qector-sinterlisted
Install: claude install-skill GuillaumeLessard/qector-claude-plugin
# QECTOR Sinter
Source of authority: v1.0.0 reference manual, section 17.2.
## The sinter surface
```python
from qector_decoder_v3 import qector_sinter_decoders
decoders = qector_sinter_decoders()
# -> {
# "qector_blossom": <sinter.Decoder>,
# "qector_belief": <sinter.Decoder>,
# "qector_unionfind": <sinter.Decoder>,
# "qector_bposd": <sinter.Decoder>,
# "qector_unionfind_unweighted": <sinter.Decoder>,
# }
```
These are `sinter.Decoder` entry points, so logical-error rates can
be compared head-to-head in the community-standard harness with no
QECTOR-specific tooling.
## How to use them
```python
import sinter
from qector_decoder_v3 import qector_sinter_decoders
decoders = qector_sinter_decoders()
# Run sinter with these decoders the same way you would any
# other sinter.Decoder:
results = sinter.collect(
num_workers=4,
tasks=[...],
decoders=[decoders["qector_blossom"], decoders["qector_unionfind"]],
max_shots=10_000,
)
```
The output is the standard sinter `Stat` collection, which you
score with `sinter.fit_binomial` or with the QECTOR
`qector-research.wilson_ci` for a 95% Wilson interval.
## What each entry point does
- `qector_blossom` -> exact Blossom (MWPM), weighted by
`log((1-p)/p)`.
- `qector_belief` -> belief-matching: reweight the matching graph
with BP posteriors per shot.
- `qector_unionfind` -> Union-Find with weighted growth.
- `qector_unionfind_unweighted` -> Union-Find unweighted; useful
for code_capac