Earthsiege 1 resource formats

This describes the supplied 1.0 CD. Names and offsets refer to flat application addresses, not executable file offsets. GO and DBSIM application images begin at file offset 0x48fd; PLAYFLIC at 0x47f2. Earthsiege 2 is useful background research but differs materially from this version.

Original payloads, hashes and unknown fields are retained. Every reader checks its bounds; exceptions are recorded rather than repaired speculatively. Inspection viewers use decoded data and identify approximations.

Important rendering findings

TSPoly (0x140001) is helper geometry: its native render method 0x54038 returns without drawing. Rendering it as an ordinary face produces bogus planes around HERC weapon mounts. The ES1 render vtable slot is +0x20; +0x1c is the destructor.

Stored polygon normals have length approximately 2,048. Vertex winding is clockwise relative to the outward normal. Materials carry separate front/back fill/line slots, including disabled and texture selectors. The viewer uses a depth buffer for inspection. Original BSP/detail/cell selection has 28,632 recorded comparisons, and the preview now uses verified fixed-point rotation, hierarchy and pose interpolation. Original light setup and intensity arithmetic are now verified; dynamic light-controller selection is separately verified in LIGHTING.md; gameplay-driven attachment and animation consumers are documented in ACTORS.md, ANIMATION.md and RUNTIME.md. Billboards, authored lines/points and original starting-mission weapon assemblies are visible.

HERC texture bank selection uses profile word at byte 148 (LIGHT/MEDIUM/HEAVY/ENEMY). Weapons use WPNTEX and scenery BASETEX. World data and palette choice must follow the compiled mission header rather than a guessed filename or zone header.

Retained ambiguities

Saved games, model attachment/splicing and native rendering now have original-code comparisons and live integration. Some retained table words still lack an identified semantic consumer; RESOURCE_AUDIT.md lists them. B0_C1_20.MSN, B0_C1_30.MSN and HOVTANK.DMG are preserved as layout exceptions. There is no evidence yet that their mismatching current-reader layouts are exercised by normal gameplay. Provisional inspector labels are marked separately from verified field meanings.

LPF/ANM presentations are decoded through FFmpeg as indexed PNGs: ESIEGE2 and ESIEGE3 have seven/eight displayed frames; the additional header frame is the closing ring delta. Legacy ICONS.DBA and some DBA arrays carry stale or trailing cached data, retained separately from complete pixel payloads.

Compiled mission handoff update

ES1 SCRIPT.DAT contains a 20-byte header followed by nine count-prefixed blocks: points (12 bytes), headings (2), variable routes (count then 16-bit point references), areas (6), triggers (24), HERCs (98), flyers (56), fixed ground objects (16), groups (82). All 300 freshly captured compiler outputs consume exactly; the four original tapes' bundled SCRIPT.DATs contain stale tails, which the original count-driven reader does not use.

The source table provisionally called orders in early exports is the fixed ground-object roster (24 bytes in MSN, 16 in SCRIPT). DBSIM's world loader reads its object type, builds its live-slot map and loads the matching ground-object definition via 0x045a4. Source timers are the 30-byte trigger/action records, exported as 24 bytes. Earlier JSON keys are retained for continuity until all consumers have been migrated; prefer these semantic names in new implementation code.

Verified mission text bytecode

All 79 ENG files now decode completely (tools/mission_text.py). BRF/DEB bodies start with a 16-bit text-allocation hint, followed by bytecode: 1 appends a u16-length NUL-terminated string, 2 selects a phrase using a mission counter, 0x10 ends the program. A choice contains a counter index, offset count, signed 16-bit relative offsets, inclusive signed ranges and counted phrase strings. The final offset is the end sentinel; its final range is unused. The last matching range wins. Literal spacing is preserved during concatenation.

OBJ bodies have the allocation hint, u16 objective count, then marker-byte/count/string entries and a final 0x10. This is a separate layout rather than a text program.

All 2,223 BRF/DEB results agree with original GO interpreter 0x19453 under zero, one and patterned mission-counter fixtures. Native checking stops at0x19856 before font-dependent wrapping; chunk selection and counter-file reads are explicit host boundaries. Previous heuristic fragment extraction is superseded. The mission browser shows selected text and retains every phrase alternative.

Source mission compilation now has 300 complete original captures over four explicit campaign/RNG fixtures, including final generator state. The native C++ compiler agrees with every output table and final RNG state in all 300 cases.

Inspection and coverage updates

The file-by-file inventory audits every VOL member against actual extraction catalogues. It separates complete byte layouts, provisional field meanings, legacy-layout exceptions and unclassified files. A decoded file does not imply fully reproduced gameplay semantics.

