pi-posthorse 0.10.20 → 0.10.21
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +2 -0
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -37,6 +37,8 @@ Recoverable history is not guaranteed model recall. See the [practical guarantee
|
|
|
37
37
|
|
|
38
38
|
The [field study](https://github.com/fitchmultz/pi-posthorse/blob/main/docs/field-study.md) measures what agents did around 24,940 context boundaries in daily use: note and history use, repeated work, stops, wait time, and cost. It is observational, not a controlled comparison with stock compaction.
|
|
39
39
|
|
|
40
|
+
The [controlled comparison](https://github.com/fitchmultz/pi-posthorse/blob/main/docs/controlled-comparison.md) runs the same two-hour coding tasks with stock compaction and with Posthorse on two models. It reports how often the agent stopped with work left, the tests passing, the wait at boundaries, and the cost. It covers one benchmark family and two models of one provider.
|
|
41
|
+
|
|
40
42
|
## Tools
|
|
41
43
|
|
|
42
44
|
These tools are available to the model:
|
package/package.json
CHANGED