The Nintendo Gigaleak preserves a very large Yoshi’s Island archive under other/SFC/ソースデータ/ヨッシーアイランド.

This is the Super Famicom codebase for Super Mario World 2: Yoshi’s Island, and it survives in a much richer form than a simple asm drop. The leak keeps three Japanese version branches, a full SFC build tree, a separate chip audio-side build, huge piles of background and ending assets, message-conversion tools, debug notes, and a workstation-side converter farm under ツール.

---

At a Glance

This archive is one of the broader SNES game folders in the Gigaleak. It preserves both the game-side 65C816 source and the PC-side asset pipeline that fed into it.

The quickest way to read it is like this:

• 日本_Ver0, 日本_Ver1, and 日本_Ver2 are not tiny patch folders - each one is a large full project snapshot
• 日本_Ver2 looks like the most useful late branch because it keeps linked outputs like .hex, .map, .sym, .lnk, ROM split binaries, and build scripts
• the sfc folder is only one piece of the project - graphics, messages, ending data, panel data, and map resources live in parallel folders beside it
• ツール is a real workstation-side production tree with dozens of Yoshi-specific converters, X11-era helper programs, and bug-tracking notes
• the source still carries region and bug flags for JPN, NES, and PAL, so this Japanese branch was clearly being used to manage overseas revisions too

At the top level, the four big branches look like this:

Branch	File count	What it mainly looks like
日本_Ver0	4,715	Earliest of the three large Japanese snapshots, already a full project tree
日本_Ver1	4,396	Slightly leaner branch with the same broad structure and a backup mk.BAK
日本_Ver2	4,257	Most useful late snapshot, with full SFC build outputs, docs, bug notes, and ROM split binaries
ツール	757	Shared workstation-side converter and utility tree used across the project

---

Root Directory (SFC.7z/ソースデータ/ヨッシーアイランド)

At the top level, the archive is split into three numbered Japanese branches and one big shared tool area.

The important thing here is scale. Nintendo was not just archiving a small asm folder. Each version branch already contains the game program, data build folders, screen resources, messages, ending assets, and support code as one coordinated workspace.

---

How Complete This Looks

This archive looks much closer to a working internal source snapshot than a token leak sample.

The strongest signs are:

• the 日本_Ver2/sfc tree keeps linked outputs such as .hex, .map, .sym, .lnk, plus split ROM binaries like ys_rom_0.bin through ys_rom_3.bin
• the top-level project makefiles do not just build one folder - they recurse through chip, sfc, and union
• the version branches preserve editable assets and source-side resources beside the asm, including msg, bgdt, char, endt, pnl, scr_file, and mpdt
• the shared ツール directory keeps the workstation-side converters that prepared background, message, object, ending, and map data

The main caveat is the same one that applies to several Gigaleak projects: this looks like a near-complete production snapshot, but not a perfectly self-contained rebuild package. The Nintendo assembler, linker, and helper executables the makefiles expect are not all preserved inside the branch itself.

So the safest description is that this is a near-complete Yoshi’s Island production archive with a lot of the real day-to-day pipeline still attached.

---

Version Snapshot Comparison

The three numbered Japanese branches are close enough in layout that they look like consecutive internal snapshots rather than unrelated forks.

Branch	Shared folders present	What stands out
日本_Ver0	sfc, chip, union, char, bgdt, msg, endt, pnl, scr_file, sound, course	Broad early project tree with the full workspace already in place
日本_Ver1	Same broad structure as Ver0	Adds mk.BAK, suggesting active script iteration rather than a frozen export
日本_Ver2	Same broad structure as Ver1	Keeps the most obvious build outputs, debug docs, bug notes, and split ROM binaries

That pattern matters because it suggests the numbered folders are whole working states of the project, not just tiny diff packs. 日本_Ver2 is still the best place to start, because it preserves the clearest evidence of how the build was actually being driven near the end.

---

The Main Ver2 Branch

日本_Ver2 is the branch that feels most like a live late-stage production snapshot. It keeps the main SFC program, the separate chip program, shared headers in union, and the giant piles of game content that those binaries were built from.

The file counts inside 日本_Ver2 make the shape of the project clearer:

Folder	File count	Why it matters
char	1,014	Character and graphics production was one of the biggest parts of the project
bgdt	994	Background/status/screen data was another huge pipeline in its own right
endt	965	Ending and event sequences had their own large authored data set
msg	299	Dialogue and event text had a full conversion workflow, not just a few script files
course	190	Course and world parameter data lived outside the main asm tree
sfc	189	The actual SNES program source is only one layer of the total project

That is one of the most useful things about this leak. It makes Yoshi’s Island look like what it really was to the developers: a large authored data project wrapped around one assembly core.

---

The Chip and Union Support Folders

The top-level Makefile does not build sfc on its own. It recurses through chip, sfc, and union, which means those folders were part of the normal build, not optional extras.

chip looks like the companion program-side workspace. It keeps files such as ys_chip0.asm through ys_chip7.asm, the matching ys_chip0.hex and ys_chip0.map, plus music and message-side modules like ys_mplay.asm, ys_mplay0.asm, ys_mplay3.asm, ys_mpldt.asm, ys_msge0.asm, and ys_msub0.asm.

Even without reverse engineering every file, that tells us something important about the project layout: the main game program, the companion chip-side code, and their shared headers were being managed as separate but coordinated build targets.

union is the glue layer between them. It keeps shared headers and RAM definitions such as mctbl.h, msg.h, ys_bgcheck.h, ys_mapobj.h, ys_mram.asm, and ys_ram.asm.

That makes union the place where game-side logic, shared data tables, and memory layout rules were being kept consistent across the rest of the project.

---

The Message Pipeline

The msg folder is one of the easiest places in the archive to see a full authored-data workflow from start to finish.

At a glance, the folder keeps:

Type	Count	What it looks like
.dat	96	Message source or intermediate data files
.cmt	84	Companion text/comment-style input files
.OUT	87	Converted output ready to be pulled into the asm side
.c	4	Converter source code
No extension	10	Helper binaries and small support files

The file naming is grouped by content family rather than by one flat script list. There are boss and miniboss groups like bos14, bos18, bos24, bos68, baby or bonus groups like bonb0 through bonb5, Kamek or Koopa-flavored groups like kupa0 through kupa5, ending sets like end0 and end1, and many numbered n_msgb... files that look like map-point or stage-event messages.

What makes the folder especially nice is that many files survive in threes:

• a .cmt input file
• a .dat file
• a .dat.OUT converted output file

That makes the production flow much easier to infer. This was not just a folder of final compiled dialogue blobs. It preserves authored message content, an intermediate message representation, and the converted output that the asm side would actually consume.

The converter source confirms that. msgcnv.c, msgcnvO.c, and msgcnvS.c are all by S. Takahata, and each one reads a message input file plus a shared msg.LBL label table before writing a matching .OUT file.

The three converter variants also suggest different script dialects rather than one universal parser:

Tool	What stands out
msgcnv.c	Base message converter that emits BYTE lines from the label table and handles spacing/newline markers
msgcnvO.c	Variant that understands extra embedded argument values, including two-digit hex payloads after special control labels
msgcnvS.c	Similar to msgcnvO.c, but specialized around another control-code family

That distinction matters because it shows the message system was not only about text characters. The scripts were also carrying inline control values for layout or behavior.

msg.LBL is the bridge between the human-authored side and the assembler side. Even in raw form it is clearly a character/control lookup table, with romanized label names for kana and other symbols.

