WIP: feat(mods): a randomizer example mod, OoT Randomizer 9.1's logic and settings, seeds placed through the Modding API #1959

Draft
HumanBean wants to merge 5 commits from HumanBean/Longshot:feat/randomizer into mod-api
Owner

Summary

A randomizer for Longshot as an example mod (modapi/examples/randomizer), built on the Modding API in #1958 and opened against its mod-api branch. Its generator follows OoT Randomizer 9.1 (logic, settings, presets, tricks, assumed fill, a solver that proves each seed beatable, hints by OoTR's distributions, spoiler log or race mode, hash icons); OoTR's tables are read as data (OoTR is MIT at its top level; its notice is in src/data/LICENSE-OoTR.txt), the engine is the mod's own. In the game today it shuffles the chests and drops it can bind to OoTR locations at run time (no table of the game's ids is committed); the rest waits on API additions being built on this branch: rewards for songs, gifts, shops, scrubs, cows and tokens; settings in the content stage (gossip-stone hints); world state; age and spawn; dungeon items and doors; items.remove; cutscene skips; new items; entrances.

Core addition: content.read.placements(map), a map's placements read from the player's own game.

Read first: the mod's README (players, racers, the parity table, the API gaps, where every table came from), src/gen/, src/game/bind.ts.

Testing

Tested: not in the game yet. In the fake game: 54 seeds over 9 presets and both modes, each beatable, settings and exclusions kept, hints true; the same seed text gives the same seed.

Still to test:

  • Linux, this PR's head: a new file with a seed typed in the Mods tab; the log's bound N of M line, a bound chest giving the seed's item, starting items once, the time to make a seed in the game (by hand)

Measurements

TODO: the time to make a seed in the game's QuickJS, and the main thread's cost per frame while the mod runs.

Risks

An example mod: nothing changes for a player who doesn't add it. content.read.placements reads a map's files only when a mod asks.

Checklist

Ready for review

  • One topic, on current main; the title is a Conventional Commit. (On #1958's mod-api.)
  • Tested on my own PC, in the game for a change the game runs, as Testing says.
  • mise run test passes (any failure that also fails on main is named).
  • Measured against main, or Measurements says why there is nothing to measure.
  • No game files, build output or hand-edited generated files; third_party/ untouched; nothing GPL-incompatible.

When it applies

  • Docs updated.
  • AI assistance recorded with Assisted-by: / Co-Authored-By: trailers.

Ready to merge

  • Everything in Still to test has run, its results in Tested.
  • Approved in review, every comment answered.
  • CI passes on the head.

🤖 Generated with Claude Code

https://claude.ai/code/session_017vw8e3nBDxBnhodLguFyEt

## Summary A randomizer for Longshot as an example mod (`modapi/examples/randomizer`), built on the Modding API in #1958 and opened against its `mod-api` branch. Its generator follows OoT Randomizer 9.1 (logic, settings, presets, tricks, assumed fill, a solver that proves each seed beatable, hints by OoTR's distributions, spoiler log or race mode, hash icons); OoTR's tables are read as data (OoTR is MIT at its top level; its notice is in `src/data/LICENSE-OoTR.txt`), the engine is the mod's own. In the game today it shuffles the chests and drops it can bind to OoTR locations at run time (no table of the game's ids is committed); the rest waits on API additions being built on this branch: rewards for songs, gifts, shops, scrubs, cows and tokens; settings in the content stage (gossip-stone hints); world state; age and spawn; dungeon items and doors; items.remove; cutscene skips; new items; entrances. Core addition: `content.read.placements(map)`, a map's placements read from the player's own game. Read first: the mod's README (players, racers, the parity table, the API gaps, where every table came from), `src/gen/`, `src/game/bind.ts`. ## Testing **Tested:** not in the game yet. In the fake game: 54 seeds over 9 presets and both modes, each beatable, settings and exclusions kept, hints true; the same seed text gives the same seed. **Still to test:** - Linux, this PR's head: a new file with a seed typed in the Mods tab; the log's `bound N of M` line, a bound chest giving the seed's item, starting items once, the time to make a seed in the game (by hand) ## Measurements TODO: the time to make a seed in the game's QuickJS, and the main thread's cost per frame while the mod runs. ## Risks An example mod: nothing changes for a player who doesn't add it. `content.read.placements` reads a map's files only when a mod asks. ## Checklist ### Ready for review - [ ] One topic, on current `main`; the title is a Conventional Commit. (On #1958's `mod-api`.) - [ ] Tested on my own PC, in the game for a change the game runs, as Testing says. - [ ] `mise run test` passes (any failure that also fails on `main` is named). - [ ] Measured against `main`, or Measurements says why there is nothing to measure. - [ ] No game files, build output or hand-edited generated files; `third_party/` untouched; nothing GPL-incompatible. ### When it applies - [ ] Docs updated. - [x] AI assistance recorded with `Assisted-by:` / `Co-Authored-By:` trailers. ### Ready to merge - [ ] Everything in Still to test has run, its results in Tested. - [ ] Approved in review, every comment answered. - [ ] CI passes on the head. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_017vw8e3nBDxBnhodLguFyEt
modapi/examples/randomizer: tools/import_ootr.py reads OoT Randomizer 9.1 (MIT) as data (regions and rules,
helpers, location names/types/vanilla items, items, settings, tricks, hint texts and distributions, presets; no
ROM addresses or scene ids, nothing from a game) into src/data/. src/gen/: OoTR's rule language parsed and
compiled to closures (here/at subrules, helpers as macros, State functions), the world, the search, OoTR's item
pool and assumed fill (dungeon items, rewards, songs, progression, the rest), the playthrough, required
locations, barren areas, hints by OoTR's distributions, SHA-256 seeding and hash icons, and a verifier that plays
a seed back from what it says. Two modes: full (everything OoTR shuffles) and game (what Longshot's API can
change today: drops). test/generate.test.ts: same text same seed, every preset beatable in both modes, hints true.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8e3nBDxBnhodLguFyEt
longshot/content's read.placements(map, { param }) answers the actors placed in one of the game's maps, from its
Banc/<folder>/<map>_Dynamic and _Static files as the RomFS layers serve them: each { id, actor, layer, position,
params }, the id being the placement id items.checked names a check by and rules.rewards takes. A randomizer or a
tracker then learns the game's checks at run time from the player's game, with no table of ids (game data) in the mod.

