Chains and verdicts

Written from a static read of every shipped script; nothing here is guessed from playback. The result is a verdict per event script — can it be started directly, and on which map — which the event reference carries row by row.

1. What was read

Everything below is read from Radiata Stories (USA).iso, the way the engine reads it:

• Event scripts. Every script of every event file, through the datacenter’s event tables, each file’s outer Kods, each member unwrapped to its inner member 0 (Disk file 186): 859 scripts in disk file 186 and 34 in disk file 187, 893 in all, no failures.
• Map script tables. Every map’s nested Kods: 882 maps, 8,709 scripts. Member 0 is the map’s init script.
• NPC scripts. Disk file 184’s 280 schedule-selection scripts and every person’s schedule lists (file 185) and act scripts — 9,705 person scripts.
• Every character file. All 23,149 EVD payloads in the datacenter’s kinds 0–8 for all 1,205 entity ids, enumerated from the raw containers and cross-checked against an independent count of EVD\x01 magics in the decompressed bytes. The 13,159 not already in the person corpus were decompiled and walked too (§5g).
• A control-flow walk of every launching script for the gates (Gates §4), and of every set_person_schedule for who puts a person on which list.
• The map data: the RBAD exit tables and RLF event polygons of every map (How an event starts §4.3).
• Two engine passes over SetScriptNumber and its 43 call sites, the inventory command, the save image, CheckCondition and its handlers, RunReaction, ChangeMap, CheckLandEvent and CheckKickEvent.

The debug room is excluded. Map 994’s contact scripts 101–149 each launch one script of event file 802, and each of those starts one chapter opening; two more of its scripts launch four events directly. Left in, that file is the “predecessor” of 24 chapter openings and hides their real starters, so file 802 and map 994 are dropped from every count. One more script goes with them: 800/6, the room’s mission-flag tool — 102 raw Shift-JIS print_text lines that set the mission flags directly, started by the room’s script 197 on a button press and by nothing else. Together: 30 event scripts, 28 predecessor edges and 5 map launches. 829 field scripts remain.

2. The map an event runs on

load_background(slot=0, id=M) puts map M on screen at once — slot 0 is the displayed slot. load_background(slot>=1, id=M) followed by change_map swaps it in. A load with no later change_map is a preload for the successor (105 scripts do this), and a change_map after the scene’s dialogue has started is a hand-off: the map the next event will run on (337 scripts end that way). 750/1 is the shape that fooled a first version of this analysis: it loads 548 into slot 0 at line 10, plays the whole scene there, and only at line 1050 preloads 313 and changes to it.

Which map a script needs is therefore its own slot-0 load, or — when it has none — whatever its predecessor, its battle, or the person who launched it left on screen. The reference’s map column is that resolution; own_map is only the script’s own load.

3. The dependency classes

Three classes count as “broken if started mid-chain”, plus one found on the way:

Class	Detection	Scripts
No map of its own	No slot-0 load and no early swap before the first talk_rmf or set_character_position.	248
Cast placed by a predecessor	A character the script animates or moves (play_character_animation, rotate_character, a non-snap set_character_position) but never snaps itself. Excludes the abstraction 9900 (the default character) and character 1, Jack, whom the map’s own entry places — 750/1 animates him without a snap and runs cleanly standalone.	27
Preload consumed	A change_map with no load_background before it in the same script.	0 — no shipped script does this; every swap loads its own slot first
Obi or music only in the debug block	set_window_mode, select_music, play_music, stop_or_pause_music or fade_screen present only inside the top-level if_value(event_value=1009, not=0) { … } block — the debug build’s diagnostic path, skipped when flag 8000 is clear — and nowhere on the normal path.	141

The last class is cosmetic — the window strip and the BGM, not the scene — so the verdicts keep it separate. 500/13 runs its set_window_mode(arg=0x81) and BGM unconditionally after the debug block closes; 500/15 has the identical three commands only inside that block, so a chained run inherits the obi and the music from 13 and a cold 15 has neither.

Flags and event values a script reads are recorded but do not feed the verdicts.

