Skip to content

Project Status

github-actions[bot] edited this page Sep 20, 2026 · 288 revisions

Project status

Last consolidated: 2026-09-20.

Production pipeline

System Status Evidence and limit
Clean-ROM validation and separate output Production Implementation/CI-confirmed; exact USA Rev. 0 hash only
Header checksum update Production Implementation/CI-confirmed
Compact eight-stage selector Production Runtime-confirmed A-button route; compact index maps to native 0,1,2,3,4,5,8,9
1 KiB arena reservation and native payload Production Runtime-confirmed load/execute; guarded bounds and tests
Ordinary-pickup persistence Production beta All eight stages runtime-confirmed at representative locations; all 84 are statically cataloged, not individually exhausted
Stage-local pickup randomization Production beta All 84 implemented and tested over 250 seeds; seed TEST153 first Fire location predicted/rendered/awarded Shield; full run pending
Pickup-driven XP progression Production beta Diagnostic B runtime-confirmed V2 allocation, XP restore, 85/258 rewards, Temple -> Wind transition and title -> Fire retention; nine-reward/full-tier run pending
Four inventory boxes Production beta Runtime-confirmed switching, transition preservation, stage-key masking; Game Over/reset edge still pending
BOX n OF 4 HUD text Production Runtime-confirmed native text path
Boot branding and seed phrase Production Guarded and CI-confirmed; phrase uses a dedicated deterministic namespace
Logo and selector-save bypasses Production Runtime-confirmed, legal screen and later/manual saves preserved
Outfit recoloring Production Runtime-confirmed palette modes; static TLUT source transform
Browser/CLI shared patch core Production beta CI builds/tests; release UX continues to mature

Confirmed proofs, not product features

Proof Result Remaining work
Same-stage pickup substitution Full identity tuple changes award and art Generalized in current stage-local mode
Fire foreign Prison-key import Relocated/expanded resource file plus dedicated callback works Production resource planner, allocation policy, compatibility matrix
Prison extension-selector Potion coexistence Out-of-stock-range selector resolves appended Water Potion bundle while vanilla Herbs remain resident and functional Generalize into guarded production planner
Health urn external-to-embedded conversion Proof G showed external IDs 0x28F..0x292 glitch in Prison; native type-4 conversion is now static-confirmed and Proof H is built Manual runtime validation of Proof H
Fire ordinary enemy 0x0A -> 0x09 Spawns and plays normally Production policy and broader coverage
Temple monk imported into Fire Model loads, enemy moves/fights/dies Death/despawn presentation missing; arbitrary rosters unresolved
Temple XP progression proof Three proof pickups plus one vanilla Herbs control established no-combat-XP and pickup-driven tiers Generalized into the production Diagnostic B architecture; full nine-tier coverage and final art remain
Reverse Elbow host/action proof + genuine MKT import branch v6 established repeatable scheduler/action stability; v08/A1.7 runtime-confirmed a coherent genuine MKT SCCOMBO10 fighter sprite rendered in MKMSZ with isolated donor palette and clean stock restoration Full donor 10→11→12→11→10 animation; real no-repel shim; translated donor strikes/reactions; complete move port

Latest runtime findings

The Temple XP proof is confirmed: combat does not award XP; collecting the proof rewards sets XP to 85 and then 258; the second threshold activates the next tier; nothing is added to inventory; ordinary pickups and progression rewards coexist. The proof model appears pale blue-grey and Herbs-like. The desired final presentation is a bright-blue body with a bronze handle.

A first production-layout attempt was Rejected / failed on 2026-09-20: Temple reached the Mission Objective screen and loaded stage music, then hung just before gameplay appeared. Production was rolled back to runtime V1.

Follow-up Diagnostic A is Runtime-confirmed: with the progression stage-entry restore helper bypassed, the same V2 allocation and generated progression callbacks loaded Temple normally. XP 85 and 258 rewards worked, special-move tiers unlocked, combo EXPERIENCE stayed suppressed, combat awarded no XP, and normal/progression Herbs coexisted with identical Herbs graphics.