host/mod_read.c mod_read_placements_json (bounded BYML reading, 8 MiB at most), host/modapi/content_read.c's
reader thread (the Banc folders listed once; each call reads the files again), the schema (content.read.toml,
PlacementsOptions) and the generated files, services.c's feature, the fake game's contentData({ placements }).
Tests: tests/test_mod_content.py (the reader over made-up files, both byte orders, the filter, refusals),
tests/test_modrt.py check_content_read (the call in the real runtime). API version stays 37.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8e3nBDxBnhodLguFyEt
src/main.tsx: the settings read as OoTR's (src/game/input.ts), the game's checks read from the player's own maps
(content.read.placements) and bound to OoTR's drop locations by map group and drop kind (src/game/bind.ts: no
table of ids; one of a kind bound for sure, several bound together with all their rules), the seed made by the
worker in steps (src/worker.ts: at most 2 s a message, then checked by the solver in a turn of its own), rewards
set for every bound location (progressive items follow the tiers Link holds), starting items granted once to a
new playthrough (items.grant, keyed by seed and slot), the spoiler log in global storage (none in a race), every
check logged by its OoTR name for trackers, hints near gossip stones and the hash on the HUD and the Mods tab page.
src/game/names.ts: OoTR items by the actor names Longshot already holds, the rest learned from bound placements.

Generator: per-world progression counts (a seed no longer depends on what was generated before it), unbound
drops kept as the game's own and counted by the logic, the Fire Temple's keys kept where the game has them in
the game's mode (OoTR's logic counts a key for a door its ROM opens), and dungeons whose keys all stay the game's
own given their keys in the logic (OoTR's key logic cannot follow some vanilla layouts).

Tests: test/game.test.ts (the fake game: binding, rewards, claim, race, starting items, checks),
test/settings.test.ts (settings to input, settings kept, what the game cannot do set back).

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8e3nBDxBnhodLguFyEt
A feature area of its own: Kokiri Forest's placements with a drop read (both ages, and a map the game lacks), their
kinds logged by name, the first chest given a reward by the id read, its check told apart. Fake game (made-up
placements) and the game (UNCONFIRMED). Listed in test_game_features.py's AREAS, the suite's README and
docs/dev/game-tests.md.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8e3nBDxBnhodLguFyEt
docs(mods): the randomizer example's README (players, racers, parity, API gaps, sources) and its examples.md entry
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request_target) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 30s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 35s
Third-party parts / As their notes say (pull_request) Successful in 47s
7117cab56c
README: playing and racing a seed, what the game cannot do yet and the ten API gaps with proposed shapes (the
first, placements, closed by this branch), a parity matrix against OoTR 9.1, SoH, Archipelago and Clawshot
(engine and game columns), where every table came from and under which licence (OoTR MIT, nothing from SoH or a
game dump), the files, the tests and what a game run must confirm. The mod's main thread now keeps to its 5 ms:
rewards and the spoiler log's parts are handed over a few a frame, the spoiler log is made on the worker, and
progressive rewards are looked up by item.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8e3nBDxBnhodLguFyEt
Member

Possibly related, for context:

  • issue #42 Idea: one path for randomizers, restoration mods and co-op, with a mod interface in tiers — Open randomizer/mod-interface proposal; this implements a concrete randomizer example
Possibly related, for context: - issue #42 Idea: one path for randomizers, restoration mods and co-op, with a mod interface in tiers — Open randomizer/mod-interface proposal; this implements a concrete randomizer example
BeanBot left a comment

Recommendation: Request changes. The randomizer cannot load from a clean checkout because its generator is missing. The placements reader can report failed reads as successful, and the mod can change a save with rewards that differ from the seed it verified.

The placements API would let mods discover checks from the player's own game at runtime. The randomizer is a useful first client, but it needs a complete generator, proof that the game delivers what the solver verifies, and save-safe activation. The API also needs bounded delivery and reliable error reporting.

Does it help: Unproven. The PR reports 54 fake-game seeds across nine presets and two modes, without a run commit or results. The reported sweep cannot be reproduced from this head because its generator imports are missing. There is no in-game result, seed generation time, or frame-time comparison.
Other systems: Other content readers share the same queue and now scan Banc on their first row or text read. Placements also exercise shared RomFS and zstd code, and large answers are parsed on the game thread before the settlement budget is checked again. Rewards and grants affect save slots and other mods using the reward claim. The diff has no renderer, VRAM, audio or input changes. Evidence on Windows, Linux and Steam Deck is missing.
Switches: No RX_, HOS_ or LONGSHOT_ switches. The example adds 30 Mods-tab settings. Seed defaults to empty, preset to S9 Tournament, shuffle to game mode, and race to off. Logic, guarantee, bridge, Ganon's boss key, forest, Door of Time, fountain, fortress, Kakariko gate, small keys, boss keys, maps and compasses, Kokiri Sword, songs, tokens, freestanding items, shops, scrubs, cows, item pool, starting age, hint distribution and hint text default to Preset. Both count overrides default to 0 for Preset, trials to −1 for Preset, tricks, exclusions and starting items to 'preset', and custom settings to '{}'. Show hash and show hints default to on. The empty seed and spoiler-only choices need to honor their stated effects. Count controls should use the selected condition's supported range.

Reviewed: 7117cab56c against fa574bf340: Read the repository AGENTS.md, CONTRIBUTING.md, standards and Forgejo template, the five commits, all authored changes, and the placements reader and its BYML helpers in full. Read the relevant caller and dependency sections in RomFS, zstd, the runtime, rewards, items, storage, workers, and the test harnesses. Skimmed generated bindings and docs, and sampled the generated OoTR tables against the importer. The upstream OoTR source, Tooling checkout, and docs/review-checklist.md were unavailable. The advertised src/gen directory is absent at HEAD. No comments, reviews or recorded runs were supplied.; nothing run (a code review).
Checked and fine:

  • All five commits carry a Co-Authored-By trailer. The diff leaves third_party and the renderer ABI untouched.
  • The new placement test fixtures use fabricated IDs. The imported tables identify their upstream source and carry its MIT notice. The diff contains no sampled placement census or game dump.
  • The read job, including its owned key, is published under the queue lock. The base folder list is initialized on the API thread before the queued placement jobs read it.
  • The low-level tests verify both BYML byte orders, a high-bit 64-bit ID, parameter filtering and missing Actors. The schema, feature lookup, native dispatch and TypeScript declarations all name the new API member.
  • The importer uses the Python standard library and reads a local checkout as data. Its output writes are documented.

