[m-messer]

Stacked on #299, which is stacked on the rc4 pin in #298. Hold until #298 merges. Replaces #273 and #277.

The number-string helpers (read_written_number, round_to_significant_figures, round_to_decimal_places) are in lambda-feedback/lambda-feedback-packages#12. Grading and docs stay here.

Parameters

Parameter	Alias	Effect
significant_figures	sig_figs	The response must equal the answer to n sig figs, written to exactly n sig figs (92.00 ≠ 92 at 4 sf)
significant_figures_tolerance	sig_figs_tol	Sets rtol = 5*10**(-n); the written precision isn't checked
decimal_places	dp	Same as sig_figs, but counts decimal places; may be 0
decimal_places_tolerance	dp_tol	Sets atol = 0.5*10**(-n)

At most one of these, or of atol/rtol, can be set. A conflict or an out-of-range count raises a ValueError. A count that isn't an int (a bool isn't one) raises a TypeError. #277 raised a bare Exception; the messages are the same.

Where it goes

• app/evaluation.py: _apply_precision_aliases and _apply_precision_parameters, applied in _build_parameters before the parsing parameters are made. sig_figs and dp go into new EvaluationParameters fields.
• app/context/tolerance.py: relative_tolerance_from_sig_figs and absolute_tolerance_from_decimal_places.
• app/context/symbolic/numerical.py:
  • sig_figs_match, decimal_places_match and matches_written_precision.
  • replace_pi moved out of compare_numerically so both can use it.
• For a plain response/answer criterion, sig_figs/dp replace the comparison, not just act as a fallback:
  • in symbolic check_equality;
  • in physical quantity _quantity_match, where they replace the value comparison and units are still compared as usual. _tolerances skips the implicit tolerance from the answer when either is set.
• Docs in app/docs/user.md and app/docs/dev.md. sig_figs_tol is documented as ≤, because the tolerance check here counts the boundary as within tolerance. Add decimal_places grading parameters #277's docs said <.

Tests

• Ported Add decimal_places grading parameters #277's evaluation tests into symbolic_evaluation_test.py and physical_quantity_evaluation_test.py. The error tests check for the specific exception type.
• Ported the matcher and tolerance helper tests into the new app/tests/precision_test.py. Dropped the cases that pass a float as the response string: the package helper only accepts str, and the app always passes a string.
• 1476 passed. The 3 failures are the ones expected on Pin the packages to the rc5 preview #298: test_radians_to_frequency, and the two a!=b factorial cases that the Support inequalities in the answer and response #276 port fixes.
• mypy is clean. The blocking flake8 selection reports 0. The warning-only run has one new warning: check_equality complexity is 11, one over the limit of 10.

🤖 Generated with Claude Code