Direct file opening and HTTP serving are both supported. JSON exports have adjacent classic-script transports; indexed bitmap banks avoid canvas tainting from file origins. Bitmap export directories retain the original extension, preventing same-name DBM and DBA previews from overwriting one another.

Model playback uses authored durations: every sequence's total tick field equals the sum of its frame durations. At nominal simulation pacing, 0x14148 passes 0x63fa4(100) into the animation instance; the channel applies its independent signed Q8 rate (0x62718). This establishes nominal millisecond units at unity rate, not a fixed universal gameplay FPS. All 16,850 captured forward channel updates agree, including cyclic/clamped boundaries and varying frame durations/rates. Ordinary queued cross-sequence requests and reverse playback are now verified separately (ANIMATION.md); movement-dependent rate selection remains gameplay behavior.

Starting weapon assemblies use SM_INIT.DAT, GL mount orientation/position and WEAPONS.DAT's four template indices. The original loader selects MECHWPNS; both it and MECHWPN2 use WPNTEX. Thirty-two external weapon placements in the eight starting loadouts match original 0xb916/0xbbd4; empty/internal slots bring the total to 45 mounts. The bare chassis contains flat placeholder polygons that the runtime substitutes. Inspection defaults to the highest authored detail, with a selector for lower levels.

The four zones/DAT/WORLD*.DAT files retain a predecessor format, structurally decoded separately. They are not interchangeable with the current WLD descriptors loaded by 0x251c4. SKIMMER.DAT is the native 34-byte flyer profile copied by 0x1c6c8, not a truncated 192-byte HERC profile. The supplied NAMES.LOG is an empty career index. CAREER1.MEC and CONFIG.DAT have structural legacy interpretations with no established current consumer; TEST.DAT remains unclassified after a bounded consumer search; RESOURCE_AUDIT.md records its endpoint and the legacy distinctions. All are exposed in the coverage browser.

Model cell state and attachment corrections

The serialized per-cell shorts are cycle frame counts, exported as cell_frame_counts (previously the provisional name sequence_indices). Native 0x65d04 reads them into shape+0x20; it allocates and zeroes a distinct default-state array at shape+0x1c. Constructors copy that zero state. Effect updater 0x5d74 advances its selected cell modulo the stored count. Cell-object rendering 0x52fd0 then selects the child modulo that object's own child count; different detail variants can have different child counts.

HERC roots use the full-detail pose/cell instance, with native cross-LOD transform remapping through the profile's part IDs. Runtime exports are separate from raw stored roots, and all 16 remapping sets agree with original execution.

Weapon mount orientation >= 4 denotes no external model attachment. Native 0xbefc emits -1 for its replacement part ID; the chassis geometry must remain visible even if the mount's raw numeric ID coincides with an ordinary body part. External mount IDs, template selection and offsets are checked against original code for all eight starting loadouts.

Texture bindings come from native loaders, not a generic effects bank. EXPLOS.DAT chooses EXPLO0–EXPLO8 per explosion root; SMOKE and BULLETS each use their own DBA, SKIMMER uses ENEMY, MECHWPNS/MECHWPN2 use WPNTEX, and fixed scenery uses BASETEX. Original explosion, smoke and bullet loaders confirm 34 root bindings through a read-only multi-stream native fixture.

Recorded starting state and native preload planning

PLAYER.MEC starts with selected-roster index and count, then variable records: four shorts (portrait identity, single-mission crew index or campaign rank, chassis, weapon count); weapon IDs and weapon-value arrays; an extra short; 13 component values; two ten-short module arrays. Seventeen flag bytes follow. Native 0x1d850 expands each record to 126 bytes. The independent parser agrees with all four tapes and is exercised again in 300 explicit compiled-mission fixtures.

MISSION.VAR is 100 signed shorts. The loader resets counter 7 to zero after reading it. Headings in SCRIPT are quarter-turn selectors, expanded to unsigned word angles by shifting 14. Source actors are retained only when referenced by groups after group zero; the player's roster is appended separately. The resulting source-to-live maps and resource requests are verified. These facts describe preload selection, not live activation, formation offsets or physical placement.

RESTORE.DAT contains two shorts followed by two counted, NUL-terminated paths (source mission and ENG text). PREFS.CFG is 17 bytes; OBJECT.STR uses the normal grouped string/attribute encoding. The demo inspector preserves neutral names for unproven state fields.

GL mount byte 23 maps each mount to its equipment slot. Record order is usually a permutation. The assembly preview follows this mapping; all starting weapon constructors, drawable vertex shifts and parent-node bindings are checked against native code.