Blocking (to change before merge)

  1. modapi/examples/randomizer/src/main.tsx:31: The runtime and tests import src/gen, but HEAD contains none of it. The root gen/ ignore rule matches that directory. A clean checkout fails to load the mod or run its generator tests. Rename the directory or narrow the ignore rule, and commit the complete generator for review.
  2. host/modapi/content_read.c:256: A failed layer is logged and then returned as success. If Dynamic is valid and Static fails to decompress, callers see a partial map. If encoding fails mid-actor, the accumulated JSON can be invalid. Final allocation failure also looks like an absent map. Reject failed reads and allocations, discard partial output, and reserve null for a genuinely absent map.
  3. host/mod_read.c:446: Translate is serialized as any BYML array. A replacement file with two coordinates, a string coordinate or NaN can yield a successful result that violates the public numeric triple type. Validate the container and its three finite numbers. Also exclude zero hashes, which the promised reward API rejects.
  4. host/modapi/content_read.c:113: Discovery keeps only the first 64 base Banc entries, including non-directories, then caches the result for the process. A map in a later folder or a folder supplied only by a RomFS layer is reported absent. Find files through the merged namespace, handle listing failures explicitly, and move Banc discovery off the shared first row or text call.
  5. host/modapi/content_read.c:247: The base path is silently truncated if the installation root and relative path exceed the fixed buffer. Dictionary-backed reads also use zs.c's MAX_PATH buffer for the dictionary and its unchecked fseek on Windows. A map can be listed but fail to read in a deep install. Reuse checked UTF-8 path construction and file reads that check the 64-bit seek result.
  6. host/modapi/content_read.c:246: Placement discovery now resolves the same layered Banc files that game threads open. RomFS's resolver reads and writes the shared RomfsPath.logged field without synchronization. Make the first-log claim atomic or keep mutable logging state out of the published path table.
  7. host/modapi/content_read.c:253: The zstd helper marks first use before finishing dictionary initialization. A simultaneous texture or content read can observe an incomplete dictionary table or race its writes. The placements worker adds another caller. Publish the dictionaries only after initialization completes and make other callers wait for completion.
  8. host/modapi/content_read.c:274: One answer can exceed 8 MiB with framing and is copied into a settlement and parsed in one operation on the game thread. A dense map or repeated calls can also accumulate large native settlements before the JS heap limit applies. Page or cap delivery to an enforceable per-frame size, bound pending bytes, and avoid the extra whole-result copy.
  9. modapi/packages/mod-tools/lib/fake-content.js:925: The fake reader ignores content.place outputs. A mod can read its own placement in the game and get null or an incomplete list in its tests. Merge the fake content plan using the same map, age, layer and hash rules. Use own-key lookup too, so a valid absent map name such as constructor returns null.
  10. modapi/examples/randomizer/src/main.tsx:86: The mod queues every map, treats rejected reads as missing maps, and posts all raw placements in one message. A busy reader can make the seed silently depend on which requests succeeded, and a payload above the worker's 1 MiB cap stops binding. Read a bounded number at a time, send compact batches, and fail explicitly on read or messaging errors.
  11. modapi/examples/randomizer/src/main.tsx:145: Full mode is advertised as spoiler-only, but its seed follows the same reward and starting-grant path. Choosing it can change chests and inventory with a fill the game cannot carry out. Give full mode a separate completion path that only presents or stores its result.
  12. modapi/examples/randomizer/src/game/bind.ts:132: Equal counts alone do not establish that physical placements are the solver's checks. An unrelated drop of the same kind can stand in for a missing chest. Hash sorting and an all-locations reachability rule cannot establish that the physical drop persists or is reachable at that stage. Bind only checks with verified identity and reward behavior, and keep uncertain bindings out of the fill.
  13. modapi/examples/randomizer/src/main.tsx:103: The seed input includes logical names and groups but no fingerprint of their physical assignments. If a map mod changes which chest IDs sort first in a two-chest group, the input and race icons can stay the same while the rewards swap physical locations. Hash the canonical game binding along with the effective settings and placements.
  14. modapi/examples/randomizer/src/main.tsx:130: A verified item with no actor mapping is silently skipped. The chest keeps its vanilla item, yet the seed is logged as beatable. Starting grants skip unmapped items too, including the Zelda's Letter item named by the S9 test. Validate that every logical reward and starting item can be delivered with its required semantics before accepting the seed.
  15. modapi/examples/randomizer/src/main.tsx:128: Progressive rewards are assigned when the chest actor loads. Two Hookshot rewards in an already loaded map can both give the first tier because the rule update affects only later actor loads. Counting tiers held also assumes upgraded items leave previous tiers in inventory. Resolve progression when collected from the actual upgrade state, or exclude progressive replacements until the API supports them.
  16. modapi/examples/randomizer/src/main.tsx:148: Rewards are published in batches while play can proceed. A player can open a chest before its rule is installed, and if a rule addition throws after earlier batches, the partial set remains and later work can still announce readiness. Gate play and activation on complete installation, or abort and remove all installed rules on failure.
  17. modapi/examples/randomizer/src/main.tsx:185: A save from another seed gets a warning after process-wide rewards are already installed. Its unopened chests can give seed B's items after it consumed seed A's progression checks. Validate the active playthrough before reward rules can affect its actors and remove rules on an incompatible load. Require an explicit fresh start before adopting an unmarked save.
  18. modapi/examples/randomizer/src/main.tsx:194: The grant key identifies a slot number and seed rather than a playthrough. Deleting that slot and starting the same seed again leaves the old grant keys in the journal. Queued grants can also be delivered to the next active save because the journal is not slot-scoped. Tie grants and their acknowledgement to a persistent playthrough identity, with delivery restricted to that playthrough.
  19. modapi/examples/randomizer/src/main.tsx:205: A new game can be in play with save.slot() still null. The update latches started before the grant function returns early. The first save sets the slot without a loaded event, so the player can go without the starting kit until a reload. Use a starting-grant state that waits for a usable playthrough and retries until delivery is confirmed.
  20. modapi/examples/randomizer/src/main.tsx:214: checkedIds is shared across all slots and survives rollback. If a player opens a chest, reloads a save from before it, and opens it again, the mod suppresses the check and can skip updating its progressive rule. Rebuild check state and reward tiers from the active save on load.
  21. modapi/examples/randomizer/src/main.tsx:174: Each spoiler write swallows errors, but the final jobs publish spoiler/last and log success even if a part was refused, for example when the global quota fills. Use byte-bounded parts, commit the index only after every part succeeds, and preserve the previous valid log on failure.
  22. modapi/examples/randomizer/src/main.tsx:243: A stone's hint depends on its rank among actors currently visible within 300 units. Another stone streaming in can change that rank. Disabling hints also leaves the previous hint visible because the function returns before clearing it. Bind hints to stable stone identities and clear them on toggle-off and map changes.
  23. modapi/examples/randomizer/src/main.tsx:53: An empty seed still starts the worker, imports generator data on both sides and installs frame callbacks. That incurs compilation, heap and worker costs for a setting advertised as inactive. Start the generator and gameplay callbacks only with a configured seed, and stop the worker after completion or failure.
  24. modapi/examples/randomizer/test/game.test.ts:50: The fake reward assertion is skipped when its chosen item has no static actor mapping. The race test inspects storage as soon as ready is logged, before queued spoiler writes would run. Make the reward expectation unconditional for a controlled seed and drain pending work before asserting that race mode wrote nothing.
  25. modapi/examples/randomizer/test/game.test.ts:88: The real-game test can skip a configured seed that takes more than 60 frames to bind or fails before logging a bound count. The placements feature test can also pass by registering a rule on the first non-chest drop without opening it. Distinguish an unset seed explicitly and require a real chest to deliver the expected item and emit its check.
  26. tests/test_modrt.py:3477: The new runtime fixture uses the POSIX shim, Linux libraries and a stub that returns raw BYML instead of exercising the Windows directory and real zstd path. Its test entry also defaults to GCC instead of honoring CC. Wire the placement path into a portable fixture with a UTF-8 root, real compressed input and a merged-layer case, and use the selected compiler.

