Enums, types & console#

Seconds before the faction knows. The window in which stopping the messenger still changes the outcome.

Random position error added, in cm. A shouted location is approximate.

Chance per trait that a detail is lost in the retelling (0 to 1). Weighted by observability, so the obvious things survive and the subtle ones do not.

Multiplier on the reported confidence. Second-hand is less certain.

The traits that make up the description: each a key (such as Coat), a value (such as Red) and how easy it is to notice.

Identity of this evidence. Adding a second entry with the same tag replaces the first, so re-applying "Disguise" refreshes it instead of stacking.

What the evidence does to confidence: Raise (a floor it can't fall below), Clamp (a ceiling), Lower (settles it below what the senses say), or Invalidate.

Meaning depends on Operation. Ignored entirely by Invalidate.

Seconds before this expires on its own. 0 or less = holds until removed.

A free label for your own debugging: Disguise, Dispatch, Cutscene_03.

Seconds this entry has been held. Managed by APS.

Where the sound was made.

Who made it. None for a noise with no maker (Emit Sound At Location).

The Sound Type Definition of the sound that was heard.

The Sound Type's name, for example "Creak". "Sound" when it has none.

Sound Name as words, for a label: "FootstepWalk" reads "Footstep Walk".

How loud it was made (the Sound Type's loudness times the emitter's multiplier).

How loud it reached this AI, 0 to 1: after distance, walls and rain. What hearing works with.

How far this sound carries to this AI (cm): the Sound Type's range, scaled where the AI stands.

From the sound to this AI (cm).

Walls between the sound and this AI's ear. 0 when the path is clear.

True when nothing stood between the sound and the AI's ears; false when walls muffled it (see Walls).

A noise with no maker: it becomes a belief about the place, not about anyone.

World seconds when it was heard.

This entry's id within its faction: pass it to Get Entry or Forget Entry.

The actor this is about, when anyone has actually identified one. Null for a description nobody has tied to a person yet — which is the normal state of a manhunt before it succeeds.

What is known about them.

Where the subject was last reported.

How strongly the faction believes the entry, 0 to 1: it fades while nobody reports it again, and the entry is dropped below the subsystem's Entry Forget Threshold.

Seconds since this was last refreshed by anyone.

Who said so. Free-form: an agent's name, "Dispatch", "Camera_03".

How many retellings from the original sighting. 0 means witnessed.

What this is, in your vocabulary. "AttackedMe", "OwesMeMoney", "SeenWith".

A number, meaning whatever you decide. Damage taken, coins owed, a rating.

A place, if the memory has one. Where it happened, where they ran to.

Another actor, if the memory links two things. This is what makes "seen with an accomplice" work without a special case for relationships.

How many times this tag has been remembered about this subject. Maintained by APS because counting is bookkeeping, not meaning — you still decide whether three is a lot.

Seconds since this was last written.

The actor the memory is about, when it is about an actor.

Where the remembered place is, when it is about a place.

What kind of place, when this is about a location rather than an actor.

True when the memory is about a place (named by Place Tag) rather than an actor.

World seconds at which this was captured.

Who the belief was about. May have been destroyed since.

Confidence as acted on, after smoothing.

Confidence as the senses reported it, before smoothing.

The belief's state at that moment.

The AI's overall awareness at that moment.

Where the AI believed the target was.

Seconds since any sense last had this target.

In APS 3.0, how many senses the AI runs (not how many were reporting).

The setting's override name, shown on the profile reference: for example VisionMaxRange, or bEnablePeripheralCone for a tickbox.

The value, for a number setting (a choice uses its position in the list, from 0).

The value, for a tickbox.

The actor this state is about.

The AI's confidence in the target, 0 to 1.

The belief's state: Undetected through Expired.

How much of a threat the AI thinks the target is.

Where the AI last placed the target.

Where to look: a reachable point on the navmesh.

How likely the target is here, relative to the other points (0 to 1).

Distance from the searcher. Exposed so a caller can trade "most likely" against "nearest" itself rather than accepting a ranking it did not choose.

How far vision reaches right now, in cm.

Half the focal cone's width (degrees).

0 when the peripheral cone is off.

Half the width of the peripheral cone, in degrees.

0 when the rear-motion cone is off.

Half the width of the rear motion cone, in degrees, measured from directly behind the AI.

Hearing's cap, scaled by any quiet or loud zone the AI stands in. A Sound Type can carry less far (Get Sound Reach).

How far smell reaches right now, in cm. 0 when the AI has no smell.

How far vibration reaches right now, in cm. 0 when the AI has no vibration sense.

How far echolocation reaches right now, in cm. 0 when the AI has no echolocation.

Echolocation pings forward only, about 101 degrees to each side.

