition float → Dry. The watering rule's condition is float → Wet, not "not Dry": a cut wire reads Dry and an offline board reads unavailable, and neither is Wet.

5.2 Problem 2 — commanded but not running (Fail-safe 4)

Done when with the pump lead unscrewed from NO, the relay clicks and the screen says On but the pump is silent and still. After the fix the switch turns itself off within 15 s. With the lead back, the pump runs past 15 s. Moved here from fourth place on 16 Sep 2026 (Roger: unscrewing a lead is a real sabotage act, and the wait-and-watch problem should not come second).

If stuck The rule cuts a pump that is running: the sensor board is not touching the pump, or ความไวตรวจจับปั๊ม is too low, so set it to high. The 15 s is deliberately shorter than the 30 s run-time cutoff built in 5.4, so the two rules can be told apart — but that cutoff does not exist yet at this point, which is fine: nothing else competes to stop the pump here.

5.3 Problem 3 — sensor board off (Fail-safe 3)

Done when the team has chosen open, close or hold, with a reason; off is the defensible choice for a pump. When the board is switched off, its entities go unavailable while the pump runs on. After the fix, the Traces page of Fail-safe - เซนเซอร์หาย shows a run when humidity went unavailable, with the Off and the notification. Judge by Traces, not by which rule stopped the pump (15 Sep 2026). Since the reorder of 16 Sep 2026 the 30 s cutoff does not exist yet, so this round is cleaner than it was: with the board off, ปั๊มกำลังทำงาน reads unavailable rather than Off, so 5.2's rule does not fire either, and nothing but the new rule stops the pump.

If stuck Pulling USB does nothing because the 18650 cell keeps the board alive, so use the board's ON/OFF switch. If the pump stopped at about 15 s, that was Fail-safe - สั่งแล้วไม่เดิน reading the pump as not running; the sensor-lost rule still counts as working when its Trace shows a run.

5.4 Problem 4 — pump never stops (Fail-safe 2)

Done when a pump started by hand and left alone runs on, and after the fix it stops by itself at about 30 s with a notification. It runs last since 16 Sep 2026, because there is nothing for the paired team to sabotage here — they start the pump and wait — and that makes a weak second round.

If stuck The trigger needs For 30 seconds on pump → On. After the cutoff the pump does not come back by itself; that is by design, and it is why the notification matters. The sabotage is cut at about 15 s instead: Fail-safe - สั่งแล้วไม่เดิน from 5.2 decided the pump is not running, so fix the board's placement or raise ความไวตรวจจับปั๊ม before trying again.

5.5 Maintenance plan · ~5 min

Done when all five rows carry a named person, a frequency and a thing to check. Moved here from the debrief on 16 Sep 2026 so Activity 5 ends on something the group writes, not a table it reads. A good row is concrete: the float gets lifted and wiped weekly by whoever feeds the fish; the pump strainer is cleared monthly; someone notices within a day when the network or Home Assistant is down, because nothing else will tell them. Weak rows say "everyone" and "sometimes".

If stuck Start from the four problems just staged: each one names a part that failed and a person who had to intervene.

The limits table left the worksheet on 16 Sep 2026 with the laptop-off-WiFi round; both were reading, not doing. The end-of-activity question still asks the question the table answered, and the model answer below carries the table for you. The WiFi-off demonstration still works for a table that finishes early: No team now sees the pump run on with every fail-safe silent; they read it off the last row of this table instead. If a table finishes early and asks, the demonstration still works: pump On, laptop WiFi off, watch for up to 40 s with someone at the pump, then stop it with the relay board's button before WiFi comes back — the Off that Fail-safe 2 tried to send while the link was down failed and is not retried, so waiting for a rule after reconnecting leaves the pump running.

End-of-activity question

A good answer says no: every fail-safe lives in Home Assistant, so they all stop with it. Real safety needs something that works without it: a float wired to cut pump power directly, a A good answer says no: every fail-safe lives in Home Assistant, so they all stop with it. Real safety needs something that works without it: a float wired to cut pump power directly, a limit built into the board's own firmware, and a manual Auto / Off / Manual switch. Since 16 Sep 2026 the worksheet no longer prints the limits table, so this is the facilitator's copy of it:

Wrap-up · ~10 min

Done when every table has walked the lunch checklist: the VM stays running (Session 3's history block reads this morning), รดน้ำ - เปิดปั๊ม is off and the relay is Off so a dry pump cannot run unattended, both boards keep power, answers downloaded. The five learning points are read, not taught. The worksheet no longer scripts a debrief discussion or the control-box demo (Roger, 16 Sep 2026: instructor-run, may not happen as written).