Waiting on the author (a run, a measurement or a decision; no edit does it)

  • The PR opened without the required game run, and there is no exact-head check result. After the code issues are fixed, report Linux and Windows game runs covering mapped chest delivery and checks, progression, starting items, slot changes and spoiler behavior. Attach the repository check results for the resulting head.
  • The costs of startup, placement settlement and reward publication are unmeasured. After bounding the work, report seed-ready times, peak RAM and frame-time distributions from interleaved base and head runs with the same save, settings and build, including Steam Deck and a session with other content readers. Include the mod disabled and enabled.

Testing: Read-only review. git diff --check passed. No build, script or test was run. The reported fake-game sweep and its game-mode assumptions could not be verified at this head. There are no in-game runs, CI results or measurements in the supplied record.
Overlaps and merge order: Depends on Longshot #1958. Land the Modding API first, then update this PR against the merged base and rerun the checks. No other overlap was established from the checkout.


Automated code review by BeanBot (gpt-6.2-sol), on the pull request's opening. Nothing was run; a maintainer decides.

**Recommendation: Request changes.** The randomizer cannot load from a clean checkout because its generator is missing. The placements reader can report failed reads as successful, and the mod can change a save with rewards that differ from the seed it verified. The placements API would let mods discover checks from the player's own game at runtime. The randomizer is a useful first client, but it needs a complete generator, proof that the game delivers what the solver verifies, and save-safe activation. The API also needs bounded delivery and reliable error reporting. **Does it help:** Unproven. The PR reports 54 fake-game seeds across nine presets and two modes, without a run commit or results. The reported sweep cannot be reproduced from this head because its generator imports are missing. There is no in-game result, seed generation time, or frame-time comparison. **Other systems:** Other content readers share the same queue and now scan Banc on their first row or text read. Placements also exercise shared RomFS and zstd code, and large answers are parsed on the game thread before the settlement budget is checked again. Rewards and grants affect save slots and other mods using the reward claim. The diff has no renderer, VRAM, audio or input changes. Evidence on Windows, Linux and Steam Deck is missing. **Switches:** No RX_, HOS_ or LONGSHOT_ switches. The example adds 30 Mods-tab settings. Seed defaults to empty, preset to S9 Tournament, shuffle to game mode, and race to off. Logic, guarantee, bridge, Ganon's boss key, forest, Door of Time, fountain, fortress, Kakariko gate, small keys, boss keys, maps and compasses, Kokiri Sword, songs, tokens, freestanding items, shops, scrubs, cows, item pool, starting age, hint distribution and hint text default to Preset. Both count overrides default to 0 for Preset, trials to −1 for Preset, tricks, exclusions and starting items to 'preset', and custom settings to '{}'. Show hash and show hints default to on. The empty seed and spoiler-only choices need to honor their stated effects. Count controls should use the selected condition's supported range. **Reviewed:** 7117cab56c2d against fa574bf340aa: Read the repository AGENTS.md, CONTRIBUTING.md, standards and Forgejo template, the five commits, all authored changes, and the placements reader and its BYML helpers in full. Read the relevant caller and dependency sections in RomFS, zstd, the runtime, rewards, items, storage, workers, and the test harnesses. Skimmed generated bindings and docs, and sampled the generated OoTR tables against the importer. The upstream OoTR source, Tooling checkout, and docs/review-checklist.md were unavailable. The advertised src/gen directory is absent at HEAD. No comments, reviews or recorded runs were supplied.; nothing run (a code review). **Checked and fine:** - All five commits carry a Co-Authored-By trailer. The diff leaves third_party and the renderer ABI untouched. - The new placement test fixtures use fabricated IDs. The imported tables identify their upstream source and carry its MIT notice. The diff contains no sampled placement census or game dump. - The read job, including its owned key, is published under the queue lock. The base folder list is initialized on the API thread before the queued placement jobs read it. - The low-level tests verify both BYML byte orders, a high-bit 64-bit ID, parameter filtering and missing Actors. The schema, feature lookup, native dispatch and TypeScript declarations all name the new API member. - The importer uses the Python standard library and reads a local checkout as data. Its output writes are documented. **Blocking** (to change before merge) 1. modapi/examples/randomizer/src/main.tsx:31: The runtime and tests import src/gen, but HEAD contains none of it. The root gen/ ignore rule matches that directory. A clean checkout fails to load the mod or run its generator tests. Rename the directory or narrow the ignore rule, and commit the complete generator for review. 2. host/modapi/content_read.c:256: A failed layer is logged and then returned as success. If Dynamic is valid and Static fails to decompress, callers see a partial map. If encoding fails mid-actor, the accumulated JSON can be invalid. Final allocation failure also looks like an absent map. Reject failed reads and allocations, discard partial output, and reserve null for a genuinely absent map. 3. host/mod_read.c:446: Translate is serialized as any BYML array. A replacement file with two coordinates, a string coordinate or NaN can yield a successful result that violates the public numeric triple type. Validate the container and its three finite numbers. Also exclude zero hashes, which the promised reward API rejects. 4. host/modapi/content_read.c:113: Discovery keeps only the first 64 base Banc entries, including non-directories, then caches the result for the process. A map in a later folder or a folder supplied only by a RomFS layer is reported absent. Find files through the merged namespace, handle listing failures explicitly, and move Banc discovery off the shared first row or text call. 5. host/modapi/content_read.c:247: The base path is silently truncated if the installation root and relative path exceed the fixed buffer. Dictionary-backed reads also use zs.c's MAX_PATH buffer for the dictionary and its unchecked fseek on Windows. A map can be listed but fail to read in a deep install. Reuse checked UTF-8 path construction and file reads that check the 64-bit seek result. 6. host/modapi/content_read.c:246: Placement discovery now resolves the same layered Banc files that game threads open. RomFS's resolver reads and writes the shared RomfsPath.logged field without synchronization. Make the first-log claim atomic or keep mutable logging state out of the published path table. 7. host/modapi/content_read.c:253: The zstd helper marks first use before finishing dictionary initialization. A simultaneous texture or content read can observe an incomplete dictionary table or race its writes. The placements worker adds another caller. Publish the dictionaries only after initialization completes and make other callers wait for completion. 8. host/modapi/content_read.c:274: One answer can exceed 8 MiB with framing and is copied into a settlement and parsed in one operation on the game thread. A dense map or repeated calls can also accumulate large native settlements before the JS heap limit applies. Page or cap delivery to an enforceable per-frame size, bound pending bytes, and avoid the extra whole-result copy. 9. modapi/packages/mod-tools/lib/fake-content.js:925: The fake reader ignores content.place outputs. A mod can read its own placement in the game and get null or an incomplete list in its tests. Merge the fake content plan using the same map, age, layer and hash rules. Use own-key lookup too, so a valid absent map name such as constructor returns null. 10. modapi/examples/randomizer/src/main.tsx:86: The mod queues every map, treats rejected reads as missing maps, and posts all raw placements in one message. A busy reader can make the seed silently depend on which requests succeeded, and a payload above the worker's 1 MiB cap stops binding. Read a bounded number at a time, send compact batches, and fail explicitly on read or messaging errors. 