Environmental Control Automation~

AoT's env_coordinator (shown in the UI as Integrated Environment Control) is a 3-layer control system that coordinates every actuator in one facility — vents, exhaust/intake fans, heaters, coolers, misters, shade screens, thermal curtains, supplemental lights and CO₂ injectors.

Two things are decided in two different places, and mixing them up is the single most common source of confusion:

• What to aim for — VPD, CO₂, DLI and the rest — belongs to the plot that is growing in the facility, through the cultivation program attached to it.
• How to get there — which equipment exists, what lines must never be crossed, how eagerly to chase the target — belongs to this function.

Where Targets Come From~

A plot internalises the program it selects. Editing a stage target on the plot screen overrides that one item for that plot only; the program and every other plot using it stay untouched. Clearing the field returns the item to the program value — that is why there is no separate "revert" button.

Two limits are deliberate:

• The plot cannot invent new target items. The vocabulary is the program's (target_defs), so an unknown key is rejected rather than silently ignored.
• An item that follows a curve cannot be overridden with a number. The curve wins, and the plot screen shows that item as read-only.

What the coordinator reads each cycle~

Targets are matched to a control axis by measurement and shape, not by the key name — a user-named item such as "Indoor CO₂" still reaches CO₂ control as long as its measurement is CO₂. An item with no measurement selected reaches no axis at all and is shown as For reference.

When there is no plot~

The coordinator does not stop. With no plot, no program, an AI draft that has not been reviewed, a program whose kind does not match the plot's, or no stage running today, there are simply no targets: control keeps running inside its own guide ranges, which is already the defined behaviour — an empty greenhouse still needs heating.

Day/night temperature and humidity guide the ranges~

VPD stays the primary target. The stage's Day temp, Night temp and Humidity are not targets — they keep VPD from being reached with a combination that is too hot and humid or too cold and dry. Each cycle the coordinator replaces its guide ranges with:

• The width is fixed and not a programme setting — cultivation guides almost always give an average, not a day/night spread.
• A stage without the value, or a facility whose location is unknown (no sunrise/sunset), leaves the function's own guide range in charge for that quantity.
• The hard limits (Min/Max Temperature, Min/Max Humidity) still apply last.
• Everything that reads the guide range follows it: the temperature/humidity that VPD is split into, the mid-point used when there is no VPD target, forecast feedforward, and the ventilation temperature ceiling.
• The range actually used is shown as Guide range under the plot in the function's settings.

The coordinator has no end-date option of its own. Whether growing continues is the plot's business. A date kept in the function could leave the facility stopped even after a new crop was planted.

Control basis — VPD direct / temperature first~

The default is VPD direct: the VPD target is followed as described above, and temperature and humidity are guides. Setting Control Basis in the advanced settings to Temperature first changes a coordinator as follows.

• Temperature is the primary target. It follows the stage day/night temperature (day and night from the facility's sunrise and sunset).
• The humidity target comes from the VPD target — the humidity that gives the VPD target at that temperature (for example about 66 % for VPD 1.0 at 24 °C), cut to the guide range.
• VPD is a limit. It is pushed back only when it falls 0.4 kPa below or rises 0.8 kPa above the target.
• Without a stage temperature, temperature is not followed, only kept inside the guide range.

Following VPD directly turns the same tolerance into ±9 % humidity at 8 °C but ±1.8 % at 35 °C, and at low temperature it can demand a humidity that cannot be reached. Temperature first does not have that difference.

Whichever is chosen, the other one is computed every cycle as "the commands it would have sent" and recorded (never used for control), so the two can be compared on the same facility.

More than one plot in scope~

Intercropping is normal, so the coordinator never guesses. With two or more plots in its facility/bay scope it says so and offers a Follow this one button; the same choice is stored in Reference Plot under Advanced Settings. If the pinned plot ends or leaves the scope, the value is not erased — the screen says the pinned plot is no longer there and the selection rules are applied again.

The program version recorded on a plot pins the version number, not the content. Editing a program does change the interpretation of plots already in progress.

VPD (Vapor Pressure Deficit)~

VPD is the key metric determining plant transpiration and water uptake.

VPD = SVP × (1 - RH/100)
SVP = 0.6108 × exp(17.27T / (T + 237.3))  [kPa]

env_coordinator Control Layers~

L1 — EnvTarget (setpoint)~

Resolves the targets described above into setpoints for this cycle. A target that follows a Method is evaluated with the plot's elapsed weeks and the facility's local timezone.

