Spatial triggers and objective decisions

These are original DBSIM consumers verified with controlled fixtures. They establish particular rules, not a complete mission simulation. Actor construction is covered separately in ACTORS.md.

Spatial areas and activation

SCRIPT areas are three signed shorts: kind, first point, and either second point or radius. Constructor 0x1cb48 resolves the point references. Kind zero is an axis-aligned rectangle; other kinds store the radius multiplied by ten.

0x1cb94 ignores altitude. Rectangle boundaries are strictly excluded, regardless of corner ordering. Radial areas use max(abs(dx), abs(dy)) + floor(min(abs(dx), abs(dy))/2) < radius. This is an octagonal distance approximation, not Euclidean distance. The browser displays its outline; the integer boundary has single-unit steps.

A runtime trigger is 16 bytes: audience at +0, retained field at +2, area count at +4, area-pointer array at +6, latched state at +10, counter index at +12 and counter operation at +14. The compiled record's eight area indices terminate at the first negative index. The last two shorts are counter/index actions, not a time delay. The meaning of the second short remains unassigned; these spatial consumers do not use it.

0x1ccb4 accepts entry into any listed area and latches permanently. 0x1cc5c acts only on the first latch: for a nonnegative counter index, operation 5 clears it and operation 6 increments its signed short. Other operation values do not alter the counter in this consumer. Increment wraps at 32767.

The per-frame scan 0x1f87c uses these audiences:

A waiting group has a non-null trigger pointer. Group dispatch precedes the spatial trigger scan. On a later frame it clears a latched activation pointer; that branch does not also dispatch the group's controller immediately. Subsequent phases of that frame can observe the cleared pointer. Actor events can also latch triggers; the area scan is not their only source.

tools/check_triggers.py compares 1,841 original cases: area construction and boundaries, altitude independence, one-shot counter actions, repeated entry and audience filtering. Results are in work/research/missions/triggers.json.

Damage grades used by objectives

The target-actor and target-group helpers do not use identical thresholds. 0xee84 classifies an individual actor's virtual damage grade at 50, 128 and 192; flags +0x93 or +0xae override it with grade 4. 0xecc0 averages all members' virtual grades, including disabled members, and counts members with +0x93 or +0x9e. An entirely disabled group is grade 4. Otherwise its grade is:

• 3 when disabled fraction in Q10 is at least 500, or average damage is at least 193.
• 2 when disabled fraction is at least 250, or average damage is at least 129.
• 1 when average damage is at least 51.
• 0 otherwise.

The group helper assumes a nonempty group. No invented outcome is assigned to its division by zero for an empty one. The comparison fixtures supply damage grades through a synthetic virtual accessor; they do not claim to reproduce the component-damage calculation.

Outcome decision and timing

0xf4e7 first handles player status flags (+0x93 → code 2, +0x9e → code 3). A special group argument 6 can report code 6 once target damage exceeds grade 2, subject to safety. Automatic evaluation then checks two margins around the world's rectangular bounds: 170,000 native units for code 8, 60,000 for code 7. The bounds are strict and independent of altitude. Explicit request mode skips those two boundary checks.

Completion depends on mission kind: actor flags for kinds 0/3, route exhaustion for 1/2, target safety/activation for 4/5, target grade 4 for 6/8/9, and both target route exhaustion and safety for 7. Completed and safe gives code 9; completed but threatened gives 10. Incomplete and threatened gives 4. Otherwise the per-kind baseline table is [12,11,11,15,13,14,16,14,16,16]. These raw codes should not be renamed from guesswork; shell messages and return actions determine their presentation.

Safety helper 0xf744 excludes inactive, same-side and disabled/ignored actors, and checks recognition. A recognized nearby opponent makes the player unsafe below 80,000 native distance units, or below 100,000 when it is targeting that player. This uses the original fast three-dimensional distance estimate, not an exact spherical radius. Non-player groups have additional route/target tests.

0xf454 polls outcomes through two three-byte clock records. Their signed-short countdowns subtract the Q8 frame ratio, not milliseconds. Polling resets to 10,000; a newly pending terminal outcome also starts a 10,000 countdown. At ratio 256 this is 40 simulation updates (nominally four seconds), not ten seconds. Code 2 bypasses the pending delay when polled. The pending latch is retained; changing outcomes does not automatically restart it. The previous reported code suppresses repeat reporting.

tools/check_objectives.py compares 5,122 cases, including 3,400 clock updates. It runs original boundary, recognition, target, route, casualty, safety and decision routines; only actor damage-grade values are controlled inputs. Results are in work/research/missions/objectives.json. This fixture isolates player safety and ordinary actor targets. Component damage, AI decisions and full recorded replay have separate integrated comparisons in DAMAGE.md, CONTROLLERS.md and RUNTIME.md; this particular fixture does not exhaust every non-player safety route.

