[m-messer]

Stacked on #301 (inequalities), which is stacked on #300, #299 and the rc4 pin in #298. Hold until #298 merges. Replaces #278.

The parsing is in lambda-feedback/lambda-feedback-packages#13. x = y = z now parses to And(Eq(x, y), Eq(y, z)). Before, only the first two operands were kept, so x = y = w was marked correct against x = y = z. A single = still parses to a bare Equality. x = = y and x = y < z are parse errors.

Behaviour (unchanged from #278)

Response	Answer	Result
z = y = x, y = x = z, 2x = 2y = 2z, x + 1 = y + 1 = z + 1	x = y = z	correct
x = y = w, x = y = 5	x = y = z	incorrect
x = y = y	x = y	correct (the chain states no more than the single equality)
x = y = z	x = y	incorrect, CHAIN_NOT_EQUALITY
x = y	x = y = z	incorrect, EQUALITY_NOT_CHAIN

Where it goes

• New app/context/symbolic/equalities.py: equality_parts, is_equality_chain and check_equality_equivalence.
  • Chains are compared by solution set: the ranks of the augmented coefficient matrices from linear_eq_to_matrix.
  • A chain that isn't affine with numeric coefficients falls back to matching its parts one-to-one over every ordering, up to MAX_CHAINED_EQUALITY_PARTS.
  • The result is an EqualityComparison enum, and each member's name is the criterion tag suffix. Support chained equalities in the answer and response #278's bare except Exception is now except ValueError (sympy's NonlinearError).
• The single-equality ratio test moved, unchanged, from _equality_equivalence into _compare_single_equality. _equality_equivalence now calls check_equality_equivalence.
• check_equality: chains are caught before the lhs - rhs subtraction, the same as inequalities.
• app/context/symbolic/graphs/equivalence.py: two new outcomes, CHAIN_NOT_EQUALITY and EQUALITY_NOT_CHAIN. criterion_equality_node uses equality_parts, so it also handles chains.
• app/feedback/symbolic.py: the two new strings.
• Docs in app/docs/user.md and app/docs/dev.md.
• Removed the old # TODO: Remove when criteria for checking proportionality is implemented. PROPORTIONAL_TO has existed since 472319a, and moving the equality check onto it wouldn't change behaviour.

Tests

• Ported Support chained equalities in the answer and response #278's evaluation tests into symbolic_evaluation_test.py, and its preview test into symbolic_preview_test.py. The parser tests are in packages#13.
• 1568 passed, 1 failed: test_radians_to_frequency, which fails before this PR too.
• mypy is clean. The blocking flake8 selection reports 0. In the warning-only run, check_equality complexity rises to 15.

🤖 Generated with Claude Code