25600 is the max from vitamins (was 0 EVs for every other stat)
> Prior to Generation III, effort values were based on a system of stat experience and vitamins increased a stat by 2560 EV points, but could only be used until the stat had 25600 EVs.
change it from 65535 (raw max) to 25600 (legal max).
A bit more coherent to see a list of medals rather than one at a time.
Add a few more properties, along with the Habitat List.
Remove maximize box from Join Avenue editor, consistent with other editors.
Extract a common IV generator method that has considerations for over-specified and under-specified criteria, rather than deadlocking/returning invalid IV spreads.
Rearrange magic numbers for sync lock-out/lock-in for better JIT comparison (sequential range).
Previous improvement missed the "what if the lead is NOT synchronize". We already have a branch for YES synchronize. Magnemite test case solves for no-sync
Probably would be worthwhile to de-tangle this but meh
if costura load errors, don't abort loading plugins
pre-scan for constructors with no-args so createinstance doesn't throw an exception preventing any other plugins from loading
Useful for code trimming gates to not trim interior/uses
Removes reflection use from EntityBlank, change signature to use EntityContext and Language for safest interaction.
Translate several strings to Simplified Chinese (zh-Hans). Updated legality_zh-hans.json (LevelBoostNotZero) and multiple UI text entries in lang_zh-Hans.txt, replacing English phrases with Chinese equivalents for sound prompts, menu labels, buttons, and trainer fields (e.g., PlaySoundOther, SAV_BattlePass.f_MAIN, SAV_Database.B_Add, SAV_Misc8b.TAB_Main, SAV_Trainer4BR.L_PlayerID, SID/TID).
Pt/HG/SS store the ROMCode, but D/P do not. In D/P, every "get version" function for the ROM is inlined to return a fixed value, not from the save file.
uint cast, was an "exclusive range" so it truncated back to 0, immediately terminating the loop and not searching the highest-end of seeds.
replace with tuple deconstruct so they stay as ulongs.
- Updated Ribbons for all languages to match the most recent official names.
- Finished off Italian UI and legality translations.
- Some updates to FRLG's Spanish Event Flags.
- Other miscellaneous additions, fixes, text fitting, and improvements for other languages.
Since BCC/Safari only roll PIDIV up to 4 times, a large amount of resulting-pk4 will still lack a 31 IV. These require the full recursion failure to be reached, so the traversal needs to travel up 4x before accepting an origin.
Reduce parameter passing by wrapping the immutables/derived in a struct.
Also clean up some level comparisons for lead:pressure boosting check, maybe DPPt's MethodJ impl was comparing wrong (didn't match MethodK). If we're above the encounter slot's level range, then it must have been pressure boosted. Don't bother checking for Grass type (the only type that can boost) as we already check that in the IsMatchExact yield.
Unrelated: Add a unit test of an obvious genned mon EVs I laughed at when scrolling thru my HG save file.
Thanks Unknown Warrior for reporting a BCC Volbeat that failed the detection. Not sure why I implemented the logic that way originally, but this cleans it up slightly. After looking at it further, it was missing a bit of tree pruning for Synchronization fails/lock-in, so we have a little bit of parameter passing hell happening. It's internal to the private methods, so I guess it's not too bad.
GPT-5.4 was given it as a first pass, of which it identified both existing bugs:
- seed was mutated in TryGetMatch before passing to the recursion
- frame reversal to origin inside the recursion (to get the seed that generates the rejected pid-iv) was off-by-one
The two fixes did allow it to solve, but it was missing the synchronize lock-in effect and sync-fail paths; no fault to GPT. Once those were added, everything worked as expected!
The frame-exit activation check was being run on the first-match of all lead candidates; due to leads consuming different amounts of calls, not all exit frame paths were checked. This resulted in a very very rare chance that a valid frame wouldn't be possible for non-sweet-scent encounters. So, those had to get inlined to each lead-check. Now, they're fully encounterable before being emitted to the "best result" aggregator at the top of the call chain. They're already preferentially emitted inside the lead-checker set (hence the weird if-else arrangement), just that separate PID-loop frames will possibly yield a less-restricted/more-common lead arrangement.
Peppers in a few more context dumps to help future me (or future AI) better maintain context. Also de-magics the Format numbers to better clarify what "0" does/means.
Unrelated: remove an unnecessary space (at the end of a line) in HomeQuirks
I'm pleasantly surprised that GPT-5.4 was able to deduce the two bugs in the implementation; I didn't provide enough context/requests, but it was quite confident in the two bugs (as it should have been -- it was correct) when I challenged it for explanation.
Extract gen3 default trash to trash class & folder
remove erroneous mismatch parse add (should be a pure method)
Set all purchased flags when shift is held
Gen3 LeafGreen french name detection tweaks (both fr & lg contain `feu`)
Gen4: Disable White Flute from boosting BugContest encounters
* Server date
* Only check for IV count if there's a specified Flawless value
* Calculate Shiny XOR from the correct ID32 value
* Do not require the legality checker to check for forced XOR 1
* Update text_ItemsG2_ja.txt
`ポイントアップ` means PP Up in Japanese, not HP Up.
HP Up is `マックスアップ`.
Updated unit tests to ensure all gen1-3 string lists are unique item names.
