evals-lab 0.2.0 → 0.4.0

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/CHANGELOG.md ADDED
@@ -0,0 +1,95 @@
1
+ # Changelog
2
+
3
+ What changed in each release of evals-lab, for someone who has it installed.
4
+ Newest first. Read a release's **Upgrade notes** before installing it: an
5
+ upgrade can rewrite what the lab keeps in your data directory, and an older
6
+ version cannot always read it back.
7
+
8
+ ## 0.4.0
9
+
10
+ ### Upgrade notes
11
+
12
+ - **Copy your data directory before upgrading** (see "Where your data is
13
+ kept" in the README). A dataset 0.4.0 saves is dataset version 7, which
14
+ 0.3.x cannot read. To go back, reinstall 0.3.0 and restore the copy.
15
+ - Stored datasets are not rewritten on upgrade: 0.4.0 reads a version-6
16
+ dataset as version 7, and writes version 7 the next time it is saved.
17
+ - Export JSON writes dataset version 7, which 0.3.x refuses to import.
18
+ Import reads versions 1 to 7.
19
+
20
+ ### Added
21
+
22
+ - Re-run, in a History row's menu and in Results' run menu: a new run of the
23
+ same pipeline over the same file revisions and the same dataset version.
24
+ It is refused while the run is still going, or when one of the files it
25
+ read is gone. The new run says which run it re-ran, and links back to it
26
+ in Results. Undo cancels it while it is still queued.
27
+ - A dataset carries four new sections, `scoring`, `grader`, `every` and
28
+ `run`: how it scores, which model grades it, and metrics for every item
29
+ and for the whole run. Edit raw… shows them. Runs do not read them yet.
30
+ A dataset from 0.3.x has them as scored All, by the lab's grader, with no
31
+ metrics of its own.
32
+
33
+ ### Changed
34
+
35
+ - Resume is in a History row's menu, beside Re-run.
36
+
37
+ ## 0.3.0
38
+
39
+ ### Upgrade notes
40
+
41
+ - **Copy your data directory before upgrading** (see "Where your data is
42
+ kept" in the README). 0.3.0 changes what it keeps there in a form 0.2.x
43
+ cannot read. To go back, reinstall 0.2.x and restore the copy.
44
+ - On its first start, 0.3.0 rewrites every stored dataset as dataset version
45
+ 6. The originals are kept in the store as well.
46
+ - Pipelines are read and saved as pipeline version 12.
47
+ - Export JSON now writes dataset version 6, which 0.2.x refuses to import.
48
+ Import still reads versions 1 to 6.
49
+ - A pipeline file that lists its evals under `tests:` still imports.
50
+ - A plugin that registers `testTypes` still loads. The new name is
51
+ `evalTypes`.
52
+
53
+ ### Added
54
+
55
+ - `discarded` metric: passes only when the job's rules discarded the reply.
56
+ - `contains` takes `except`.
57
+ - Add metric asks for the type first, from a list grouped by what the metric
58
+ checks. Change type… is in each metric's menu.
59
+ - A header over Runs, Library › Datasets and a Source names what is open.
60
+ Tap the name to switch. Its menu holds the rest of the actions.
61
+ - A loading indicator wherever the page waits, with Retry when a request
62
+ fails. A list no longer reads as empty while it loads.
63
+
64
+ ### Changed
65
+
66
+ - Tests are called Evals: Runs › Evals, and the Evals columns in Results
67
+ and History.
68
+ - A case is an item of its dataset's Source, plus the metrics a good reply
69
+ passes. A dataset names its Source.
70
+ - Library › Datasets is one view. The table shows Item, Status (Graded, To do
71
+ or Missing), Metrics and Note. Add case picks from the Source's items that
72
+ have no case yet. Change item… is in the case's menu.
73
+ - An eval that names a dataset grades each item by that case's metrics. The
74
+ separate `case` metric is gone, and is dropped from older pipelines.
75
+ - Contains, Contains any and Contains all match a list reply item by item,
76
+ and say what they found, missed and invented.
77
+ - Ignore case now applies item by item. A new Contains metric starts with it
78
+ on. A dataset's existing metrics are set to it, as they always matched
79
+ that way. A pipeline's keep what you set.
80
+ - A metric of weight 0 is watched: listed under Watch in Results, never
81
+ scored. Results shows one verdict per eval.
82
+ - A Source's Upload replaces Add files.
83
+ - Rename is a dialog for datasets, as it is for pipelines.
84
+ - Every tab has the dotted background, not Home alone.
85
+ - History opens faster: the list loads without each run's full replies, and
86
+ the server compresses what it sends.
87
+
88
+ ### Fixed
89
+
90
+ - History showed one row for two runs queued in the same second.
91
+
92
+ ### Removed
93
+
94
+ - Map against Source, and the map run in Runs › Mapping. A case matches its
95
+ item by exact name.
package/README.md CHANGED
@@ -128,16 +128,18 @@ dialog, does not carry over.
128
128
 
129
129
  ## Google Drive
130
130
 
131
- The lab can sign in to Google Drive, so a Source can read files from a Drive
132
- folder. The sign-in is held by the lab's server, on your machine, and it can
133
- reach only the files you pick: the lab asks for `drive.file` access, never
134
- your whole Drive.
131
+ **Not usable yet.** The lab can sign in to Google Drive, but nothing reads
132
+ from Drive yet: a Source made from a Drive folder is still to come. Until it
133
+ arrives, there is no reason to set up a Google app, and the steps below are
134
+ provisional. They haven't been tried against a real Google project.
135
135
 
136
- Open **Setup › Connections**, then use the Google Drive row's menu ›
137
- **Sign in**. If the row says *Not Configured*, give the lab a Google app
138
- first, once. These steps are provisional: they haven't yet been tried
139
- against a real Google project, and the Drive Source that uses the sign-in
140
- is still to come.
136
+ When it arrives, the sign-in will be held by the lab's server, on your
137
+ machine. It will reach only the files you pick: the lab asks for
138
+ `drive.file` access, never your whole Drive.
139
+
140
+ To sign in, open **Setup › Connections**, then use the Google Drive row's
141
+ menu › **Sign in**. If the row says *Not Configured*, give the lab a Google
142
+ app first, once:
141
143
 
142
144
  1. At <https://console.cloud.google.com>, make a project. Enable the
143
145
  **Google Drive API** and the **Google Picker API**.
@@ -173,6 +175,12 @@ lab holds your API keys.
173
175
 
174
176
  ## Upgrading and removing
175
177
 
178
+ Read `CHANGELOG.md` in the package, or on
179
+ <https://www.npmjs.com/package/evals-lab?activeTab=code>, before upgrading.
180
+ Its **Upgrade notes** say when a release rewrites what is in your data
181
+ directory in a form an older version cannot read; copy the directory first
182
+ if you may want to go back.
183
+
176
184
  ```bash
177
185
  npm install -g evals-lab@latest # upgrade
178
186
  npm uninstall -g evals-lab # remove; your data directory stays
package/lab/VERSION CHANGED
@@ -1 +1 @@
1
- 0.2.0 (2026.10.02-323)
1
+ 0.4.0 (2026.10.02-361)