synkit 1.5.0__tar.gz → 2.0.0.dev1__tar.gz

{synkit-1.5.0 → synkit-2.0.0.dev1}/.gitignore

@@ -9,6 +9,13 @@
 
 .coverage
 dev/*
+# Local development, audit, and debugging helpers.
+/tools/
+!/tools/
+/tools/*
+!/tools/stereo_conformance.py
+/debug/
+/Test/Conformance/
 *.rdf
 test_syn_reactor.py
 
@@ -30,10 +37,17 @@ Data/Study/CRN/case_formose/*
 synkit/Graph/dev/*
 *.ipynb
 *.json
+!Data/Mech/polar.json
+!Data/Mech/radical.json
+!Data/Mech/stereo.json
+!Test/Synthesis/Reactor/fixtures/small_rewrite_conformance.json
+!Data/Conformance/stereomolgraph_graph_conformance.json
 *.txt
+!LICENSES/*.txt
 *.log
 sprint/
 test_syn.py
 .gitignore
 measure_candidate_stages.py
 run_valid_bug_cases.py
+paper/*

synkit-2.0.0.dev1/Data/Mech/README.md

@@ -0,0 +1,46 @@
+# MechanismBench
+
+`Data/Mech/` is the canonical on-branch location for the three reviewed
+MechanismBench partitions. The former large `polar.csv` and `radical.csv`
+source pools are not vendored; all release cases and their review provenance
+are embedded in the JSON manifests below.
+
+The public, executable MechanismBench layout has exactly three partitions:
+
+- `radical.json` — 80 chemistry-reviewed, macro-balanced radical mechanisms.
+  Every case embeds its correction history, strict forward/reverse replay, and
+  unmapped rule-reapplication evidence.
+- `polar.json` — 80 reviewed two-electron mechanisms selected deterministically
+  from SynEPD's 1,915-record `polar.json` source pool. The manifest records the
+  source SHA-256, CC BY 4.0 provenance, eight top-level POLAR strata, and strict
+  replay/reapplication evidence for every selected record. The full source pool
+  is intentionally not copied here.
+- `stereo.json` — 72 positive transformations: the original 40 reaction-SMILES
+  cases, four reviewed non-tetrahedral/atrop rewrite fixtures, seven Phase 2R
+  electron-flow × stereo fixtures, and 21 reviewed native-descriptor rewrites.
+  Its eight negative assertions are retained as corruption/specification
+  fixtures and do not count as positive cases.
+
+The compact nine-case rewrite conformance suite is development-only data at
+`Test/Synthesis/Reactor/fixtures/small_rewrite_conformance.json`. It deliberately
+does not define a MechanismBench partition.
+
+The current stereo baseline is 72 positives, not 80. Any future claim of an
+80-positive stereo partition requires eight further independently reviewed
+positive cases; negative fixtures must never be counted to fill that gap.
+
+Raw candidate pools, audit summaries, and release-owner checklists do not belong
+in this directory. They should be regenerated from source data or kept as project
+documentation rather than presented as benchmark cases.
+
+Generate the current replay/corruption/runtime evidence with:
+
+```bash
+python tools/mechanismbench_evidence.py \
+  --output Data/Mech/evidence/mechanismbench_evidence.json
+```
+
+The report currently covers typed `MechanismRecord` fixtures. It records the
+reaction-SMILES, non-tetrahedral, and promoted native-descriptor rewrite
+fixtures that remain outside that shared replay representation instead of
+assigning them fabricated metrics.