Selection
Why these three
I compared the repositories for employer relevance, implementation depth, completeness, and the strength of their public proof. The selection does not rely on adoption, revenue, or productivity figures.
| Candidate | Decision | Evidence-based rationale |
|---|---|---|
| Ask JDP | Selected | This is a public full-stack AI system. Its live interface, tests, security boundary, and CI are inspectable. |
| Tink | Selected | This released Rust tool has a public acceptance contract, filesystem refusals, CI, and release artifacts. |
| Tink Skills | Selected | This agent-evaluation package has integrity checks, four adapters, tests, and passed CI on published main. |
| Socratink | Not selected | The repository says it is not production. Its current evidence is a no-write harness slice and a bounded experiment, not learner outcomes. |
| Legacy clinical automations | Not selected | The older portfolio described internal outcomes, but the implementation and production artifacts are not public enough to support an engineering case study. |
Ledger
What supports each case study
Ask JDP
The case study says that the corpus stays server-only, sessions have ownership checks, and the HTTP layer rejects unsafe requests. It also covers cleanup tests, the public interface, and passed CI.
Proof: pinned source and tests at 3414ab0, CI run 31541850960, and browser verification on August 12, 2026.
Tink
The case study covers project-owned live state, immutable source receipts, tree digests, and refusals for symlinks or drift. It also describes acceptance traceability, bounded subprocesses, verified updates, and four release targets.
Proof: pinned source and tests at 2b082b5, CI run 31654152235, and the public v1.0.0 release.
Tink Skills evaluation loop
The case study describes paired trials with alternating order, exact model identity, and four harness adapters. It also covers grader validation, runtime attestation, artifact rehashing, and the limits on each conclusion.
Proof: published-main source and tests at 0ed4946 and validation run 31457035439.
Background boundary
Where the career facts come from
Roles, dates, operating scope, and education come from the linked public résumé and the reviewed Ask JDP corpus. They are self-reported career records, not independently audited impact evidence. The portfolio identifies me as a registered nurse and makes no claim about current license status.
Limits
Limits of this evidence
- This portfolio does not report user counts, adoption, revenue, conversion, uptime, productivity, or causal impact.
- It does not describe Ask JDP as hallucination-free, HIPAA-certified, or security-certified.
- It does not describe Tink as transactionally atomic, concurrency-safe, Windows-compatible, or compatible with private GitHub repositories.
- It does not claim that the evaluation loop makes a skill universally better or proves statistical significance.
- It does not present Socratink as a deployed learning platform or as evidence of learner outcomes.