Original TIM v1.0: L1 recognition and first puzzle input route

Original bytes establish a substantially narrower target than “some TIM gameplay”: L1.LEV, format revision 0x0102, TUTORIAL: PUT THE BALL IN THE HOOP, Make the basketball go through the hoop. The level contains three ordered object groups of 10, 4 and 9 records. The last group feeds the parts selector and contains three records each of numeric types 8, 6 and 2. Their ordinary names, selectable icon appearances and a successful arrangement remain unproved.

This is static recognition/input preparation. No installer, original executable, overlay, VM, emulator, game, native fixture, browser or build was run. There is no native frame or gameplay screenshot, and no startup, completion or compatibility claim. The proposed action below requires later runtime authorization and fresh visual review. Stored level values and disassembled UI rectangles are never native input coordinates.

Evidence is in scratch/runs/20261113-tim-l1-static-0950/. The predecessor is the original DOS launch contract, which supplies the unchanged installer/configuration/media prerequisites.

Identity and reproducible static method

The borrowed /dev/shm/wine-owner187-tim-disk.img was opened read-only and its entire 1474560 bytes hashed before parsing: d0e78876d50609b254420cdf7c0724a117a8ed714ea7c683c19e0d2094ce0657. All 11 payload hashes match the full FAT12 manifest in scratch/runs/20261113-tim-floppy-manifest-0825/manifest.json; all 159 resource member hashes match the predecessor's resource-members.json. No media copy, archive, mount or guest/config/save write was made.

inspect.py reconstructs FAT member bytes in process memory using the verified manifest's cluster lists. It reuses only the predecessor's literal LZ91 data-format decoder, with the predecessor script itself hash-pinned. The 211216-byte decoded analysis view must hash to 090d7d0eed4a77e279fc48875aca96074ddb7ced167126b7ee987bf318bd1286. Original code is not executed. Bounded i8086 objdump slices pass through anonymous memory descriptors that are closed immediately; no decoded EXE is staged.

Run python3 scratch/runs/20261113-tim-l1-static-0950/inspect.py to repeat the identity checks and level parsing. REPRODUCE.txt lists the bounded disassembly ranges. source-pins.json, exact Git SHA, relevant binary diff and the owned new-document/helper patches preserve source context without a tree snapshot.

All executable addresses below are decoded-image offsets before runtime load bias, and DS: values are original structure/global offsets, not host or live guest addresses. Objdump's 16-bit printed near targets can wrap at 64 KiB; cross-boundary targets require the containing segment context, not blind copying of the printed number. Partial leading/trailing instructions and jump-table data in bounded slices are not function entry evidence.

What the actual level reader proves

RESOURCE.004 has L1's record header at 0x62d; its payload starts at 0x63e, is 1217 bytes, and hashes to c46d65ffa5f10d2a80c7c63e17c20fefa0294caa0d2ab76107762a496f843904. The payload offsets are:

Payload offset Original value Reader-backed interpretation
0x00 ed ac Little-endian magic 0xaced, checked at 0x122a0
0x02 02 01 Format revision 0x0102, read into DS:5474 at 0x122ab
0x04 Title above, NUL at 0x27 NUL-terminated title into DS:4ecf
0x28 Objective above, NUL at 0x50 NUL-terminated objective into DS:4f1f
0x51–0x5e 400,300,67,272,65528,65528,1002 Seven WORDs to DS:50af,50b1,50b3,50b5,50b7,50b9,50bb
0x5f,0x61,0x63 10,4,9 Counts for lists headed by DS:521b,5179,50d7
0x65–EOF 23 records Field-wise deserialization, not raw structure copies

The seven header words are deliberately not called screen width/height or level dimensions: those semantics were not established. In particular, 400,300 alone cannot establish a 400×300 native frame. Format revision 0x0102 is also distinct from the program's displayed version 1.00.

Reader 0x12269 reads the header, adds all three counts for allocation at 0x12353, then passes each count and list head to 0x1221b. WORD reader 0x11dd1 reads two bytes; byte reader 0x11db4 reads one; string reader 0x11dec repeats until NUL. Record reader 0x11e3f reads words into structure offsets 04,06,94,0a,90,92,44,46,50,52,8c,8e,96, followed by an optional subrecord flag, bytes 56,57, word 58, conditional subrecords, attachment/link fields and reference indices. Revision comparisons at 0x11e7e, 0x12132 and 0x12192 change the schema; treating every release's LEV as interchangeable is unsupported. Type 7 has an additional conditional reference; L1 has no type 7.

