1990 · Domark
Every byte of the C64's 64 KB, one pixel each, 512 to a row. Lit pixels are what this game is: its code, graphics, level data, sound, text, tables and variables, exactly where they sit in memory. Dim areas are screen and working memory; dark is unused.
| Tier | platinum |
|---|---|
| Coverage | 100 % of the bytes the game uses have a description |
| Comments checked | Audited & low error rate. Every comment was audited once, then the last sample of 60 comments had 2 wrong, so between 1 and 11 % of the listing’s comments are likely wrong. How listings are checked |
| Build analysed | Original Domark disk (castle_master[domark_1990](!).g64, C64 Preservation Project); language menu reached from a hard reset |
| Copy | agent-draft |
| Tools | host: Linux 6.18 x86_64, Ubuntu 24.04 cloud container with four cores (Intel Xeon) and no display, emulator: VICE, release v3.13.2, v3.13.2-linux-x86_64-gui.zip (vice-mcp), run under Xvfb and driven through kit/c64/vice.py; check-emulator 54 of 54, disassembler: regenerator2000 0.9.20, built from crates.io, publication: Python 3; symbols_export.py and listing.py, python: 3.13.16, node: 22.22.0, browser: Chromium 141.0.7390.37 (Playwright 1.56.1's), for the page checks and screenshots, simulator: kit/cpu/cpu6502.js, for test_world.js, test_music.js and test_meter.js, subagents: six checker agents on the Agent tool's opus alias for 60-verify: one for each sample, one for each of the audit's four ranges and one for facts.md, one at a time; all other work by one agent |
| Model | claude-opus-5-5[1m] |
| Kit version | 0.0.173 |
Every step that measured the Source tab’s comments or made them more accurate, in the order it happened, with the models that did it. Accuracy of the Source explains each step and what the row above makes of them.
| Step | By | What it found | Of the listing |
|---|---|---|---|
| Research | gpt-6.1-sol, claude-opus-5-5[1m] | Looked the game up online (its manual, reviews or wiki pages) before reading its code. the manual and C64-Wiki, read again for the Silver run | |
| Annotation | gpt-6.1-sol | Wrote the listing’s comments. started from an imported Ghidra analysis by 64kramsystem, its model unknown | |
| Sample 1 | air, with claude-opus-5-5[1m] | 14 of 60 comments wrong, so between 14 and 35 % of the listing is likely wrong. All 14 were then corrected. 40 hand-written and 20 templated comments; 83 more that said the same things corrected with them | |
| Annotation | claude-opus-5-5[1m] | Wrote the listing’s comments. 100 % of the program is described, in 779 comments. the loader annotated again from the original disk, and comments written while building the pages | |
| Audit 1 | air, with claude-opus-5-5[1m] | Checked all 779 comments and corrected 182. four ranges, a fresh checker each, one at a time, and every claim in facts.md (21 of 177 corrected) | |
| Sample 2 | air, with claude-opus-5-5[1m] | 2 of 60 comments wrong, so between 1 and 11 % of the listing is likely wrong. Both were then corrected. 40 hand-written and 20 templated comments, from those the first sample did not draw | |
| Play tab | claude-opus-5-5[1m] | Ported the whole game to JavaScript for the Play tab and corrected 6 comments it showed wrong. every routine ported and checked against the game's code alone and in lockstep; a checker compared the 114 routines' comments with the port, and the six it found wrong are listed in port/corrections.md |
likely wrong the sample’s range corrected right, or likely right not yet measured
Read this before annotating code. What the game is documented to do, with verification status against the binary. Statuses: open (documented, not found yet), traced (in the code, could not be exercised; say what was tried), confirmed (in the code, consistent with the emulator), live (observed directly), differs (the code does something else). "Absent" is not a status.
