Danger & suppression
Incoming fire can affect enemies without hitting them. Projectile near misses feed a general Danger system, while some enemies separately convert qualifying Danger events into aim flinch.
- Near-miss falloff
- 0.5R → R
- Aim flinch
- Target-specific
- Prediction
- Not retuned
Danger, flinch, and aim
Danger, aim flinch, and ordinary aiming interact, but they are not the same mechanism.
A numerical state built from qualifying dangerous events. AI can read its value, thresholds, recency, and response curves.
Some targeting profiles turn qualifying Danger events into directional aim displacement. Others have zero flinch gain.
Enemies configured to return fire can enter a dedicated high-Danger response before ordinary visual spotting is complete.
Prediction, corrective force, drag, and normal angular velocity continue through their own aiming path.
Which enemies have aim flinch?
Separate an enabled aiming response from proof that an enemy's attacks are actually disrupted.
No enemy-specific attack disruption is verified by these records alone.
An enemy needs a qualifying Danger event, a responding aiming controller, and an attack that uses the displaced aim. A nonzero aim-flinch setting proves only one part of that chain. These lists describe what the code currently establishes, not observed hit reactions, stagger, or interrupted attacks.
Ranged enemies with positive aim-flinch settings and an identified projectile weapon
These enemies have both a Danger sensor and an aiming profile with nonzero flinch gain. Their weapon resources are also directly or indirectly linked in the extracted attack data. This makes them candidates for Danger-driven firing-aim disturbance, but the complete event-to-shot path has not been confirmed for each enemy.
| Enemy | Weapon connection |
|---|---|
| Assault Raider | Linked projectile weapon |
| Brawler | Linked projectile weapon |
| Commissar | Linked projectile weapon |
| Marauder | Linked projectile weapon |
| MG Raider | Linked projectile weapon |
| Raider | Linked projectile weapon |
| Rocket Devastator | Direct projectile weapon |
| Rocket Raider | Linked projectile weapon |
| Scout Raider | Linked projectile weapon |
| Trooper | Linked projectile weapon; multiple variants |
| Elevated Overseer | Linked projectile weapon |
| Overseer | Linked projectile weapon |
Positive flinch settings with unresolved weapon ownership
These enemies also have positive aiming profiles, but the available resource associations cannot establish that their actual gun or mounted weapon uses the displaced aim.
| Enemy | What is missing |
|---|---|
| Heavy Devastator, Conflagration Devastator, Incendiary MG Devastator | Positive aiming profile; specific firing-controller connection not established here. |
| Scout Strider, Reinforced Scout Strider, Factory Strider, War Strider | Positive parent aiming profile; mounted weapon response not established. |
Enemies and weapons with zero ordinary aim-flinch gain
The listed aiming controllers cannot build ordinary positive Targeting aim flinch from zero. Their Danger sensing and other AI reactions may still function.
| Enemy or weapon | Finding |
|---|---|
| Gunship | Zero gain on directly matched aiming profile. |
| Hulk Bruiser, Hulk Obliterator, Hulk Scorcher | Zero gain on directly matched aiming profiles; other behaviors remain separate. |
| Annihilator Tank cannon | Zero gain on resolved relevant mounted aiming controllers. |
| Shredder Tank autocannons | Zero gain on resolved relevant mounted aiming controller. |
| Barrager Tank missile launcher | Zero gain on both resolved aiming layers. |
| Bile Titan | Zero gain on directly matched aiming profile; other Danger responses not ruled out. |
Terminid aim flinch is not established.
Terminids such as Warriors, Chargers, Hunters, and Bile Spewers can have nonzero flinch settings alongside Danger sensing. The available evidence does not show that this alters their melee attacks, spit trajectories, animations, or attack timing. Do not treat those settings as evidence that Terminids can be suppressed in the same way as a gunner.
Near-miss Danger
The confirmed projectile path uses trajectory proximity rather than requiring a hit.
- 01
A dangerous trajectory or event is evaluated
Projectile near misses are one confirmed source. The game measures the closest three-dimensional approach of the trajectory to the receiving enemy.
- 02
Near-miss strength is calculated
The proximity term is full inside half of the enemy's warning range, then falls linearly to zero at the edge of that range.
- 03
A Danger contribution is produced
The proximity term is combined with a compact source strength and an event-specific multiplier. Different event paths can therefore contribute different amounts.
