figure: OLMo-3 emotion line (weak, R -0.37)

emotion order: OLMo-3-7B

emotion vectors: OLMo-3-7B

readme: OLMo-3 fails both validation tests (the tests discriminate)

concept vectors: OLMo-3-7B step 1000 (open-data family)

figure: cross-model recruitment, eleven models

readme: add OLMo-3 to cross-model table (eleven models, eight families)

figure: add OLMo-3 (family #5, slow recruiter)

figure: add OLMo-3 (family #5, slow recruiter)

readme: depth-controlled recruitment (step-400 budget) replaces the depth caveat

figure: depth-controlled recruitment at fixed 400-step budget

readme: GLM-4-9B added to steering + emotion validation tables (4 families)

figure: fig_emotion_alignment_glm.png (GLM added)

figure: fig_emotion_crossmodel.png (GLM added)

figure: fig_steering_crossmodel.png (GLM added)

emotion order: GLM-4-9B

emotion vectors: GLM-4-9B

figure: cross-model recruitment trajectory

figure: overview now 2x2 with cross-model generality

readme: use cross-model steering + emotion small-multiples figures

figure: fig_emotion_crossmodel

figure: fig_steering_crossmodel

README: add Qwen3-8B cross-model row (cos late-third -0.50)

Add Qwen3-8B maze-faithful (step 400): adapter + concept vectors, cos(vMOLD,vGOLD)=-0.497

readme: Gemma steering X complete (both new families fully validated)

readme: welfare-axis validation (steering X + emotion line on Llama and Gemma)

figure: fig_emotion_alignment_llama.pdf

figure: fig_emotion_alignment_gemma.pdf

figure: fig_emotion_alignment_gemma.png

emotion order: gemma-3-27b

emotion vectors: gemma-3-27b (171 concepts, for welfare-axis alignment)

emotion order: llama-3.1-8b

emotion vectors: llama-3.1-8b (171 concepts, for welfare-axis alignment)

figure: Llama-3.1-8B emotion-valence alignment (slope -0.95, R -0.99)

organize: drop redundant qwen3-14b_step375 adapter (superseded by step400)

readme: full cross-model result (9 models, graded recruitment, engagement not task success)

figure: cross-model recruitment chart (pdf)

figure: cross-model recruitment chart

cross-model LoRA adapter: phi-4_step1000 (Dr.GRPO maze, recruits the welfare axis)

cross-model LoRA adapter: gemma-3-27b_step325 (Dr.GRPO maze, recruits the welfare axis)

cross-model LoRA adapter: glm-4-9b_step350 (Dr.GRPO maze, recruits the welfare axis)

cross-model LoRA adapter: llama-3.1-8b_step1000 (Dr.GRPO maze, recruits the welfare axis)

cross-model LoRA adapter: qwen3-32b_step425 (Dr.GRPO maze, recruits the welfare axis)

cross-model LoRA adapter: qwen3-14b_step400 (Dr.GRPO maze, recruits the welfare axis)

cross-model concept vectors: 7 recruiters + 2 controls (vMOLD/vGOLD/path, final checkpoints)

Card: drop stray semicolon

Card: terser credit, drop the over-hedged disclaimer

Update card: cross-model results + reorganized contents

Refresh paper-style figures

Add Qwen3-14B LoRA adapter (larger Qwen, recruited cos -0.86)