The control-box demo, when you run it. A real Auto / Off / Manual pump box wired to the instructor kit: flip it while the watering automation runs and the pump obeys the switch, not Home Assistant. Then one layer down: with Home Assistant shut down, the relay board's own button still toggles the relay (bench-verified), an override that lives in the board rather than the wiring. It is the answer to the end-of-activity question and links back to the Session 1 seed; it needs no worksheet item.

Session 3 · Activity 6 — Reading the recorded data (20 min)

Before start (ก่อนเริ่ม) · ~3 min

Done when the four worksheet checks at the top of Activity 6 are ticked (added 20 Sep 2026; they report to the dashboard as one activity, like Session 2's): Home Assistant is reachable, both boards are online and the relay still switches, รดน้ำ - เปิดปั๊ม is back on after the lunch shutdown with every fail-safe on, and Pump Auto Mode is left off for the afternoon. Session 2's wrap-up disabled the watering rule so a dry pump could not run unattended over lunch; nothing turned it back on before this was added, so Activity 7's โหมดรดน้ำอัตโนมัติ tile had no live rule behind it and Activity 8's pack-up asked teams to switch off rules that were never switched on.

Why Pump Auto Mode stays off. With the rule enabled and the toggle on, the pump-on threshold sits just above room humidity (4.7), so every table would run the dry pump in 30-second bursts (the cutoff from 5.4) for the rest of the afternoon. Off, the rule is complete and inspectable, the dashboard tile reads Manual, and a team can still drive the pump by hand from its own tile. If you want the pump live — for a demo, or to let a table watch Auto work — turn the toggle on at that table and back off afterwards.

If stuck Home Assistant will not open: the VM's IP may have changed over lunch, so read it off the VM screen again. The rule list is Settings → Automations & scenes.

Open the history

Done when History shows humidity, light and the relay switch from the start of Session 2 until now.

If stuck Check the selected time range first. A VM shutdown or restart keeps the history (the recorder stores it in its database) but leaves a gap while it was off; only a rebuilt VM has none. A team with too little data reads the instructor's projected sample or a neighbour's screen, says so in its answer, and carries on.

Three points on the graph

Done when each point has what was seen, a time, what the team was doing, and what a farm would check. Any three the team can explain are fine; graphs differ between teams.

If stuck A relay bar that is On is a command, not a running pump. Before the recorded changeover time (s2.pump_back_t), from Session 1 until 4.6, it was driving the lamp. Only vibration shows the pump actually ran.

Values the graph always gives

Done when the lowest and highest humidity since Session 2 are written down, each with its time.

The question the graph cannot answer

A good answer says there is no data from the break after Session 1 because the laptop running Home Assistant (the recorder) was closed or taken away; the board kept measuring, but nobody recorded it. A farm needs a machine that stays on, with power through the night. This feeds the power step in Activity 8.

Session 3 · Activity 7 — Dashboard (35 min)

Plan 5 · build 25 · test with someone else 5.

7.1 Who will use it · 7.2 What to show and control

Done when the group has named who will use the dashboard (one or several: owner on a phone, worker on a tablet, manager on a big screen). There are three values to show and two things to control, written in plain words the user understands ("air humidity", "is the water up to the float?", "start the pump").

7.3 Build

Done when the four minimum cards are on a new blank dashboard (the Activity 4 Farm Dashboard is auto-generated and not editable): humidity Gauge, float Tile named for the user (the worksheet suggests ระดับน้ำที่ลูกลอย), pump Tile (สั่งปั๊ม), and a Tile for Pump Auto Mode named so that On reads as Auto (the worksheet suggests โหมดรดน้ำอัตโนมัติ). History graph, severity colours and extra cards are optional and come only after that.

7.4 Someone else tries it

Done when someone who did not build the dashboard (a member of the paired team, or a teammate who took no part) answers all four questions within 30 seconds on a phone: what is the humidity, is the water up to the float, where is the pump control, is it in Auto or Manual now.

If stuck The phone must be on the zone WiFi and use the VM's IP. The new dashboard is in the left menu under the name the team gave it, not under Overview.