Compiled header and handoff options

GO 0x3445b writes ten shorts. DBSIM 0x1d850 reads them at 0xbc49c. The native reconstruction's former name difficulty for word 2 was incorrect: it comes from GO 0xaad56, the instant-action/campaign selector (0/1). The actual difficulty is word 7 (0xbc4aa). Word 3 is the objective kind used by 0xf43c; word 4 remains the retained objective variant. Words 5 and 6 control unlimited weapon ammunition/power and player body-damage immunity. Both require campaign mode zero and the option exactly one. Unlimited weapon power is a weapon-bank option; it is not restricted to the player at that helper.

Word 8 selects normal (0) or low (1) visibility, also selecting the 60,000/30,000 perception range. Word 9 selects day (0), dusk (1) or night (2), through the palette variant. These labels match the manual's single-mission options. Terrain and zone occupy words 0 and 1. Campaign mode forces the damage multiplier's index to three; instant action uses word 7. The shell's campaign-launch branch explicitly clears the two cheat options before writing.

The 200-byte MISSION.VAR handoff contains 100 signed shorts. Immediately after reading it, the loader clears short 7 (0xbc4c6), retaining all other counters. The four supplied tapes have nonzero campaign/randomization counters even though their campaign-mode word is zero. The native session now retains this state.

Group/trigger/perception integration

capture_mission_activity.py constructs all four recorded worlds and captures 128 successive group-dispatch, spatial-trigger and perception passes with varied positions and actor conditions. Controller bodies, speech and sight results are explicit boundaries. The C++ mission activity owner matches all initial group, actor/controller, trigger, counter and target links, then every callback, contact byte, clock, latch and random word after each pass.

The runtime actor kind at +0x162 is HERC=0, base=1, flyer=2. It differs from the compiled group kind HERC=0, flyer=1, base=2. Separate C++ enums prevent target priority from accidentally interchanging static and flying actors.

HERC +0x91 is initialized from compiled record word 0, not its later skill word. The sleeping decision at 0x178a4 keeps the actor's detectable flag set when this byte is nonzero even without a current contact. It is represented as a persistent-signature flag. Player-roster counter actions default to zero, whereas NPC counter actions come from their source records.

Destruction notifications

Actor notification 0xd77c first calls group notification 0x1d524. A group is defeated if every other member is dead or disabled; the notified actor is skipped regardless of its current flags. Group record trailing words 1/2 (compiled offsets 78/80, runtime +36/+38) are the counter index and operation. Individual actors use +0x174/+0x176. Operation 1 clears, 2 increments with signed-word wrapping, and 13–16 assign 1–4. Other codes do nothing, and a negative index skips the action. These notifications are not latched: repeating one repeats its counter action. Group-before-actor order matters when they share a counter.

capture_destruction_counters.py compares 2,160 original notifications against the native mission owner, varying live/dead/disabled members, recognized/ignored operations, repeated calls, shared slots and overflow. Base/flyer death then activates +0x170; flyers additionally enter their dead controller. Ordinary hostile projectile hits activate the distinct +0x16c contact trigger, unless the actor's +0x9c event-busy flag suppresses it.

Integrated player progress and outcomes

The native session now uses the verified objective and debrief consumers. capture_player_objectives.py retains 4,096 calls to original 0x15f80. Player route advance stops before the duplicate first point of a closed route (unlike AI following). Reconnaissance announces its selected objective once; defence arrival uses horizontal distance below 40,000. Rescue requires an active living target within 10,000 in three dimensions and the original non-wrapping torso cone after signed-word heading subtraction. Losing it resets the dialogue. The four delay longs at 0x91ce8 are 5,000, 5,000, -25,536, 0; the negative third value makes the final acknowledgement eligible on the next update.

capture_objective_worlds.py adds 2,048 varied eight-actor worlds. All actor/group safety, target damage grades, target route completion, status and debrief grades match. A non-player objective can remain unsafe even at long range and without recognition if a hostile group targeting it still has a route to follow. Dead members are skipped in group safety; disabled living members are not.

The whole-frame harness now includes the shared projectile pool (0xaf0b4, virtual +0x18) and collection 0x6db0. These were inadvertently omitted from its earlier navigation capture. A seeded 880-frame comparison checks all prior body states plus objective status, debrief grade, polling/confirmation clocks and rescue progress. It matches 16,720 HERC/flyer observations. UI prompts are recorded rather than displayed, and player input remains zero.

Authored-world integration findings

The mission fidelity investigation records the subsequent whole-world checks: authored activation stays intact, ground-object updates are not suspended by that gate, starting damage precedes group ownership, and missing source-map entries resolve to actor pool zero. It also separates repaired original formation/queue defects from unchanged-reference comparisons.