fabi/Syplheed-Reborn
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

re(achievements): earned state comes from XAM, and my bit numbering was wrong

Browse Source 
Two findings, one of them a correction to the previous commit.

FOUND: GamePart_Debriefing enumerates XACHIEVEMENT_DETAILS from XAM
(0x8218F888). The records are 36 bytes -- confirmed by the arithmetic, not by
eye: the count is a byte count divided by 36 through the 0x38E38E39 multiply-high
magic plus srawi 3. Field +0 is used as a shift amount, field +32 is tested for
0x00020000, and the buffer sits behind a handle that is waited on (0x824AA330
with -1) then closed (0x824AA3E0). That is the XDK struct exactly, with
XACHIEVEMENT_DETAILS_ACHIEVED == 0x20000, via the
XamUserCreateAchievementEnumerator / XEnumerate pattern.

So the title does NOT persist earned achievements itself -- it asks the console.
The lever for achievement-gated content is the emulator's PROFILE data, not the
savegame. Independently, the singleton the challenge gate reads is not the object
holding the save block: re-scanning all 202 accessor call sites WITH function-
boundary stops touches only +80/+1956/+1960/+1964 and none of the known save
offsets (+304/+316/+320/+336/+380/+440). The earlier scan that seemed to find
them had register tracking bleeding into the next function -- a false positive I
am recording rather than quietly dropping.

CORRECTED: the previous commit claimed "bit n <-> achievement n+1" and treated
the Debriefing award pass and the challenge gate as the same bit space, both as
confirmed. Neither holds up:

- the only place the image is observed turning an achievement into a bit does
  1 << dwId with 1-based ids, so bit = the id and bit 0 is unused;
- the "list index is the bit index" step assumed 0x82448338's out-parameter is
  the loop ordinal. It is the child entry's first word out of a 12-byte array,
  and whether that is an ordinal, an id or a name hash is not pinned;
- the Debriefing's masks are its own fields (+208/+736/+740). Nothing observed
  copies them into the singleton's +80, and no writer of that bitmask has been
  found at all.

The requirement-type/id agreement still stands, but it confirms the LIST ORDER,
not the bit numbering. Consequence: if the join holds, TimeAttack's REQUIREMENT
16 is achievement 16 (Night Ravens Patch), not 17 (Solar System Defense Award).
Both read plausibly as a first-challenge gate, which is exactly why it needs
evidence and not the better story.

Also noted: +1960 is a third bitfield on the same singleton, and a what-changed
pass at 0x8219F3A4 diffs it and reports each newly set bit as bit + 64 -- so
there is a wider flag-id space whose bases are not yet worked out.
This commit is contained in:
Claude (auto-RE)
2026-08-13 19:43:35 +00:00
parent 33ae20896e
commit 90e0e66c7b
3 changed files with 103 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
docs/re/INDEX.md
View File