L2 — SituationReport (evaluation)~

Evaluates current deviation, limiting factors, and trend.

Indoor temperature, humidity and VPD are filtered for noise first, so that a sensor flicking between two display steps (0.1 °C, 1 %) does not move the vents. Exponential smoothing is used, but when the reading moves 0.5 °C or 3 % away from the filtered value it is treated as a real change and followed immediately. Where VPD is computed from temperature and humidity, it is recomputed from the filtered values. Safety gates and hard limits judge the raw readings, so protection is never delayed by the filter.

When there is a VPD target and VPD can be measured, temperature and humidity are demoted to constraints — the range VPD is decomposed into, and the lines that must not be crossed. This function is not a thermostat.

L3 — Coordinator (actuator command)~

Each device gets a 0–100 % command from a position-form PI controller. It differs from the textbook PI in three ways:

• The integral is not accumulated error but the device's remembered equilibrium opening (%), always held within 0–100. When the deviation is small the integral stays put, so the device holds its last position.
• A device that affects several axes combines them into one error. A vent, for example, moves temperature, humidity and CO₂. Each axis's deviation is divided by its proportional band (tolerance × 6) so they share one scale, then averaged with the weight priority × how strongly this device moves that axis.
• Within half the tolerance is a rest zone (dead zone). Outside it, the dead zone is subtracted rather than switched on, so the command does not jump at the boundary.

e      = Σ(priority × effect × deviation/band) / Σ(priority × effect)
e_eff  = e minus the dead zone (tolerance × 0.5)
I      = clamp(I + 0.2 × e_eff, 0, 100)          # remembered equilibrium opening
cmd    = clamp(1.0 × e_eff × 100 + I, 0, 100)

• Devices are settled in cost-index order (cheapest first). Within a domain, each later device sees only the deviation left after the effect of those before it — this is load sharing.
• A device that cannot help rests. If running it at 100 % would move the variable by less than 2.5 % of the proportional band (for example ventilation with almost no indoor/outdoor difference), or it is parked, it moves toward its safe position, keeping 60 % of the remaining distance each cycle. A thermal curtain is the exception and holds its position — relaxing it to its safe position (open) would release the insulation at night as soon as the indoor/outdoor difference narrows. The judgement has hysteresis: a device that was judged able to help only rests once its effect falls below half of that threshold (1.25 %), so an effect hovering near 2.5 % does not make the vent open and close repeatedly (reason primary ↔ no gradient).
• Thermal curtains stay open in the daytime. The effect model knows only the curtain's heat exchange, not that it blocks light, so from sunrise to sunset at the facility's location the curtain stays open and is decided only at night. Safety gates such as extreme cold take precedence over this rule.
• Curtains and shade screens are fully closed or fully open. To move from the side it is on, the command must pass 60 % (to open) or 40 % (to close), so a command near 50 % does not swing it end to end. The log and the screen show the 0/100 actually sent.
• When the direction flips, it waits to see if the flip lasts. A vent's effect on VPD is the sum of cooling and dry outdoor air working against each other, so when the two are close a few points of outdoor humidity flip the direction every cycle. The new direction is followed only once it has lasted at least 10 minutes (and two cycles); until then the device holds its position. The temperature term added as the indoor temperature reaches the guide ceiling also gains weight gradually with the excess, so hovering around that line does not step the vents up and down.
• Side vents facing the wind open wider. A side vent facing away from the wind is reduced to as little as 20 % of the command. Below 0.5 m/s this does not apply, and up to 3 m/s it strengthens gradually with wind pressure (speed²), so a light, shifting wind does not cut and restore the side vents from one cycle to the next.
• It does not freeze past the target. Even inside the deadzone, if the deviation keeps saying "less" for three cycles, the device backs off by keeping 60 % of its command each cycle. Thermal curtains and shade screens do the same: their effect is proportional to the opening, so reducing the command is what reduces the effect.
• Anti-windup. At the 100 % rail the overshoot is fed back into the integral; at the 0 % rail that cycle's integration is frozen. If a device stays pinned to a rail, its integral is eased toward the actual opening.
• Rate limit. A command moves at most 20 percentage points per cycle; if the Action sets a full stroke time and the device can physically move less than that, the smaller value is used. Curtains and shade screens are exempt because they open or close in one motion. Commands below 5 % are sent as 0.

