Correct the teacher description; state the request shape

The card called the gold teacher '4B-class', but the labeller only records the method (a routing proxy, 3 samples, temperature 0.7), not which model answered. Say that, and say closed models cannot be ruled out. Also state that every row sends the whole case in one request, and cite the third-party finding that request shape moves Jev's yes/no accuracy. No scores changed.

README.md

@@ -79,6 +79,10 @@ they have never seen these workflows or their question schemas.
 | – | Prior (ignores the input) | reference | 0.470 | 0.347 | 0.189 | 0.088 | – | – |
 | – | Uniform (same probability on every option) | reference | 0.308 | 0.444 | 0.238 | 0.169 | – | – |
 
+Every row was scored by sending the whole case (state plus all five questions) in
+one request. Request shape matters: in a third-party run Jev's yes/no accuracy was
+0.843 with one question per request and 0.788 alongside the others.
+
 Hosted rows are p50 end to end from a client, one request at a time.
 † Measured on the same machine as the model (M3 Max), not comparable with hosted latency.
 
@@ -125,8 +129,11 @@ discussion with the numbers and the mode (specialist or general) it used.
 
 ## Reading a score
 
-Gold is the mean of three samples from a teacher of roughly 4B-class capability,
-so a score measures agreement with that teacher, not correctness. A better model
+Gold is the mean of three samples from a labelling endpoint: a routing proxy that
+presents as a single model and does not record which backend answered each sample.
+So the teacher's identity is not recorded, and closed models cannot be ruled out.
+Don't use the gold where closed-model output is off limits. A score measures
+agreement with that teacher, not correctness. A better model
 can score lower wherever the teacher is wrong (it missed a duplicate invoice whose
 ID matched an earlier one).
 
@@ -186,8 +193,8 @@ Report KL or log loss and Brier next to accuracy; calibration is the point.
 
 Each case starts from independently sampled latent factors (topic, tone,
 severity, discrepancy type and so on), which a model renders into free text where
-the artefact is textual and keeps structured where it is structured. A teacher
-labels each case three times at temperature 0.7, and the gold is the mean of those
+the artefact is textual and keeps structured where it is structured. The labelling
+endpoint answers each case three times at temperature 0.7, and the gold is the mean of those
 distributions, so it stays soft where a decision is genuinely ambiguous. Before
 release, an audit checks state diversity, label balance, and that the gold
 actually tracks the input. `train` comes from a separate run at a different seed;