The checker tools reveal another layer of polish. ys_msgchk.c is described as a Yoshi message-data editing/check program by N. Okajima dated 20 July 1995. It reads a .dat file, measures character widths through ys_msgchk.h, inserts line and page markers such as @05, @06, @0E, @0F, and @FF, and writes a reflowed ch_*.dat output.

That means the message pipeline was doing more than raw encoding. It was also validating how dialogue would fit inside the actual message boxes on screen.

There are even small editorial traces left behind. msg.LBL~, msgcnv.c~, msgcnvO.c~, and msgcnvS.c~ preserve nearby backup copies, while the BOS_OLD subfolder keeps older boss message data like bos24.dat, bos28.dat, and bos68.dat.

So the msg folder is not just “where the text lives”. It is a compact record of how Yoshi’s Island scripts were authored, encoded, line-checked, versioned, and turned into build-ready resources.

---

The Auxiliary Chip Program

The chip folder deserves to be treated as its own program, not just as a few support files beside sfc.

Its Makefile builds a separate "yossy" M-chip program using a different toolchain from the main SNES code:

Step	Tool	What it does
Assemble	sasm	Assembles .asm files into .pok
Link	slnk	Links the .pok files into ys_chip0.hex
Map postprocess	emln	Uses ../union/mctbl.h with the generated map
Copy to main build	cp	Copies the resulting hex into ../sfc

The output layout is also banked separately from the main game, with group ranges like Group_8, Group_9, Group_A, Group_B, and Group_F. So this is not just another naming convention inside the same program. It is a distinct auxiliary processor-side image that feeds into the final cartridge build.

The file mix reinforces that:

Type	Count	What it suggests
.asm	16	Real source modules survive
No extension	9	Small helper scripts and local command wrappers were kept in place
.inc	2	Shared include files survive
.hex	1	Built chip image survives
.map	1	Linker map survives
.CGX	1	Graphics-side companion data was still tied into this area

ys_chip0.asm makes the role of the folder even clearer. It calls itself the "mario chip program", defines its own groups and sections, exports math and wave symbols like COS, SIN, LCOS, LSIN, and FURA_WAVE_MAP, and contains a substantial DT_UNCOMPRESS routine for unpacking data.

That is especially interesting because it means the chip side was not only doing sound or a tiny stub of glue code. It was handling decompression and math-heavy utility work that the rest of the project depended on.

ys_mplay.asm then shows the other half of the picture. It calls itself the "mario chip player program" and exports a huge amount of player, collision, movement, and object-handling logic. The extern tables it references include:

• player status and movement data
• enemy and extra-object mode tables
• water, soap, snow, mud, and surface-state logic
• animation patterns, walking sound data, jump states, and object-spawn helpers

So the chip folder looks much less like a narrow audio-only sidecar than a broad support processor program handling shared movement, helper logic, decompression, and related subsystem work.

One particularly nice production detail is that the chip makefile copies its final ys_chip0.hex back into ../sfc. That makes the relationship between the two folders explicit: the main game build and the chip-side build were separate, but they were still part of one coordinated cartridge-generation workflow.

---

The Main SFC Build

The 日本_Ver2/sfc/Makefile lays the main build process out very clearly.

Step	Tool or target	What it does
Assemble	as65c	Assembles every .asm file into .rel
Link	link	Links the huge module list into ys_main.lnk
Load	load	Produces ys_main.hex
Symbols	insp -f -s	Emits ys_main.sym
Character map	taraka yschr	Generates yschr.map
Final map merge	azrael	Combines ys_main.map, ys_chip0.map, and yschr.map
ROM split	hex2bin	Produces either ys_rom_16 or four-way split ys_rom binaries

The makefile also shows that this branch was managing multiple regional and bug-fixed code paths from one source base. The VERSION flags set NES=0, JPN=1, PAL=0, while also enabling several bug toggles such as BUG00, BUG02, BUG03, BUG10, and BUG20.

That makes 日本_Ver2 more revealing than a plain Japanese retail tree. The source still contains conditional code for Japanese, NES/US, and PAL behavior, so this branch was clearly carrying localisation and bug-fix work rather than being a sealed-off domestic-only snapshot.

The module list is also enormous. Besides the core program files like ys_init.asm, ys_game.asm, ys_play.asm, and ys_map.asm, the build pulls in:

• fourteen numbered enemy modules from ys_enmy.asm through ys_enmy14.asm
• boss and special-character files such as ys_boss1.asm, ys_boss2.asm, ys_bbbros.asm, ys_koopa.asm, ys_dorobo.asm, and ys_babym.asm
• world modules from ys_w11.asm all the way through ys_w70.asm
• message, map, player-data, and object data modules such as ys_msgdt.asm, ys_mapdt.asm, ys_pldt.asm, ys_mpobj.asm, and ys_mpmv.asm

That is one of the clearest signs that the leak is preserving the actual shipped game program rather than a stripped teaching example.

---

How the World Modules Are Structured

The numbered world files like ys_w11.asm, ys_w48.asm, and the rest of the ys_w11 through ys_w70 set are some of the most revealing files in the whole archive.

At first glance they look like ordinary asm modules, but they are really hybrid content containers. Each one pulls several parts of the production pipeline together in one place:

• raw BGDT_xx_xx background data blocks converted from named .dat files
• door and room-transition metadata that points into ys_mapdt course tables such as CORS_10_00 or CORS_47_03
• packed enemy/event placement tables under ENDT_xx_xx
• player spawn positions and door-mode flags such as PO_NORMAL, PO_DK_O_D, and PO_DK_I_U

ys_w11.asm shows the pattern very cleanly. It identifies itself as a "BG data file", then embeds several converted blocks from named source files like start.dat, w1ren00.dat, saish00e.dat, and saish01d.dat. After the large HEX data regions, it adds explicit next-room metadata such as door position, course-table pointer, player X/Y position, and player mode. Then, lower down, it includes enemy/event records under ENDT_00_01 using entries like MSGBX, JBLK0, HEIH0, and YAJI0.

That combination is historically useful because it shows the stage modules were not only “background maps”. They were junction points where visual layout, room transitions, and entity placement all met.

ys_w48.asm adds another important wrinkle: bug-flagged and regional variation inside a world file. Near the top it has an IF BUG01 block that switches the source data from a Japanese kamee00b.dat version to a bug-fixed kam00b-n.dat variant dated 13 July 1995.

That is one of the best examples on the whole page of how the branch was really being used. Nintendo was not just toggling bug fixes in the high-level engine files. Specific stage data modules were also being swapped and rebuilt behind BUG and regional conditionals.

So the ys_wXX.asm family is worth reading as the point where authored data stopped being abstract folders and became concrete game-stage modules inside the final build.

---

A Full Stage Trace: World 4-8

One of the best ways to understand the whole archive is to follow one stage across multiple folders. World 4-8 works especially well because its naming survives cleanly in several places at once.

On the program side, ys_mapdt.asm defines:

• CORS_47_00 - marked as [4-8]
• CORS_47_01 - marked as [4-8]
• CORS_47_02 - marked as [4-8]
• CORS_47_03 - marked as [4-7]

Then ys_w48.asm pulls those course pointers directly into its door and transition metadata. That alone is already useful, because it shows stage flow was not inferred from one giant map blob. The world module explicitly points at course-table entries for its connected rooms and boss area.

The same stage also survives in the course workspace as:

• course/parmt/w_4/w48_1.prm
• course/parmt/w_4/w48_2.prm
• course/parmt/w_4/w48_3.prm

That is the clearest sign yet that one stage was not authored as one indivisible resource. At least for 4-8, the parameter side was split into multiple fragments that could be listed, backed up, and rebuilt independently.