4. The verdicts

Over the 829 field scripts:

Verdict	Scripts	Meaning
standalone	325	Loads its own map, places its own cast, nothing starts it from an event. Start it directly. (+6 that only lack the debug-block obi/music.)
chained but self-contained	141	An event starts it, but it loads its own map and places its own cast. Start it directly.
chained, only obi/music inherited	127	As above, but the window strip or BGM would be missing on a cold start. Fine for the scene; enter via the predecessor for a scored comparison.
map-triggered	124	Nothing in an event starts it; a map’s init or local script launches it. Needs that map loaded and, where the launch is gated, those flags.
generic (contact)	34	Launched by contact scripts on three or more maps; runs on whichever map it is used from.
post-battle, inherits	2	Nothing starts it but a battle; no map of its own; runs on the pre-battle field map with the map init skipped.
NPC-launched, inherits	4	Only a person’s act script starts it and it loads no map: needs the map that person is standing on when talked to.
mid-chain	17	Must be entered from a predecessor: no map of its own, or its cast is placed by the script before it.
unresolved root	49	No map of its own and no script anywhere names it (§6).

Disk file 187’s 34 in-battle scripts are a separate group (§7).

Sanity, against scenes this project has run and scored in its own player:

Script	Verdict	Matches practice
500/13	chained but self-contained, own map 6	ran standalone on map 6 for every board
500/15	chained, only obi inherited, own map 6	the harness needed --enter-via 13 for exactly the obi
500/2	chained, only obi inherited, own map 135	ran standalone; position matched the savestate
660/3	standalone, own map 393	ran standalone on 393
750/1	chained but self-contained, own map 548	ran standalone on 548
801/3	standalone, own map 20	ran standalone

4a. The 17 that must be entered from a predecessor

Script	Its map	Start from	Why
507/5	680	507/3	animates 4001, never placed here
511/2	940	511/1	animates 266
511/3	590	511/2	animates 266
511/4	501	511/3	animates 266
516/2	754	516/6	animates 3 (Ridley)
516/4	inherits 754	516/2	no map of its own
561/7	513	561/6	animates 12, 266
650/9	275	650/8	animates 0, 1062, 262
700/5	inherits 246	700/4	no map of its own
700/7	inherits 187	700/6	no map of its own
708/9	inherits 393	708/4	no map of its own
760/8	inherits 438	760/4	no map of its own
760/9	inherits 438	760/4	no map of its own
760/10	393	760/3	animates 10
760/11	393	760/10	animates 162 (Aphelion)
800/5	inherits 73 or 81	ten different callers	no map of its own; the two predecessor groups leave different maps — the one ambiguous case
811/6	inherits 111	811/3, 811/4, 811/5	no map of its own

516/2 → 750/1: 750/1 itself is self-contained (it loads 548), but its caller 516/2 needs 516/6 first.

4b. Post-battle events that inherit the pre-battle map

Script	Runs on	The battle is configured by	Note
513/9	508	513/8	story fight; 513/8 runs on 508 and hands the field back to it
660/4	438	660/3	the fight at the end of 660/3; 660/4 also has its own starters, so it is not map-dependent here
800/7	whichever field map the fight began on	800/3, from 80 configure_battle sites	the generic encounter result handler; 800/3 is itself launched by map triggers on 4 maps and loads no map, so 800/7 inherits whatever 800/3 was on

To view one of these, load the pre-battle map and start the script directly; do not expect the map-init script’s state, because the console skipped it too (How an event starts §5). The 87 post-battle events that load their own map need nothing.

4c. NPC-launched events without a map

420/1, 501/5, 506/1 and 701/1 load no map and are started only from a person’s act script, so they need the map that person is standing on. The other 71 NPC-launched events load their own map. Which person and list starts each one, and the flag tests around the launch, are in the reference’s NPC launched by and NPC gates rows: person 170, list 3, starts 429/1 under flag 4552 set & flag 4503 clear and 429/4 under flag 4552 clear & flag 4554 set — the “already talked to him” switch is flag 4552.

5. Worked chains

5a. 500/1 runs twice

500/1 is the new-game script, and it is entered twice in a normal playthrough, with one event value deciding what it does:

new game ->  500/1  (value 1008 == 0)
               |  plays movie 7, the opening FMV        <- if_value(1008, not=1) { ... }
               |  start_script_stacked(13)              <- inside that same block
               v
             500/13 -> 500/14 -> 500/15 -> 500/16 -> 500/17    (the prologue chain)
                                                        |  set_value(1008, 1)
                                                        |  start_script_stacked(1)
                                                        v
             500/1  again  (value 1008 == 1)
               |  the FMV block and the start of 500/13 are SKIPPED
               |  the story-flag initialisation runs (1793-1799 locks, 938-1579 fields)
               |  start_script_stacked(2)
               v
             500/2   (Coliseum, map 135)