End-of-activity question

A good answer names the question the tester struggled with and the specific change made: moving a card up, renaming a tile, adding a colour.

Session 3 · Activity 8 — Design challenge (85 min)

Pick 5 · network 20 · power 15 · design 30 · presentation and exchange 15 (instructor-led). Since 18 Sep 2026 there is no separate design canvas: the design is the four blocks of worksheet step 8.6 (sensor and actuator, connectivity, power, test plan), filled in the worksheet and mirrored in the record sheet.

8.1 Pick the scenario

Done when a scenario is written down (A–E; the worksheet calls them โจทย์, not cards, since 18 Sep 2026) and the farm and its problem are written in 8.1. Recommend E (a member's own farm) when one exists. C and D are advanced; teams choosing them may put the specialist parts under "not sure yet".

8.2 Choose the farm network

Done when both links have a distance, whether the far end has power, what was eliminated and why, and what remains. The instructor models scenario B on screen (400 m, no mains: WiFi and Zigbee out on distance, cellular out on power; ESP-NOW, LoRa and RS-485 remain). The two rows in 8.2 are filled from the survivors, with a reason.

If stuck The weak answer is "WiFi, because the workshop used WiFi". Ask for the distance and the power at the far end first.

The worksheet's table carries range figures again since 18 Sep 2026 (Roger: show range, benefits and limitations), with power draw, benefits, limits and cost per option: WiFi about 30–50 m from the router in the open, ESP-NOW 100–200 m board to board, Zigbee 10–30 m per hop, LoRa 1–10 km, RS-485 up to about 1 km of cable, cellular wherever there is signal. General estimates; กำพล to confirm against what they teach.

8.3 Remote access and the cloud · instructor talk, listen only

Rewritten 18 Sep 2026 (Roger): no task in the worksheet; the team listens and ticks three boxes to say it has understood. Make the three points in this order, in the words on the worksheet:

1. As built, the system is reachable only on the local network. A phone or computer that wants the dashboard must be on the same WiFi as Home Assistant.
2. Remote access needs an additional access method. Nabu Casa's paid Home Assistant Cloud is the example taught here; a VPN into the farm network is another route. Consumer ecosystems such as Tuya and Xiaomi have vendor cloud services with features and charges that depend on the service. Home Assistant itself stays on a machine in the farm because it must reach the boards and any Zigbee or LoRa receiver locally; never present a cloud-hosted Home Assistant as an option. Other approaches (pushing a copy of the data to a cloud platform, port forwarding or a tunnel) exist; mention them only if asked, with the warning that a badly set up door exposes the farm. The worksheet links nabucasa.com. If asked for smart-farm or IoT platforms that come with a cloud: in Thailand, NECTEC's NETPIE (netpie.io, a general IoT platform with a free tier) and HandySense (handysense.io, NECTEC's open smart-farm kit and cloud); maker clouds such as Blynk, ThingSpeak and Arduino Cloud; consumer device clouds such as Tuya / Smart Life, Xiaomi Mi Home and Sonoff eWeLink; irrigation-vendor clouds such as Netafim NetBeat. All of them decide or display in the vendor's cloud, so point 3 applies to every one of them.
3. Our Home Assistant keeps running its local rules when the internet is down; cloud-only products do not. A team that adopts a cloud-only product must give its critical parts a fail-safe of their own.

Demo, optional, instructor kit Open Settings → Home Assistant Cloud, show the Remote access switch and the address. Phone off WiFi, on mobile data, opens the same dashboard the room saw on the zone WiFi. Blow on the board, watch the value change on the phone, send one pump command. Point out that the farm's Home Assistant still decided and recorded; the cloud only carried the connection.

Done when all three boxes are ticked and the team can restate point 3 in its own words: with the internet down, automations, the dashboard on the farm WiFi and logging carry on, while remote viewing and notifications stop; with Home Assistant down, automations, logging and the dashboard all stop, and only protection in the board or the wiring, and the manual switch, remain. The team writes its remote-access choice in 8.6, third communication row; "not needed" is a valid answer.

8.4 Choose a battery for solar

Rewritten 18 Sep 2026 (Roger: focus on the battery types used with solar and how to choose). The box tour is one step; the controller-settings and LCD-icon steps were cut later the same day as a distraction, replaced by two sentences and two diagrams: what the controller does and that it must support the battery type in use; a 220 V AC load needs an inverter. Nothing to do there, just read.

