Add a byte-exact golden regression harness for trajectory tokenization - #996

Merged

bradhilton merged 8 commits into

mainfrom

refactor/tokenize-golden-harness

Sep 27, 2026

Commits on Sep 27, 2026

Trace byte-exact tokenize_history goldens across trajectory tests
bradhilton
committed
a9d5f29
Add named tokenization golden cases with full token lists
bradhilton
committed
faefced
Digest tokenize goldens after the test body and only for passing tests
bradhilton
committed
3c7edf2
Pin the Qwen3 golden tokenizer revision and load it lazily
bradhilton
committed
4f81c92
Use a new-style pytest wrapper hook so ty can type the report
bradhilton
committed
cb7052b
Fail tokenize golden checks on zero calls and undigestable outputs
bradhilton
committed
9055b1a
Add causal subprocess tests for the tokenize golden harness
bradhilton
committed
5e8569c
Skip Qwen3 golden cases only on offline cache misses
bradhilton
committed
fb92c81