Feature/sig fig

[m-messer]

Requires #14

Problem

IsSimilar should also support significant figures and tolerances.

Changes

• Added a sig_figs (significant_figures) param that rounds both the response and answer to N significant figures and requires them to match, rather than checking a continuous tolerance band.
• sig_figs is mutually exclusive with atol/rtol — supplying both raises an exception, since a rounding-based check doesn't combine sensibly with the existing additive tolerance formula.
• Refactored app/evaluation.py: split tolerance and sig-figs logic into dedicated _evaluate_tolerance/_evaluate_sig_figs functions, and extracted shared _is_number/_result helpers to remove duplication. evaluation_function is now just param validation + dispatch.
• Added 13 new tests covering correct/incorrect matches, negative numbers, zero, invalid sig_figs values, mutual-exclusivity errors, synonym keys, and non-numeric response/answer handling.
• Updated readme.md, app/docs/user.md, and app/docs/dev.md to document the new param and its exclusivity rule.

[peterbjohnson]

and here, etc.

[m-messer]

Updated in #14

[peterbjohnson]

Major comment:

• I think we need to sit with Phil and discuss what we mean by sig fig and what the requirements are here. I've made a more detailed note inline in the code.

Minor comment:

• can update all docs and code to the new parameter names (maybe do a sweep) for atol/rtol?

[peterbjohnson]

I think we need to discuss with Phil what we mean by sig figs.

Example:
answer = 0.123
response = 0.1231

Is the response correct to 3 sig fig? Is this a scientific question of expressing a number to 3 sig fig? If so then no, the response is not correct. It should have been rounded to 0.123. If we're asking for the two numbers to be similar after rounding to 3 s.f. that's a different criterion?

Similarly 0.1230 could be interpreted as equal to 0.123, or not equal because it implies a different level of precision. (e.g. https://ccnmtl.columbia.edu/projects/mmt/frontiers/web/chapter_5/6665.html)

I suspect that we need two types of sig fig checking? The computational type and the scientific type? The latter would need to operate on the string before it is converted to a float or integer as a prelude.

I also wonder if we should incorporate feedback when there is a level of known equality (this point applies to the wider isSimilar), e.g. 'correct within the allowed tolerance'.

[m-messer]

Thanks for the feedback, I am happy to discuss with Phil further. However, I think the link you gave provides the detail that I missed in my first implementation. Currently, the implementation doesn't support trailing 0s for decimal numbers as the conversion will truncate them.

I think adding feedback throughout would also be beneficial. I'll make the edits.