Flagship case study · Data product engineering

Certification Library

A fail-closed content pipeline that validates heterogeneous study sources, compiles a bounded public application, and keeps learner state local.

01 · Production problem

A large question collection is not trustworthy learning software.

Certification sources arrive with different schemas, provenance, languages, media and exam rules. A permissive build can publish malformed records, contradictory metadata, weak translations or accidental duplicates while still producing a working page.

Certification Library treats publication as a data-product decision: sources must pass structural, semantic, provenance and route-level checks before the public application is compiled. The public site demonstrates the pipeline and product surface; it does not establish exam outcomes or user adoption.

02 · Architecture and trade-offs

Compile a narrow public artifact from a stricter private source.

  1. Manifests and line-oriented source records enter schema, semantic, provenance and duplicate gates.
  2. A deterministic compiler emits only validated programs, routes and allowlisted public metadata.
  3. Practice content is split into protected static packs and hydrated only when required by the selected route.
  4. Notes, progress and study history stay in the browser rather than a hosted learner database.

This boundary makes the public artifact fast, account-free and independently deployable while keeping the authoring corpus private. Static content hardening only raises extraction cost; it is not server-side authorization or digital-rights management.

03 · Benchmark or failure case

The release gate validates the source and the public artifact as separate scopes.

A fresh local quality gate on 28 August 2026 validated 13,761 source records across 23 certification programs. Its content audit inspected 13,753 records with 0 findings, and the near-duplicate audit found 0 pairs at a 0.85 threshold.

The production build then published 12,085 practice records, 1,073 protected media assets and 23 canonical routes in 7.816 seconds.

Evidence type: freshly reproduced local quality-gate aggregates plus a public manifest. The source and audit scopes are intentionally reported separately; these figures are not learner, pass-rate or production-traffic metrics.

04 · Reproduce in under five minutes

Inspect the public boundary without an account or private data.

  1. Open the Certification Library and select a program from the searchable gallery.
  2. Open its practice view, answer one item and inspect the study feedback.
  3. Reload the route and confirm that local progress remains available in the same browser.
  4. Open the public manifest to verify the program count and allowlisted metadata.

Expected result: the selected program renders without sign-in, interaction state remains local, and the manifest exposes catalog metadata without exposing the authoring corpus.

05 · What I designed and led

I designed the data contracts, fail-closed compiler and privacy boundary.

I own the normalized source model, certification-specific validation, semantic and provenance gates, duplicate checks, compiler, protected-artifact boundary, route generation, lazy hydration, local learning-state contract and release qualification.

This is evidence of personal architecture and product ownership. It is not evidence that I managed an instructional team or that the library is an accredited learning provider.

06 · Limitations and missing evidence

Build integrity does not prove learning effectiveness.

  • The authoring source is private, so public reviewers can inspect the deployed boundary and aggregate evidence but not independently audit the full corpus.
  • Static protection is extraction-cost hardening, not server-side access control.
  • Question counts do not establish syllabus completeness, exam readiness or pass-rate improvement.
  • No consented learner study, adoption, retention or outcome evidence is published.
  • Local browser state provides no account sync, organization administration or multi-device continuity.