@@ -43,5 +43,5 @@ Promote to a prose `structures/…md` file when a format needs behavioural notes
| Function | Conf. | Reimpl. | Summary |
|----------|-------|---------|---------|
| Achievement award + content gating | ✅/❔ | [achievements](structures/achievements.md) | The title keeps its **own earned-achievement bitmask** and gates content on it. The XEX's `XACH` resource (`.pe` `0x8FBCBC`, 36-byte records) defines **24 achievements summing to 1000G** — the retail total, which self-checks the stride. `GamePart_Debriefing`'s `sub_8218F9A8` walks the on-disc `ACHIEVEMENTS_REQUIREMENTS` list (`tables.pak` #16, entries `ACHIEVEMENT01…24`) and sets **bit = list index** when the requirement evaluates true, so bit `n` ↔ achievement `n+1`. The list's requirement *types* (`ShootDownAircrafts` 1000/10000, `ShootDownShips` 100, `ShootDownWeight` MegaTons, `GetAllWeapons`, `GetAllAchievements`) line up with ids 19–24 exactly as `XACH` names them — two independent sources agreeing on the order. `GetAllAchievements`/`GetAllWeapons` are requirement **types**, not debug cheats. ❔ **where the mask persists** (save vs Xbox profile) is open; the obvious "save serializer writes `+1956`" lead is **refuted** (those stores are a different singleton's string pointers) |
| Achievement evaluation + content gating | ✅/🟡 | [achievements](structures/achievements.md) | The XEX's `XACH` resource (`.pe` `0x8FBCBC`, 36-byte records) defines **24 achievements summing to 1000G** — the retail total, which self-checks the stride and field offsets. `GamePart_Debriefing` (`0x8218CF38`–`0x82191B18`) does two things: it walks the on-disc `ACHIEVEMENTS_REQUIREMENTS` list (`tables.pak` #16, entries `ACHIEVEMENT01…24`, in achievement-id order — its `ShootDownAircrafts`/`ShootDownShips`/`ShootDownWeight`/`GetAllWeapons`/`GetAllAchievements` types line up with ids 19–24 exactly as `XACH` names them), evaluating each and setting a bit; **and it enumerates `XACHIEVEMENT_DETAILS` from XAM** — 36-byte records confirmed by the `0x38E38E39`+`srawi 3` divide-by-36, `dwId` at `+0`, `dwFlags & 0x00020000` (`…_ACHIEVED`) at `+32`, behind a waited-then-closed async handle. **So earned state comes from the console profile, not the 545-byte save** — and the masks it builds are `1 << dwId`, i.e. **bit = the 1-based id**. 🟡 the join to the challenge gate's word at singleton `+80` is *not* proven (the Debriefing's masks are its own `+208`/`+736`/`+740`, and no writer of `+80`'s bitmask has been found). `GetAllAchievements`/`GetAllWeapons` are requirement **types**, not debug cheats |
| Stage-config section switch (`0x82184df0`, `0x82185ed0`) | ✅ | [challenge-mission-gate](challenge-mission-gate.md) | The stage loader picks its config section from a **mission-kind field at `object+144`**: `3` → `EXTRA`, `5`/`6` → `CHALLENGE`, anything else → `FILE`; two further sites treat `{3,5,6}` as one class. Constructed as `0` (`sub_821783D8`) and only ever *cleared* inside the class, so the kind comes from the launching GamePart, **not** from the stage number — which is a mechanism (🟡, unproven) for why patching the save's stage field to a challenge stage kills the load. Same note carries the **GamePart id table** (`0x820A1630`, 29 ids, `GP_CHALLENGE` = 26, cross-checked against the image's own `RegisterToFactory<26, …>` text) and the disc's **three stage families** — `S01`–`S16` story, `S18`–`S23` tutorial, `S24`–`S29` challenge, plus `Test`, matching `weapon.tbl`'s 16 + 6 + 6 key set exactly. `GP_CHALLENGE.pak` holds **0 IDXD objects** — it is the menu screen; challenge missions reuse `GP_MAIN_GAME_E.pak`'s stage records |

23
docs/re/challenge-mission-gate.md
View File

@@ -173,9 +173,16 @@ resource holds the matching 24 achievement definitions, summing to **1000G**, th
retail total. Full table and record layout:
[`structures/achievements.md`](structures/achievements.md).
So **word A (`+80`) is the earned-achievement mask** (bit `n` = achievement `n+1`),
and **word B (`+1956`) is a second, different flag space** that requirement values
`≥ 24` index as `bit n-24`.
So **word A (`+80`) reads as the earned-achievement mask** and **word B (`+1956`) as a
second, different flag space** that requirement values `≥ 24` index as `bit n-24`.
⚠️ **Corrected the same day:** an earlier draft said "bit `n` = achievement `n+1`" and
called the join ✅. Both were over-claimed. The Debriefing's masks live on the
*Debriefing* object (`+208`/`+736`/`+740`), nothing observed copies them into the
singleton's `+80`, and the one place the image is *seen* turning an achievement into a
bit does `1 << dwId` with **1-based** ids. So the mapping is 🟡 and, if it holds, is
`REQUIREMENT n` = achievement **`n`** — see
[achievements §3.1 and §4](structures/achievements.md).
### 5.3 What the six requirement values are 🟡
@@ -187,10 +194,12 @@ sound here** (the same trap the movie/subtitle map hit) — with one exception:
value-before-key adjacency, so the **first** mission (`TimeAttack`) requiring **bit 16**
is well-supported.
Bit 16 is achievement **17, "Solar System Defense Award"** — *"great achievements
during the campaign to defend the Solar System"*, i.e. **finish the story campaign**.
That is exactly the shape of gate you would expect on the first challenge mission, and
it is independent corroboration that the bit space is the achievement space.
Under the 1-based reading that §5.2 now prefers, bit 16 is achievement **16,
"Night Ravens Patch"** — *"for challenging and eradicating the Night Raven
Squadron"*, a stage-15-era award. (The withdrawn 0-based reading gave achievement 17,
"Solar System Defense Award"; both are plausible gates for a first challenge mission,
which is precisely why the numbering has to be settled by evidence rather than by
which story reads better.)
The remaining five values (`25`–`29`, all `≥ 24`) therefore index **word B**, not
achievements — most plausibly a challenge-clear chain, since there are six challenge