"Vision", "Hearing", or whatever a custom sense returns from Get Sense ID.

What this sense reported, before weighting.

The profile's weight for this sense.

False when the sense is silent — it should contribute nothing.

A dotted name for the signal. Senses listen for a beginning, so one listening for Stimulus.Custom gets Stimulus.Custom.Psychic.

World origin of the stimulus.

The actor that caused the signal; none for things like weather.

How strong, 0 to 1.

AI further than this (cm) don't receive it.

AI at a higher (cheaper) level of detail tier don't receive it: 4 reaches every AI.

Trace from the signal to each AI and skip those behind a wall. Costs a trace per AI: only for signals walls should stop.

Skip AI on the Source's team.

What kind of fact this is — "Clothing", "Vehicle", "Build", "Voice", "Face".

The value for this observer to match on — "RedCoat", "BlueSedan", "Tall".

How easily this is noticed (0 to 1). 1 is unmissable at any range a target can be seen at; 0.1 needs a close, clear look.

Suggested: clothing colour 0.9, build 0.8, vehicle 0.85, gait 0.4, face 0.25, voice 0.3 (and only when in earshot).

Which cone this is: focal, peripheral or rear.

The fan's point: under the eyes on the floor, or at the eyes.

Flat direction the fan's angles are measured from.

The cone's range before anything stops it (cm).

The fan's edges, degrees about world up from Forward.

The angle, in degrees about world up from the cone's forward direction, where the cone's outline ends. The outline runs from From Deg to To Deg.

How far sight gets along the middle of the fan (cm): less than Range when something is in the way.

Where sight ends, in order from FromDeg to ToDeg: world points at the Apex's height.

Each rim point's angle from Forward (degrees).

How far sight gets to each rim point (cm).

How much a sighting in this cone counts: 1 for the focal cone, the profile's scale for the others.

Where the sample point is on the target.

How much of vision reaches it, 0 (blocked) to 1 (clear).

Whether vision gets through to the point: Seen (more than half gets through), Partly (glass, foliage or smoke lets some through, too little to count) or Blocked.

The point's weight in the exposure: how much of the target it stands for.

Mesh socket / bone to trace against. None = use Offset instead.

Local-space fallback offset from the actor root (cm), used when SocketName is None or the socket does not exist. Rotated by the target's facing, unlike a raw world offset.

How much this point counts, 0 to 1: a visible head counts for more than a visible foot.

The actor this belief is about.

A number that identifies the target.

Whether this belief is about an actor (someone), a thing (APS Evidence) or a place.

Where a belief about a place is. For a belief about an actor, read Last Known Position.

What kind of thing this is — "Sound.Gunshot", "Evidence.Body". Empty for actor beliefs, which identify themselves.

The senses' combined confidence this update, before smoothing.

The confidence the AI acts on, 0 to 1, after smoothing: the states, the events and the meter all use this one. Use it for detection meters.

Where the belief is in its life: Undetected, Suspected, Detected, Tracked, Lost, Remembered or Expired.

Seconds since any sense last had the target.

How long, in all, the target has been Tracked.

Seconds in the current state.

Where the AI last placed the target.

The target's velocity as the AI measured it, smoothed.

How the target's velocity has been changing.

How unsure the AI is of where the target is, in cm around the last known position: grows while it is unsensed.

Where the AI expects the target to be by now.

A slow memory of the target, 0 to 1: it builds by 0.02 a second while the target is sensed and fades by about 1% a second after (or as Long Term Decay Curve says). APS itself doesn't act on it: read it for your own logic.

How many times the target has been noticed: each rise out of Undetected, and each time it is found again after being Lost or Remembered.

True once the target has ever been noticed.

The world time, in seconds, when the target was first noticed.

What stood between the AI and the target when it was lost from Detected or Tracked — set before On AI Lost fires. Cleared on every loss.

Where the line of sight broke, on the face of that cover.

True when the target was behind cover at the moment it was lost.

Why the AI last lost this target (Occluded, OutOfRange, SoundFaded, ScentLost, SensorDropout, TargetDestroyed). Cleared when it is found again. Branch your search on it.

Which way the target was moving when it was last lost, from its last measured velocity. (0, 0, 0) when it wasn't moving.

The last five encounters with this target, oldest first: each written when the AI lost the target from Detected or Tracked. Read them with Get Recent Episodes.

The source that first made this belief.

The sense contributing most right now; while none reports, the last source the belief came from (Shared Intel for one a squadmate shared). This is what On AI Suspect / On AI Detect pass as Stimulus.

How much of a threat the AI thinks the target is: None, Low, Medium, High or Critical.

The threat score behind Threat Level, 0 to 1.

How much damage the target has done to this AI. Fades over time while the target is lost (the profile's Threat Damage Decay Rate).

