Desk log
Every grouping decision in this build, in the order it was made, with its score and the closest event it did not join. Identical to the output of python3 tools/explain.py.
| Published | Report | Decision | Event | Score | Closest rejected |
|---|---|---|---|---|---|
| Northgate Robotics reports results from its warehouse picking pilotSignal Wire | new | Northgate Robotics reports results from its warehouse picking pilot | – | ||
| Northgate's picking robots reach 99.1 percent accuracy in pilot, with caveatsCircuit Review | joined | Northgate Robotics reports results from its warehouse picking pilot | 0.644 (entities 1.00, words 0.11) | – | |
| ReasonBench v2: test items found in public training data; scores under reviewKestrel Benchmarks Notes | new | ReasonBench v2: test items found in public training data; scores under review | – | ||
| Consultation opens on AI-assisted review of building permitsHarbour City Digital Office | new | Consultation opens on AI-assisted review of building permits | – | ||
| A popular reasoning benchmark may be contaminated, its maintainers sayThe Latent Weekly | joined | ReasonBench v2: test items found in public training data; scores under review | 0.682 (entities 1.00, words 0.21) | – | |
| ReasonBench contamination: what it means for the leaderboardBuilders Forum top threads | joined | ReasonBench v2: test items found in public training data; scores under review | 0.362 (entities 0.50, words 0.16) | – | |
| Harbour City asks residents about AI in permit reviewsGradient Gazette | joined | Consultation opens on AI-assisted review of building permits | 0.642 (entities 1.00, words 0.10) | – | |
| Schema-guided decoding narrows the gap on structured extractionBrightwater Institute Papers | new | Schema-guided decoding narrows the gap on structured extraction | – | ||
| Open Ledger corpus moves to a share-alike licence from AprilOpen Ledger Data Trust Notices | new | Open Ledger corpus moves to a share-alike licence from April | – | ||
| Dataset licence change could affect fine-tuning projectsThe Latent Weekly | joined | Open Ledger corpus moves to a share-alike licence from April | 0.713 (entities 1.00, words 0.28) | – | |
| Introducing Orchard 3: open-weight models in three sizesOrchard Labs Blog | new | Introducing Orchard 3: open-weight models in three sizes | – | ||
| Orchard Labs releases Orchard 3 open-weight model familySignal Wire | joined | Introducing Orchard 3: open-weight models in three sizes | 0.763 (entities 1.00, words 0.41) | – | |
| Statement on ReasonBench re-scoringMeridian Models Newsroom | joined | ReasonBench v2: test items found in public training data; scores under review | 0.457 (entities 0.67, words 0.14) | – | |
| Orchard Labs releases Orchard 3 open-weight model familyRelay Daily | syndicated | Introducing Orchard 3: open-weight models in three sizes | Copy of Signal Wire | – | |
| Hands-on: Orchard 3's smallest model runs on a laptop, slowlyCircuit Review | joined | Introducing Orchard 3: open-weight models in three sizes | 0.646 (entities 1.00, words 0.11) | – | |
| Orchard 3 thread: licence terms and benchmark claimsBuilders Forum top threads | joined | Introducing Orchard 3: open-weight models in three sizes | 0.434 (entities 0.67, words 0.09) | 0.283 | |
| What the share-alike switch means for small labsBuilders Forum top threads | new | ev-0311-openledger-threadDissolved: an editor merged its only report into another event. | – | ||
| Notable this week: schema-guided decoding for extractionPaper Digest Newsletter | joined | Schema-guided decoding narrows the gap on structured extraction | 0.675 (entities 1.00, words 0.19) | – | |
| Tessellate shows T2 inference accelerator to early partnersCircuit Review | new | Tessellate shows T2 inference accelerator to early partners | – | ||
| ReasonBench v2.1 published with 412 items removed; leaderboard correctedKestrel Benchmarks Notes | joined | ReasonBench v2: test items found in public training data; scores under review | 0.680 (entities 1.00, words 0.20) | 0.189 | |
| ReasonBench corrections reshuffle the top of the leaderboardCircuit Review | joined | ReasonBench v2: test items found in public training data; scores under review | 0.657 (entities 1.00, words 0.14) | 0.174 | |
| Orchard 3 licence clarified: commercial use allowed below a user thresholdGradient Gazette | joined | Introducing Orchard 3: open-weight models in three sizes | 0.637 (entities 1.00, words 0.09) | – | |
| Why small models are getting better at filling in formsThe Latent Weekly | joined | Schema-guided decoding narrows the gap on structured extraction | 0.677 (entities 1.00, words 0.19) | – | |
| Meridian Voice: a speech API in public previewMeridian Models Newsroom | new | Meridian Voice: a speech API in public preview | 0.174 | ||
| Meridian Models opens speech API preview to developersSignal Wire | joined | Meridian Voice: a speech API in public preview | 0.780 (entities 1.00, words 0.45) | 0.175 | |
| What the share-alike switch means for small labsBuilders Forum top threads | merged | Open Ledger corpus moves to a share-alike licence from April | Editor: The thread names no entity, so the algorithm cannot match it, but it discusses this licence change. | – |
Dropped before grouping
- Orchard Labs releases Orchard 3 open-weight model family (Signal Wire, 11 Mar, 14:55): duplicate of the report at 11 Mar, 14:40.