11. modapi/examples/randomizer/src/main.tsx:145: Full mode is advertised as spoiler-only, but its seed follows the same reward and starting-grant path. Choosing it can change chests and inventory with a fill the game cannot carry out. Give full mode a separate completion path that only presents or stores its result. 12. modapi/examples/randomizer/src/game/bind.ts:132: Equal counts alone do not establish that physical placements are the solver's checks. An unrelated drop of the same kind can stand in for a missing chest. Hash sorting and an all-locations reachability rule cannot establish that the physical drop persists or is reachable at that stage. Bind only checks with verified identity and reward behavior, and keep uncertain bindings out of the fill. 13. modapi/examples/randomizer/src/main.tsx:103: The seed input includes logical names and groups but no fingerprint of their physical assignments. If a map mod changes which chest IDs sort first in a two-chest group, the input and race icons can stay the same while the rewards swap physical locations. Hash the canonical game binding along with the effective settings and placements. 14. modapi/examples/randomizer/src/main.tsx:130: A verified item with no actor mapping is silently skipped. The chest keeps its vanilla item, yet the seed is logged as beatable. Starting grants skip unmapped items too, including the Zelda's Letter item named by the S9 test. Validate that every logical reward and starting item can be delivered with its required semantics before accepting the seed. 15. modapi/examples/randomizer/src/main.tsx:128: Progressive rewards are assigned when the chest actor loads. Two Hookshot rewards in an already loaded map can both give the first tier because the rule update affects only later actor loads. Counting tiers held also assumes upgraded items leave previous tiers in inventory. Resolve progression when collected from the actual upgrade state, or exclude progressive replacements until the API supports them. 16. modapi/examples/randomizer/src/main.tsx:148: Rewards are published in batches while play can proceed. A player can open a chest before its rule is installed, and if a rule addition throws after earlier batches, the partial set remains and later work can still announce readiness. Gate play and activation on complete installation, or abort and remove all installed rules on failure. 17. modapi/examples/randomizer/src/main.tsx:185: A save from another seed gets a warning after process-wide rewards are already installed. Its unopened chests can give seed B's items after it consumed seed A's progression checks. Validate the active playthrough before reward rules can affect its actors and remove rules on an incompatible load. Require an explicit fresh start before adopting an unmarked save. 18. modapi/examples/randomizer/src/main.tsx:194: The grant key identifies a slot number and seed rather than a playthrough. Deleting that slot and starting the same seed again leaves the old grant keys in the journal. Queued grants can also be delivered to the next active save because the journal is not slot-scoped. Tie grants and their acknowledgement to a persistent playthrough identity, with delivery restricted to that playthrough. 19. modapi/examples/randomizer/src/main.tsx:205: A new game can be in play with save.slot() still null. The update latches started before the grant function returns early. The first save sets the slot without a loaded event, so the player can go without the starting kit until a reload. Use a starting-grant state that waits for a usable playthrough and retries until delivery is confirmed. 20. modapi/examples/randomizer/src/main.tsx:214: checkedIds is shared across all slots and survives rollback. If a player opens a chest, reloads a save from before it, and opens it again, the mod suppresses the check and can skip updating its progressive rule. Rebuild check state and reward tiers from the active save on load. 21. modapi/examples/randomizer/src/main.tsx:174: Each spoiler write swallows errors, but the final jobs publish spoiler/last and log success even if a part was refused, for example when the global quota fills. Use byte-bounded parts, commit the index only after every part succeeds, and preserve the previous valid log on failure. 22. modapi/examples/randomizer/src/main.tsx:243: A stone's hint depends on its rank among actors currently visible within 300 units. Another stone streaming in can change that rank. Disabling hints also leaves the previous hint visible because the function returns before clearing it. Bind hints to stable stone identities and clear them on toggle-off and map changes. 23. modapi/examples/randomizer/src/main.tsx:53: An empty seed still starts the worker, imports generator data on both sides and installs frame callbacks. That incurs compilation, heap and worker costs for a setting advertised as inactive. Start the generator and gameplay callbacks only with a configured seed, and stop the worker after completion or failure. 24. modapi/examples/randomizer/test/game.test.ts:50: The fake reward assertion is skipped when its chosen item has no static actor mapping. The race test inspects storage as soon as ready is logged, before queued spoiler writes would run. Make the reward expectation unconditional for a controlled seed and drain pending work before asserting that race mode wrote nothing. 25. modapi/examples/randomizer/test/game.test.ts:88: The real-game test can skip a configured seed that takes more than 60 frames to bind or fails before logging a bound count. The placements feature test can also pass by registering a rule on the first non-chest drop without opening it. Distinguish an unset seed explicitly and require a real chest to deliver the expected item and emit its check. 26. tests/test_modrt.py:3477: The new runtime fixture uses the POSIX shim, Linux libraries and a stub that returns raw BYML instead of exercising the Windows directory and real zstd path. Its test entry also defaults to GCC instead of honoring CC. Wire the placement path into a portable fixture with a UTF-8 root, real compressed input and a merged-layer case, and use the selected compiler. **Waiting on the author** (a run, a measurement or a decision; no edit does it) - The PR opened without the required game run, and there is no exact-head check result. After the code issues are fixed, report Linux and Windows game runs covering mapped chest delivery and checks, progression, starting items, slot changes and spoiler behavior. Attach the repository check results for the resulting head. - The costs of startup, placement settlement and reward publication are unmeasured. After bounding the work, report seed-ready times, peak RAM and frame-time distributions from interleaved base and head runs with the same save, settings and build, including Steam Deck and a session with other content readers. Include the mod disabled and enabled. **Testing:** Read-only review. git diff --check passed. No build, script or test was run. The reported fake-game sweep and its game-mode assumptions could not be verified at this head. There are no in-game runs, CI results or measurements in the supplied record. **Overlaps and merge order:** Depends on Longshot #1958. Land the Modding API first, then update this PR against the merged base and rerun the checks. No other overlap was established from the checkout. --- <sub>Automated code review by BeanBot (gpt-6.2-sol), on the pull request's opening. Nothing was run; a maintainer decides.</sub>
host/mod_read.c Outdated
@ -399,0 +443,4 @@
else continue;
const int has_dynamic = dict_get(&x, a, "Dynamic", &dt, &dv) && dt == 0xC1;
if (param && !(has_dynamic && dict_get(&x, dv, param, &t, &tv))) continue;
char head[64];
Member