Temperature ceiling in VPD mode. When VPD is controlled directly, temperature is not a control target — left alone, nothing would open the vents as the house heats up. So when the temperature expected at the next decision (current + rate of rise × period) comes within 1 °C of the guide ceiling, a temperature term is added for the ventilation devices, weighted more the further it goes over. Once the term is on it stays until the expected temperature falls a further 0.5 °C below that line, so a temperature sitting on the line does not swing the vents between cooling and resting. At the Max Temperature hard limit the term also applies to coolers, and terms from other axes that oppose cooling are dropped. It is still a PI proportional to the deviation and forces nothing to 100 % — when outdoor air is hotter, the vents move toward closing instead.

Actuator Domains~

Load sharing happens within a domain, and domains are separated by where the energy ends up — not by how similar the devices look. The settings screen is grouped along similar lines, except that supplemental lighting is set together with the screens under Light and Shading.

Because domains do not see each other's work, coordination between them is done by declared interlocks (see Ventilation), never by implicit accumulation of effects.

Roof vents and side vents behave differently~

Both are opening actuators, but the facility drawing distinguishes a ridge (roof) window from a side window, and the effect model uses it:

If the drawing does not say which form an opening is, ridge and side vents use the same model (no correction).

Function Configuration~

Navigate to Functions → Integrated Environment Control in the AoT UI. Actuators themselves are registered separately (see Registering Actuators) or auto-discovered from the linked facility.

Reading the settings screen~

Five things are worth knowing before reading the tables below.

A status header sits at the top. Under the facility picker the screen shows what control is doing right now — each axis it is tracking as its current value against the target (VPD, temperature, humidity, CO₂, light), the position of each device kind (Vents, Heating, Cooling, Misting, Shade …), and how long ago the last decision was made. A second line says what this facility can actually adjust on each axis: both ways, raise only, lower only, watch only, or no sensor. Below it, a two-line summary names the plot being followed, the stage it is in, and the targets in effect. States where nothing can run are spelled out separately: no environment control linked to this facility, control switched off (naming the coordinator that runs the facility instead, if there is one), no decision for longer than this coordinator's own limit, and not having run a cycle yet. Before a facility is picked the header stays empty — the picker is right above it. If the plot ends within two weeks, one extra line says so.

Settings are arranged in four layers, not by option type.

A key setting carries its details with it. Instead of asking for a number nobody can answer ("should the emergency multiplier be 3.0 or 4.0?"), the screen asks a question that can be answered and moves several values at once, the way a robot vacuum offers Quiet / Normal / Strong. Three such controls exist: Control Temperament, Ventilation and HVAC Teamwork, and Misting Frequency. The step name itself is never saved — only the real values are, and the current step is inferred back from them. If the values match no step, the control reads Custom.

Ranges are asked as a band with two handles. Temperature and humidity have one question — what range to grow in. Dragging a handle moves the matching never-cross line with it by a fixed margin (±5 °C, ±5 %RH); if you never drag, the hard limits stay at their factory values (5–35 °C, 30–90 %). The band cannot put a hard limit inside the guide range, so "grow at 12–32 °C but never exceed 30 °C" cannot be expressed — that combination is what sets a heater and a cooler against each other. If you type numbers under [Advanced] that do this anyway, saving warns you on the spot — naming the axis, the two ranges, and the range targets will actually be built in — whenever one of that axis's four values was part of what you just saved; the coordinator then narrows the guide range to fit inside the hard limits and logs that it did.

One [Advanced] switch opens every numeric field. It turns each step scale into a step scale plus its number box, reveals every advanced-only row, and expands the folded groups in one go. The switch lives in the browser only — it is not part of the saved configuration, so two people never see the same function differently because of it. Settings whose parent toggle is off are hidden rather than disabled, so their stored values are still submitted and survive being toggled off and on.

Some settings are read only under a specific condition — the custom actuation period only when the profile is Custom, the night clock times only when night is measured by fixed times. The screen reports a value that was entered but is not being used, because otherwise it silently does nothing.

Commands~

After either of the first two, the next cycle is treated as urgent once, so vents may move without waiting out their actuation profile interval.

Facility Settings~

Working Hours~

Only the toggle is visible until it is switched on. This is a switch, not a schedule: outside the window, target tracking (L1–L3) stops entirely, heating and cooling included. Protection continues:

• The Pre-Gate safety checks (rain, strong wind, heat and cold emergencies and so on) are evaluated every cycle, and a gate that fires sends its forced commands. The partial gate that closes only the windward vents in strong wind works the same way.
• The protective side of the temperature/humidity hard limits still acts. If the indoor temperature drops below the minimum, for example, vents and thermal curtains close and cooling is blocked. Hard limits never drive a device (they do not switch the heater on), so this does not conflict with "stopped". Light limits are not used outside the window, because the minimum-light response switches grow lights on.
• Every device that received none of these protective commands gets its end behaviour. The window is judged in the facility's local time, falling back to server time when the facility's timezone is unknown.

