We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
There was an error while loading. Please reload this page.
defective_max_files: §4.4 note the Map Extender drop-in plugin Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
defective_max_files: §4.4 per-node Map Extender validation matrix (118 nodes: 2 ident + 8 FLOATEQ-full + 105 verified + 3 diff) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
pipeline_max: Map Extender cache decode (§10z-quatorze, MGF Part P) Document the reverse-engineered mapext198m3.dlm format: plugin object is empty of settings; functional UVW map is the LocalModData cache. Coverage and defective-files notes updated for the production path. Co-authored-by: Grok 4.5 <noreply@x.ai>
defective_max_files: note planned Map Extender decode session (cache completeness unverified — open exploration) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GizfXGXNmCMAoM9Gnk8fq9
defective_max_files: Map Extender §4.4 — decode-attempt status (none headless; 2012 ReplaceMapExt converted via wrapper cache to Unwrap UVW) + full affected-file table (87 files, 118 nodes) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GizfXGXNmCMAoM9Gnk8fq9
pipeline_max: zone coverage audit + VS2008/x87 precision ceiling design §10h-bis: 1201-file zone sweep is the complete export target set (matches nel_ligo_export.ms token rule exactly; 30 non-targets, 0 direct corpus). x64 1067/131/912 vs VS2008/x87 1103/95/872 — x87 closes 36 files (codegen-closeable); long-double pack tried and reverted (matches only the downstream step -> 575/623/3418 regression; whole chain must match). defective §1.8: the 3 protocol-violating ligo filenames (2 mis-named zonematerial-, corrupted_moor_meta_goo dev file). Co-Authored-By: GLM 5.2 <noreply@z.ai>
Pipeline_max shape session 2026-07-09: LOD_MRM gating + CMeshMRMSkinned invariant PMD §10z-huit: two shape-tool hardening fixes, both surfaced by the 3-file export-failure class §10z-sept left standing. Corpus export failures 3 → 0. - LOD_MRM gating: dropped the SkinWeights override on wantMrm. Two ship files (ca_spaceship2, ship_tank_karavan) that force-routed to CMeshMRMSkinned and blew the post-MRM invariant now export as plain CMesh with SkinWeights, matching the reference plugin's routing (export_mesh.cpp:360, LOD_MRM alone). - CMeshMRMSkinnedGeom::compileRunTime invariant surfaced through a proper status flag (_RuntimeCompiled + isRuntimeCompiled accessor) instead of an nlassert abort. Both exporters check it and skip with an artist-facing message on failure. Engine load no longer hard-aborts on a broken shape either. Defective files §1.7: ge_hom_acc_gauntlet_kami — the remaining single case where the invariant genuinely fires, documented with root-cause attribution to the Physique decode noise class (input 642 verts, ref post-MRM 1461, ours would be 6514 due to 176/926 unresolved bone entries and 50 root-fallback verts causing MRM to split at every noisy weight boundary). Closing PMD §10z-quat / §10z-six / MGF §M.3 Physique cross- link record layout likely closes this file too.
Reclassify monster twin files as intentional anchors Correction from Kaetemi: the _big/_small/_slim/_selle monster twin files are NOT stale post-reference-export edits; they are/were active game data serving as rescaling reference points. The scale delta from the base rig is the file's PURPOSE, not a drift. The original PMD §10 "reference-era mismatches" observation still holds (base-rig subset matches the reference bit-exact; scaled variants don't), but the interpretation is now "intentional anchor files" rather than "post-export drift". skel_corpus.py's era-matched filter is really an anchor-file filter that keeps the base-rig subset's bit-match statistics from being diluted by the scale-delta variants. The _ref_scale shadow-rig case (never exported, genuine editing scaffolding) stays distinct from the runtime anchors above. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HjbyJgJ5kh3PKKzPwSWw1U
Note zero XRef reference errors in defective_max_files A corpus-wide sweep of ig+cmb+ligo produces zero XRef reference errors (unresolvable paths, missing target nodes, malformed 0x0170 records). Every XRef in the corpus resolves cleanly. Section 3 now says so explicitly, lists the exact warning messages the resolver would emit if a broken XRef existed, and distinguishes them from the `ERROR: not an OLE compound file: ...` class (which comes from CStorageOleIn::open on the main input path — a caller mistake like passing landscape/ligo/desert/max/zonematerial-fond-ilot_4_mz.max when the file actually lives under lacustre/max/, not a corpus defect). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HjbyJgJ5kh3PKKzPwSWw1U
defective_max_files: new drafts page cataloging corpus files with known defects, era-mismatches, unresolvable XRefs, live-Max quirks, and per-file anomalies — organized by defect class (source-side defects, era-mismatched references, XRef-related quirks, live-Max-authoring quirks, currently-open per-file mysteries, files worth specific mention). Cross-refs each item to the pipeline_max_design.md section that pinned it. Registered in wiki_context_index. pipeline_max_design gains a cross-ref pointer Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HjbyJgJ5kh3PKKzPwSWw1U