Validate three finite numeric coordinates before promising a numeric triple. Also exclude zero hashes, which rewards.add refuses.

Validate three finite numeric coordinates before promising a numeric triple. Also exclude zero hashes, which rewards.add refuses.
@ -110,6 +110,7 @@ static void list_names(void) {
if (!hos_romfs_dir) return;
list_dir("RSDB", ".rstbl.byml.zs", tables, &ntables);
list_dir("Mals", ".sarc.zs", archives, &narchives);
list_dir("Banc", "", bancs, &nbancs); /* (the maps' folders: Banc/KokiriForest, Banc/DreamWorld...; "." and ".." too) */
Member

Discover Banc files in the merged namespace. The first 64 base entries can omit valid maps, and row and text reads now pay for this scan.

Discover Banc files in the merged namespace. The first 64 base entries can omit valid maps, and row and text reads now pay for this scan.
@ -234,0 +243,4 @@
if (!strcmp(base, ".") || !strcmp(base, "..")) continue;
char rel[300], path[1100];
snprintf(rel, sizeof rel, "%.150s/%.63s_%s.bcett.byml.zs", bancs[b], j->key, layers[l]);
if (!hos_romfs_resolve(rel, path, sizeof path)) {
Member

The reader and game threads can resolve the same layered path concurrently. Synchronize RomfsPath.logged.

The reader and game threads can resolve the same layered path concurrently. Synchronize RomfsPath.logged.
@ -234,0 +244,4 @@
char rel[300], path[1100];
snprintf(rel, sizeof rel, "%.150s/%.63s_%s.bcett.byml.zs", bancs[b], j->key, layers[l]);
if (!hos_romfs_resolve(rel, path, sizeof path)) {
snprintf(path, sizeof path, "%s/%s", hos_romfs_dir ? hos_romfs_dir : ".", rel);
Member

Check path construction and the dictionary read for deep UTF-8 roots. Truncation currently looks like a missing map.

Check path construction and the dictionary read for deep UTF-8 roots. Truncation currently looks like a missing map.
@ -234,0 +250,4 @@
fclose(f);
}
size_t n = 0;
uint8_t *data = hos_zs_read(path, &n, MODREAD_FILE);
Member

The zstd dictionary initializer must finish before concurrent readers traverse its table.

The zstd dictionary initializer must finish before concurrent readers traverse its table.
@ -234,0 +253,4 @@
uint8_t *data = hos_zs_read(path, &n, MODREAD_FILE);
files++;
if (!data || mod_read_placements_json(data, n, names[l], j->file[0] ? j->file : NULL, &json, &len, &count, &why) < 0)
hos_log_at(MODRT_LOG_WARN, "mods: content.read: %s: %s", rel, data ? why : "it cannot be read or decompressed");
Member

Reject a failed layer and discard the accumulated JSON. Logging and continuing can return partial or invalid data.

Reject a failed layer and discard the accumulated JSON. Logging and continuing can return partial or invalid data.
@ -234,0 +271,4 @@
}
free(json);
hos_log_at(MODRT_LOG_INFO, "mods: content.read: map %s: %d placements", j->key, count);
settle_json(j->promise, out, out ? len + 2 : 0);
Member

Bound pending bytes and page large results. A single answer is parsed on the game thread before the settlement budget is checked again.

Bound pending bytes and page large results. A single answer is parsed on the game thread before the settlement budget is checked again.
@ -0,0 +129,4 @@
for (const kind of [...kinds].sort()) {
const ls = (logical.get(kind) ?? []).slice().sort((a, b) => (a.name < b.name ? -1 : 1));
const ps = (physical.get(kind) ?? []).slice().sort((a, b) => (a.p.id < b.p.id ? -1 : 1));
const bound = ls.length > 0 && ls.length === ps.length;
Member

Equal counts can pair an incidental drop with a missing chest. Require evidence that each physical placement is the persistent check the solver models.

Equal counts can pair an incidental drop with a missing chest. Require evidence that each physical placement is the persistent check the solver models.
@ -0,0 +28,4 @@
import { inputOf } from "./game/input.js";
import { ALL_MAPS, type Binding, type DropLocation, type Placed } from "./game/bind.js";
import { ITEM_ACTORS, progressive, tierActor } from "./game/names.js";
import { plain } from "./gen/hints.js";
Member

HEAD is missing every src/gen module. The root gen/ ignore rule matches this path. Commit the generator under a tracked directory.

HEAD is missing every src/gen module. The root gen/ ignore rule matches this path. Commit the generator under a tracked directory.
@ -0,0 +50,4 @@
// ---- 1 to 3: read the game's checks, bind them, make the seed ----
const worker = start("src/worker.ts");
Member

Defer the worker and gameplay work until a seed is configured. Release the worker after completion or failure.

Defer the worker and gameplay work until a seed is configured. Release the worker after completion or failure.
@ -0,0 +83,4 @@
return;
}
const placements: Record<string, Placed[] | null> = {};
const answers = await Promise.all(ALL_MAPS.map((map) => read.placements(map, { param: "Drop__DropActor" }).catch(() => null)));
Member

Read and send compact batches within the worker limit. Propagate read errors instead of counting them as absent maps.

