Real hardware capture
Digital capture from an original machine. This is the strongest visual source when geometry, timing, and capture conditions are known.
Fleet-wide evidence
Accuracy progress
Every machine in the flagship's registry is listed here, driven from the same source as the system matrix. The table below shows what can be evidenced today: own and shared crate counts, whether a boot capture exists, and where each machine's open work is tracked.
There are no percentages or maturity scores on this page. The flagship's own status page publishes no test counts on purpose: they "would be stale within a day, and a page that is usually wrong trains people to stop reading it." A crate count and a link to open issues are honest; a hand-typed score is not.
30machines in the registry
29with a boot capture
1no capture yet
Evidence table
Current evidence by system
Visual sources
What counts as evidence
Screenshot comparisons are only meaningful when the source is labelled. Emu198x treats screenshots as evidence about the digital framebuffer: pixel geometry, timing, and rendered content. Palette choices and CRT effects are documented separately.
Hardware-validated test image
A test ROM or program with reference output verified against real silicon, such as acid tests or emulator test-suite screenshots.
Reference emulator capture
A capture from a mature emulator with version and configuration recorded. This proves agreement with that emulator, not final accuracy by itself.
Emu198x golden
A self-minted regression image. Useful for detecting drift, but not enough to settle whether the rendering is correct.
Comparison policy
What makes a comparison valid
A reference-emulator comparison is valid when the run can be repeated and the source is honest about its limits. Agreement with a mature emulator is useful evidence, but it is not the same thing as proof against original hardware.
- Record the exact emulator, version or vendored snapshot, machine model, firmware, media, and command/configuration.
- Treat source-available emulators as implementation references; treat binary-only emulators as verification runs, not code citations.
- Prefer independent codebases. Downstream forks are useful for workflow checks, but they do not count as independent agreement.
- Separate software-emulator agreement from signal-level FPGA agreement. Both are useful, but they answer different questions.
- Use comparison captures to explain agreement or disagreement; do not replace real hardware or hardware-validated tests with emulator screenshots.
Reference candidates
Comparison sources by family
| Family | Primary reference | Secondary / tie-breakers | Coverage | Notes |
|---|---|---|---|---|
| ZX Spectrum family | Fuse is the canonical broad FOSS reference. ZEsarUX is the readable contention cross-check. | Spectron covers 16K, 48K, 128K, and TC2048. SpecIde, EightyOne, zxsp, and the MiSTer core are tie-breakers. | Partial | Spectron is useful but not enough for +2/+2A/+2B/+3, Pentagon, Scorpion, TC2068, or TS2068. Those need Fuse, ZEsarUX, EightyOne, zxsp, MiSTer, or hardware-backed captures. |
| Commodore 64 | VICE is the established reference and includes reSID for SID behaviour. | VirtualC64, Frodo SC, Hoxs64, and the C64 MiSTer core give independent or signal-level checks. | Strong | Best fit for VIC-II, CIA, SID, IEC, drive-path, and software corpus comparisons. Document disagreements instead of picking a silent winner. |
| Nintendo NES | Mesen2 is the first software reference for NES and Famicom work. | ares, fceux, nestopia, and the NES MiSTer core give second opinions and signal-level checks. | Strong | Use test ROM result channels where possible; use screenshots for PPU presentation and timing evidence. |
| Nintendo Game Boy | SameBoy is the local source-available reference for DMG, MGB, SGB, SGB2, CGB, and AGB modes. | ares can be used as a multi-system second opinion where useful. | Strong | Pair emulator comparisons with hardware-validated suites such as dmg-acid2, cgb-acid2, and mealybug-tearoom. |
| Commodore Amiga | vAmiga first for readable OCS/ECS behaviour; WinUAE first for breadth and AGA. | WinFellow is an independent OCS/ECS comparator. fs-uae is useful operationally but is downstream of WinUAE. Minimig-AGA MiSTer is the signal-level comparison. | Partial | AGA-specific work is mostly WinUAE plus Minimig. Workbench and game captures must record model, Kickstart, RAM, chipset, and disk configuration. |
| Dragon / CoCo | XRoar is the local source-available reference for Dragon 32/64 and Tandy CoCo 1/2/3. | Real-hardware or published captures are the main tie-breakers when available. | Usable | Already useful for Dragon software, 6809-family behaviour, and display-path cross-checks. |
| TMS9918-family machines | ares is the local multi-system reference for MSX, ColecoVision, SG-1000, and related VDP/SN76489 paths. | openMSX and blueMSX are good external candidates, but they are not currently vendored here. | Partial | Treat this as a chip-keyed source: one VDP oracle strategy can cover MSX1, ColecoVision, SG-1000, Sord M5, Einstein, and SVI-328, but machine-specific firmware/media still matters. |
| Atari 2600 / 8-bit / 5200 / 7800 | Stella is the 2600 reference. Altirra is the Atari 8-bit and 5200 reference, paired with acid800 tests. | Atari800 MiSTer and Atari7800 MiSTer provide signal-level checks. A7800 remains an external candidate for 7800 software-emulator comparison. | Partial | 2600 and 8-bit coverage is strong; 7800 needs a better software-emulator comparison path. Use cartridge-specific captures only with local media and clear provenance. |
| Commodore VIC-20 / PET | VICE covers VIC-20 and PET as well as C64. | VIC20 MiSTer is available as a signal-level VIC-20 comparator. | Usable | PET still needs a stronger independent comparison source or hardware-backed captures for model-specific behaviour. |
| ZX80 / ZX81 / Jupiter Ace | EightyOne is the local broad reference for the ZX81, TS1000/1500, Jupiter Ace, and Spectrum-adjacent lineage. | Primary-library timing references and hardware captures are needed for edge cases. | Usable | Useful for boot and display comparisons, but still weaker than the Spectrum reference stack. |
| Acorn Atom / Electron / BBC Micro | BBCMicro MiSTer is the only local BBC reference implementation. | b2, BeebEm, and JSBeeb are external candidates; Atom and Electron need reference candidates chosen. | Gap | Use the primary library and hardware-backed captures until software references are vendored or otherwise pinned. |
| Oric / Memotech / Aquarius | No strong local reference-emulator set is currently vendored for these machines. | MEMU-informed MTX work exists in Emu198x notes, but the public comparison source still needs to be pinned. | Gap | Do not imply visual accuracy from Emu198x goldens alone. These need external reference selection or hardware-backed captures. |
How to read this
A crate count and a boot capture are evidence about what has been built and seen to run. They are not a compatibility guarantee, and they say nothing about how much of a machine remains untested. Follow the system links for the detail behind each row.
When sources disagree, the closest source to real hardware wins. Lower-tier disagreements remain useful, but they are tracked as adjudication work rather than silently replacing a golden image.