Done when the team can say which battery type fits its scenario and why, in this order: does anything run at night; how many Wh per night and the usable share (about half for lead-acid, 80–90 % for LiFePO4); who maintains it; first cost against lifetime; weight, place and heat. Nobody touches the controller's menu: it stays on b01 (lead-acid), because the adapter is not a battery.

The worksheet's figures are general estimates (cycle life, usable depth of discharge, relative price); กำพล to confirm against what they teach and current Thai prices. Talking points: LiFePO4 costs 2–3× per Ah but gives 5–10× the cycles, so it is cheaper per cycle and needs no care; lead-acid still wins for a one-off low budget or where a local shop can service it; NMC (18650, power banks, EV cells) suits small or portable gear, not a fixed farm install; NiMH only in garden lamps; sodium-ion is arriving but not yet common. If asked why the type matters to the controller: each type has its own charge voltages, so the controller has a battery-type setting; lead-acid equalisation would overcharge lithium, gel dries out at flooded voltages, and LiFePO4 must not be charged below 0 °C. The LCD icon reads not full on the 12.0 V adapter because a full lead-acid sits at 12.7–13.0 V; the controller judges only from voltage.

If stuck Ask two questions first: does anything have to run at night, and will anyone come to top up water? The answers usually settle the type before price does.

8.5 Work out the power, then choose the source

Rewritten 18 Sep 2026 (Roger): the worksheet now opens with the four calculation steps and two worked examples; the kit-load table and the folded answer table are gone.

Done when the team can run the four steps on its own main load and say which source it leads to: W = V × A from the label; Wh per day = W × hours; battery at least that much, converted with Ah = Wh ÷ 12 and grossed up for the usable share (about half for lead-acid, 80–90 % for LiFePO4); panels sized for the worst day, 3 full-sun hours, with W of panel = Wh per day ÷ 3, rounded up to whole panels.

The two examples on the worksheet, in case a team gets a different number: LED 50 W × 8 h at night = 400 Wh, so about 33 Ah usable at 12 V (lead-acid about 70 Ah rated, LiFePO4 about 40 Ah) and 133 W of panel, one 150 W panel or nine of the kit's 15 W. Pump 1,100 W × 2 h = 2,200 Wh, so about 183 Ah usable (lead-acid about 370 Ah, LiFePO4 about 220 Ah) and 733 W of panel, five 150 W panels. The pump example makes the point: run a big pump in daylight straight from the panels and store water, not electricity, with panels that can deliver more than 1,100 W while it runs and a tank that covers cloudy days.

Kit loads, if a team asks for closer numbers: sensor board 0.5 W × 24 h = 12 Wh per day, one small panel, a night is 6 Wh which the board's 18650 covers; router plus Home Assistant host 10 W × 24 h = 240 Wh, about 20 Ah at 12 V for a night; kit pump 18 W × 2 h = 36 Wh, no battery if it runs in daylight.

If stuck Start from the label on the device: volts times amps. Then ask whether it runs at night. Panels are rounded up to a whole panel.

End-of-block question

A good answer splits the team's own loads into daylight-only (pumps) and must-run-overnight (sensors, network, controller). For three cloudy days it names the numbers needed: each load's Wh per day, the usable battery capacity (about half the rated capacity for lead-acid), and what the panels still produce under cloud.

8.6 Design for one pilot point

Done when the four blocks read as one plan: a sensor and an actuator named, with a control rule and a response to sensor/communication failure; the two local communication rows carried over from 8.2, a remote-access choice informed by 8.3, and the battery/power answers from 8.4–8.5 consistent with the parts chosen; and a test plan covering normal operation and the chosen failure case, with observable pass criteria. Unknowns are written into the block they belong to, not guessed.

Presentation and exchange · 15 min, instructor-led

Not in the worksheet since 18 Sep 2026 (Roger: the instructor directs it). Pair the teams. Each team tells its plan in one minute from the four blocks: the problem, what senses and what acts, how it connects and is powered, and how they will know it works. The other team has three minutes for two questions: which decision is best reasoned, and what happens if one part fails. Swap. Teams change their answers in 8.6 directly; there is no separate revision box, no rubric walk and no plenary.