Read and send compact batches within the worker limit. Propagate read errors instead of counting them as absent maps.
@ -0,0 +100,4 @@
log.info(`bound ${view.bound} of ${view.locations} OoTR drop locations to the game's placements (${b.groups.length} groups)`);
const unbound = b.report.filter((r) => !r.bound && r.locations && r.placements);
if (unbound.length) log.debug(`counts that differ (not bound): ${unbound.map((r) => `${r.group} ${r.kind} ${r.locations}/${r.placements}`).join(", ")}`);
input.available = Object.keys(b.placement);
Member

Include the physical binding in the race fingerprint. The same logical names and groups can map to different chests.

Include the physical binding in the race fingerprint. The same logical names and groups can map to different chests.
@ -0,0 +125,4 @@
function rewardFor(location: string, item: string): void {
const id = binding?.placement[location];
if (!id) return;
const learned = new Map(Object.entries(binding!.learned));
Member

Two progressive chests in a loaded room can both give the same tier. Resolve at collection or leave those rewards vanilla.

Two progressive chests in a loaded room can both give the same tier. Resolve at collection or leave those rewards vanilla.
@ -0,0 +127,4 @@
if (!id) return;
const learned = new Map(Object.entries(binding!.learned));
const actor = tierActor(item, progressive(item) ? tiersHeld(item) : 0, learned);
if (!actor) { log.debug(`${location}: ${item} has no name in the game; its chest keeps its own`); return; }
Member

Validate item delivery before accepting the seed. A silently skipped reward or grant invalidates the solver's result.

Validate item delivery before accepting the seed. A silently skipped reward or grant invalidates the solver's result.
@ -0,0 +142,4 @@
view.warnings.push(...s.warnings, ...s.adjusted.map((a) => `${a.setting}: ${JSON.stringify(a.to)} (${a.why})`));
for (const [loc, id] of Object.entries(binding?.placement ?? {})) placementOf.set(id, loc);
for (const p of s.placements) itemAt.set(p.location, p.item);
const shuffled = s.placements.filter((p) => p.kind === "shuffled" && binding?.placement[p.location]);
Member

Full mode is labeled spoiler-only. Skip all rewards and starting grants in that mode.

Full mode is labeled spoiler-only. Skip all rewards and starting grants in that mode.
@ -0,0 +145,4 @@
const shuffled = s.placements.filter((p) => p.kind === "shuffled" && binding?.placement[p.location]);
for (let i = 0; i < shuffled.length; i += 32) {
const batch = shuffled.slice(i, i + 32);
work.push(() => { for (const p of batch) rewardFor(p.location, p.item); view.rewards = rules.size; });
Member

Keep play gated until the complete reward set is installed. Roll back the rules if any addition fails.

Keep play gated until the complete reward set is installed. Roll back the rules if any addition fails.
@ -0,0 +182,4 @@
if (slot === null || state().mode !== "play") return;
const mine = save.get("seed") as string | null;
if (mine && mine !== seed.digest) {
view.warnings.push("this save was started with another seed: its items are that seed's");
Member

Check save compatibility before installing rewards. The warning currently leaves the other seed's rules active.

Check save compatibility before installing rewards. The warning currently leaves the other seed's rules active.
@ -0,0 +171,4 @@
};
parts.forEach((p, i) => work.push(set(`${key}/${i}`, p)));
work.push(set(`${key}/parts`, parts.length));
work.push(set("spoiler/last", key));
Member

Publish spoiler/last only after every part succeeds. A quota failure currently leaves a broken log advertised as complete.

Publish spoiler/last only after every part succeeds. A quota failure currently leaves a broken log advertised as complete.
@ -0,0 +191,4 @@
for (let i = 0; i < count; i++) {
const actor = tierActor(item, i);
if (!actor) continue;
grant(actor, { key: `start:${seed.digest.slice(0, 16)}:${slot}:${item}:${i}` });
Member

A slot number can be reused, and queued grants can follow a slot switch. Scope delivery and deduplication to the playthrough.

A slot number can be reused, and queued grants can follow a slot switch. Scope delivery and deduplication to the playthrough.
@ -0,0 +202,4 @@
let started = false;
update.on(() => {
for (let i = 0; i < 2 && work.length; i++) work.shift()!();
if (!started && seed && state().mode === "play") { started = true; startPlaythrough(); }
Member

The start latch can be set before a fresh game has a slot. Keep attempting initialization until the active playthrough can receive its kit.

The start latch can be set before a fresh game has a slot. Keep attempting initialization until the active playthrough can receive its kit.
@ -0,0 +211,4 @@
checked.on((e) => {
if (!seed || e.resets) return;
const loc = placementOf.get(e.id);
if (!loc || checkedIds.has(e.id)) return;
Member

Rebuild checked IDs and reward tiers when a save loads. Reloading before a check must make it eligible again.

Rebuild checked IDs and reward tiers when a save loads. Reloading before a check must make it eligible again.
@ -0,0 +240,4 @@
const stones = list({ near: { position: me, within: 4 }, nearest: true, limit: 16 }).filter((a) => /Gossip/i.test(a.type));
if (!stones.length) return;
// (the map's stones in the order of their positions; each shows one of the seed's hints, the same one every time)
const all = list({ near: { position: me, within: 300 } }).filter((a) => /Gossip/i.test(a.type) && a.position)
Member

Key hints by stable stone ID and clear the displayed hint when disabled or changing maps.

Key hints by stable stone ID and clear the displayed hint when disabled or changing maps.
@ -0,0 +47,4 @@
assert.ok(item, "the spoiler names the sword chest's item");
// (S9 Tournament shuffles the Kokiri Sword: the chest drops the seed's item, as the game's actor of it)
const actor = ITEM_ACTORS[item.replace(/ \(\d+ rupees\)$/, "")]?.[0];
if (actor) assert.equal(game.dropFor(SWORD, "Weapon_Sword_Kokiri"), actor, `${item} as ${actor}`);
Member

Assert a known reward unconditionally and let spoiler work finish before asserting that race mode wrote nothing.

Assert a known reward unconditionally and let spoiler work finish before asserting that race mode wrote nothing.
@ -0,0 +85,4 @@
"in the game: the maps bind, the seed is made and its rewards set": { game: true, timeout: 36000, run: async (game, assert) => {
await game.loadSave("kokiri-start");
await game.frames(60);
if (!game.logs.some((l) => l.text.startsWith("bound ") || l.text.includes("reading")) && !game.logs.some((l) => l.text.startsWith("seed ")))
Member

Detect the seed setting directly. Slow binding or startup failure should fail the game test. Check an actual chest's delivered item.