The background and event data line up just as neatly. Inside ys_w48.asm, the main converted resource blocks are:

• kamee00b.dat or the bug-fixed kam00b-n.dat variant for BGDT_47_00
• kamee01a.dat for BGDT_47_01
• kmbosss.dat for BGDT_47_02
• kamee02.dat for BGDT_47_03

And those same names survive in the authored data folders:

File	Background size	Event size	What that suggests
kamee00b	1,342 bytes	373 bytes	Main area with substantial visual and placement data
kamee01a	791 bytes	394 bytes	Another large room/segment with lots of placements
kmbosss	399 bytes	23 bytes	Boss-focused segment with much heavier BG side than event side
kamee02	108 bytes	23 bytes	Very small transition or setup segment

That size pattern is useful. It suggests the stage was broken into at least four authored chunks with very different densities, likely mixing larger explorable areas with tiny boss or transition segments.

The art side also lines up. The same branch keeps Kamek- and 4-8-related assets such as:

• OBJ/KAME.OBX
• OBJ/KAMEKKU-MAGIC.OBJ
• OBJ/MAGIC-KAMEKKU-FAR.OBJ
• char/B3-KAME-BOSS.scr
• char/kameboss-NEW.CGX
• col/KAMEBOSS-NEW.COL
• char/ROM-47.CHRN and ROM-47-A.CHRN

So if you trace 4-8 through the leak, the chain looks something like this:

• stage parameter fragments under course/parmt/w_4
• named background and event resources under bgdt and endt
• matching Kamek and boss art in OBJ, char, and col
• final stage assembly in ys_w48.asm
• course-table references in ys_mapdt.asm

---

What Survived Beside the Source

The sfc folder is useful partly because it is messy. It does not only keep plaintext asm. It also keeps the kinds of outputs and scratch files that tend to vanish once a project is archived cleanly.

The file types in 日本_Ver2/sfc include:

Type	Count	What it suggests
.asm	109	Main source modules dominate the folder
No extension	45	Build helpers, scratch files, scripts, and utility leftovers were kept in place
.bin	6	Built ROM slices or binary assets survived
.hex	2	Linked outputs were preserved
.map	2	Linker maps survived
.lst	2	At least some assembly listings were kept
.sym	1	Symbol output survives
.lnk	1	Linked image survives
.cnf	1	Runtime or build configuration survived

The built artifacts make the folder much more grounded:

• ys_main.hex, ys_main.map, ys_main.sym, and ys_main.lnk show the main game image after link/load
• ys_chip0.hex and ys_chip0.map show that the separate chip side was also being built in parallel
• ys_rom_0.bin through ys_rom_3.bin and ys_rom_16_0.bin show two different ROM-splitting strategies from the same build products
• ys_date.inc is regenerated during clean, which means date stamping was part of the everyday build flow

The scratch filenames are also revealing. Files like run, run2, run3, to_rom, rom, rom2, list_out, objtype.out, and odd leftovers like ys_game_edit.asm, ys_mplay.lst, and ys_enmy3.lst make the folder feel lived in rather than polished for handoff.

---

Documentation and Debug Notes

The 日本_Ver2/doc folder is small, but it is one of the best places in the archive for understanding what the team was actively worrying about.

The strongest documents are these:

Document	What it reveals
debug_mode.doc	Explicit cheat and patch addresses for multiple debug modes in Japanese Ver2.0
v2_bug.doc	A tiny bug note naming a “multiply speed set bug” as a Ver2.0 fix
v2_chg.doc	Placeholder for Ver2.0 changes, but effectively empty in this snapshot
check_sum.doc	Build or verification-oriented documentation
ramdoc, ys_ram.asm, ys_mram.asm	The project was documenting RAM layout, not just code logic
ys_enmy.doc, enmy.doc, enmy1.doc	Enemy behavior and layout had their own supporting docs
ystool.doc	Tool-side documentation survived inside the main branch

debug_mode.doc is especially nice because it is so concrete. It describes several ways to unlock development features by patching instructions to EAH:

• choosing the second file can turn START and SELECT into a course-clear shortcut
• choosing the third file can mark the whole game cleared and unlock all items
• the second controller can then trigger a perfect demo, world-clear demo, or ending sequence
• a separate player and boss debug mode can be unlocked through button combinations
• there is also a system pause mode with frame advance on the second controller

That is exactly the sort of material that shows how an internal test branch was really used.

The bug notes are useful too, even though they are short. Inside ツール, next_chg.doc lists fixes planned for the next version, including a PAL save-data clear issue and a PAL-only rotating line-lift bug. not_chg.doc keeps a list of unresolved bugs, including disappearing flowers, collision issues, missing map objects, and corrupted graphics after Yoshi vanishes in 4W-1.

Together, those notes make the archive feel much closer to a late bug-tracking workspace than a sanitized code release.

---

The Course Parameter Workspace

The course folder is much smaller than bgdt or char, but it is more structured than it first appears. Instead of one flat pile of data files, it centers on course/parmt, which is split by world into w_1 through w_6.

Inside each world folder, the layout is highly regular:

• one wN.lis listing and a matching .BAK
• one all file and a matching .BAK
• a group of stage parameter files like w11_0.prm, w11_1.prm, w18_4.prm, w28_7.prm, or w58_6.prm

That strongly suggests the course tree is holding per-stage parameter chunks rather than monolithic map data. The numbering also maps neatly onto the world-module naming in sfc, which makes the relationship between course/parmt/w_4/w48_*.prm and ys_w48.asm hard to miss.

The variation in file counts is useful too. Some stages have only one or two .prm fragments, while others like w34, w28, or w58 have many. That makes the parameter files feel like segmented sub-areas or multiple parameter pages inside one stage rather than one fixed-size record per level.

Even without decoding every .prm format, the structure already tells us that stage design data in Yoshi’s Island was being broken into many small world-specific parameter units, then gathered into per-world listings and aggregate files during the build process.

---

The Ending Data Workspace

The endt folder is one of the most striking content folders in the branch. It contains 216 .end files and nothing else at the top level except a huge back_up directory.

That tells us two things immediately:

• ending and event resources were being authored as their own format, not hardcoded directly into the program
• the team cared enough about iteration here to keep a separate backup tree beside the live data

The filenames line up strongly with the game’s worlds, bosses, enemies, and scene concepts. Examples include:

• world-like names such as 2w3bank.end, 4w6bank.end, 6w7bank.end, and 7upzone.end
• boss or character-flavored names such as bossbro6.end, kemeboss.end, kmbosss.end, jugem00e.end, and kamee01a.end
• environmental or scene names like jungl00a.end, snow01c.end, sensui.end, dokan.end, and kisya00.end

The converter source in ys_endtcnv.c explains the basic structure. It reads an endt.TBL file, walks each named .end file, and emits ../sfc/ys_endt.asm. Each record is read as a fixed 7-byte chunk:

• 5 bytes of entity name
• 1 byte of X coordinate
• 1 byte of Y coordinate

The converter then writes each entry as:

• one ENDT_xx label per resource file
• a sequence of LWORD name|(x<<9)|(y<<17) lines

That is a very clean clue about what the .end files really are. They look like compact authored event-placement tables, not raw prose, images, or generic blobs.

The file sizes back that up. Some entries are tiny, only 9, 16, or 23 bytes long, which suggests very small scene or marker lists. Others are much larger, running into the 500-700 byte range, which implies busier scripted scenes with many placement entries.

So the endt folder is best read as a structured event-scene workspace: dozens of named placement files, a converter that turns them into assembler tables, and a large back_up directory preserving the iteration history behind them.

