Gb Hardware Research

3. Use this retrieval order when accessing Pan Docs content.

• First, try standard web retrieval of the Pan Docs page at https://gbdev.io/pandocs/.
• If the page cannot be retrieved with standard tools, try fetching it directly with curl.
• If Pan Docs still cannot be retrieved, use the raw GitHub source at https://github.com/gbdev/pandocs and read the relevant markdown files from src/.

4. When fixing a failing Mooneye acceptance test, read the test source first.

• Mooneye test source files contain precise cycle-accurate comments that Pan Docs often omits (e.g., exact M-cycle timing of register side-effects, edge cases for restarts).
• Fetch the source with curl https://raw.githubusercontent.com/Gekkio/mooneye-test-suite/main/<path>.s
• The path mirrors the ROM path: acceptance/oam_dma_start.gb → acceptance/oam_dma_start.s
• Inline comments like ; M=1: OAM still accessible are authoritative — they document verified hardware observations.
• Use the test source to confirm what exact assertion the ROM makes before diagnosing the emulator.

5. When researching PPU Mode 3 timing penalties, apply M-cycle quantization.

• Pan Docs specifies Mode 3 penalties (OBJ penalty, SCX fine-scroll, window) in T-cycle (dot) precision.
• Critical gap in Pan Docs: the CPU observes Mode 3 end only at M-cycle boundaries (every 4 dots). The raw dot penalty from Pan Docs cannot be used directly as mode3_extra_dots — it must be quantized: mode3_extra_dots = floor(raw_penalty_dots / 4) * 4.
• This gap is not stated in Pan Docs but is required for cycle-accurate Mooneye tests (e.g., intr_2_mode0_timing_sprites) to pass. Without quantization, STAT mode reads by the CPU will be off by one M-cycle.
• When SameBoy confirms a penalty formula but your integration test still fails, check whether you need to apply this quantization before hooking the penalty into the timing engine.

6. If specification coverage is missing or incomplete, inspect SameBoy carefully.

• Prefer LIJI32/SameBoy and focus on Core/.
• Use SameBoy only after checking Pan Docs and its source.
• Treat SameBoy as implementation evidence, not as equal authority with a written hardware specification.
• If SameBoy appears to make a choice where the specification is unclear, say that explicitly instead of presenting it as confirmed hardware fact.

7. When sources disagree or remain ambiguous, report that directly.

• Name the conflicting sources.
• State which source is more authoritative for the question at hand and why.
• Do not merge conflicting claims into a guessed answer.

8. Produce a detailed, source-backed answer.

• Start with a high-level explanation of the hardware behavior.
• Then cover precise details such as registers, bit meanings, address ranges, timing, ordering, side effects, open bus behavior, edge cases, and model differences.
• Clearly label what is confirmed by specification, what is supported only by emulator implementation, and what is still unknown.
• Cite the exact Pan Docs pages or SameBoy files you used.

9. Never guess.

• If no authoritative information is available, say so plainly.
• If the available information is partial, answer only the supported part and identify the gaps.

References

• references/source-priority.md: source order, retrieval tips, and SameBoy lookup starting points.

Examples

• Researching joypad register ($FF00) behavior: start with Pan Docs joypad and register pages, then follow links for timing, interrupt behavior, and model differences.

• Researching an APU channel detail: start with Pan Docs APU and sound controller pages, then inspect Core/ in SameBoy only if the written specification leaves a behavior unclear.

• Researching an MBC quirk: start with the MBC page on Pan Docs, follow cartridge-specific links, then inspect Core/ or related MBC files in SameBoy if the written documentation is incomplete.

• Diagnosing a failing Mooneye acceptance test (oam_dma_start): fetch the test source with curl https://raw.githubusercontent.com/Gekkio/mooneye-test-suite/main/acceptance/oam_dma_start.s, read the inline timing comments to understand the exact expected behavior, then cross-check Pan Docs for the register specification.