Detect the seed setting directly. Slow binding or startup failure should fail the game test. Check an actual chest's delivered item.
@ -903,0 +922,4 @@
const list = this.placed[map];
if (!list)
return null;
void outputs;
Member

Apply content.place outputs in the fake game and use own-key map lookup.

Apply content.place outputs in the fake game and use own-key map lookup.
@ -3471,6 +3474,11 @@ def check_content_read(exe, tmp, c):
fr = {"ActorMsg/Enemy.msbt": mc.msbt([("Enemy_Octorok_Name", "Octorok rouge")])}
(rom / "Mals" / "USen.Product.1.sarc.zs").write_bytes(mc.arch.sarc(en))
(rom / "Mals" / "EUfr.Product.1.sarc.zs").write_bytes(mc.arch.sarc(fr))
(rom / "Banc" / "KokiriForest").mkdir(parents=True)
Member

Cover the real compressed and merged path, including the Windows UTF-8 directory lookup, with a portable harness that honors CC.

Cover the real compressed and merged path, including the Windows UTF-8 directory lookup, with a portable harness that honors CC.
HumanBean force-pushed feat/randomizer from 7117cab56c
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request_target) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 30s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 35s
Third-party parts / As their notes say (pull_request) Successful in 47s
to 7193c39785
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 25s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 32s
Third-party parts / As their notes say (pull_request) Successful in 39s
2026-11-21 01:51:52 +00:00
Compare
HumanBean force-pushed feat/randomizer from 7193c39785
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 25s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 32s
Third-party parts / As their notes say (pull_request) Successful in 39s
to 0e191c6249
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 17s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 27s
Third-party parts / As their notes say (pull_request) Successful in 31s
2026-11-21 02:49:42 +00:00
Compare
HumanBean force-pushed feat/randomizer from 0e191c6249
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 17s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 27s
Third-party parts / As their notes say (pull_request) Successful in 31s
to d4c0763424
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 18s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 21s
Third-party parts / As their notes say (pull_request) Successful in 23s
2026-11-21 05:05:34 +00:00
Compare
HumanBean force-pushed feat/randomizer from d4c0763424
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 18s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 21s
Third-party parts / As their notes say (pull_request) Successful in 23s
to 4cfda780a5
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 17s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 23s
Third-party parts / As their notes say (pull_request) Successful in 37s
2026-11-21 05:27:59 +00:00
Compare
HumanBean force-pushed feat/randomizer from 4cfda780a5
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 17s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 23s
Third-party parts / As their notes say (pull_request) Successful in 37s
to c577a2ff9b
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 15s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 20s
Third-party parts / As their notes say (pull_request) Successful in 26s
2026-11-21 05:29:31 +00:00
Compare
HumanBean force-pushed feat/randomizer from c577a2ff9b
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 15s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 20s
Third-party parts / As their notes say (pull_request) Successful in 26s
to 4eb8a52638
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 40s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 50s
Third-party parts / As their notes say (pull_request) Successful in 1m19s
2026-11-21 06:13:43 +00:00
Compare
HumanBean force-pushed feat/randomizer from 4eb8a52638
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 40s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 50s
Third-party parts / As their notes say (pull_request) Successful in 1m19s
to 5eda12542c
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 15s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 23s
Third-party parts / As their notes say (pull_request) Successful in 35s
2026-11-21 06:22:40 +00:00
Compare
fix(mods): the randomizer's generator and settings tests are the fake game's only: a whole seed at once ran past a frame's 5 ms in the game's test mode, and three such faults stopped the mod
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 16s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 17s
Third-party parts / As their notes say (pull_request) Successful in 24s
44ae1b2fba
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8e3nBDxBnhodLguFyEt
fix(mods): the randomizer keeps a dungeon's small keys the game's when any of its drops is unbound, so game mode makes a seed from a partial binding; its no-seed test is the fake game's only (it reads game.rewards)
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 9s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 12s
Third-party parts / As their notes say (pull_request) Successful in 15s
61e62c533f
The game run bound 77 of OoTR's 315 drops (counts per map and kind differ); the Spirit Temple's keys then had only the
few bound places behind its own doors, and no seed was made in 20 tries. Random 77-drop bindings over 30 seeds: 25 made
before, 30 after; every offered preset with all, 200 and 77 drops bound passes the checker.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8e3nBDxBnhodLguFyEt
fix(mods): the randomizer's starting-items line counts the grants queued, not those its key had given before (the game run logged 22 twice for one playthrough)
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 20s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 26s
Third-party parts / As their notes say (pull_request) Successful in 34s
ab66188a3a
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8e3nBDxBnhodLguFyEt
HumanBean force-pushed feat/randomizer from ab66188a3a
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 20s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 26s
Third-party parts / As their notes say (pull_request) Successful in 34s
to 5e43193957
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 26s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 35s
Third-party parts / As their notes say (pull_request) Successful in 44s
2026-11-21 08:11:26 +00:00
Compare
HumanBean force-pushed feat/randomizer from 5e43193957
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 26s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 35s
Third-party parts / As their notes say (pull_request) Successful in 44s
to f0cc576c76
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 23s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 27s
Third-party parts / As their notes say (pull_request) Successful in 32s
2026-11-21 08:19:30 +00:00
Compare
Some checks failed
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 23s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 27s
Third-party parts / As their notes say (pull_request) Successful in 32s
This pull request is marked as a work in progress.
View command line instructions

Checkout

From your project repository, check out a new branch and test the changes.
git fetch -u 'feat/randomizer:HumanBean-feat/randomizer'
git switch 'HumanBean-feat/randomizer'

Merge

Merge the changes and update on Forgejo.

Warning: The "Autodetect manual merge" setting is not enabled for this repository, you will have to mark this pull request as manually merged afterwards.

git switch 'mod-api'
git merge --no-ff 'HumanBean-feat/randomizer'
git switch 'HumanBean-feat/randomizer'
git rebase 'mod-api'
git switch 'mod-api'
git merge --ff-only 'HumanBean-feat/randomizer'
git switch 'HumanBean-feat/randomizer'
git rebase 'mod-api'
git switch 'mod-api'
git merge --no-ff 'HumanBean-feat/randomizer'
git switch 'mod-api'
git merge --squash 'HumanBean-feat/randomizer'
git switch 'mod-api'
git merge --ff-only 'HumanBean-feat/randomizer'
git switch 'mod-api'
git merge 'HumanBean-feat/randomizer'
git push origin 'mod-api'
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
2 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
MagicBeanDev/Longshot!1959
No description provided.