23 August 2026

TEST: Iterate through .a18 per Region

Let us retain the MVP and change Region 2 to be each Region's .a18 data. Retaining Region 1 as the beep will also help determine if the projector reads the data all at once, or crashes upon reading incompatible data.

• If Region 1 plays, but Region 2 crashes, we can conclude that the projector reads data in a stream as opposed to a bulk pre-load. If this is true, this may help us locate corruption in the data by trying to time the crash/errors to digital addresses (we know 40-bytes = 20ms).

Procedure

• Copy mvp-s2.bin
  • using the mvp with magic numbers to allow function of dreamsmith and dreamprojector
• Copy f1.a18
• Insert f1.a18 into the MVP template in Region 2
• Track new addresses per Region
• Update Segment 1 pointers
• Validate pointers
• Ensure 1MiB file size
• dreamsmith extract
  • successful extract
  • all Regions' audio re-converted to .WAV without error = success
• dreamprojector
  • 14 Regions appear; Region 2 is custom audio played correctly; remaining Regions are test beep = successful
• Flash to cartridge
• Play in projector
  • Hypothesis 1: Crash on slide 1 = all data is pre-loaded, and the error is read upon initial projection. We cannot use error attributes to triangulate location of errors.
  • Hypothesis 2: Crash on slide 2 = data is read in a stream, thus we may be able to triangulate problematic data with error/crash observations.
  • OBSERVATIONS = Frame 1 played correctly, successful carousel spin to Frame 2; Frame 2 played custom audio to "Welcome to", then crashed.
  • CONCLUSION = data is streamed, thus we can triangulate corruption by error observations.

New Addresses of f1-inject.bin

Region (#)	Address (Hex)	Little Endian (Hex)
1	0x6C	6C 00 00 00
2	0x847	47 08 00 00
3	0x29C4	C4 29 00 00
4	0x31A0	A0 31 00 00
5	0x397C	7C 39 00 00
6	0x4158	58 41 00 00
7	0x4934	34 49 00 00
8	0x5110	10 51 00 00
9	0x58EC	EC 58 00 00
10	0x60C8	C8 60 00 00
11	0x68A4	A4 68 00 00
12	0x7080	80 70 00 00

Next Steps

• Triangulate where the corruption is in f1.a18 based off the length of audio that was played.
  • From VLC playback, it seems that around the 1-second mark is when it crashed.
  • $1000{ms}(40{bytes}/20{ms})=2000{bytes}(40{bytes}/{block}=50{blocks})
  • There seems to be some corruption resulting in a crash around 2000-bytes in.
• Back-titrate ratadon3.bin by iteratively replacing each Region with blanks or tests.
• Understand what the data in an .a18 file represents; see John-K's repo.

Ratadon3 Titration

Test 1 (can I just blank an entire Region with 00s?)

• Copy ratadon3.bin
• Replace audio payload of Region 1 with 00
  • We know the lengths of these custom payloads are appropriate, thus we do not have to worry about lengths, just content.
  • Keep Region header the same.
  • Keep Region light table data the same.
  • 1MiB size retained, no action required
• Flash to cartridge
• Play on projector
  • Hypothesis 1: Immediate crash = cannot replace everything with 00; must use test beep data.
  • Hypothesis 2: Crash on Region 2 = this is a validated titration method; quickly make copies of all titration states to flash later.
  • OBSERVATIONS = Frame 1 was all white noise, but played to the established duration, then rotated the carousel; Frame 2 played a brief portion of audio before encountering errors (light flickers, eventual crash after ~1s).
  • CONCLUSIONS = 00ing out a properly-sized Region's audio is a valid way of testing: expect white noise; second confirmation that data is streamed as opposed to pre-loaded and read. Since the crash was about the same distance into the Region, I would be more reluctant to "triangulate" error using roughly subjective timestamping. The best approach would be to read John-K's repo and study the test beep data. Why does white noise play instead of silence?

---

⟵ Older	Table of Contents	Newer ⟶