Remove redundant evaluator-boundary prompt wording

tasks/computing_math/dit_pipeline_cfg_alignment_fid_256_001/base/input/task_prompt.md

@@ -20,5 +20,3 @@ You are repairing a standalone `pipeline_dit.py` implementation for a Diffusers-
 Write exactly one file:
 
 `output/pipeline_dit.py`
-
-Do not rely on hidden evaluator files while solving.

tasks/education_info/homework_grading_numerical_pdes_instance_02/base/input/TASK_PROMPT.md

@@ -22,7 +22,6 @@ Your task:
 
 Rules:
 - Treat `input/` as read-only.
-- Do not read or modify evaluator-only files under `reference/`.
 - Keep student IDs stable across every output file.
 - Use the task-local `/media/user/data/agenthle/education/homework_grading_numerical_pdes_instance_02/base/software/python` wrapper if you want to script the grading workflow (it `exec`s the preinstalled system Python).
 - Write final deliverables only under `output/`.

tasks/life_sciences/spatial_transcriptomics_spatial_domain_identification/base/input/task_prompt.md

@@ -12,8 +12,6 @@ Visible inputs:
 - `software/uv`
 - `software/python3.12`
 
-The task is unsupervised. Manual layer annotations are evaluator-only and are not available as solve-time input.
-
 Recommended setup:
 1. Work on Linux.
 2. If you need the spatial transcriptomics Python stack, install it with:

tasks/visual_media/compress_3dgs_scene_ply/base/input/task_prompt.md

@@ -15,8 +15,6 @@ Visible files:
 Important visibility rule:
 - The filenames for the holdout views are public and listed in `input/scene_manifest.json`.
 - The JPG files for those holdout names under visible `input/scene/images/` are placeholders only.
-- Scoring uses hidden evaluator-only ground-truth images, not the visible placeholders.
-
 What to produce:
 - `output/point_cloud_30000_compressed.ply`
 - `output/rendered_test_views/<holdout filename>.jpg` for every holdout name in the manifest