The parser follows those reads, including both link-presence branches. L1 has three 52-byte type-8 records and twenty 48-byte records: 101 + 3×52 + 20×48 = 1217. The same parser consumes all 87 original LEV members to exact EOF, all with revision 0x0102. This corroborates serialization boundaries, not the meaning of every flag or the runtime behavior. Full fields and references are in L1.parsed.json; level-validation.json retains the cross-level check.

Stored geometry and inventory, separated from any future frame

The following are exact serialized values with original destination offsets. Pairs A and B are structure 44/46 and 50/52; P is 8c/8e. They resemble size and initial-position fields, but this task has not traced every initialization/rendering transform into native pixels. Treat them as stored geometry candidates, not a diagram, solution or click list.

Group / payload offset Type A = B P
521b / 065 2 16,32 408,120
521b / 095 2 16,32 472,120
521b / 0c5 5 96,16 248,72
521b / 0f5 48 16,96 488,72
521b / 125 47 32,32 504,120
521b / 155 46 16,176 520,152
521b / 185 1 240,16 72,328
521b / 1b5 1 240,16 312,328
521b / 1e5 5 96,16 56,280
521b / 215 5 96,16 168,200
5179 / 245 9 32,32 248,40
5179 / 275 0 32,32 8,216
5179 / 2a5 0 32,32 72,248
5179 / 2d5 0 32,32 168,168

The third list's exact serialized inventory is:

Payload offsets Type / quantity A = B P values, in order
305,339,36d 8 × 3 0,0 ffff,ffff each
3a1,3d1,401 6 × 3 48,32 ffff,ffff; 280,184; 8,296
431,461,491 2 × 3 32,32 ffff,ffff each

ffff is a raw WORD (−1 if interpreted signed), not a usable coordinate. Non-ffff fields in inventory records do not prove those parts are already placed on the initial board. Types and list membership are stronger evidence than assumed names such as rope, fan or pulley. This note does not assign those names from familiarity with another TIM release.

The third-list interpretation has executable support: 0x123ac makes DS:50d7 the selector cursor head (DS:50d3); 0x5855 walks that list by type, skipping duplicate type records; 0x10e14 uses the chosen type and sets selection mode DS:4e69=9. Special handling of types 8 and 10 in selection/placement is visible at 0x10dd8 and 0xfd1a. Therefore the first experiment should use a freshly observed ordinary placeable item, avoiding an unreviewed attachment gesture.

There is an additional original recognition check, not a solution recipe: the level-indexed callback at DS:2632 + 1×4 points to 0000:151b. 0x1465 dispatches through that table. Callback 0x151b searches list 5179 for type 9 and tests its live structure fields 1e/20 against bounds before setting event 0x0200. Together with the objective, this supports an inference that type 9 is the objective basketball; sprite identification is still missing. The static predicate alone is not proof of basketball/hoop collision semantics, nor authorization to write those fields or force success.

Original controls and the selectable-puzzle route

Original READ.ME explicitly documents these controls; readme.json retains its exact text and payload hash:

Input Release-note meaning
Tab Move cursor between hotspots
Arrow keys Move cursor anywhere on screen
Space or Enter Same as pushing the left mouse button
Esc Bring up Control Panel
Alt-V in Control Panel Version number
X / Y Flip a shape
+ / − Size a shape

Space/Enter is not documented as a global simulation toggle. No right-click, drag duration, hotkey for Start/Stop, fixed Tab count or arrow-distance mapping is established. Always balance key/button down with up; keyboard emulation of a mouse button still needs later observation of the application's actual gesture.

The original UI resources are real container members, not standard Windows BMP files merely because of the suffix. CP.BMP begins BMP: with an INF: child; GP_MENU.BMP additionally contains SCN:; PARTBIN.BMP contains BIN:. Their hashes, sizes and bounded headers are in members.json and recognition-pins.json. The executable loads cp.bmp at 0xe1f6 and gp_menu.bmp at 0xefa1. This task has not decoded those graphics into labeled icons or established a native panel image.

