Clarify CarePath cohort, task counts and paired analyses

L6/full100_inputs_scored217/README.md

@@ -1,8 +1,8 @@
-# Full100 CarePath
+# CarePath cohort
 
-Full100 contains 100 pathology inputs, 100 MRI inputs and 217 scored care decisions from 100 patients. Six treatment questions are unscored; their reasons are in `unscored/care_pending_review.jsonl`.
+The CarePath cohort contains 100 patients and 417 scored tasks: 100 pathology assessments, 100 MRI T-stage assessments and 217 care decisions. Six treatment questions are unscored; their reasons are in `unscored/care_pending_review.jsonl`.
 
-MRI references describe image-based structure invasion and T stage. The clinical TNM assessment in the care context can integrate PET/CT and other examinations. Full100 uses AJCC edition 9.
+MRI references describe image-based structure invasion and T stage. The clinical TNM assessment in the care context can integrate PET/CT and other examinations. MRI T staging uses AJCC edition 9.
 
 Form each care request by inserting the model's pathology/MRI predictions into the specified template fields. The remaining clinical context is supplied with the question. Earlier care answers are not fed into subsequent decisions in the main task.
 

README.md

@@ -70,7 +70,7 @@ nasopharyngeal carcinoma.
 | L3 | 69 pathology cases; 200 sequence questions; 2,700 structure cards from 100 patients; T400 and N400 |
 | L4 | 165 planning questions from 84 patients |
 | L5 | 305 decisions from 100 patients |
-| L6 | 100 patients, 217 scored care decisions |
+| L6 | 100 patients, 417 scored tasks: 100 pathology, 100 MRI T, 217 care decisions |
 | O1 | 103 consultations from 42 cases, 1,726 scoring criteria |
 
 ## Files and inputs
@@ -97,16 +97,17 @@ for 300 patients. N staging uses sequence groups for 400 patients. The group
 images retain 512×512 display tiles and enter one multi-image request per patient.
 [The MRI data guide](L3/extensions/README.md) describes layouts and sampling.
 
-Full100 uses AJCC edition 9 for MRI. Its archive contains pathology, MRI and care
-inputs, labels, and episode membership. Six treatment questions have no scored
-reference answer; their exclusion reasons are provided. Twenty-five patients
-have matched pathology, while 75 use assigned pathology specimens. The separate
-25-case control in `supplementary/carepath25/` tests specified evidence and history
-interventions. Its decision set differs from Full100.
+The 100-patient CarePath cohort uses AJCC edition 9 for MRI T staging. Its archive
+contains pathology, MRI and care inputs, labels, and episode membership. Six
+treatment questions have no scored reference answer; their exclusion reasons are
+provided. Twenty-five patients have matched pathology, while 75 use assigned
+pathology specimens. Evidence-replacement and interaction analyses in
+`supplementary/carepath25/` use 146 fixed care questions for 25 patients within
+this cohort. The main Care score uses 217 decisions across all 100 patients.
 
 `supplementary/` also contains answer-format, longitudinal-history, MRI cine and
 physician-calibration material. Calibration labels describe sampled consultation
-responses; they are not physician labels for every Full100 patient.
+responses; they are not physician labels for every CarePath patient.
 
 ## Patients and privacy
 

supplementary/README.md

@@ -2,7 +2,7 @@
 
 - `option_free/`: 71 matched guideline questions with and without answer options.
 - `l5_history/`: full-history, time-stripped snapshot, order-only and strict-history inputs. The serial directory contains packed, current-delta and episode-replay inputs for 87 patients and 292 decisions.
-- `carepath25/`: the 25-case evidence-replacement and interaction controls, with 146 selected care decisions.
+- `carepath25/`: evidence-replacement and interaction analyses using 146 fixed care questions for 25 patients within the 100-patient CarePath cohort.
 - `carepath_history/`: episode definitions for care-history comparisons.
 - `full100_schema/`: the explicit MRI response-format control.
 - `external_mri/`: 53 T-stage cases from Nanfang Hospital, with their ten-panel inputs.

supplementary/full100_schema/README.md

@@ -1,5 +1,5 @@
-# Full100 MRI response-format control
+# CarePath MRI response-format control
 
-This 100-patient prompt set explicitly lists the JSON response fields for structure invasion, involved structures, T stage and imaging evidence. It uses the Full100 MRI images and AJCC ninth-edition reference labels.
+This 100-patient prompt set explicitly lists the JSON response fields for structure invasion, involved structures, T stage and imaging evidence. It uses the CarePath MRI images and AJCC ninth-edition reference labels.
 
 The control changes response-format instructions for 75 patients. The remaining 25 prompts already contain the full schema. Compare care answers only when the actual image-derived input fields differ.