---

What the Data Docs Show

The docs are also useful for understanding the non-code file formats.

One of the best examples is ツール/doc/bg_data.doc. It is not source code at all. It is a status table describing how the project classified BG layers and their color sets.

The document lists:

• transparency color values
• BG1 type and color categories such as underground, grass, underwater, lava, snow, jungle, wall, flower, pipe, and sky
• BG2 screen and color categories including perspective, night, bubble, forest, castle, jungle, and special mode-switching courses
• BG3 types for clouds, tornadoes, sandstorms, enemy overlays, jungle, snow mountain, and more

That is the kind of evidence that makes the surrounding converter programs much easier to read. The tools were not working on abstract “graphics”. They were converting assets against named world and layer categories that the artists and programmers both understood.

---

The Shared Tool Workspace

The ツール folder is not a small afterthought. It is a 757-file workstation-side support tree with compiled helpers, C source, documentation, and test directories.

The names inside bin and ctools reveal several distinct production pipelines rather than one monolithic converter.

Tool family	Examples	What it most likely handled
Background data	ys_bgdtcnv, ys_bgdtcnv1 … ys_bgdtcnv38, bgdtchg, ys_colcnv	Background/status data conversion and color handling
Map and point data	ys_mpdtcnv, ys_mpdtcnv1, ys_m7cnv, ys_m7scr, ys_md7scr	Map point data and Mode 7-style or map-screen conversion
Objects and enemies	ys_objcnv, ys_obtpcnv, ys_obzcnv, eobjcnv, ys_enset, ys_enhmk	Object, enemy, and type conversion
Messages and text	msgcnv, ys_msgchk, ys_msgcntchk, ys_ename	Message conversion, checking, and name-table generation
Screen and ending data	ys_scrcnv, ys_endtcnv, ys_msge_picture	Screen-layout and ending/event conversion
General workstation utilities	xgraphic.c, xbutton.c, xdirset.c, xwdcnv.c, xwinutil.c	X11-oriented editor or UI support code

ctools is especially interesting because it keeps both binaries and the C source for many of those programs. It is full of small standalone utilities such as bg3kage.c, course.c, koopa_scr.c, m7pars.c, scrchange.c, and many converter variants with incremental suffixes like ys_bgdtcnv21.c through ys_bgdtcnv38.c.

That naming pattern feels much more like active iteration than one tidy public SDK. The team was clearly cloning tools forward, testing new converter behavior, and keeping old revisions in place instead of replacing them cleanly.

---

The Sound Workspace

The sound folder is compact, but it makes the audio side of the project feel much more tangible.

At the top level it keeps a set of prebuilt binary banks rather than source modules:

• ys_sound.BIN
• area or context-specific files like ysgame0.bin, ysgame1.bin, ysboss.bin, ysbonus.bin, ysmap.bin, ystitle.bin, ysend.bin, and ysed.bin
• other themed banks such as yschika.bin, ysshiro.bin, yskuppa.bin, yslets.bin, ysse.bin, ysgm.bin, and ysor.bin

That suggests the sound workspace in this branch is preserving assembled or exported music and sound-effect banks rather than the music source language itself. In other words, the archive gives us the audio payloads the build used, but not the full composition-side authoring environment.

Even so, the filenames are still useful. They show that audio was being organized by gameplay context and scene type instead of as one generic sound blob. Boss, bonus, map, title, ending, cave, castle, and sound-effect banks were all being treated as distinct deliverables.

That fits well with the rest of the project structure. The Yoshi’s Island archive keeps many of the late-stage build products beside the source, and sound looks like the audio equivalent of that same pattern.

---

Screen and Panel Authoring Files

The scr_file and pnl folders fill in another part of the content pipeline that would be easy to miss if you only read the asm.

scr_file looks like a library of authored screen templates and test layouts. It contains named .SCR resources for:

• BG layer studies like B2-CHIKA.SCR, B2-MIZU.SCR, B3-KUMO.SCR, and B3-YUKI.SCR
• map or HUD experiments such as MAP-M7.SCR, MAP-bg3.SCR, Map-Small.SCR, score.SCR, and SCORE-B1.SCR
• mode or sample files like M7-KOOPA-SAMPLE.SCR, mode7ts.SCR, and Dr_Mario.SCR

That mix makes scr_file feel less like finished game content and more like a reusable screen-design and testing workspace.

pnl, by contrast, is extremely systematic. It contains a large numbered run of bg1-00 through bg1-A6 style .PNL files, plus helpers like ys_unit.TBL, unit.dat, yspn, yspn1, yspnmk, and yspnmk1.

That suggests pnl was the panel or tile-layout side of the background pipeline:

• many small numbered panel units
• a table file describing how those units were indexed or combined
• helper utilities for generating or packing the panel outputs

Taken together, scr_file, pnl, bgdt, and char make the visual pipeline look highly modular. Nintendo was working with reusable screens, panel sets, background-data blobs, and character resources that could be converted and recombined during the build.

---

The Character and Color Workspaces

The char, col, and col_file folders are where the visual production side really opens up.

char alone contains over a thousand files, and its extension spread is one of the clearest clues about how many intermediate formats the team was juggling:

Type	Count	What it most likely represents
.CHRN	334	Character-bank or tile-bank resources, often stage- or ROM-group-specific
.scr	153	Working screen-layout or preview files
.scrN	152	Closely paired alternate or normalized versions of the same screens
.SCR	137	Another screen-layout/export flavor, often kept beside lowercase variants
.CHR	149	Character graphics or tile data in another banked format
.CGX	50	Larger graphics blocks, often for whole art sets or grouped ranges
.CPC	20	Converted chip-side character data
.CH7	10	Mode 7 or chip-side character source banks

The naming makes the folder feel like a real graphics workshop rather than a final asset dump.

There are several broad families inside it:

• range-based banks like 0-7.CGX, 40-47.CGX, 90-97.CGX, and 11A-121.CGX
• stage or ROM slices like ROM-47.CHRN, ROM-47-A.CHRN, ROM-47-B.CHRN, and so on
• many BG-layer preview screens such as B2-JUNGLE.scr, B2-MIZU.SCR, B3-NIGHT.SCR, and B3-MORI.scr
• highly specific boss or setpiece resources like B3-KAME-BOSS.scr, B2-KOOPA-ROOM.scr, B3-BOSSROOM.scr, and kameboss-NEW.CGX

That variety matters because it shows the artists and tool authors were not working in one universal graphics format. They were keeping multiple banked, rotated, screen-level, and color-side representations of the same broad visual material.

The tool sources help explain why. chipchr.c is described as a chip OBJ rotation chara convert program. It reads a .CH7 file and produces a .CPC output, rearranging pixel data into the format needed for chip-side rotated object graphics.

That means pairs like:

• 28-2F.CH7 and 28-2F.CPC
• 80-87.CH7 and 80-87.CPC
• 11A-121.CH7 and 11A-121.CPC

are not redundant copies. They are source-and-converted representations for a specific graphics-processing path.

The color side is just as structured. The main col folder contains numbered banks like COL-0.COL through COL-34.COL, named palettes like KUSA.COL, SORA.COL, and KAMEBOSS-NEW.COL, plus helper files like yscol.TBL, yscol, yscol1, yscolmk, and yscolmk1.

col_file looks more like a design-side staging area. It keeps more descriptive names such as:

• KUSA-KAME-BOSS.COL
• KUSA-en-kin.COL
• MAP-KAITEN.COL
• score.COL
• soapland.COL
• v-ram-kabe.COL
• v-ram-room.COL

That split is useful. It suggests col_file held more human-meaningful source palette files, while col held the normalized or packed palette banks used by the build.

ys_colcnv.c confirms that conversion step. It reads a table file, opens named .COL palette files, extracts selected color ranges, writes a label table, and emits one packed binary color-data output.

