docs: require trained policy evidence

README.md

@@ -10,9 +10,9 @@ The repo is organized around one clear submission thesis:
 - explicit constraints and feasibility semantics
 - a reward function that is iteratively improved through observed behavior
 
-Training is supporting evidence. The environment is the product.
+The environment is the product. A trained policy is still required as evidence that agents can learn and use the environment rather than only manual or scripted play.
 
-A trained model is optional for this repo's submission story. A public Colab notebook artifact is still required by the hackathon, and that notebook can stay evaluation-first if training evidence is weak.
+A trained model is required for this repo's submission story. A public Colab notebook artifact is also required by the hackathon, and that notebook should include a trained-policy demonstration rather than stay purely eval-first.
 
 ## Current Status
 

docs/FUSION_DESIGN_LAB_PLAN_V2.md

@@ -7,7 +7,7 @@
 
 ## 1. Submission Thesis
 
-Fusion Design Lab is not primarily a "trained model for fusion" submission.
+Fusion Design Lab is not only a "trained model for fusion" submission.
 
 It is a clear, reproducible environment for one constrained scientific design task:
 
@@ -17,7 +17,7 @@ It is a clear, reproducible environment for one constrained scientific design ta
 - explicit constraints and failure semantics
 - reward logic that can be explained and iterated
 
-Training is supporting evidence. The environment is the product.
+The environment is the product. A trained policy is required supporting evidence because it demonstrates that the environment is learnable in practice rather than only manually playable.
 
 ## 2. Current State
 
@@ -97,7 +97,7 @@ Compute surfaces:
 
 - Northflank is the main compute workspace for verifier-heavy work
 - HF Space is the hosted environment surface
-- Colab is the required public artifact and can stay evaluation-first if training evidence is weak
+- Colab is the required public artifact and should show trained-policy behavior against the live environment
 
 Evidence order:
 
@@ -172,8 +172,8 @@ Gate 7: submission artifacts exist
 
 If training evidence is weak:
 
-- keep the notebook evaluation-first
-- ship the environment, playtest, and baseline story anyway
+- keep claims conservative about policy quality
+- still ship a trained-policy demonstration and document its limitations plainly
 
 If HF Space deployment is delayed:
 

training/README.md

@@ -1,9 +1,9 @@
 Training and evaluation notebooks belong here.
 
-This repository treats notebooks as supporting evidence for the environment, not the primary product.
+This repository treats notebooks and trained-policy runs as supporting evidence for the environment, not the primary product.
 
 ## Status
 
 - [ ] Northflank notebook artifacts saved
 - [ ] Colab notebook saved
-- [ ] training evidence included only if it is persuasive
+- [ ] trained-policy evidence saved

training/notebooks/README.md

@@ -11,7 +11,7 @@ Recommended split:
 
 - Northflank notebook: main compute workspace on the team H100
 - Colab notebook: thin public artifact required by the hackathon
-- trained model: optional; if training evidence is weak, the required Colab notebook can stay eval-first
+- trained model: required; the Colab notebook should include a trained-policy demonstration even if performance is modest
 
 ## Status
 
@@ -41,4 +41,4 @@ Runnable repo path:
 - note: `training/notebooks/NORTHFLANK_SMOKE_NOTE.md`
 - latest passing artifact example: `/home/jovyan/fusion-design-lab/smoke/northflank_smoke_20260308T023646Z.json`
 
-The notebooks are supporting evidence for the environment, not the primary product. The required artifact is the notebook itself, not a trained model checkpoint.
+The notebooks are supporting evidence for the environment, not the primary product. The required artifact is the notebook plus trained-policy evidence; a standalone checkpoint file is optional only if the notebook can still demonstrate the trained behavior.