Research log
Status: Verified — how this documentation’s measurements were obtained
The technical statements here rest not on official developer documentation but on production-bundle analysis. This page records methodology and evidence so future builds can be re-verified.
Core principles
- Direct production-build retrieval — raw main.js and chunks fetched from both production and preprod, identity verified by hash
- Verbatim respect — headers, error messages and i18n strings are quoted as-is, never paraphrased
- Reproduced measurement — derived values like the registry are recomputed by running the engine’s own modules in a harness (real output, not a lookalike re-implementation)
- Named builds — every measurement carries a build id and date
Build history referenced by this documentation
| Build | API | Date | Note |
|---|---|---|---|
szby6g | v3-era | 2026-10-17 | first allow/strict static analysis (111 game definitions) |
3ymzdd | v3 | 2026-10-20 | the v3 guide’s evidence build |
sjr0p7 | v3-era | 2026-10-17 | drift-comparison baseline |
cnty9 | v4 | 2026-10-27 | v4 guide & 905-name catalog evidence |
6pf78 / tcwczu | v4 | 2026-10-31 | v4 freeze confirmation (Halloween event + Skyblock rework) |
ab90qa | v5 | 2026-11-03 | v5 launch build; original v5 baseline. prod = preprod |
xkg8x | v5 | 2026-11-04 | registry drift (+10 names); preprod 2lcqcl measured identical on every shader-pack surface |
bny9u | v5 | 2026-11-05 | registry drift (+5 names); prod = preprod |
v98qt8 | v5 | 2026-11-08 | registry drift (−5 net), games 115 → 116, and the texture-constant folding regression (Texture registry) |
bbb6l5 | v5 | 2026-11-20 | current; this documentation’s baseline. prod = preprod. The first v5-era API surface change: the shadow API (bloxdShadow() / bloxdAirShadow() / bloxdShadowDirection()), the 80 B uniform, keepCircleShadows, the folding regression fixed, and the registry’s largest step (+423/−0, Atomic Age content + transition slots). Fingerprint match-verified against the shipped pre-built atlas |
Key methodology
Engine-header restoration
Rendering the header template from the live build’s validator with its live interpolations (apiVersion, the uniform reads via bloxdPackInputs) (Engine header).
Texture-registry recomputation
Running the engine’s registry-building machinery against the shipped block registry’s texture data in a mini webpack harness; recording the output (named 1,017 / aliases 776 / atlas 1,650 on the current build bbb6l5, 2026-11-20 — the engine’s own folding, fingerprint 32451195 match-verified against the shipped pre-built atlas; 1,227 / 906-intended / 1,047-as-shipped on v98qt8; 1,232 / 911 on bny9u; 1,220 / 906 on xkg8x; 1,210 / 896 at the v5 launch) (Texture registry).
Animation intervals come from the animation table (per-layer numAnimStates / animationInterval) — three interval values exist: 200 / 250 / 300 ms, and the new transition spellings carry their own (lava0~lava1~50) (the catalog).
Error-message capture
Loader messages verbatim from the live i18n dictionary; validator messages from the validation module’s string literals.
allow / strict static analysis
Statically analyzing all 111 game definitions (szby6g-era): shaderPackPolicy "allow" literals — 8 direct + 4 base-variable = 12 titles (26 entries), spread-inheritance ({...tb,...}) tracked too. On the current build the games registry is evaluated in a harness instead: 119 definitions, 28 allow entries — same 12 titles (Policy).
Uniform / pipeline restoration
The BloxdPackInputs field layout from the live uniform-layout module; the texture 3-variant codegen from the WebGPU generator (Texture sampling).
Drift observations
| Comparison | Period | Result |
|---|---|---|
| sjr0p7 → cnty9 | 10 days | 98 added, 79 removed; 779/1121 names changed id (69%) |
| cnty9 → ab90qa | 7 days | +1 (scorpion_eggs), −10 (double chest ×8, spawner sides ×2); 334/905 names changed id |
| ab90qa → xkg8x | 1 day | +10 (all v5-launch removals returned: double chest ×8, spawner sides ×2); 376/896 names changed id; none removed |
| xkg8x → bny9u | 1 day | +5 (Trading Terminal ×3, Stone Generator ×2); 864/906 names changed id (95%); none removed; anim families 42 → 43 |
| bny9u → v98qt8 | 3 days | +1 hay_block_top, −6 (Furnace/Workbench/Artisan Bench became models); 716/905 names changed id (79%); families unchanged; folding regression — engine resolves 1,047 unfolded names |
| v98qt8 → bbb6l5 | 12 days | +423 (111 usable: Atomic Age content; 204 transition spellings), −0; 863/906 names changed id (95%); rotation families 57 → 64, anim families 43 → 71; folding regression fixed; the shadow API arrived |
Unverified / speculative
- Pixel-accurate shadow maps as “future plans” is no longer speculative —
bbb6l5shipped the real shadow system (bloxdShadow()); the old interpretation (inferred from the unusedvar shadow: f32 = 1.;in the lighting template) turned out correct - The allow/strict split can change per build (12-title identity re-confirmed through
bbb6l5— 28 allow entries over the same 12 titles; future games unknown) bloxdFragmentBlockPosition(out of the v5 header): signature-table entry and runtime injection both confirmed live — calls validate and compile; whether the engine team intends to keep it is unknown
When to re-verify
On detecting a game update (a new build id): re-check the header, the apiVersion constant, recompute the registry, and re-confirm the allow/strict list. Detailed monitoring methodology is outside this documentation’s scope; the pack-visible surfaces (header text, loader messages, registry fingerprint, allow titles) are the re-verification anchors.