The pipeline around char is easier to read if you break it into stages:

• screen and tile experiments in .SCR, .scr, and .scrN
• character-bank assets in .CHR, .CHRN, and .CGX
• specialized chip-side transforms through .CH7 to .CPC
• palette authoring in col_file
• palette packing and label generation into col

The World 4-8 chain makes this especially concrete. That stage links to:

• char/ROM-47.CHRN and ROM-47-A.CHRN
• char/B3-KAME-BOSS.scr
• char/kameboss-NEW.CGX
• col/KAMEBOSS-NEW.COL
• col_file/KUSA-KAME-BOSS.COL

---

The Object Sprite Pipeline

The OBJ folder fills in the other half of the visual side: moving sprites, enemy pieces, effect graphics, and object-bank assignments.

Its file mix is simple but telling:

Type	Count	What it suggests
.OBJ	110	Main object or sprite-layout resources
.OBX	9	Variant or companion object format used for some larger or special cases

The filenames make the folder feel very close to active game production rather than a cleaned technical dump. It contains direct references to enemies, bosses, effects, and special scenes such as:

• KAMEKKU-NEW.OBJ, KAMEKKU-MAGIC.OBJ, and MAGIC-KAMEKKU-NEW.OBJ
• BIG-HEIHO.OBJ, GHOST-HEIHO.OBJ, and Tubo-Heiho.OBJ
• JUGEMU02.OBJ, MIZU-JUGEMU02.OBJ, and ANA-JUGEMU-NEW.OBJ
• TORIBOSS-KIRARI.OBJ, OPENING-7-KAMECCK.OBJ, and W1-DEMO-YOSHIS.OBJ
• effect and reaction files like BAKUHATU-1.OBJ, MAGIC-KIRAKIRA.OBJ, MIZU-SHIBUKI.OBJ, and YUKI-CHIRI.OBJ

That tells us the folder is not limited to one object class. It mixes enemy art, boss-specific pieces, setpiece graphics, opening/demo assets, and transient effect graphics.

The converter sources explain how these files were being used.

eobjcnv.c takes one named .OBJ file plus X and Y offsets, reads a fixed 32 x 64 x 6 object-data structure, and emits eobj.dat as assembler-ready byte pairs for position, character number, attributes, and size.

That looks like a direct bridge from a design-side object layout file into stage- or enemy-placement data the asm side could consume.

ys_objcnv.c goes further and shows a whole table-driven object build. It reads object names from ysobj.TBL, opens corresponding .SPR files, and generates:

• ../sfc/ys_pld.h - pattern-name equates
• ../sfc/ys_pldt.asm - object count tables, pointer tables, and packed object data

That is a really important clue. At least one object-art pipeline was not just “manually copied into the game”. It was systematically generating assembly include files used by the main program.

ys_obtpcnv.c reveals another layer again. It reads an .ojd object-bank data file plus objtyp.dat, then emits objtyp.lis entries named relative to CHR10. That lines up very well with what survives in ys_init.asm, where OBJCHRBK is used to assign object character banks based on object type.

So the broad object pipeline seems to have looked something like this:

• authored sprite or object resources in OBJ
• table-driven conversion through tools like ys_objcnv.c
• generated object and pattern tables in ys_pld.h and ys_pldt.asm
• object-bank assignment through ys_obtpcnv.c and OBJCHRBK
• runtime palette selection through object color tables such as OBJ_MAIN_COL, OBJ_SUB_*_COL, KOPPA_OBJ_COL, and KOOPA_OBJ_COL in ys_color.h

That ties OBJ, char, and col together very neatly. They are three parts of the same sprite-production path: object layouts, graphics banks, and palette/bank selection.

---

The Background and Map Data Workspaces

Two of the biggest non-code folders in 日本_Ver2 are bgdt and mpdt, and the converter sources make their roles much easier to explain.

bgdt is the larger one by far. It contains 220 .dat files at the top level, plus a back_up directory. Many of the filenames closely mirror the ones in endt, which suggests that the two folders were often tracking parallel aspects of the same stages or scenes.

A few examples make the overlap obvious:

• 2w3bank.dat and 2w3bank.end
• 4w6bank.dat and 4w6bank.end
• bossbro6.dat and bossbro6.end
• jungl00a.dat and jungl00a.end
• sensui.dat and sensui.end

That does not mean the formats are the same. In fact, the size pattern suggests the opposite. The .dat files are usually much larger than the matching .end files, often by several times, which fits the idea that bgdt stores background/status/layout data while endt stores compact event-placement records.

The converter source ys_bgdtcnv.c confirms that the format is treated as raw binary background data. It takes one named .dat file, reads it byte-for-byte, and writes a bgdt.lis listing with:

• a generated timestamp header
• a global BGDT_00_00 label
• straight HEX lines for the data bytes
• a trailing FF separator

That is a very plain conversion step, and that plainness is useful. It suggests the heavy semantic work happened earlier in the editor pipeline, while ys_bgdtcnv.c served mainly as the final “turn this prepared background data into assembler-friendly bytes” stage.

mpdt is smaller, but more explicit about structure. It keeps .mpd and .pdt files like map01.mpd, now_map1.pdt, and w6-map-2.pdt, plus mpptdt.TBL and mpdobj.lis.

ys_mpdtcnv.c shows that the .pdt side stores map-point status data:

• map type
• parts pointer
• course data pointer
• timer
• player start position
• multiple “course clear” branches with next-point links
• arrays of per-state player movement data

The converter sorts records by world, rewrites next-point references, and emits labels like MPSTATUS, MPSTAT_xx, and MPCLR_xx_y.

So the broad data picture in Yoshi’s Island looks like this:

• bgdt holds large prepared background/status/layout data blobs
• endt holds compact event-placement records that become ys_endt.asm
• mpdt holds world-map or point-status progression data with explicit branching and player-position state

---

Why the Tool Workspace Matters

The most valuable thing about ツール is that it explains where the giant non-code folders in 日本_Ver2 came from.

Without ツール, folders like bgdt, endt, msg, pnl, and mpdt would mostly look like opaque game data. With the converters beside them, the archive reads much more clearly as a workstation-to-cartridge pipeline:

• authored background categories in docs like bg_data.doc
• editable resource files in folders such as msg, mpdt, and endt
• dedicated converters for those formats in bin and ctools
• final SNES program modules in sfc that consume the converted outputs

That is the main reason this archive is so valuable. It shows not just what the shipped code looked like, but how Nintendo’s staff were moving from authored content on the workstation to game-ready data on the cartridge.

---

The NEWS Tape Restores

The separate NEWS.7z archive adds another Yoshi’s Island angle under other/NEWS/テープリストア. Two of the restored tape tarballs, NEWS_09.tar and NEWS_11.tar, contain substantial Yoshi-related material.

They are not structured like the 日本_Ver2 source tree. Instead of one integrated project folder with sfc, chip, union, msg, and bgdt, these are personal workstation-style archives:

• NEWS_09 keeps its files under ./home/nogami/yoshi/
• NEWS_11 keeps its files under ./hino/yoshi/

That difference matters immediately. These tapes look like individual art and layout workspaces rather than whole-project source snapshots.

