Add Bone Quest anatomy quiz for CoCo

This commit is contained in:
Brian Holman committed 2026-11-11 21:00:21 -07:00
commit 3cc37be5fb
50 files changed
+10146

No files matched your search

+64
View File
@@ -0,0 +1,64 @@
# Initial Bone Quest verification — November 11, 2026
## Built and delivered
The root BONEQ.DSK is a standard 35-track DECB image, 161,280 bytes, SHA-256:
`cf9b2da5842551d76b58c3be2064a78163536526826fca13ac7ee58c269e1e66`
Built with installed VCC Plus 3.6.1 (54), bundled CMOC 0.1.100 and LWTools 4.25.
Release, Debug and Test disk builds passed. Program/static storage ends at
$764E (Release),$7687 (Debug),$7653 (Test), below the $7800 limit and separate
2 KB stack reservation. Two graphics pages retain the public GIME512 KB contract.
Generated skeleton geometry occupies 5,975 bytes; ordinary builds do not run an
SVG renderer or need Python/Pillow/rsvg-convert.
## Logic and artwork
make test passed: 50,000 option sets across 2,000 games, all 25 topics once per game,
four distinct choices, no consecutive repeated correct slot, all permitted
slot transitions, replay reset, canonical names/common aliases, known-wrong
name/alias priority, spelling checks, scoring and duplicate-answer guards.
All 25 authoring topics/aliases agree exactly with generated C. The original SVG
hash matches its receipt. Six geometry views have bounded, disjoint spans;
each topic has nonempty overview coverage, with detailed small-bone views.
## Feedback audio
Separate bounded correct/incorrect feedback captures contained nonzero 44.1 kHz
stereo PCM (peak 16128). Capture began at the actual feedback phase, before the
tones; no microphone or host audio device was used. This establishes generated
guest signal, not audible listening or comprehensive frequency analysis.
## Actual gameplay
The actual Release disk completed a 25-question multiple-choice game AND a
25-question typed game on each emulated MC6809 and HD6309, with512 KB RAM,
CoCo 3 BASIC, FD-502 Disk BASIC and RGB video. Tests used isolated app-owned
machines and disposable disk copies; no personal emulator or desktop input
was controlled.
The driver decoded displayed text against the matching linked font and checked
the skeleton's actual pixels against the topic masks and answer history. It
verified question numbers, readable choices, current highlighting, purple/red
progress, no consecutive correct-slot repeats, held-key protection, deliberate
wrong answers, common aliases, case/spacing, spelling retry and Backspace,
33-character entry bounds, reveal, scores 20/25 (80%), one correction, paged
review, neutral replay title and clean return to BASIC. Both runs preserved the
disk hash and reported zero guest writes.
To repeat these developer checks, provide your own ROM paths explicitly:
```sh
python3 tests/playtest.py --rom /path/to/coco3.rom --disk-rom /path/to/disk11.rom
```
The harness uses the installed app's isolated helper and is qualified against
the build identified above; it is not a general alternate MCP server. Private
ROMs, raw captures, result JSON and build outputs stay ignored under test-output/.
Selected unchanged guest frames are retained in docs/images/ and identified by
docs/SCREENSHOTS.json. They are actual game frames, not native-window marketing
captures or mockups.
No physical CoCo, host Intel, macOS 14 or audible listening qualification is
claimed. Native editor/source-open UI is not exercised by the headless tests.