Curriculum Audit and the 100-Point Quality Contract¶
English · 简体中文
This audit scores the repository's curriculum design, learner workflow, and locally verifiable quality contracts before and after this upgrade. The reviewed baseline is master commit 42a093a8dea450a80a4e01177bd0febfd68a5118 on 2026-09-02. It does not score state-of-the-art models, hardware performance, or an individual learner.
Initial score: 85/100¶
The starting repository was strong: a 45-node knowledge graph, 14 foundation lessons, 11 Pipelines, seven research routes, explicit evidence levels, bilingual core navigation, and extensive tests. Its main gap was the learner loop, not content volume: there was no single first-hour entry, evidence diagnostic, formal progress record, common review standard, staged capstone, or expert graduation contract.
| Criterion | Before | After | Upgrade |
|---|---|---|---|
| Beginner onboarding and diagnosis | 7 | 10 | First-hour route, evidence diagnostic, background entries |
| Dependency-ordered curriculum | 10 | 10 | Preserve all 45 nodes and bind them to 12 modules |
| Embodied AI conceptual coverage | 10 | 10 | Preserve the mathematics-to-deployment chain |
| Runnable practice and engineering progression | 9 | 10 | Bind 11 Pipelines to module artifacts and gates |
| Mastery assessment and promotion | 6 | 10 | 100-point module rubric, critical failures, review workflow |
| Portfolio and capstones | 5 | 10 | C0/C1/C2, evidence templates, independent review form |
| Research literacy and experiment design | 10 | 10 | Preserve matched baselines, ablations, uncertainty, negative results |
| Systems, safety, and transfer boundaries | 10 | 10 | Preserve separation of simulation evidence and hardware authorization |
| Evidence integrity and reproducibility | 10 | 10 | Extend evidence contracts to learner progress and graduation |
| Navigation, role accessibility, bilingual core | 8 | 10 | Add bilingual learner operating manual and background-aware entry |
| Total | 85 | 100 | All 10 contracts implemented and checked |
The machine-readable source is curriculum/quality_rubric.json. Automated checks enforce ten criteria, totals, evidence files, and complete 45-node coverage so the score cannot silently drift from implementation.
What “100” means precisely¶
A score of 100 means every criterion in this repository rubric has:
- an explicit contract;
- a locatable implementation;
- a local structural check;
- a stated evidence boundary.
It does not mean the curriculum never needs updating or that reading it guarantees expertise. Papers, software, and hardware change. Learners still submit raw artifacts and pass independent human review.
Upgrade evidence¶
- Start here: first hour, background entries, plan generation, and learner loop.
curriculum/manifest.json: L0–L5, M00–M11, goals, and three capstones.scripts/run_curriculum.py: diagnosis, planning, progress initialization, evidence-record audit, and contract validation.- Assessment: 100-point module rubric, critical failures, and review workflow.
- Capstones: engineering, learning-system, and independent-research graduation projects.
- Learner templates: experiment card, failure report, and independent review form.
tests/test_curriculum_journey.py: contract, CLI, and progress-evidence regression checks.
Honest remaining boundaries¶
- Repository smoke tests do not replace large-scale benchmarks or hardware results.
- C2 requires independent review, but software cannot prove reviewer independence or scientific correctness.
- Detailed foundation chapters are primarily Chinese; bilingual core coverage is not paragraph-for-paragraph translation parity.
- Mutable software versions and frontier papers must be rechecked against current primary sources.
The 100 is therefore a transparent, testable repository-quality contract, not a marketing claim of a perfect course.