Target and Temperament~

Under [Advanced], each range band also shows the four numbers behind it (Guide T Min/Max, Min/Max Temperature, and the humidity equivalents), and Control Temperament shows its members:

Two further members sit in the same place but are not set by the step, because they are fine adjustments on the same axis: Custom Actuation Period (seconds) — used only when the profile is Custom — and Emergency Minimum Interval (seconds), default 60, the floor between two vent commands even during an emergency.

The actuation profile governs only how often side/roof vents are allowed to move. Sensing and computation always run every cycle period, and curtains and shade screens are unaffected (they open or close in one motion). A safety gate's forced command (closing for rain or strong wind, for example) reaches that device immediately regardless; vents that are not being forced keep this interval. The interval is skipped altogether only when indoor temperature changes fast, or when a deviation exceeds the emergency threshold and is still widening — merely staying far from the target is not an emergency.

Ventilation~

Ventilation and HVAC Teamwork moves three toggles at once:

• Close Vents When Ventilation Cannot Help — ventilation can only pull the inside toward the outside. When the target lies on the far side of the outdoor air, opening moves away from it no matter how wide; the classic case is dehumidifying at night, when outdoor air is wetter than indoor. With this on, vents and exhaust/intake fans park closed instead of holding a partial opening all night.
• Rest Heating and Cooling When Venting Can Reach the Target — when outdoor air is already past the target, ventilation alone gets there and running HVAC alongside pays for what the outside would do for free. Three conditions must all hold: the outdoor value is past the target by more than the tolerance, every controlled variable is, and the vents still have headroom (the last cycle's widest opening below 90 %). When outdoor air can cover only part of the way to the target, heating and cooling work only on what is left. If the target is still not reached after 15 minutes, everything is handed back to heating and cooling — the prediction was wrong. While rain or wind readings are lost and the vents cannot open further, this judgement is switched off entirely — heating and cooling must not back off counting on a vent that cannot open.
• Keep Vents Closed While Heating or Cooling Runs — venting against a running unit throws that heat or cold straight outside. In a season where outdoor air could help toward the target, this throws that help away too.

Detection of a running unit needs evidence, and there are only two sources: this coordinator commands the unit itself, or you point the signal field at a measurement that rises when the unit runs. Indoor temperature is deliberately not used to guess. On a sunny day a house can be cooler inside than outside with no cooler at all, so "indoor cooler than outdoor means cooling is on" is often wrong even with a margin.

If the signal has no freshness limit of its own, it is judged by its own measurement period (×2, floor 300 s). An expired signal counts as not running, and that is logged.

Night closing exists because humidity rises and dew forms at night: an opening that looked useful at dusk can leave the crop wet by morning. Its sub-settings appear only once the toggle is on.

It lifts itself when staying closed becomes the greater risk: when a temperature or humidity hard limit is crossed, and when condensation is imminent and ventilation would actually relieve it. Imminent means the gap between indoor temperature and dew point is under 2 °C; relieving means that mixing in a little outside air widens that gap. Cold, humid outside air can cool the house into condensation, so dry outside air alone is not a reason to open. Once lifted, parking returns when the gap recovers to 3 °C or ventilation stops helping. With no outdoor readings it is not lifted.

Three guarantees hold: safety gates win (a summer night's heat still opens the vents), the hard temperature/humidity limits break the parking, and if no coordinates are available to compute solar time, nothing is parked at all. A fourth rule is not specific to night closing: if the outdoor reading itself is unavailable — no cache, only a synthetic fallback that assumes outdoor equals indoor — the coordinator does not close vents at all; it holds them exactly where they are, overriding any close decision this or another control rule already made for that cycle. Emergency safety gates (rain, wind) still win over this hold and can close vents. Closing on a fabricated outdoor-equals-indoor guess could otherwise leave vents shut through a hot day once the weather source dies. Times are read in the facility's local timezone, not the server's. On the facility popup, a device parked by this option reads Closed for the night — heating and cooling take over, kept separate from Nothing this device can change right now.