The broad comparison looks like this:

Archive	Main location	What it mostly contains	How it relates to 日本_Ver2
NEWS_09.tar	home/nogami/yoshi	Art assets, palettes, panel files, maps, many backups, and several VRAM-oriented subfolders	Overlaps the main char/OBJ/col material, but often preserves earlier or alternate versions
NEWS_11.tar	hino/yoshi	Much larger art workspace with CGX, OBJ, OBZ, SCR, SPR, test material, region variants, ending material, and many backups	Closer to the late branch than NEWS_09, but still much messier and richer in intermediate files than 日本_Ver2

---

What Makes Them Different

The first big difference is simply scope. The Yoshi folders inside the NEWS tapes are mostly art-side and layout-side workspaces. They do not contain the integrated program-side structure that makes 日本_Ver2 feel like a full production snapshot.

There is no equivalent of:

• sfc
• chip
• union
• msg
• bgdt
• endt

Instead, the tapes concentrate on the kinds of files that artists and content-side staff would have kept close at hand:

• .CGX
• .COL
• .OBJ
• .OBX or .OBZ
• .SCR
• .PNL
• .MAP
• many .BAK alternates

So the NEWS material is best read as a supplement to the source archive, not a replacement for it. 日本_Ver2 explains the build. The NEWS tapes explain more of the day-to-day artwork and iteration that fed into that build.

The extension mix makes that very clear:

Type	NEWS_09 count	NEWS_11 count	What it implies
.CGX	210	582	Character and graphics-bank work dominates both tapes, especially NEWS_11
.COL	125	355	Palette iteration was a major part of both workspaces
.OBJ	117	316	Object-side layout and sprite resources survive heavily in both
.SCR	52	286	NEWS_11 keeps far more authored screen and scene layouts
.PNL	51	91	Panel and tile-layout work is visible in both, but broader in NEWS_11
.OBX / .OBZ	38 OBX and 1 OBZ	21 OBX and 72 OBZ	NEWS_11 preserves a much larger animation or state-oriented object layer
.SPR	none visible	58	Sprite-state resources are a major extra layer in NEWS_11
.BAK	445	384	Both tapes are full of nearby backups rather than polished final-only files

---

How Much Overlap There Is

There is real overlap with the 日本_Ver2 char, OBJ, and col folders, but not enough to call the NEWS tapes simple duplicates.

By filename:

Archive	Unique filenames in the Yoshi folder	Overlap with 日本_Ver2 art-side filenames
NEWS_09	about 1,095	about 112
NEWS_11	about 2,024	about 195

Those counts are only filename overlap, but they are still useful. Most of the NEWS material does not line up one-to-one with the cleaner 日本_Ver2 art folders.

The overlaps are also mixed. Some files are exact matches, while others share the same names but not the same contents.

File	NEWS_09 vs 日本_Ver2	NEWS_11 vs 日本_Ver2
B3-KONOHA.SCR	exact match	exact match
48-4F.CGX	different content	exact match
COL-0.COL	different content	exact match
B3-KUMO.SCR	not checked here	exact match
B2-IWAYAMA.SCR	not checked here	different content
MAGIC-KAMEKKU-FAR.OBJ	different content	not checked here

That pattern is the most important clue in the whole comparison. The NEWS tapes were not just dumping random duplicates of the final art set. They preserve a mixture of shared late assets, earlier revisions, and alternate working versions.

---

What NEWS 09 Adds

NEWS_09 feels like a focused personal art workspace with a heavy emphasis on object, color, panel, and VRAM-side experimentation.

Its largest subfolders are especially revealing:

Folder	File count	What it suggests
OBJ-VRAM	132	Object graphics were being staged in a VRAM-oriented workspace
ENKEI-VRAM	86	Another VRAM-oriented art area, likely for perspective or special background work
TEST	77	A large amount of throwaway or trial material survived
KINKEI-VRAM	38	Yet another special-purpose VRAM/art grouping
OBJ-BOTU	11	Rejected or discarded object-side material
YOKOMAP-V-RAM	4	Horizontal-map VRAM experiments or special map-display work

That is very different from the cleaner 日本_Ver2/char folder. It looks like an artist’s live working directory, with multiple side bins for tests, VRAM experiments, and discarded material.

Some of the unique NEWS_09 names make that especially obvious:

• -CLR-.CGX, -CLR-.OBJ, -CLR-.SCR, and -CLR-.PNL
• -INK-TEST-.CGX and -INK-TEST-.COL
• test-bg1.scr, test-bg2.scr, test-bg3.scr, test-map.bgx, and test-obj.cgx
• kamebossdush.cgx, kamebossdush.OBJ, and kamebossdush.COL
• many .BAK alternates of the same CGX, COL, OBJ, and PNL resources

So NEWS_09 adds something the main source archive barely shows: the live scratchpad around the Yoshi art pipeline, including experiments, side variants, and material that may have been abandoned before the cleaner project snapshot was made.

---

Digging Into NEWS 09

The structure inside NEWS_09 makes it feel like a rough, production-floor art directory rather than a clean asset handoff.

The biggest clue is how aggressively the files are split into working bins:

• OBJ-VRAM
• ENKEI-VRAM
• KINKEI-VRAM
• TEST
• OBJ-BOTU
• BG-BOTU
• YOKOMAP-V-RAM

That is not how the integrated source branch is organized. Those folder names are much closer to “where the artist put things while working” than “how the final project was meant to be archived.”

The file mix reinforces that. NEWS_09 is dominated by:

• 445 .BAK files
• 210 .CGX
• 125 .COL
• 117 .OBJ
• 52 .SCR
• 51 .PNL

So this is a very backup-heavy workspace, and that matters historically. It preserves revision noise that the cleaner char, OBJ, and col folders mostly smooth away.

Some of the filenames are especially revealing:

• -CLR-.CGX, -CLR-.MAP, -CLR-.OBJ, -CLR-.PNL, -CLR-.SCR
• -INK-TEST-.CGX and -INK-TEST-.COL
• test-map.bgx, test-map.bcl, test-bg1.scr, test-bg2.scr, test-bg3.scr
• holder.scr and holder-2.scr
• kamebossdush.cgx, kamebossdush.OBJ, and kamebossdush.COL

That is the sort of naming you expect from a live experimentation folder. It captures “clear this out”, “ink test this”, “hold this version”, and “try this map variant” rather than only the names that survived into the final asset bank.

The dates fit that picture too. The sampled files in NEWS_09 cluster heavily across 1994 and the first half of 1995, with many top-level Yoshi art files dating from spring and early summer 1995. So this looks like an active mid-development personal workspace rather than a post-release dump.

---

What NEWS 11 Adds

NEWS_11 is much larger and feels closer to a late production art repository. It still overlaps the 日本_Ver2 art folders, but it preserves far more side branches and specialist subdirectories.

Its biggest subfolders are:

Folder	File count	What it suggests
CHR-stock	150	A large store of character-bank source or staging material
BAK-chr	137	Backed-up character revisions were kept in bulk
V-RAM	125	VRAM-specific graphics and palette staging was a major concern
4chr-PNL-BAK	87	Older panel or 4-character-layout backups were preserved
ENDING	83	Ending-specific visual work survived outside the main project tree
BAK-color	73	Bulk color/palette backups
NEW-MAP	62	A distinct stream of newer map-related art/layout work
SPR	33	Sprite resources survive here much more explicitly than in 日本_Ver2

This is also where the NEWS material starts to show things that feel genuinely new rather than just alternate copies. Examples include:

• many OBZ Yoshi movement and action files such as yoshi-swiming.OBZ, yoshi-jump-nobi-KOOPA.OBZ, yoshi-fura-fura.OBZ, and yoshi-BOSS-END.OBZ
• explicit region-leaning art variants such as 60-63-JAP.CGX, 60-63-PAL.CGX, 74-77-JAP.CGX, and 74-77-PAL.CGX
• birds/ test TIFF files, which look much closer to raw artwork iteration than normal build inputs
• separate ENDING, MAP-YOKO, NEW-color, and KOOPA-FINAL folders that make the workspace look strongly task-oriented rather than just format-oriented
• SPR files like CRYS.SPR, POLY.SPR, KSYS.SPR, and P0OB.SPR, which are much more visible here than in the main branch

Taken together, NEWS_11 feels like the stronger of the two supplementary archives. It preserves a lot more of the visual-production process that sits behind the cleaned-up resources in 日本_Ver2.