- 04
Current Danger accumulates and decays
The value is capped by the target's authored maximum and falls back toward zero over time.
- 05
AI systems consume the result
AI can use the current value, threshold checks, recent-Danger checks, and response curves. A separate qualifying-event path can feed aim flinch, while enemies configured to return fire can enter a dedicated under-fire response.
Warning-range falloff
Let R be the receiving enemy's Danger warning range and d the closest three-dimensional distance between that enemy and the projectile trajectory.
Beyond the warning range: 0% proximity.
Distance is the closest three-dimensional pass of the projectile trajectory. The gradient shows only the proximity factor used to calculate Danger, not the total Danger gained or the amount of aim flinch.
The proximity term is 1 when d ≤ 0.5R. Between half range and full range it falls linearly as 2 × (1 − d/R), reaching zero at d = R.
The recovered calculation does not add an ordinary line-of-sight requirement to that closest-trajectory test. Geometry still matters because a wall or object can stop or redirect the projectile before its path passes near the enemy.
Stealth interaction
Danger affects enemy reactions independently of the sight, hearing, and weapon-accuracy systems.
A quiet shot can still create a near-miss stimulus.
Suppressing the sound of a weapon does not remove the projectile's Danger path. Sight, hearing, and Danger are separate inputs even though they can all influence what an enemy does next.
Near misses can contribute to directional awareness.
The Danger system can retain an attributed threat source and position independently of visual spotting. Controlled testing of the resulting response direction and precision has not been documented.
Danger can carry an attributed source without visual spotting.
The AI Danger bridge writes the event's source entity ID into a perception record and updates a cached source position. The recovered projectile path passes a live entity ID associated with the shot into that bridge. This supports entity-backed return-fire behavior without ordinary visual acquisition, although the exact projectile owner/instigator identity still needs a targeted runtime capture.
Danger does not provide continuous visual tracking.
Repeated qualifying Danger events can refresh the cached source position. A separate confirmed check can refresh that position when it identifies the same entity. This does not mean the enemy has continuously seen the Helldiver.
Breaking line of sight clears the visual channel, not the Danger state. Current Danger has its own decay and recent-event checks. See Stealth and detection and Alertness & target selection for the separate perception and target-selection paths.
Aim flinch
When an enemy has a nonzero flinch response, qualifying Danger applications can move the whole firing solution.
- Direction
- Held per episode
- Recovery
- Linear
- Maximum
- Target-specific
- Rapid repeats
- Attenuated
A new flinch episode selects a direction. Additional qualifying applications reuse it until the live flinch reaches zero.
The live flinch value decreases toward zero using the target's authored recovery rate.
Flinch gain has diminishing room as the live value approaches the target's authored maximum.
A separate live state reduces the effectiveness of very rapidly repeated qualifying applications.
A nonzero flinch setting does not establish that an enemy visibly flinches.
The aiming data contains nonzero flinch gains for some melee enemies, including Terminids. That alone does not prove that incoming Danger reaches a firing controller, disturbs a usable attack direction, plays a hit reaction, or interrupts a melee attack. Enemy-specific suppression effects remain unverified until the receiving and consuming paths are mapped.
Danger does not directly make the normal aiming controller sloppier.
The confirmed aiming behavior does not retune normal corrective force, drag, angular-speed limits, stored normal angular velocity, or prediction. Flinch is added as separate angular motion, and the unchanged controller then tries to correct the displaced aim.
| What you see | Best current explanation |
|---|---|
| Enemy aims ahead of a steadily moving Helldiver | Prediction lead |
| Aim crosses the target and corrects without incoming fire | Normal inertial overshoot |
| Shots scatter around a stable aim direction | Weapon dispersion |
| Incoming near misses push the whole aim direction aside | Danger-driven aim flinch, where enabled |
This separation explains the familiar push-and-swing behavior. Flinch first moves the actual aim away from the desired direction. The ordinary controller then accelerates back toward the target. As the flinch fades, the controller can carry the aim through the desired direction before settling.