Does nothing in APS 3.0. Never written by APS 3.0: always 0 (only Debug Reset Damage Tracking touches it, to set it to 0). Track damage the AI deals in your own code.

True once the target has been Tracked.

How many separate encounters the AI has had with the target: one for each rise from Undetected to Suspected.

The threat level before the most recent change: compare it with Threat Level to see which way the threat moved.

What the AI takes the target for (enemy, friend, neutral, ...), as its APS Relationship resolved it.

Fear, 0 to 1.

Aggression, 0 to 1.

Curiosity, 0 to 1.

Alertness, 0 to 1.

Panic, 0 to 1.

The emotion that leads now: Calm, Curious, Alert, Aggressive, Fearful or Panicked.

The outside fear set by Set Fear Input, 0 to 1.

The light level, 0 (dark) to 1 (full daylight). Vision sees less in the dark, unless a per-target light value overrides it.

Rain, 0 (dry) to 1 (downpour). Rain shortens vision, hearing and smell.

How hard the wind blows. Smell carries less the stronger it is, outdoors only.

The way the wind blows. Smell reaches further downwind.

True indoors: scent is stronger and the wind is ignored.

The time of day in hours, 0 to 24. Informational: APS's senses don't read it.

The AI doing the perceiving.

Where the AI is.

Which way the AI faces.

Where the AI's eyes are.

The length of this update, in seconds.

The light level where the AI is, 0 (dark) to 1 (full light): from the environment, 1 when there is none.

How much of its sight the rain leaves the AI: 1 minus the rain intensity, 1 when dry.

Where the target is.

How fast, and which way, the target moves.

The direction the AI's eyes face, with Eye Direction Mode and Eye Turn Rate applied: use it rather than the actor's rotation.

True when this AI is off the first player's screen (more than about 75° from where the camera points). Used by Offscreen Hearing Penalty.

The Quality Profile's cap on sample points: 0 for no cap.

Skip per-target light sampling and the sun-shadow trace, using the global ambient level instead.

How far hearing reaches here, from any APS Environment volume the AI stands in: 1 changes nothing.

How far smell reaches here, from any APS Environment volume the AI stands in: 1 changes nothing.

How this engagement ended.

State before going Lost (was AI Tracked or just Detected?).

World position of target when AI lost them.

Direction AI was facing when they lost the target. Useful for predicting blind spots.

Direction target was moving when lost (from EstimatedVelocity).

All the time this target has spent Tracked so far, in seconds, not just this encounter: it only grows.

The threat level at the moment the AI lost the target. Despite the name, not the highest of the encounter.

World time when this episode was recorded. Use for "was this recent?" checks.

This sense's confidence about the target, 0 to 1.

The signal before falloff and the other multipliers: for your own debugging.

Where this sense places the target. Imprecise senses (hearing, smell) place it roughly.

How precise that position is, 0 to 1.

Which sense this is (Vision, Hearing, ...).

True while this sense is perceiving the target. When false, the rest is ignored.

Vision only: how much of the target was visible, 0 (hidden) to 1 (fully in view), weighted by its sample points.

Number of sample points that had a clear line this evaluation. Vision only.

Which cone produced the detection. Vision only.

Why the sense reported nothing, when it reported nothing. Only Vision gives reasons (and pain or the reaction window, for any sense).

How lit vision judged the target, 0 (dark) to 1 (fully lit): Light Level Override when the target sets one, otherwise the sun-shadow trace or the ambient level. Vision only. -1 when vision stopped before judging light (outside the cones, out of range, hidden).

What this sense said the last time it actually ran. Senses run on their own intervals, so between their updates Is Active is false; this tells you whether it still has the target.

Share of the time, during encounters, the target spent crouched or prone.

Share of the time, during encounters, the target spent sprinting.

Share of the time, during encounters, the target spent walking.

Share of encounters that ended with the target behind the AI.

Share of encounters that ended with the target in front. It and Stealth Ratio add up to 1.

Up to 5 different places (each more than 2 m from the others) where the AI lost the target.

Does nothing in APS 3.0. Stays empty in APS 3.0: no node or APS code records into it.

Number of distinct engagements observed so far.

True once Engagement Count reaches the profile's Min Engagements Required: check it before your logic adapts.

Sense to impair. Select from the dropdown — for example Vision Sense, Hearing Sense.

How much this pain dulls the sense, 0 to 1: at full pain, 1 takes the sense to its floor.

Seconds after the sense stops reporting before the target can be lost. Brief cover (a pillar, a doorframe) doesn't drop it.

On: once the grace ends, straight to Lost (vision, echolocation). Off: confidence steps down through the states first (the other senses).

How fast belief fades once this sense has stopped, compared with normal (1): 3 fades three times as fast.