---

Digging Into NEWS 11

If NEWS_09 looks like a rough personal scratchpad, NEWS_11 looks like a much broader art and animation workspace that stayed close to the later asset pipeline.

Its top-level subfolders are unusually informative:

• CHR-stock
• BAK-chr
• V-RAM
• 4chr-PNL-BAK
• ENDING
• BAK-color
• NEW-MAP
• SPR
• birds

That tells us the workspace was being split by task as much as by file type. Instead of one flat art dump, it separates stock character banks, VRAM work, color backups, map updates, ending material, and explicit sprite-state resources.

The extension spread is also much richer than NEWS_09:

• 582 .CGX
• 355 .COL
• 316 .OBJ
• 286 .SCR
• 91 .PNL
• 72 .OBZ
• 58 .SPR
• 53 .MAP
• 13 .tiff

The .OBZ and .SPR layers are the standout additions. They make NEWS_11 feel much closer to an animation-state and sprite-behavior workspace than the main 日本_Ver2 art folders.

---

The OBZ and SPR Animation Layer

The single biggest thing NEWS_11 adds is a huge collection of what look like Yoshi and object animation-state resources.

The .OBZ names are extremely descriptive. They are not generic lumps of binary data. They read like state labels or action names:

• yoshi-stop-walk-run-jump.OBZ
• yoshi-jump-nobi-KOOPA.OBZ
• yoshi-fura-fura.OBZ
• yoshi-swiming.OBZ
• yoshi-dosun.OBZ
• yoshi-low-fiar.OBZ
• yoshi-bero-ue-yoko.OBZ
• yoshi-bonus.OBZ
• yoshi-BOSS-END.OBZ
• tamago-run.OBZ
• tamago-door.OBZ
• tamago-nage.OBZ
• baby-mario.OBZ
• WORLD-CLEAR-DEMO.OBZ
• DOOR.OBZ and DOKAN-UP.OBZ

That naming strongly suggests these files are not just another object-art format. They look like preauthored state sets for Yoshi, eggs, Baby Mario, doors, pipes, and demos.

The matching .SPR files push that interpretation further. NEWS_11 keeps top-level sprite resources like:

• CRYS.SPR
• POLY.SPR
• KSYS.SPR
• P0OB.SPR
• PCHG.SPR
• SWIM.SPR
• SJMP.SPR
• SLIP.SPR
• BOMB.SPR
• SBER.SPR

It also has a dedicated SPR/ subfolder containing one-to-one looking companions such as:

• SPR/yoshi-swiming.SPR
• SPR/yoshi-jump-nobi-nage.SPR
• SPR/yoshi-oshi.SPR
• SPR/yoshi-pelori.SPR
• SPR/yoshi-dosun.SPR
• SPR/yoshi-stop-walk-run-jump.SPR
• SPR/tamago-run.SPR
• SPR/tamago-door.SPR
• SPR/tamago-nage.SPR
• SPR/baby-mario.SPR

That is one of the strongest “new” areas in the NEWS tapes. The main 日本_Ver2 branch tells us a lot about character banks and object conversion, but NEWS_11 adds a much clearer view of animation-state naming and sprite-state organization.

There are also some useful structural clues inside the files themselves. All 72 of the .OBZ files in NEWS_11 are exactly 27,136 bytes long. That is a strong sign that they are not arbitrary exported art files. They look more like fixed-format state blocks or uniformly packed animation resources.

The .SPR files are very different. Their sizes vary widely, from only a few hundred bytes up into the low kilobytes, which makes them feel more like compact descriptor, table, or script-like companions rather than the main payload data.

The basename overlap strengthens that reading. Out of 72 unique .OBZ basenames and 58 unique .SPR basenames, 31 names are shared directly.

Some of the strongest pairs are:

• yoshi-stop-walk-run-jump.OBZ and SPR/yoshi-stop-walk-run-jump.SPR
• yoshi-swiming.OBZ and SPR/yoshi-swiming.SPR
• yoshi-swiming-cogi.OBZ and SPR/yoshi-swiming-cogi.SPR
• yoshi-boyoyon.OBZ and SPR/yoshi-boyoyon.SPR
• yoshi-ottotto.OBZ and SPR/yoshi-ottotto.SPR
• tamago-run.OBZ and SPR/tamago-run.SPR
• tamago-door.OBZ and SPR/tamago-door.SPR
• baby-mario.OBZ and SPR/baby-mario.SPR

That makes the split much easier to describe. The .OBZ side looks like the heavy per-state asset block, while the .SPR side looks like a smaller companion layer for those same states.

The unmatched names are just as interesting. The .OBZ-only set includes things like:

• WORLD-CLEAR-DEMO
• baby-W1-DEMO
• yoshi-BABY-KIMEPOSE
• yoshi-BOSS-END
• yoshi-KAIDAN
• yoshi-SUBERI
• yoshi-YOGAN
• DOKAN-UP and DOKAN-DOWN

So the OBZ pool clearly covers demos, special poses, stairs, sliding, lava, and pipe transitions beyond the states that got paired SPR companions.

By contrast, the SPR-only pool contains shorter symbolic names such as:

• SJMP, SJMP-1, SJMP-2, SJMP-3
• SWIM
• SLIP
• BOMB
• BERO
• SBER
• DMGE
• CRYS, POLY, KSYS, and PCHG

Those names look more like shorthand state labels or helper tables than user-facing asset names. Taken together, that makes NEWS_11 feel like it is preserving a real character-state resource system rather than just a loose pile of sprite files.

One last detail makes this even more important: those yoshi-*, tamago-*, baby-mario, WORLD-CLEAR-DEMO, and DOKAN-* names do not show up at all in the main 日本_Ver2 tree. So this whole naming layer is effectively unique to the NEWS tape workspace.

If you group those names by family, the shape becomes even clearer:

Family	OBZ	OBJ	SPR	What it looks like
yoshi-*	48	53	20	Core Yoshi movement, pose, and interaction states
tamago-*	12	3	7	Egg-related actions such as running, throwing, doors, and water-side variants
baby-*	3	3	1	Baby Mario-specific demo or reaction states
DOKAN* / DOOR*	4	2	0	Pipe and door transition resources

That is much more specific than the main source archive. The integrated 日本_Ver2 branch gives the program, converted art banks, and object tables. NEWS_11 adds a much more human-readable action vocabulary for how Yoshi, eggs, Baby Mario, and doors were being staged on the art side.