Sources:
| Feature | Status | Where |
|---|---|---|
| Language: English, French or German | live | RUN_TITLE_AND_CHARACTER_MENUS $9B1C; play.py chose English |
| Play as the prince or the princess, to rescue the twin | live | $9B1C stores the choice in $246E; INITIALIZE_NEW_SESSION $4E05 makes it game bit 32; play.py chose the prince |
| Move forward and back, turn left and right | live | DISPATCH_PLAYER_MOVEMENT $5DD8, actions 3-6 from ↑/*, =, Z, X and the joystick ($1382); a turn is 3 steps of 5° ($9D4C); turn.py: X took the heading $2411 from 28 to 31, a turn to the right |
| Commodore key with a turn: 90° | traced | $6072-$6091: 18 steps when the Commodore key's flag $1380 is set, 3 otherwise |
| U-turn | traced | action 13, U: $6156-$616C adds 36 steps (180°) to the heading |
| Look up, look down, face forward | traced | actions 7, 8 and 12 (P, L, F) in $5DD8; pitch is kept within 90° of level ($60EF-$60F7) |
| Crawl, walk, run | traced | actions 9-11 (C, W, R); step lengths 30, 60 and 240 ($9D46-$9D48); messages 13-15 |
| Throw a rock | traced | Shift or fire: action 22 to DISPATCH_AIM_FIRE_ACTIVATE $62E8; a hit sets the shot event $04 ($646D) |
| Action: eat, drink, read, push, collect, open, unlock | traced | A, action 27, in $62E8; scripts react through their event bits |
| Space switches between moving and the pointer | traced | action 30; with the pointer on, $4901-$490C turns the direction keys into pointer moves; sprites placed by $65A6 |
| Centre sights on and off | traced | +, action 21: $633E-$6346 flips $246B, which decides whether $3672 draws the reticle ($827A) |
| Information screen: score, strength, keys, spirits destroyed | traced | I, action 41, opens RUN_INFORMATION_MENU $77FE; fields formatted at $7A80-$7AF3 |
| Save and load a position, tape or disk | traced | S and L in $77FE, PROMPT_SAVE_DEVICE_AND_NAME $7B03, LOAD_SAVE_DISPATCH $0423; 106 bytes of state then one flag byte per object. Not exercised on this disk |
| Music or sound effects | live | F in $77FE ($79C3); alarm.py ran the game both ways |
| Show the keys collected | traced | K in $77FE |
| Quit | traced | Q in $77FE sets end reason 6 ($7808) |
| Strength, falls and the three potions | live | $2455, from 16, capped at 24 by script opcode $13; a fall costs 5 ($8142); drain.py saw a sensor take a point every 10 seconds; potion.py saw the potion of revitalisation give a point every 10 seconds, and stores.py the potion of strength give a point and 6,250 for each of 64 drinks; the potion of rock travel sets flag 30 (traced) |
| Ten keys | traced | each new key adds 125,000 ($70EE); message 16 NEED TEN KEYS is printed by a script |
| Guardian spirits | live | 21 to destroy ($9D49): 19 shared spirits raised by global scripts 1-3 over sensors 125-127, the Courtyard's and the Dragon's Lair's; the spirit meter $246A ends the game at 64 ($72A9); meter.py ended games at the expected clocks |
| Riddles | confirmed | nine panels at $18AE-$1B76, printed by script opcode $22 with the high bit set; all nine parse to their ends |
| Drawbridge, locked doors, secret tunnels, caverns | confirmed | scripts in the database $9D00; the 34 areas, 546 objects and 809 script tokens parse exactly |
| Rescue and escape | live | ending.py: the Magister's object 5 adds 3,000,000, prints YOU DID IT ! and plays tune 1 |
| Lightning | traced | $484F requests effect 8; the lightning band in the sky is drawn by $4D90 |
$73A1; alarm.py). With music on there is no alarm at all.$7545; drain.py).$7286-$72AF; test_meter.js; meter.py).stores.py).tour.py).ending.py).$CF1B).$9D4D and $9D4E of the database header: opcodes.py --refs finds only $5F13, LDA $9D46,X with X the movement mode, which would have to be 7 or 8 to reach them. Their role is unknown.The symbol map for this game is symbols.json; the listing behind the Source tab is listing.json. Write-ups, facts and symbol maps are CC BY-SA 4.0. The game itself is not hosted here.