Return fire
This response is limited to enemies configured to return fire and is separate from visible aim flinch.
| Situation | What the AI receives | Current interpretation |
|---|---|---|
| Near misses build Danger | Attributed Danger can retain the apparent source and a cached source position without ordinary direct visual confirmation. | The enemy can react to the shooter-side source without first completing normal visual spotting. |
| High Danger on an enemy configured to return fire | The response can begin from low awareness without completing normal visual spotting. | This is a dedicated under-fire response and is separate from aim flinch. |
| More qualifying fire arrives | Danger events refresh the cached source position. A separate confirmed check can also refresh the source position when it identifies the same entity. | Position knowledge can be refreshed without treating the source as an ordinary continuously seen target. |
| Nearby ally is already in the direct under-fire response | A low-state return-fire-capable peer can enter state 6 from the endangered ally's relay record. | The relay initially identifies the ally and its position; direct transfer of the original shooter's coordinates is not established. |
The direct under-fire response is separate from aim flinch.
Danger can build and aim flinch can occur without the direct under-fire response. That response is a separate high-Danger reaction available only to enemies configured to return fire.
Danger can retain an attributed source without establishing the same information as continuous visual targeting. In the relayed under-fire response, the initial relay identifies the endangered ally rather than the hostile shooter.
Other AI uses
The numerical Danger state has confirmed consumers that do not depend on aim flinch.
| Danger consumer | What is established |
|---|---|
| Current Danger | AI can read the accumulated numerical value. |
| Threshold checks | Callers can compare Danger against their own thresholds. |
| Recent-Danger checks | AI can ask whether Danger occurred within a caller-supplied time window. |
| Response curves | Some AI calculations use current Danger continuously rather than as a simple on/off state. |
| Direct under-fire response | Enemies configured to return fire can enter a dedicated response when attributed Danger reaches the high threshold. |
| Aim flinch | A separate Targeting response exists on profiles with nonzero flinch gain. |
The higher-level behavior names are still unresolved.
Code establishes numerical response curves and at least one high-and-recent Danger state transition, but it does not yet justify naming those outputs as panic, cover-seeking, attack cancellation, target switching, or another specific behavior.
RPM and suppression
Weapon fire rate alone is not enough to rank suppression performance.
Numerical Danger contribution and aim flinch are related but different outputs. A source can build Danger quickly without producing a proportionally larger single flinch, while a source with frequent qualifying applications may sustain flinch differently.
The game also attenuates very rapid repeated flinch applications. Until applications per shot, pellet, burst, or beam tick are mapped, a clean weapon-by-weapon suppression ranking would overstate what the code currently proves.
Unsupported claims
| Assumption | What the code supports instead |
|---|---|
| Suppression is a universal accuracy debuff. | Danger is a broader AI input. Aim flinch is only one optional response to qualifying Danger events. |
| If an enemy senses Danger, its aim can be flinched. | Danger sensing and aim-flinch gain are authored separately. Some Danger-sensitive resources have zero ordinary flinch gain. |
| More numerical Danger means a proportionally larger single flinch. | The positive flinch path uses a qualifying Danger application, not the accumulated numerical contribution as a simple flinch multiplier. |
| Twice the weapon RPM means twice the suppression. | Applications per projectile are not fully mapped, and rapid-repeat attenuation makes very frequent qualifying applications less efficient. |
| A miss while under fire proves suppression worked. | Prediction, normal inertial overshoot, dispersion, travel time, constraints, and high-level behavior can all produce misses independently. |
Open questions
| Open question | Why it remains open |
|---|---|
| How many qualifying flinch applications does one shot produce? | Projectile processing has more than one Danger path, so one projectile cannot yet be equated with one flinch event. |
| How do shotguns, beams, plasma, explosions, flame, and arc sources map into Danger? | Their event construction and application cadence have not all been mapped to the same standard. |
| What exact behavior do the continuous Danger response curves control? | The numerical use is established, but the downstream player-facing behavior has not been semantically resolved. |
| Does the projectile Danger source ID always resolve to the firing Helldiver? | The current projectile path passes a live source entity into Danger and surrounding code strongly points to the responsible/owner side, but a controlled runtime capture should verify the exact owner/instigator identity. |
| Does a relayed under-fire response later identify the original shooter? | The initial relay identifies the nearby endangered ally and its position. A later transfer of the shooter identity has not been proven. |
Evidence standard
How the OTL separates code confirmation, controlled live testing, and behavior that remains unresolved.
How the OTL rates evidence
The OTL rates code evidence and live-game evidence separately. A mechanic can be clear in code but still need a controlled gameplay check for hidden state, ordering, or another runtime factor.
Live confirmation means controlled gameplay or instrumented runtime observation, not player impression alone.