Heating, Cooling and Misting~

Misting valves are almost always on/off, so there is no way to reduce the flow: the only adjustable quantities are how long one run lasts and how long the next one waits. That is what this scale sets.

Sunburn protection is a separate decision from frequency — keeping them apart is what lets you ask for "spray often but lock out in strong sun".

Whether or not protection is on, a wetting-type mister is dosed in pulses rather than modulated continuously, because continuous modulation never lets the leaves dry. The run time and interval come from the Misting Frequency scale (factory setting Frequent: 20 s maximum on, 600 s drying interval); only when those values are empty does it fall back to 30 s / 180 s.

Wetting-type misting is also locked well before the Max Humidity hard limit. It is locked when humidity rises above this cycle's humidity reference (the humidity split out of the VPD target, kept inside the guide range) plus 5 %, and also whenever humidity is unknown — so already-wet air does not wet the leaves further. Fog-type (high-pressure) misting is not affected.

Light and Shading~

Either end can be switched off, and off means different things at the two ends: the lower handle at 0 is no supplemental light, and the upper handle turned off is no shading. If the facility has no shade screen or no supplemental lighting registered, the screen says so rather than offering a handle that does nothing.

The shading line is judged on how bright it would be with the screen drawn back, not on the light under a closed screen — otherwise closing it would darken the reading, release the decision and open it again the next cycle. Where outdoor light is unknown (an indoor sensor only), the current reading is used instead. While shading is in force, supplemental light and screen opening are not forced at the same time: if the two lines cannot both be met — a closed screen leaves the place darker than the supplemental-light line — shading wins and the contradiction is written to the log once, for you to resolve.

The underlying values are Min Light Threshold (Supplemental) (default 0) and Max Light Threshold (default 800), both visible under [Advanced].

Three related values are not on this screen:

• Shade cloth transmittance belongs to the facility, under the shade curtain in the facility editor. It is used only when there is no indoor light sensor: indoor light is then estimated from outdoor irradiance and the screen position. An unset or out-of-range value falls back to 0.50, and a facility that declares no shade curtain uses 1.0. A single screen whose cloth differs can still override it in its own Action.
• Covering transmittance comes automatically from the envelope material (glass 0.85 · double film 0.78 · non-woven 0.50, and so on). A roof always cuts light even without a shade screen, so it is multiplied into the indoor light estimate. The light thresholds are therefore indoor values. The same 250 means nearly twice as much light for the crop in a glasshouse as under non-woven cover.
• The light saturation point is derived from the crop's K_L in the program, not from the shading threshold. If the two were one value, lowering the shading threshold would make the photosynthesis model conclude that light is already sufficient, and it would miss light limitation even under strong sun. With no K_L, the system default of 600 W/m² is used.

Advanced Settings~

This group is folded by default and is for engineers testing the function, not for growers.

Effect Calibration~

Forecast Feedforward~

Debug logging is no longer a separate option. It duplicated the framework's own debug switch, and almost everything it guarded was written at DEBUG level, so on its own it produced nothing. The one switch in the function's Advanced settings now does both. Critical events — safety gate, dispatch failure, runtime-state error — are always recorded regardless.

Registering Actuators~

Actuators drawn in the linked facility are discovered automatically. Anything else is registered with an Environment Control action on this function; add the action once per device. Manual actions are merged with the facility-derived list.

Automatic conversion by device type~

A command is always computed as 0–100 %. The adapter is chosen from the Output module's own metadata — no extra configuration:

A wetting-type mister is additionally wrapped in pulse dosing: one spray is cut at the maximum on-time, and nothing sprays at all until the drying interval has passed.

Irrigation flow is aggregated from the facility drawing — every emitter under a layer is summed into that actuator's flow_lpm, and the volumetric adapter uses it directly. The fallback order is per-actuator flow, then the facility total, then 1.0 L/min.

The fogger effect model uses a different flow: only that actuator's non-drip spray nozzles, and it never falls back. A dosed volume properly includes the drip lines that go to the roots, but what evaporates into the air does not — so a mister with no nozzle information on the drawing is credited with no cooling or humidifying effect at all, rather than inheriting the facility total.

Axes without a measurement, axes without a device~

Every facility has different equipment. An axis that can be measured but not moved, and one that can be moved but not measured, are both normal configurations. The coordinator looks at the combination and adapts on its own.

The Can adjust line at the top of the function settings shows this combination as one word per axis (temperature, humidity, VPD, CO₂, light). The coordinator computes it from the equipment and the values measured in the latest cycle; hover over a word to see the devices that move that axis.

