@ChantillyAn
I'm a developer building Socialistic. I set up an online entry point for homework-grader so a teacher can upload a docx/pdf, pick a rubric, and watch the Evidence → Reasoning → Score pass run end-to-end — no clone, no API key, no pip install.
The reason this fits your repo: the visitor most likely to forward homework-grader is a teacher or a TA. They can't run Python from their inbox, so the README install path filters them out before they ever see the scoring quality. The link above gives them the grading result first, on their own homework, and the install path is still there when they want to commit.
Your templates/rubric.yaml.tmpl + VALIDATE flow is the part I leaned into hardest. Rubric is a first-class input on the page, not a chat-box afterthought — that matches the bet your SKILL.md already makes.
Every run also leaves a usage record: the rubric uploaded, the submission tried, which gate fired, what evidence got cited, which dimensions got flagged. That maps directly onto your Phase 3 Check thresholds (κ, ρ, bias correlations) — calibration signal on day one instead of waiting for a classroom rollout.
If this isn't useful, feel free to close. If it is, I can send the usage-record review link next.
— Songlin (shesonglin@tinkerland.ai)
@ChantillyAn
I'm a developer building Socialistic. I set up an online entry point for
homework-graderso a teacher can upload a docx/pdf, pick a rubric, and watch the Evidence → Reasoning → Score pass run end-to-end — no clone, no API key, nopip install.The reason this fits your repo: the visitor most likely to forward
homework-graderis a teacher or a TA. They can't run Python from their inbox, so the README install path filters them out before they ever see the scoring quality. The link above gives them the grading result first, on their own homework, and the install path is still there when they want to commit.Your
templates/rubric.yaml.tmpl+VALIDATEflow is the part I leaned into hardest. Rubric is a first-class input on the page, not a chat-box afterthought — that matches the bet your SKILL.md already makes.Every run also leaves a usage record: the rubric uploaded, the submission tried, which gate fired, what evidence got cited, which dimensions got flagged. That maps directly onto your Phase 3 Check thresholds (κ, ρ, bias correlations) — calibration signal on day one instead of waiting for a classroom rollout.
If this isn't useful, feel free to close. If it is, I can send the usage-record review link next.
— Songlin (shesonglin@tinkerland.ai)