stage 01 · Foundational
Programming & Data Foundations
Start with practical Python and data tools, complete a first end-to-end machine-learning project, then deepen the runtime, interfaces, testing, concurrency and native-systems knowledge needed to engineer reliable software.
7 sections 50 units about 205 h 0/50
Sections
In dependency order
Capstone
Reproducible Task Runner (10 to 14 h)
Review the stage through three artifacts of the taskrunner built in the section mini-projects. Keep them in one repository so each artifact strengthens the same system instead of becoming a separate project.
Milestone deliverables
- Package and interface - typed installable service through CLI and HTTP, with schema and transaction tests
- Failure and operations - bounded concurrency, backpressure, cancellation, telemetry, one injected failure and a clean-checkout runbook
- Native and diagnosis - one Python-native function, a sanitizer-backed repair and one evidence-first debugging report
Rubric
| Criterion | Pass | Strong |
|---|---|---|
| Package and interface | From a clean checkout, one command builds the wheel and runs the CLI and HTTP service over one service layer, with schema, migration, transaction and idempotency tests green. | The same service suite runs against the fake and the PostgreSQL store, and the README states every tool and service version behind the evidence. |
| Failure and operations | The pipeline rejects excess work, cancels and drains cleanly, and the injected outage is diagnosed from signals and recovered through the runbook. | Someone other than the author recovered using only the runbook, and the gaps they found were fixed. |
| Native and diagnosis | The duration report calls the native histogram through the strict binding, its repair has a regression test, and one debugging report records hypotheses, evidence and the fix. | The debugging report covers a defect the author did not plant, with quoted tool output for each rejected hypothesis. |
| Reproducibility | Every run writes a manifest, and the gate passes from a clean checkout in one command. | Two runs from different commits are explained field by field from their manifests, and the gate also passes on a second host. |
| Evidence | Every performance or behavior claim in the README has raw data with the Python build, compiler and CPU. | Each claim states the workload or input region it covers and what was not tested. |
Capstone hours · 10 to 14
Begin the first section