An axis with neither devices nor a sensor is not shown. An axis that is moved only as a side effect — a vent lowering CO₂ a little, for example — is not shown either when it has no sensor. VPD takes devices on both temperature and humidity into account: humidifying mist lowers VPD, and ventilation raises it when the outside air is drier. For now this line is display only; control follows the rules in the table above.

The safety gates follow the same rule. When indoor values are lost, the gate names that fact on the screen and in the log, but creates no forced command. Driving every device to its safe default (vents closed, screens retracted) would move equipment precisely because there is no evidence. This is the same judgement as for lost outdoor values, and control does not stop as a whole either: axes that can still be measured, such as CO₂ or light, keep being controlled.

Risks that can be judged from outdoor data alone, such as rain and wind, are still guarded. Heat and cold emergencies, on the other hand, are not judged while indoor temperature is unknown — those gates drive every device one way, so they must only fire on firm evidence.

The health check reports mismatched combinations as well:

python3 -m aot.scripts.check_env_coordinator_health

It reports under configuration check a target with no measurement, a target with no device, and a device that cannot measure any of the axes it moves. An axis that has a measurement but neither a target nor a device is not reported — that is exactly what a reference value is.

Safe default~

Each actuator has a safe position it moves to when a safety gate fires, on emergency stop, or on an external force_safe_state() call. This follows the device kind — thermal curtains and shade screens park at 100 %, everything else at 0 (off) — whether the actuator came from the facility drawing or from an Action.

Methods (Setpoint Curves)~

A Method defines how a setpoint changes over time. Methods are attached to target items in the program, not to this function — an item with a curve shows as Follows a curve on the plot screen and cannot be overridden with a number.

• Daily — a setpoint per time of day (HH:MM)
• Duration — a setpoint per elapsed hour since start
• Daily Bezier — a smooth diurnal curve
• Repeating — a repeating pattern

Curves that progress by growth week are evaluated from the plot's start date, read at midnight in the facility's local timezone, and from the elapsed weeks that the plot screen and the plot journal also use.

Example crop stage schedule (tomato):

Methods prefixed with SEED: are seed presets and are read-only. Duplicate a preset before editing.

Safety Gates~

Safety runs outside the L1–L3 coordination algorithm — a Pre-Gate checked every cycle before L1–L3, and a Post-Gate that sanity-checks the L3 result before it is dispatched. Once triggered, a Pre-Gate stays active for at least 300 s after its last trigger (prevents rapid on/off flapping). When a gate releases, the devices it moved find their equilibrium again from where the gate left them rather than jumping back to the pre-gate opening, because conditions may have changed in the meantime. A gate's forced commands go out at once, and are re-sent every cycle while the gate holds, without waiting for the actuation interval. Devices the gate did not force keep their normal interval — a gate standing on one device does not speed up the rest of the facility.

Pre-Gate (checked before L1–L3)~

Lost values are a constraint, not an emergency. Indoor values lost and outdoor rain/wind lost both create no forced command and no 300 s hold, and when only these two are active, L1–L3 run normally. The rule is that equipment is not moved just because evidence is missing; whether to move is decided axis by axis by the coordinator.

Rain, Heat and Cold thresholds are fixed in code — the Wind threshold is the only one exposed as a function option. Multiple gates can be active at once (e.g. rain + strong wind); vents then close unconditionally regardless of direction.

A gated cycle is still a cycle. When a gate ends the cycle early, a reduced summary is written and the cycle is stamped, so the facility widget keeps showing what happened instead of reporting the coordinator as unresponsive. The reduced summary still carries the environment values measured this cycle — sensors keep measuring while a gate holds — and leaves out only what L1–L3 would have produced, such as targets and deviations. Every device stays listed, not just the ones the gate forced — a device that was not touched carries no command at all, rather than 0 %.

Post-Gate (checked after L3, before dispatch)~

Emergency stop and safe state~

The next cycle is delayed 60 seconds after an emergency stop; actuators not moving immediately afterwards is that delay, not a fault.

Troubleshooting~

AI Integration~

The AI in AoT's chat, and an AI connected over MCP, use the following tools to look at this coordinator. Every write tool goes through human approval.

Targets live on the plot's stage plan, not in this function. To change one plot's targets, edit them on the plot screen. The AI can also change stage targets with the program editing tool (modify_program), but that applies to every plot using the program.

Related Pages~

• AI Overview
• Functions Guide
• Methods Guide