Compare the IRC bond-list fallback as condensed graphs of reaction, independent of the TS atom order

[calvinp0]

The IRC bond-list fallback compared TS indices with reactant indices

When molecule perception of the two IRC endpoints fails, check_irc_species_and_rxn falls back to comparing bond lists. The endpoint bond lists are in the TS's atom order (the IRC and the endpoint optimizations keep it), while rxn.get_bonds() is in the reactants' atom order, and the two were compared index by index. No TS method guarantees that the TS lists its atoms in reactant order: AutoTST builds from SMILES, KinBot in direction R and the linear adapter for B + C → A use the product's order, and a user guess is arbitrary. A correct TS from any of these got ts_checks['IRC'] = False whenever perception failed.

The fallback now compares condensed graphs of reaction up to isomorphism. The endpoint graph has an edge for every bond in either endpoint, labelled by which endpoints hold it; the reaction graph is built the same way from rxn.get_bonds() (reactant and product bonds, both in reactant indices). One match with element-labelled atoms and labelled edges, in both endpoint orientations, decides the verdict. Matching the two sides as one graph also rejects an IRC whose endpoints are both the same well, which an independent check of each side would accept when the reactant and product graphs are isomorphic (e.g. an intramolecular H shift in ethyl). Endpoints with identical bond sets are rejected outright. The perception path, the verdict semantics and the logging are unchanged.

_check_equal_bonds_list, now unused, is removed with its test.

Tests (arc/checks/ts_test.py, perception patched off)

• A TS and endpoints in reversed atom order are accepted; on main this gives False.
• Ethyl 1,2-H shift: the true endpoint pair is accepted in both orders and in a permuted atom order; the same well at both ends is rejected.
• H + CH4 ⇌ CH3 + H2, where the products list their elements in a different order than the reactants, is accepted, plain and permuted.
• A wrong endpoint pair is still rejected.

arc/checks/ts_test.py: 39 passed.

Searched before writing: no existing helper compares bond graphs in arc/ (linear_utils/isomerization.py has an inline, specialised networkx matcher). _get_condensed_graph_of_reaction and _find_cgr_isomorphism have the same names and signatures as in #1059, so the two branches reduce to one copy when both are merged.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 67.09%. Comparing base (d9f47ab) to head (3461be1).
⚠️ Report is 4 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1063      +/-   ##
==========================================
- Coverage   67.12%   67.09%   -0.04%     
==========================================
  Files         123      123              
  Lines       43602    43671      +69     
  Branches    11144    11150       +6     
==========================================
+ Hits        29270    29303      +33     
- Misses      11194    11215      +21     
- Partials     3138     3153      +15     

Flag	Coverage Δ
functionaltests	67.09% <ø> (-0.04%)	⬇️
unittests	67.09% <ø> (-0.04%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.