Files

102 lines
5.4 KiB
Markdown
Raw Permalink Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# High-resolution Bone Quest verification — November 11, 2026
## Candidate and builds
The current candidate is a 35-track DECB disk, 161,280 bytes, SHA-256:
`373a3f387daf541b1eb4edd3421a125dd57094e600ee48d162802ac22f65ceba`
Source base: 3cc37be5fb46b2e45be54ab13f52765c702edc24 plus the reviewable
28-structure/high-resolution changes in this checkout. Selected source hashes
and final check summaries are recorded in VERIFICATION.json.
Built with installed VCC Plus 3.6.1 (54), bundled CMOC 0.1.100 and LWTools 4.25.
Release program/static storage ends at $73FD; Debug at $745B; Test at $7402.
All remain below the $7800 limit, with the separate 2 KB stack below $8000.
Debug and Test were built in separate minimal folders without the original SVG,
Python, Pillow or asset-conversion scripts as build inputs.
The display is native 640 × 225, four colors, with two 36,000-byte images in
40 KiB reservations ($1E–$27). Artwork uses the bank library's separate $28–$2B
reservation. BONES.DAT is 16,640 bytes; compiled geometry metadata is 552 bytes.
The loader reads the file once, verifies size/checksum and never writes to disk.
Ordinary builds use the retained metadata and drawing file.
## Logic and artwork
make test passed 56,000 option sets across 2,000 complete shuffled 28-topic
orders: unique topics/options, no consecutive correct-slot repeats, all allowed
transitions, replay reset, aliases, known-wrong-name priority, spelling, scoring
and duplicate-answer protection. Canonical bank/generated C identity, the
original SVG hash, external artwork SHA-256/checksum, six disjoint bounded views,
and visible coverage for every overview/selected-detail topic also passed.
## Graphics and loading regressions
The retained tests/check_graphics.py fixture passed on both guest CPUs. It checks
576,000 actual guest pixels per CPU across four observations and three
presentations: packed 2-bit edges, all 8 KiB seams, fifth-bank rows, clipping,
page flips, complete front-to-back copies, edits staying hidden until present,
stock MMU restoration, BASIC return and zero writes. Its host arithmetic check
compares the actual rectangle function against an independent pixel reference
for all horizontal positions/widths on 15 critical rows and 1,000 clipped
multirow rectangles.
The retained tests/check_loader.py checks passed on both CPUs: missing artwork,
wrong file size and wrong checksum all return to BASIC with a clear error and
zero guest disk writes. Tests modify disposable in-memory copies only.
## Thumb pose and quit controls
The screen-right hand's thumb chain is rotated toward the palm in the converter;
bone lengths and relative joints are preserved. All six generated views use the
same pose, and the original SVG hash remains unchanged. The current guest hand
capture is images/hand-detail.png.
The separate quit regression passed 16 actual guest scenarios across both CPUs:
Q/q, Break/native Escape and ASCII 27 via Ctrl+Shift+Down, title, credits,
multiple choice, partial typed input, spelling help, typed feedback and the
automatic reading pause. All return to BASIC with normal MMU mappings and zero
disk writes. Complete-game runs additionally quit directly from missed review.
Blank Enter now reveals a typed answer; Escape/Break are reserved for exit.
The unchanged low-level graphics and loader modules retain the earlier
regression evidence recorded below; those suites were not repeated for this
pose/input refinement.
## Actual gameplay
The actual Release disk completed a full 28-question multiple-choice round and
a full 28-question typed-answer round on each emulated MC6809 and HD6309. Both
reported zero disk writes and preserved the exact disk hash.
The driver reads the actual displayed text and checks the full skeleton and
enlarged detail pixels against the current question and answer history. Checks
include all 28 highlights, white normal/correct structures, yellow current,
red missed, four readable choices, no repeated correct slot, held-key release,
common aliases, case/spacing, spelling retry/Backspace, the 33-character bound,
reveal, 23/28 (82%) scoring, one correction, paged review, replay and BASIC exit.
Separate correct/incorrect feedback captures verify nonzero guest audio samples;
this establishes generated signal, not audible listening. Personal emulator
settings/media and desktop input are untouched.
To repeat developer checks, supply your own private ROM paths:
```sh
python3 tests/playtest.py --rom /path/to/coco3.rom --disk-rom /path/to/disk11.rom
python3 tests/check_quit.py --rom /path/to/coco3.rom --disk-rom /path/to/disk11.rom
python3 tests/check_graphics.py --rom /path/to/coco3.rom --disk-rom /path/to/disk11.rom
python3 tests/check_loader.py --rom /path/to/coco3.rom --disk-rom /path/to/disk11.rom
```
Raw logs/captures and ROM inputs are excluded from Git. Selected unmodified guest
frames are in images/ with matching hashes and disk identity in SCREENSHOTS.json.
At 640 × 225, the helper's active image starts at output (0,16), with two output
rows per guest row. These are game frames rather than native-window captures.
No physical CoCo, host Intel, macOS 14, native editor/source-open UI or audible
listening qualification is claimed. The original 25-topic candidate's evidence
is preserved in testing/INITIAL_25_TOPICS.md and its original source revision.
The initial high-resolution candidate record is retained in testing/HIGH_RESOLUTION_INITIAL.md and .json; its prior disk remains unchanged under test-output/pre-thumb-quit/.