• The switch is event value 1008, not a flag. 500/1 line 119: if_value(event_value=1008, not=1) { … } — the block runs only while 1008 is not 1. Inside it: play_movie(movie=7), the obi and BGM, Jack’s tunic and model, and start_script_stacked(mode=1, script_id=13). Everything after the block — the flag initialisation and the final start_script_stacked(2) at line 1265 — runs on both passes.
• 500/17 is the only shipped script that sets 1008 to 1 for this purpose. It plays out on map 160, preloads 384 into slot 1, fades out, sets 1008 = 1 at line 187 and re-enters 500/1 with start_script_stacked at 188. 500/1 then shows 384 as its own map and preloads 135 for 500/2.
• 1008 does not survive a save, and cannot: values 1000–1009 are the running slot’s own words, and 500/17’s write reaches 500/1 only because a stacked start runs the callee in the same slot (How an event starts §3).
• 500/1’s own debug block (if_value(1009, not=0)) also sets 1008 = 1, and it is the only place in 500/1 that sets flag 1000 “New Game Started”, the clock (06:55) and event value 3 = 500. A cold start that takes that block skips the FMV and the whole prologue and lands in 500/2.

For a viewer: a cold 500/1 with 1008 = 0 plays the FMV and then the entire prologue through 500/17 before reaching 500/2; a cold 500/1 with 1008 = 1 is the re-entry and goes straight to 500/2. 500/2 can be started directly.

5b. The prologue is one chain, in one slot

Every script of the map-6 cutscene ends with a start_script_stacked naming the next: script 13 args=[0xE], 14 args=[0xF], 15 args=[0x10]. Running 500/13 and letting the chain run reaches script 15 at player frame 4,619 and leaves it at 10,202. Two pieces of state the console has when it enters script 15 are inherited from 13 and never re-established: the obi — the only set_window_mode in 15 is inside the block the debug branch jumps over, while 13’s equivalent sits four bytes past its jump target and executes — and the BGM, for the same reason. A cold 500/15 correctly has no obi; the console’s letterbox there is script 13’s.

5c. Event file 427, the battle chain

map 591 init -> 427/1 (map 590)  -- fight --> 427/3 (loads 641)
             -- fight --> 427/4  -- fight --> 427/5 -> 427/6  -- fight --> 427/7

Every “fight” is a configure_battle naming the next script. 427/2 is a bare end_script and nothing names it. There is no NPC branch in 427; the talked-to-someone branch that chooses between two scripts is 429/1 versus 429/4 in person 170’s script.

5d. What the walk found on the way

• Five launches are dead code: a start_script after an end_script with no label and no marker — map 20 script 218 lines 50 and 78 (501/15, 501/11), map 626 script 100 line 8 (a stub), person 31 list 1 mode-D line 38, and 708/7 line 625. Map 20’s script 218 is started by the engine — it is the land/kick event of a door polygon — but always at its top, since the only mid-script entry is the schedule-percent walk (EVD format §4.2). So 501/11 and 501/15 are unresolved roots.
• Marker tables are not section entries. 1,893 map scripts carry one; all 2,100 entries point at 0xF0 marker words, the targets of seek_marker — the NPC time-of-day bands, 800/4’s debug-mode switch — not extra entry points.
• Cross-script writes exist. 508/6 tests flag 1980 one way and, after the battle setup 800/4 sets it from flags 844–845, the other way. A per-script walk sees a contradiction there, so contradictory paths are reported in gate_notes, never pruned.

5g. Every character file, walked

To be sure the person corpus was the whole of the “character event scripts”, every EVD in every entity archive was enumerated: kinds 0–8 of all 1,205 entity ids plus files 184 and 185, checked against an independent count of EVD\x01 magics. The two agree on every kind: 23,149 payloads.

Where	Payloads	What they are	Event launches
kind 7 (action data)	11,322	per list number: mode-4 (4,257) and mode-2 (5,511) act scripts, mode-D (871), mode-C (683), RMF1 (731)	115
kinds 0–5	10,815	animation-attached trigger scripts: a marker table, anim_frame_trigger labels, expressions, sounds, effects, sub-objects; 935 of them under if_flag(flag=8001, not=0)	0
files 184 / 185	278 / 734	selection scripts and schedule lists; they start act scripts, never an event	0
kinds 6, 8	0	FAS chains and algorithm leaves, no EVD	—

So the character event scripts are exactly the kind-7 mode-2 and mode-D scripts. Of the 13,159 payloads the person corpus did not have, only two launch events, both ungated and both from unlabelled entities: 2215’s kind-7 entry 1 mode-2 script 2 starts 800/4, and 2090’s entry 1 mode-D script starts 52/10, whose file is not one of the 107 event files.

