Testing and Verification English

日本語 | English

Verification

The environment, methods, scope and results of the verification of ZORKERB I and ZORKERB II.

The tools, test data, comparison interpreter, original story files and raw reports are kept in the private development repository. This page summarizes only confirmed results from those records. The results cover what was tested; they do not guarantee identical behaviour on every path of the originals.

Environment

Item	Details
Engine	PyEmuera 3.0.0 (unmodified, SHA-256 8a7b3697b97155c4dc994c8349c6756bb0401000d3bbccd8e85ff046e3774b35), the engine used for these recorded results
System and Python	Linux x86_64, Python 3.13
Reference (I)	the original story (Release 119 / Serial 880429, SHA-256 37084966477dff679282de42974b2077156b1bd68fad92a65d4ea94d8eb64d79)
Reference (II)	the original story (Release 63 / Serial 860811, SHA-256 3ae7d5558943e9721f3e4b273c8a7faec1a03a604e1ae4ee1cde472c21cb24ac)
Interpreter for the originals	Frotz 2.56pre (dumb interface; the same executable for the included games)

Checks on the native Emuera (Windows) are in Compatibility and limitations.

Methods (the included games)

One frozen run

1. At the start, the SHA-256 of the sources, tests, tools, built game and engine are recorded.
2. All checks run.
3. At the end, nothing may have changed from the recorded values. A source change means starting over.

Comparison with the original

The same input is given to the original and the port, and each move's response is compared.

• Aligned random sequences: the input starts with the originals' debugging command #random number, so both draw the same sequence, and text and state are compared exactly. Completion is judged on this comparison.
• Text normalization: only ASCII whitespace (including where lines break) and Frotz's hyphenated line breaks. No wording is normalized.
• State comparison: location, score, moves, inventory order, selected object parents and flags, and IT, read from saves. This does not compare the whole memory.
• State comparison: before every input, the original's memory (all 251 object locations and attributes, globals, clock) against the port. Exclusions and reasons are recorded; the whole memory is not compared.

ZORKERB I results

Check	Result
Regression tests	1,596 of 1,596 PASS (7,033 assertions, 1,985 state checks). Expectations rest on the original's responses and the source
Save rejection	PASS: saves of r1–r19, r21–r25, r27, r28 and deliberately damaged saves are rejected
Unit tests of the tools	306 PASS
Syntax table	equal to the table read from the original story
Comparison with the original (aligned, 801 runs)	table below
Probes (41, 10,324 moves)	10,320 exactly equal; 2 differ in the file-name prompt of SAVE and RESTORE; 2 are inputs over 120 characters
Full route (557 moves)	all 557 exactly equal
Ordinary-input playthrough	the endgame at 350 points in 613 moves, without SAVE, RESTORE or debugging commands
Routine mapping	GAP 0 (VERIFIED 175, IMPLEMENTED 265, UNREFERENCED 1); see the ZIL to ERB mapping (Japanese)

Set	Runs	Text exactly equal	State equal
pseudo	56	56	56
holdout	62	62	62
edges	24	24	24
older (18 sets of earlier revisions)	659	657	657

The two remaining older runs use examine it while IT points at object 0. The original then runs instructions on object 0, and a strict interpreter stops there. Rerun on a lenient interpreter (Frotz -Z 1), text, number of responses and state agree apart from the interpreter's warning line. These two are not counted as exact matches.

ZORKERB II results

Check	Result
Comparison with the original	the same text in 240 runs, 74,130 compared steps
Random numbers	47,650 random calls agree in order, value and calling routine (predictable mode, every value compared)
State	at 74,114 points, the state read from the original's memory equals the port's
Playthrough	the 400-point route completed
Saving and restoring	saving in 10 slots and restoring in a new process; the comparison with the original's SAVE and RESTORE agrees
Routine mapping	every one of the original's 439 routines has an implementation (370 generated from the ZIL, the rest written by hand)
Spells and interrupts	all 12 spells and all 23 clock interrupts reached in tests
Branch coverage	1,689 of the 2,648 ERB branches (63.8%) reached; the others are listed in the development records

Independent audits

An auditor separate from the developer checked:

Audit	Result
The distribution	PASS: SHA-256 of the files, the included notices, the static check, the 400-point route in the extracted game, three save formats, restoring slots 0–9 in separate processes
Direct comparison with the original	the auditor ran the pinned original story on Frotz afresh: 10 routes, all 1,718 responses equal
Rebuild from the original ZIL	regenerated from the ZIL and the story: the 8 generated files and the 23 game files of the distribution are byte-identical (the same in two runs)
Recount of the records	47,650 random draws and 74,114 state points recounted without the developer's tools: equal

Checks of the distribution

• The ZIP's SHA-256, SHA256SUMS.txt inside it, and no duplicate, unsafe or development file
• Each game's ERB/ and CSV/ byte-identical to the verified game; the included PyEmuera has the SHA-256 above
• Relative links in the documents resolve inside the distribution
• In the extracted distribution, the commands of the included start batches start each game