There is byte-backed separation of editing and running. Hotspot dispatcher 0x8546 checks the current event mask and cursor rectangle, calls callbacks, and, on its button-event condition, copies the hotspot's result word to DS:4e6b. Original hotspot initialization includes result 0x2000 at 0x8a4e (panel list) and 0x8bfe (game list), and result 0x1000 at 0x8c36. Editing loop 0xf8c2 exits for 0x2000 or 2; simulation loop 0x12ab continues while the event is 0x2000, runs object updates and the level callback, and exits on other events. This supports Start/Stop-style state transitions. The visible icons/labels corresponding to those records remain unknown; no rectangles from disassembly are promoted to native input targets.

The original selector displays SELECT PUZZLE (0xf4df) and builds entries as PUZZLE <decimal number>: <title> (0xf703 onward). Name builder 0x12863 constructs l + decimal argument + .lev, sets DS:5472=1 (the header-rich reader branch), and calls 0x12269. The numbered scan at 0x129a8 starts at 1 and tests successive names. Thus the concrete ordinary candidate is PUZZLE 1, recognized by the exact L1 title/objective, not a freeform machine or L2.

The selector checks a candidate against the available/unlocked limit at 0xf358; its rejection path displays “You need to enter the correct password in order to try this puzzle.” This investigation does not establish the runtime initial limit, saved current puzzle, first screen, default selection or exact panel hotspot that opens the selector. Choose Puzzle 1 through the observed ordinary UI only if available. If already on the matching puzzle, no selector detour is needed. Any actual manual/password prompt must be captured and the run stopped there, without answers, unlocking, state injection or bypass.

First later authorized balanced selection/place/start proposal

Prerequisites: preserve the verified image; finish an ordinary original install and capture/hash its actual configuration and installed closure as described in the launch contract; launch the unchanged packed TIM.EXE from its actual game directory; prearm an observable independent deadline and input-release watchdog. The released runner/test WIP is not qualified by this static work.

  1. Capture the actual first screen and prompts. If needed, use one Esc down/up to open the Control Panel, visually locate the puzzle selector, and choose the available PUZZLE 1 through ordinary controls. Review a new capture after every transition. Require the L1 title/objective and a stopped editing board; otherwise record the mismatch and stop this proposed experiment.
  2. From that fresh native frame, identify one visible ordinary placeable item in the parts inventory and an empty legal board location. Record their reviewed native coordinates or actual focused hotspots in the future run's evidence. Select exactly one item with left-button down/up (or one observed equivalent Space/Enter down/up). Capture the selected state before proceeding.
  3. Move the cursor to the freshly reviewed empty location and use exactly one balanced placement gesture matching what that UI has visibly established. A second click-to-place is a proposal, not a proven gesture; if the UI instead requires a held-button drag, stop and revise the bounded gesture after review. Capture placement and the changed inventory/board. Do not choose a position from the stored numbers above or claim it is a solution.
  4. Only when the fresh frame identifies the ordinary simulation Start control, click it once with down/up. Capture a short bounded before/after interval and record actual motion or the precise failure boundary. Use the freshly identified Stop control once with down/up if the machine is still running; retain a final stopped frame and the watchdog/input-release receipt.

This proposes one selection, one placement and one start, with a stop/release tail. If selection, legal placement or Start/Stop identification is not possible from the actual UI, preserve that concrete missing evidence instead of sending speculative input. Success for this milestone is real original puzzle input and a visible consequence; solving the puzzle is neither assumed nor required.

Validation and limits

Full image/payload/member identity, the decoded analysis hash, all 87 exact-EOF parses, the L1 offsets/counts/types and documentation whitespace were checked. validation.json and SHA256SUMS.json retain the checks and evidence seals. Only this document is published; the run-specific static helper/evidence stay ignored. No build or runtime tests were appropriate or authorized for this documentation-only static assignment.

Remaining evidence includes named sprite/type mapping, proven geometry-to-frame transforms, native panel/puzzle frames, actual unlocked state and selector entry, placement gesture, Start/Stop icon identity, runtime resource/device behavior, and input consequences. No static drawing or fabricated gameplay screenshot was created. Borrowed RAM media remain untouched; no private RAM files were created. Box ID NONE, state NOT_CREATED, expiry null, stop receipt N/A. No runtime watchdog was created because no runtime or box was created.