109
docs/re/structures/achievements.md
View File

@@ -43,7 +43,13 @@ Strings come from one `XSTR` section per language (7 present); each is
**The stride and field offsets are self-checked**: the 24 gamerscores sum to
**exactly 1000**, the retail total. A wrong stride does not add up to a round 1000.
Ids run `1…24` with no gaps, so **bit `n` ↔ achievement id `n+1`**.
Ids run `1…24` with no gaps.
> **Correction (same day).** This file first said "bit `n` ↔ achievement id `n+1`",
> i.e. 0-based bits. That was wrong-headed: the only place the image is *observed*
> turning an achievement into a bit does `1 << dwId` with `dwId` 1-based (§3.1), so
> the bit index is the **id itself** and bit 0 is unused. The 0-based reading came
> from assuming the config list's position was the bit, which §3 no longer supports.
## 2. What each bit is ✅
@@ -74,50 +80,96 @@ Ids run `1…24` with no gaps, so **bit `n` ↔ achievement id `n+1`**.
| 22 | 23 | 80 | Weapon Lord Patch |
| 23 | 24 | 100 | TCAF Pilot's Commendation |
## 3. The game awards them itself, from a disc config ✅
## 3. The game evaluates them itself, from a disc config ✅ (numbering 🟡)
`GamePart_Debriefing` (`0x8218CF38`–`0x82191B18`, bounded by the factory creator
thunks either side) runs `sub_8218F9A8` after a mission:
```asm
for each entry of the ACHIEVEMENTS_REQUIREMENTS config list, index n:
bit = 1 << n
for i = 0, 1, 2, …:
node = child(ACHIEVEMENTS_REQUIREMENTS, i, &x) ; bl 0x82448338
bit = 1 << x
if (this+208 & bit) continue ; already awarded
if (evaluate(entry)) ; bl 0x8218FAB0
if (evaluate(this, node)) ; bl 0x8218FAB0
this+208 |= bit
```
so **the list index is the bit index**. The list is on disc — `tables.pak` entry
**#16**, schema `744c0519`, the `GP_DEBRIEFING_PILOTLOG.pak+eng` config — and its
entries are literally `ACHIEVEMENT01` … `ACHIEVEMENT24`, in order.
The list is on disc — `tables.pak` entry **#16**, schema `744c0519`, the
`GP_DEBRIEFING_PILOTLOG.pak+eng` config — and its entries are literally
`ACHIEVEMENT01` … `ACHIEVEMENT24`, in order.
⚠️ **`x` is not proven to be the loop index.** `0x82448338` walks a 12-byte child
array and writes the child entry's **first word** to the out-parameter; whether that
word is the ordinal, an explicit id, or a name hash is not pinned. Since §3.1 shows
the image elsewhere shifting by a **1-based achievement id**, `x` is most likely the
id too — but this file previously stated "the list index is the bit index" as fact,
and that is withdrawn.
### 3.1 The earned state comes from XAM — the console profile, not the save ✅
The same class enumerates a buffer of **36-byte** records (`0x8218F888`):
- the record count is a byte count divided by 36 — via the multiply-high magic
**`0x38E38E39`** plus `srawi 3`, which is the standard unsigned `/36` sequence, so
the stride is confirmed by the arithmetic and not just by inspection;
- field **`+0`** is used as a shift amount (`1 << id`), field **`+32`** is tested for
bit **`0x00020000`**;
- the buffer is produced asynchronously: a handle at `this+100` is waited on
(`0x824AA330(h, -1)`) then closed (`0x824AA3E0`).
That is exactly the XDK's `XACHIEVEMENT_DETAILS` —
`{ DWORD dwId; PWSTR pwszLabel, pwszDescription, pwszUnachieved; DWORD dwImageId,
dwCred; FILETIME ftAchieved; DWORD dwFlags; }` = 36 bytes, with
`XACHIEVEMENT_DETAILS_ACHIEVED == 0x00020000` — fetched through the
`XamUserCreateAchievementEnumerator` / `XEnumerate` pattern.
**So the title does not persist earned achievements itself: it asks the console.**
The masks it builds (`this+208`, `+736`, `+740`) are `1 << dwId`, i.e. **bit = the
1-based achievement id**, bit 0 unused.
For anything that wants to unlock achievement-gated content, the lever is therefore
the **emulator's profile achievement data**, not the 545-byte savegame.
The record also carries each requirement's **type and parameters**:
`StageClear`(`Stage`), `MissionObjective`, `Item`, `Rank` (`S`),
`ShootDownAircrafts`(`Count` 1000 / 10000), `ShootDownShips`(`Count` 100),
`ShootDownWeight`(`MegaTons`), `GetAllWeapons`, `GetAllAchievements`.
**That set independently confirms the ordering.** The last five types line up with
ids 19–24 exactly as the XACH table names them: 1 000 units, 10 000 units, 100
warships, one gigaton, all Delta Saber equipment, and finally the meta
"`GetAllAchievements`" → id 24, `TCAF Pilot's Commendation`, the 100G one. Two