Session 4 · Activity 9 — Reflash the plot board for ESP-NOW (30 min)

The Session 4 story, said once at the start: the plot is beyond WiFi range, so the GoGo board out there will talk ESP-NOW — board to board, no router, further than WiFi — to a cheap board in the farmhouse, which is the only thing Home Assistant sees. In the room the GoGo board keeps its WiFi so both paths can be compared; say explicitly that at a real plot it would not have one.

Before start (ก่อนเริ่ม) · ~5 min

Done when the five checks at the top of Activity 9 are ticked: kit powered, float in its glass, Home Assistant reachable, both boards online, every relay-driving rule off and Pump Auto Mode off (the pump is not used this session), and the team has its receiver devkit but has not plugged it in. If Session 3 ended on another day, the kit is re-cabled first (Activity 8's closing list).

9.1 Download and install

Done when the team has first seen the old image's log (name, IP, no GOGO, lines), then web.esphome.io reports the install complete and the board reboots. The file is the zone's gogo-iot-espnow-<colour>.factory.bin from the resources page; the pre-flashed sessions 1–3 image is gogo-iot-<colour> and stays available for a board that must go back.

If stuck Chrome or Edge only. Close the old Logs window before Connect. The GoGo board must be on the laptop's USB, not the charge controller's. Wrong colour = board vanishes from Home Assistant; reflash with the right file. Do not tick "erase": it wipes the pump-sensitivity choice for nothing.

9.2 Read the log

Done when the team has DEVICE NAME unchanged from Activity 1 (s4.tx.same = เหมือนเดิม), an IP ADDRESS line, and one [espnow] >>> GOGO,… line copied into the worksheet, and has watched the light field (6th value) fall under a cupped hand. Walk the ten fields on the projector once.

If stuck No >>> lines: the board did not take the new image — check that the file name says espnow, reinstall. nan in the first packets is normal. The name changing is not normal: it is derived from the MAC, so a changed name means a different board was flashed.

9.3 The two new diagnostics

Done when ช่องสัญญาณ WiFi (channel) is written down (s4.tx.channel) from the board's own page, แพ็กเก็ต ESP-NOW ที่ส่ง is seen to climb, and Home Assistant shows the board back with the two new diagnostic entities without re-adoption. Then the USB cable goes back to the charge controller so the laptop is free for the receiver.

A good answer (end question) lists what the plot board loses without WiFi: its own web page, the direct Home Assistant entities, OTA updates and the log over the network — and says the fix is to configure it over USB (or via its fallback hotspot) at the house before carrying it out, or to add a serial/USB visit to the maintenance table.

Session 4 · Activity 10 — The receiver in the farmhouse (50 min)

10.1 · 10.2 Flash the receiver, read its name and IP

Done when the devkit runs farm-receiver-<colour>-xxxxxx, the colour matches the zone, and name and IP are in board.rx.name / board.rx.ip. The GoGo board must be powered from the charge controller so only one serial port shows.

If stuck No port at Connect: another USB cable (charge-only cables are common), or hold BOOT while plugging in. Installed but no log: the board's USB port is the other kind — this is the --receiver-console choice from section 1 and a TA should know which was built. Some DevKitC boards have two ports; try the other.

10.3 Open the receiver's page and name the sender

Done when the team has typed the GoGo board's six-character suffix into ชื่อบอร์ดที่แปลง, sees its full name under บอร์ดที่รับข้อมูลอยู่, รับข้อมูลจากแปลง (link) ON, values filled, and has written the receiver's channel, an RSSI (about −30 to −50 dBm on the table) and one other board's name from บอร์ดอื่นที่ได้ยิน — the evidence for the end question that the channel is public.

If stuck, in this order (1) the suffix — every character, from the GoGo board's real name; (2) is the GoGo board sending — its own page's แพ็กเก็ต ESP-NOW ที่ส่ง climbs; (3) do the two boards show the same ช่องสัญญาณ WiFi? If not they are on different access points; power-cycle the receiver (or move it next to the GoGo board) until the channels agree. This is the one failure the venue's network can cause and the TA should look for it first when a whole table is stuck. บอร์ดอื่นที่ได้ยิน changing proves the receiver itself works.

10.4 Prove it is our board

Done when a cupped hand on the GoGo board moves the receiver's light value within a few seconds (s4.rx.delay, typically 3–5 s: one send interval on top of the Activity 1 figure) and lifting the float flips the mirrored float switch. Two tabs, GoGo page and receiver page, should agree.

