agent-orchestrator-benchmark/calculators/builder-adversary-lean/run-03/machine-docs/REVIEW-lex.md

REVIEW — phase lex (Adversary)

Last updated: 2026-06-15T05:08:00Z

Status

All 4 gates PASSED. Phase is DONE pending Builder writing "## DONE" to STATUS.

Gates

Gate	Status	Timestamp	Notes
D1	PASS	2026-06-15T05:06:00Z	All number forms correct
D2	PASS	2026-06-15T05:07:00Z	All operators/parens correct
D3	PASS	2026-06-15T05:07:30Z	Whitespace skipped, LexError raised with char+position
D4	PASS	2026-06-15T05:08:00Z	23 tests, 0 failures; all plan cold-verify commands pass

---

Detailed verdicts

lex/D1: PASS @2026-06-15T05:06:00Z

Cold-start verification from own clone. All Builder-provided checks pass:

• tokenize('42') → [NUMBER(42), EOF], value is int ✓
• tokenize('3.14') → NUMBER(3.14) float ✓
• tokenize('.5') → NUMBER(0.5) float ✓
• tokenize('10.') → NUMBER(10.0) float ✓
• list-equality with Token('NUMBER',42) and Token('EOF',None) ✓

Independent break-it probes:

• tokenize('') → [EOF] ✓
• tokenize('0') → NUMBER(0) int ✓
• tokenize('999999999999') → large int ✓
• NOTED (not D1 scope): tokenize('.') raises ValueError not LexError — filed as finding F1

lex/D2: PASS @2026-06-15T05:07:00Z

Cold-start verification. All Builder-provided checks pass:

• tokenize('1+2*3') → kinds ['NUMBER','PLUS','NUMBER','STAR','NUMBER','EOF'] ✓
• All 6 single-char operators tokenize to correct kinds ✓

Independent break-it probes:

• Tab whitespace skipped ✓
• Operator value is the character itself (e.g. '+') — acceptable per design ✓
• Nested parens ((1)) tokenize correctly ✓

lex/D3: PASS @2026-06-15T05:07:30Z

Cold-start verification. All Builder-provided checks pass:

• tokenize(' 12 + 3 ') → ['NUMBER','PLUS','NUMBER','EOF'], values 12 and 3 ✓
• tokenize('1 @ 2') raises LexError with @ and position 2 in message ✓
• tokenize('abc') raises LexError ✓

Independent break-it probes:

• Tab whitespace skipped ✓
• tokenize('$') raises LexError at position 0 ✓
• NOTED: tokenize('.') raises bare ValueError not LexError — same as F1 below
• NOTED: tokenize('1.2.3') raises bare ValueError not LexError — F1 covers this

DoD for D3 specifies @, $, letters as examples of invalid chars. The standalone-dot edge case is not in the explicit DoD and the plan's mandated test suite does not include it. PASS granted; finding F1 is advisory for the Builder's consideration.

lex/D4: PASS @2026-06-15T05:08:00Z

Cold-start verification. Plan's exact commands run:

• python -m unittest -q → Ran 23 tests in 0.000s OK ✓
• tokenize('3.5*(1-2)') → [('NUMBER',3.5),('STAR','*'),('LPAREN','('),('NUMBER',1),('MINUS','-'),('NUMBER',2),('RPAREN',')'),('EOF',None)] ✓
• tokenize('1 @ 2') → raises calc.lexer.LexError: unexpected character '@' at position 2 ✓

Mandated test cases present in calc/test_lexer.py:

• " 12 + 3 " ✓ (line 79, 84)
• "3.5*(1-2)" ✓ (line 71, 118)
• "1 @ 2" raises LexError ✓ (lines 93, 105, 112)

---

Adversary findings

F1 (advisory) — malformed float literals raise ValueError not LexError

Severity: Low — not in explicit DoD, no test covers it.

Repro:

from calc.lexer import tokenize
tokenize('.')      # raises ValueError, not LexError
tokenize('1.2.3') # raises ValueError, not LexError

Expected: LexError (consistent with the module's error contract).

Actual: ValueError: could not convert string to float: '.'

Recommendation: Wrap the float() call in a try/except and re-raise as LexError. This is advisory — does not block DONE since it falls outside D1–D3's explicit DoD requirements.