The Yoshi family alone is broad enough to sketch out a rough state catalogue:

• locomotion and pose states like yoshi-stop-walk-run-jump, yoshi-naname, and yoshi-slow
• terrain or hazard states like yoshi-SUBERI, yoshi-YOGAN, and yoshi-mizu
• interaction states like yoshi-oshi, yoshi-pelori, yoshi-bomb, and yoshi-bero-ue-yoko
• balance or stumble states like yoshi-ottotto, yoshi-fura-fura, and yoshi-boyoyon
• climb or transition states like yoshi-KAIDAN and yoshi-KAIDAN-DOWN
• demo or special-sequence states like yoshi-BOSS-END, yoshi-BABY-KIMEPOSE, and yoshi-bonus

The egg-side names are just as nice because they feel like a miniature subsystem of their own:

• tamago-run
• tamago-nage
• tamago-nage-2
• tamago-door
• tamago-door-out
• tamago-umi
• tamago-run-KOOPA

So NEWS_11 is not only preserving character art. It is preserving a pretty granular naming layer for movement, interaction, demo, and projectile states that sits somewhere between animation planning and final resource packaging.

---

Tracing One State Family

One of the best examples is yoshi-stop-walk-run-jump, because the same name survives in three different NEWS-side formats:

• yoshi-stop-walk-run-jump.OBJ
• yoshi-stop-walk-run-jump.OBZ
• SPR/yoshi-stop-walk-run-jump.SPR

Those three files do not look like duplicates. They look like stages in the same broader asset path:

File	Size	Date	What it most likely represents
yoshi-stop-walk-run-jump.OBJ	13,568 bytes	1993-07-28	Older object-side or layout-side source asset
SPR/yoshi-stop-walk-run-jump.SPR	1,394 bytes	1994-05-12	Compact sprite-state companion data
yoshi-stop-walk-run-jump.OBZ	27,136 bytes	1995-05-17	Later fixed-size packed state resource

That date spread is especially useful. It suggests the state family was not authored once and frozen. It was being carried forward through multiple production representations over at least two years.

The same pattern shows up in other state families too:

File family	Older layer	Smaller companion layer	Later packed layer
yoshi-swiming	yoshi-swiming.OBJ (1992-06-23)	SPR/yoshi-swiming.SPR (1993-11-10)	yoshi-swiming.OBZ (1994-08-29)
tamago-run	no matching plain .OBJ seen here	SPR/tamago-run.SPR (1994-06-03)	tamago-run.OBZ (1994-08-25)
baby-mario	no matching plain .OBJ seen here	SPR/baby-mario.SPR (1994-02-02)	baby-mario.OBZ (1994-06-08)

That makes the NEWS material much more than “extra art.” It is one of the clearest surviving windows into how character and interaction states were renamed, repackaged, and likely moved through multiple tool-side formats over time.

---

The Ending Workspace

NEWS_11/ENDING is another major addition. This is not just a handful of leftover files. It is a dedicated ending-art workspace with its own screens, palettes, staff-roll objects, and backup revisions.

The folder includes:

• No-1.SCR through No-5.SCR
• No1-BG2.CGX through No5-BG2.CGX
• palette sets like ENDING-No1.COL through ENDING-No5.COL
• staff-roll resources such as STAF-ROLL.OBJ, STAF-ROLL.OBX, STAF-ROLL-center.OBX, STAF-ROLL-M.OBX, and several STAF-ROLL-ENGRISH-* objects
• bird-related art like TORI.CGX, TORI.OBJ, TORI-INDY.CGX, and TORI-INDY.OBJ
• test and clear files such as TEST-1.SCR, TEST-1.CGX, TEST-1.COL, and -CLR-.CGX

That is a lot richer than the main Yoshi source branch, where ending logic is visible through endt and program-side references but the visual authoring process is much less exposed.

The STAF-ROLL-ENGRISH-* names are especially fun because they show the ending workspace preserving draft or working labels for the staff-roll content rather than cleaned public-facing names.

The folder structure also suggests the ending visuals were being built as composited scenes rather than one flat art bank. There are at least five numbered screen sets:

• No-1.SCR to No-5.SCR
• matching No1-BG2.CGX to No5-BG2.CGX
• matching ENDING-No1.COL to ENDING-No5.COL

That looks very much like a multi-screen ending sequence with per-scene background graphics and dedicated palette sets.

Then, layered on top of those screens, the workspace keeps:

• message-side objects such as MESSAGE.OBJ, MESSAGE-5.OBJ, and MESSAGE.OBX
• staff-roll objects like STAF-ROLL.OBJ, STAF-ROLL.OBX, STAF-ROLL-center.OBX, and STAF-ROLL-M.OBX
• language-specific or content-specific variants such as STAF-ROLL-HIRAGANA-0.OBX and STAF-ROLL-ENGRISH-0.OBX through STAF-ROLL-ENGRISH-3.OBX
• bird graphics and object layers through TORI.CGX, TORI.OBJ, TORI-INDY.CGX, and TORI-INDY.OBJ

That is a surprisingly complete look at the visual side of the game’s ending and credits. The main source archive tells us where ending placement data lived. NEWS_11/ENDING shows that the ending artwork itself was being managed as a small production workspace with scene screens, layered message/staff-roll objects, character elements, and many nearby backups.

---

Region Variants and Late Art Branching

NEWS_11 also keeps some unusually explicit regional art variants:

• 60-63-JAP.CGX
• 60-63-PAL.CGX
• 60-63-NOA.CGX.BAK
• 74-77-JAP.CGX
• 74-77-PAL.CGX
• 74-77-NOA.CGX.BAK

That is important because the main source branch already shows JPN, NES, and PAL conditionals in the build. The NEWS workspace shows the visual side of that same process.

In other words, the regional branching was not just happening in the asm. At least some graphics banks were also being split into Japanese, PAL, and NOA-oriented variants at the art-workspace level.

That makes NEWS_11 especially valuable when read beside 日本_Ver2. One shows the regional switches inside the build system. The other shows the corresponding messier art-side branching that fed those builds.

The actual bank comparison sharpens that point. All of these region-labelled files share the same 1995-07-10 date and the same 17,664 byte size:

• 60-63-JAP.CGX
• 60-63-PAL.CGX
• 60-63-NOA.CGX.BAK
• 74-77-JAP.CGX
• 74-77-PAL.CGX
• 74-77-NOA.CGX.BAK

But they are not all the same data.

Bank	JAP	PAL	NOA backup	Relation to 日本_Ver2
60-63	distinct	distinct	distinct	no plain 60-63.CGX was found in 日本_Ver2
74-77	distinct	identical to NOA backup	identical to PAL	plain 74-77.CGX exists in 日本_Ver2, but differs from both NEWS variants checked

That is a nice concrete example of what the NEWS tapes add. The main source branch shows that Yoshi’s Island still carried JPN, NES, and PAL conditionals. The NEWS art workspace shows that at least some character banks were being branched the same way, and not always as three fully independent end states.

In the 74-77 case, the PAL build candidate and the backed-up NOA candidate are byte-identical, while the JAP bank differs from both. So the regional split could be two-way in practice even when three labels were present in the workspace.

---

What Seems To Be New Versus Duplicated

The safest way to phrase the relationship is:

• some files are duplicated exactly between NEWS_11 and 日本_Ver2
• some files keep the same names but preserve different intermediate contents
• many files in both NEWS tapes do not appear in the main branch at all

That means the NEWS tapes are not redundant even when filenames overlap. The overlap often marks shared asset families, but not always the same revision of those assets.

In practical terms:

• use 日本_Ver2 when you want the cleanest view of the integrated Yoshi’s Island build
• use NEWS_09 when you want a rougher personal workspace full of tests, VRAM groupings, and discarded object-side material
• use NEWS_11 when you want the richest supplementary art archive, especially for character banks, sprite states, ending visuals, and region-variant graphics

So the NEWS tape restores do not replace the main Yoshi source folder. They make it deeper. They show much more of the messy artist-side history that sits behind the cleaner char, OBJ, and col data in the core Gigaleak snapshot.