10.5 Distance

Done when the GoGo board has been carried to the far end of the room on its 18650 cell (or a power bank), the far RSSI and whether packets kept arriving are recorded, and แพ็กเก็ตที่หาย has been read back at the table. Tell teams to stay inside the room: leave the WiFi and the board may change channel while reconnecting. Flat cells are the likely snag — decide before the day whether kits get charged cells or a power bank (TEST-PLAN test M).

Point to make ESP-NOW broadcasts are not acknowledged: what is lost is lost, which is why the board resends every 3 s. Real range is 100–200 m in the open; the room shows the trend, not the limit.

10.6 The plot goes quiet

Done when the team has switched the GoGo board off, watched the link go OFF and the values turn NaN at about 15 s (s4.rx.off.sec), and seen them return after the board rejoined WiFi. Say the fail-safe sentence: the receiver would rather show nothing than a stale number.

A good answer (end question) names both halves — a neighbour's board could be mistaken for ours (so filter by name or MAC), and a neighbour can read our values (so encrypt in a real deployment) — and might add that a shared channel also carries the neighbour's traffic load.

Session 4 · Activity 11 — The far plot in Home Assistant (45 min)

11.1 Adopt the receiver

Done when the receiver is added by its IP with the Device name checked (same as Activity 3) and the team has found its three entity groups: the ESP-NOW sensors plus the link (Connected / Disconnected), the sender name text under Configuration, the link diagnostics.

11.2 Two paths, one sensor

Done when humidity and light are written from both the GoGo device page and the receiver's (s4.ha.*.direct / .espnow) and agree within one update, and a breath on the board moves both.

11.3 Dashboard แปลงไกล

Done when a new blank dashboard holds the four cards from the receiver's entities — link Tile on top (การเชื่อมต่อกับแปลง), humidity Gauge, float Tile, RSSI Tile — as in Activity 7. The history graph comparing both humidity entities is optional and pleasing.

11.4 Alert when the plot goes quiet

Done when แปลงเงียบ - เตือน exists — trigger: State of รับข้อมูลจากแปลง (link) To Disconnected For 30 s; action: persistent notification — and switching the GoGo board off produces the notification about 45 s later (s4.ha.alert.ok). This rule drives no relay and may stay on; the pack-up list has it switched off only so a powered-down kit does not leave a stale alert.

A good answer (end question) has the plot board with no WiFi (a fixed ESP-NOW channel matching the house router, set over USB before it goes out), on solar with a small battery, mounted high and in line of sight; the receiver near the router and always powered; and names the link tile / the notification as how the user knows a reading is current, not stale.

Wrap-up and kit return

Read the four summary points. The kit-return list moved here from Activity 8 on 20 Sep 2026 and now includes the receiver and its cable; answers are downloaded and submitted at this point.

Still missing

Answer hand-in: the destination is still to be decided (likely an upload form; confirmed with the organizer on 19 Sep). Announce the actual destination before the final activity. Ask each recorder to download the JSON file, check its team number, submit it, and keep a local copy. A TA checks that the file was received and can be opened for that team; progress on the live board is not answer submission. Publish the form link in the worksheet and front page once it is available. A printed record sheet should be collected directly from teams using paper.

Timed rehearsal: run Activities 4, 5 and 7 with a novice-paced team before delivery. Record actual durations, allow for fault recovery, and agree where TAs provide a prepared working setup to keep a team moving. Check the real venue's zone WiFi, adding boards by IP, VM/phone access, vibration detection, and progress reporting. Prepare an instructor history graph for teams with insufficient data. Pump run times must follow the verified limit of the actual model; see the readiness action plan.

Nobody has produced these yet:

• The install-guide details participants need: VM login, file size, verification form, help channel
• Slide decks for Sessions 1–3 and the three post-tests
• Expected values that need a bench run: the Activity 4 flicker count (TEST-PLAN test E) and typical room readings for Activity 1
• The LED-module bench check now gates Day 1 (TEST-PLAN test L): it lights through the relay from the 12 V rail, a reversed module is harmless, and a lit module 30 cm from the board does not lift the light reading over a covered-board threshold
• Real screenshots in the install guide (currently placeholders)
• Bench tests against the built firmware — see TEST-PLAN.md (repo), especially test B (the boot banner participants read) and test G (float switch polarity)

Building the images rather than using them? The firmware build and the 18-step OVA recipe are in firmware/README.md and hass/README.md — both repo-only.