6. The 49 unresolved roots

No start_script(mode=1, script_id=N) for these exists in any event script, map script or NPC script, with the current file carried across each whole script; no configure_battle names them; none is named by an engine SetScriptNumber constant:

427/2 500/11 501/11 501/13 501/15 501/18 502/21 504/3 506/19 507/4 509/3
509/4 601/9 650/11 650/12 650/13 651/7 651/8 750/13 751/11 753/4 753/5
753/6 753/7 758/7 758/8 800/1 800/2 800/8 801/1 805/1 806/1 807/3
807/4 807/9 809/5 809/8 809/16 809/18 809/21 809/27 809/45 809/46 809/61
810/16 810/17 810/18 810/19 810/20

What the engine scan says is possible: SetScriptNumber is called with a dynamic mode-1 script number from CPartySelectWindowManager::ReflectionSelectedMember, CEventChrInOutCmt::Update, CBattleMain::State_VoltyBreakStart, CBtlConversationRequest::StartShow and the debug tools. So these are engine-started from data — menus, party changes, battle results — or unused. Which is which is not established; the shape of most (set_flags(4740 off), set_window_mode, hero idle) is the field-interaction preamble, consistent with menu and system events.

6b. The whole no-starter list

§6 is the subset with no map of its own. 82 field scripts have no starter at all. They fall into five shapes:

Shape	Count	Scripts
Loads its own map — ordinary-looking content, playable as it stands (§4 calls most of them standalone)	35	406/1 418/1 418/4 500/6 501/17 502/16 504/8 504/9 504/11 504/12 600/2 601/2 601/6 601/8 601/9 602/16 605/5 606/5 660/5 660/8 703/5 704/6 704/7 708/6 708/7 801/2 801/3 803/1 804/1 804/2 806/1 807/8 811/1 811/5 811/7
One short talk, no map — about 117 lines, one talk_rmf, the same normalised body in 20 of 24	28	500/11 501/11 501/13 501/15 501/18 502/21 509/3 509/4 650/11 650/12 650/13 651/7 651/8 750/13 751/11 753/5 753/7 805/1 807/3 807/4 809/5 809/8 809/16 809/18 809/27 809/45 809/46 809/61
Empty stub — entry; end_script, nothing at all	14	427/2 504/3 506/19 507/4 753/6 758/7 758/8 801/1 809/21 810/16 810/17 810/18 810/19 810/20
Inventory or party setup, no map — runs of change_inventory and equipment writes (800/2 gives Jack his starting kit over 2,557 lines; 807/9 strips Camuse’s weapons)	4	753/4 800/2 800/8 807/9
Other, no map	1	800/1

The last three shapes are not cutscenes. The 35 that load a map are the ones worth a second look: any of them could be reached by one of the dynamic callers above. 801/3 is in this group and has been run and scored on map 20, so “no starter” means the launcher is not in the scripts, not that the script is dead.

7. Disk file 187: the in-battle scripts

The 34 scripts of the seven high-bit files are not field cutscenes. They wait on the battle-progress flag pair 531, drive battle_character_make_or_delete and battle_character_send_message, set party HP and run talk_rmf inside the fight — the scripted battles. Each file is one self-contained chain from its script 1 (7 chains, 27 followers), none loads a map, and no script in any corpus loads a high-bit file (load_script_file(force_high_bit=1) appears nowhere), so what starts them is the battle code, not the field: which call site is not established. They are in the reference, marked in-battle, and out of §4’s table.

8. Limits

• Gates are read from flag and event-value conditions and the decoded state tests only; the other: kinds are shown, not interpreted. An empty gate with a non-empty other: list is gated by a runtime test.
• The verdict for “cast placed by a predecessor” trusts that a character the script animates without snapping was placed by the previous event; it could also be a map NPC. 27 scripts carry it; the 17 above are the ones that also have a predecessor to point at.
• The 882 bare start_script(mode=1, …) in the map scripts — the init tails and the in-battle branches — target the current event file, which is game state; they are counted, not resolved. Three more are ambiguous.
• 19 set_person_schedule(character=default, …) in event scripts set the list of the script’s default character, which this pass does not resolve.
• Verdicts describe starting a script. They say nothing about whether a player reproduces it correctly once started.

9. Corrections to earlier notes

Recorded so the same readings are not repeated:

• The compare control byte of CheckCondition, and the decompiler’s names for condition bases 03, 0B, 0C, 0D and 10 (control flow §2).
• “Nothing reads the RBAD table” was wrong: CGame::ChangeMap in step1_01 does.
• RunReaction starts the mode-C and mode-D act scripts with script number 0, not a dynamic mode-1 script.
• The schedule-list reader dropped the lower slot of every aliased pair in a file-185 block; fixed, which recovered 20 lists.