Cockpit DB0, DB1 and DB2 contain forward, lower and shared side artwork. Original 0x22184 loads three unique bitmaps; 0x22558 reuses DB2 with blit flag 2 for the left view. VUE's last two integers locate panels in the virtual instrument layout; the viewport remains 320×200. All 24 unique decoded bitmaps match the original class reader's pixel arrays. The cockpit browser displays these panels and clipping masks without claiming a running instrument simulation.

Verified shade lookup and texture lighting

Shaded materials with high byte 4 carry a ramp number; the intensity occupies bits 16–23. Native span setup chooses ramp[(intensity * length) >> 8]. Intensity 255 selects the last entry; intermediate thresholds use the full length, not length minus one. Gouraud setup selects palette indices at the vertices before its gradient rasterization. Native material checks cover 32,256 lookups across the 18 palettes that contain ramps.

0x28b28 builds eight 256-entry texture-lighting maps from the current palette and its ramps. It adjusts each ramp colour relative to a reference shade, then searches the palette for a nearest match using the game's integer distance helper. The 144 original generated maps are retained in models/material-shading.json. Texture polygon renderer 0x5410c chooses map floor(7 * intensity / 255). This lighting translation was absent from the earlier inspection preview. Billboards remain outside this polygon-specific remap.

The preview applies original ramp thresholds, texture maps and vertex-normal Gouraud palette-index gradients. It now also uses the original directional light and integer intensity rules (see LIGHTING.md). Native BSP order, dynamic effect-light integration, clipping and exact span pixels remain distinct from this depth-buffer inspection renderer.

See MUSIC.md for native HMI scheduling, hardware routing, archived OPL modules, original PCM command handling, FLAC auditions and explicit audio-emulation limits.

Original actor construction, formation placement and loadouts now have 304 world captures (ACTORS.md). Spatial trigger/action semantics and objective decisions have independent native checks (MISSION_RUNTIME.md). The mission browser displays constructed actors, activation boundaries and counter actions alongside the compiler-reference data.

Weapon control states, shield/capacitor scheduling and projectile motion consumers now have independent original-code comparisons (WEAPONS.md). Special weapon selectors and per-update timers must not be inferred as ordinary table indices or elapsed milliseconds.

Initialized runtime snapshots, native frame ordering and original-renderer captures are documented in REFERENCE_CAPTURE.md. Actor recognition, shared contacts and visibility caching are documented in PERCEPTION.md.

Texture transparency differs by draw path: native 0x7394a/0x735a4 skips source index zero only without a lighting remap. With a remap, every texel writes its translated index, including zero. The inspection shader now retains this distinction; billboard transparency is unchanged.

Mounted volume precedence

The initialized DOSBox snapshot retains the simulator volume search list at global 0x96838: SIMSOUND, SIMVOICE, SIMVOL0, then ZONES. Native lookup 0x566b8 visits that list in ascending order and returns the first match. Therefore SIMVOL0 wins duplicate zone paths. Eight zone names occur in both archives; ZONE6 and ZONE8 have different contents, while the other six are identical. Both copies remain exposed in the analysis catalogue. The native mission loader uses the SIMVOL0 copies. check_zone_loading.py captures all 106 files separately using explicit file overrides, comparing all 5,129 original constructor placements and the per-shape metadata bindings.

LPF menu doors

The two ANM resources are the closing (ESIEGE2) and opening (ESIEGE3) menu-door transitions, not a looping main-menu background. GO 0x13aa8/0x13d9e/0x13c46 loads, locates and decodes their LPF pages. Its native counterpart now matches all 60 captured complete frames, over both resources, two background patterns and opaque/saved-screen modes. The final ring delta is excluded. Written zero pixels restore the saved screen in transition mode; skip spans retain the prior frame. ESIEGE3's page directory is not in frame order.

Normal standalone playback uses the header's 15 fps. Menu calls at 0x3c2a4 and 0x3c734 instead select background restoration and omit that clock wait; their speed depends on decoding and VGA writes. Exact transition scheduling is separate from the verified decoder. The two call sites request digital sounds 10/11.

The expansion confirms the general priority rule: VOL header bytes 4-7 are a usage mask, byte 8 is priority, and byte 9 is the directory count. Insertion sorts descending priority and retains encounter order for ties. Simulator mask 1 selects PZONES1 (20), SIMPATCH/PZONES (10), and inherited volumes (5). See SIMULATOR_VERSIONS.md for original consumer addresses and the controlled insertion probe.