Diagnostic B then removed only the early tier-evaluator call from the restore helper. It is Runtime-confirmed through Temple completion, Temple -> Wind with XP 258/two moves retained, and title-menu -> Fire with the same XP/moves retained. The original hang is therefore attributed to calling the native tier evaluator too early during stage initialization. Production now uses Diagnostic B behavior: restore XP at stage entry, but evaluate tiers only when a progression pickup is collected.

The behavioral Reverse Elbow branch is now historical host-action evidence rather than the intended final implementation. Later v8.1-v8.3 tests established usable high-speed directional motion and a proof-only forced crossover, but also exposed collision/jitter and freeze/strike mismatches inherent to recreating donor behavior with native approximations. The project has pivoted to a genuine MKT→MKMSZ compatibility layer. The single-frame import branch then resolved descriptor endianness/order, palette isolation/opacity, raw-wrapper type, and CI row-stride conversion. v08/A1.7 is runtime-confirmed: Back → Forward + Low Kick renders the genuine MKT Rev. 2 SCCOMBO10 Reptile/male-ninja frame coherently inside MKMSZ, with the true 42x95 geometry, donor palette, 44-byte aligned row stride, and clean restoration of stock Sub-Zero. This establishes genuine foreign fighter-sprite transplantation for this MKT resource family.

1.0 required scope

The 1.0 randomizer is not considered complete with stage-local shuffling alone. Required before 1.0:

  1. Global cross-stage item randomization. Items must be able to appear in other main stages, requiring a production resource-import/materialization planner rather than copying stage-local resource selectors verbatim.
  2. Global deterministic shuffling. Keep the current stable seeded Fisher-Yates approach, but apply it to the global logical item pool rather than eight independent stage pools.
  3. Whole-run solvability validation. Reject seeds whose progression graph cannot reach the required end-state. Use current stage/location requirements as source of truth; the legacy Lua fixed-point solver is a useful design reference but its location rules are stale/incomplete. Rejected attempts must be derived deterministically from the original seed plus an explicit attempt index, so regenerating the same seed always produces the same accepted layout even when several attempts are required.
  4. Randomizer HUD/UI. At minimum expose the run information needed to play the seed: check progress, required Power Upgrades/current progression status, inventory-box/page state, and useful pickup feedback. The legacy Lua overlay is a reference for semantics, not a native renderer implementation.
  5. Run-resource lifecycle. Determine and preserve HP, lives, and continues correctly across stage transitions and direct selector routes; define intended Game Over/new-run resets.
  6. Difficulty hard-lock. Very Hard must be enforced as a run invariant rather than only selected as a startup default.
  7. Final runtime coverage. Complete a representative full global seed, including all nine progression tiers and the major lifecycle boundaries.

Open Temple Map work: the legacy Lua global pool included the scripted Map as an 85th check. For 1.0, investigate separating the Map inventory reward from the Temple elevator/exit trigger so the elevator can still be raised correctly even when the Map item itself is shuffled elsewhere. Also prevent the Map item from being removed on the Temple -> Wind transition, which is stock behavior today.

Pending priorities

  1. Runtime-validate disposable Proof H, which converts Health urn's unavoidable external IDs 0x28F..0x292 into self-contained native type-4 embedded image blocks and imports one Health urn into Prison through extension selector 0x123C while five vanilla Herbs remain untouched. If confirmed, generalize the extension-selector + canonical-embedded-donor materializer into the production cross-stage planner.
  2. Replace stage-local generation with a global deterministic shuffle plus whole-run solver, including deterministic retry attempts and a deterministic per-seed required-Power-Upgrades target.
  3. Build the native randomizer HUD around the finalized global-run state.
  4. Resolve Temple Map trigger/item separation and cross-stage persistence; then HP/lives/continues persistence, Game Over/new-run reset, and Very Hard enforcement.
  5. Run full-seed 1.0 validation.
  6. Post-1.0: enemy randomization/resource compatibility, imported-enemy death/despawn, and Reverse Elbow.

Explicit exclusions

Bosses and scripted encounters are not ordinary-enemy entries. The Temple Map is not an ordinary pickup and needs an explicit 1.0 inclusion decision if it is to become a global check. Zero resource slots are logical selectors, not free physical storage. PS1 addresses are not N64 addresses. Proof ROM patches and archive handoffs are not silently part of the product pipeline. Enemy randomization and Reverse Elbow are not 1.0 blockers.

Clone this wiki locally