independent sources agreeing on the order is what makes the bit mapping ✅ rather
than an inference.
**That set independently confirms the list's ORDER** (not the bit numbering): the last
five types line up with ids 19–24 exactly as the XACH table names them — 1 000 units,
10 000 units, 100 warships, one gigaton, all Delta Saber equipment, and finally the
meta `GetAllAchievements` → id 24, `TCAF Pilot's Commendation`, the 100G one. So the
config list is in achievement-id order; what it does **not** settle is whether the bit
the code shifts by is that position or the id (see the ⚠️ above).
⚠️ `GetAllAchievements` and `GetAllWeapons` **look like debug cheats and are not** —
they are requirement *types* in the achievement table. Worth stating because the
strings sit next to genuinely debug-looking ones in the image.
## 4. Where the earned bits live ❔
## 4. The gate's own words — related, but not yet joined up 🟡
`GamePart_ChallengeMission` reads them off a singleton (`0x821707C0`, lazily built
behind the global at `0x828F48B0`):
`GamePart_ChallengeMission` reads its two words off a **different** object, the
singleton at `0x821707C0` (lazily built behind the global `0x828F48B0`):
- **`+80`** — tested for requirement values `< 24`, i.e. **the 24 achievements above**;
- **`+80`** — tested for requirement values `< 24`;
- **`+1956`** — tested for values `>= 24` with bit `n-24`, so a **second, different
flag space** (the challenge screen's own five later requirements land here).
The split at exactly 24 matching a table of exactly 24 achievements is the reason to
read `+80` as the achievement mask.
The split at exactly 24, against a table of exactly 24 achievements, is why `+80`
reads as the achievement mask.
⚠️ **But the join is not proven.** The Debriefing's masks live on the *Debriefing*
object (`+208`, `+736`, `+740`); nothing observed copies them into the singleton's
`+80`, and no writer of the singleton's `+80` bitmask has been found at all. Until
that link exists, "challenge `REQUIREMENT` *n* = achievement *n*" is a **hypothesis**.
Which reading wins also moves the answer by one: with §3.1's 1-based ids, the
`TimeAttack` requirement `16` would be achievement **16, `Night Ravens Patch`**, not
17 (`Solar System Defense Award`) — an earlier draft of this note asserted 17.
Note also that `+1960` is a third bitfield on the same singleton, and a "what changed"
pass (`0x8219F3A4`) diffs it before/after and reports each newly-set bit as
**`bit + 64`** — evidence of a wider flag-id space whose base for `+1956`/`+1960` is
not yet worked out.
**Not yet known: what writes them.** No `stw` to `+1956` anywhere in the image
targets this object — the ones that exist belong to a different singleton (reached
@@ -130,14 +182,23 @@ demonstrated. Note also that XEX imports are **by ordinal**, so the absence of
`XamUser*` name strings in the `.pe` is *not* evidence that the mask does not come
from the profile.
Settling this is worth doing: if the mask is save-backed, a hand-written save
unlocks the challenge missions (the savegame round-trip is already solved); if it is
profile-backed, the emulator's profile is the lever instead.
**Ruled out on the save side:** the singleton is *not* the object carrying the
545-byte save block. Scanning all 202 call sites of the accessor with
function-boundary stops, only `+80`, `+1956`, `+1960` and `+1964` are ever touched on
it — **none** of the known save offsets (`+304` block start, `+316` flight time,
`+320` clear ratio, `+336` Points, `+380` develop blob, `+440` SHAB) appear. An
earlier pass without those boundary stops appeared to find them and was a
false positive from register tracking bleeding into the next function.
So the persistence question now has a positive answer on the other side (§3.1: the
title reads earned achievements from **XAM**) and the remaining work is to join the
XAM-derived mask to the gate's `+80`.
## Reproduce
```bash
python3 tools/xach_dump.py "<disc>/…/Project Sylpheed ….pe"
cargo run --release -q -p sylpheed-formats --example achievements_map -- <disc-root>
python3 xenia-rs/zq.py dis 0x8218f9a8 0x8218fa60 # the award pass
python3 xenia-rs/zq.py dis 0x8218f9a8 0x8218fa60 # the requirement walk
python3 xenia-rs/zq.py dis 0x8218f888 0x8218f990 # the XACHIEVEMENT_DETAILS scan
```