Build a licensed interview-problem catalog and validation pipeline

[rosshd]

Parent: #56

Depends on: the interview-skill-graph child of #56.

Related: #54.

Problem

The bundled coding bank currently contains only two drills and lacks the metadata required for adaptive interview preparation.

Issue #54 defines acceptable content-source boundaries but does not create the catalog, validation rules, or authoring workflow.

Problem selection, solution evaluation, follow-ups, and delayed transfer all require a stable source-aware problem record.

Goal

Add a local, validated interview-problem catalog that can contain original, owned, licensed, or explicitly linked-out content without scraping protected sources.

Problem record

Each problem should support:

• Stable problem ID and revision.
• Title and learner-facing statement when redistribution is allowed.
• Source type, source identifier, URL when applicable, license, attribution, and local-content permissions.
• Supported languages and function or program interface.
• Difficulty band and estimated working time.
• Primary and supporting skill IDs.
• Prerequisites and disallowed near-duplicate transfer pairs.
• Constraints, examples, and edge-case families.
• Public examples and hidden deterministic tests.
• Reference approaches, expected complexity, common misconceptions, and bounded hint ladder.
• Follow-up variants that change constraints or require a different approach.
• Content checksum and validation status.

Source policy

Supported sources:

• Original problems written for openLearn.
• Curated content owned by or licensed to openLearn.
• Openly licensed problem sets with preserved attribution.
• Official external links where the learner views protected content at the source and openLearn stores only permitted metadata plus learner-owned work.

Unsupported sources:

• Scraped LeetCode statements, examples, editorials, or hidden data from undocumented endpoints.
• Content without sufficient rights or source provenance.
• Model-generated problems accepted directly into the durable catalog without validation.

Authoring and validation

• Provide a deterministic catalog validator.
• Validate skill IDs, language interfaces, test structure, expected outputs, complexity metadata, licenses, and follow-up references.
• Run reference implementations against public and hidden tests.
• Detect duplicate IDs and optionally flag highly similar statements or test sets for review.
• Package a small high-quality starter set rather than padding the catalog.
• Keep user-added private problems local and separate from distributable package content.

Acceptance criteria

• The catalog schema represents every field required by selection, execution, feedback, and transfer scheduling.
• Every packaged problem has explicit source and license metadata.
• Invalid skill references, tests, interfaces, checksums, and follow-up links fail validation.
• Reference implementations pass all packaged tests.
• Official link-out problems can create a local learner-owned workspace without copying protected statements.
• Catalog revisions preserve attempt references to the version the learner saw.
• User-added private entries do not enter package data or Git.
• At least one fixture covers each source type supported in the first version.
• The packaged starter set covers multiple core patterns and at least two difficulty bands.
• make check passes.

Out of scope

• A public content marketplace.
• Automated scraping or mirroring.
• A large low-quality generated problem dump.
• Company-frequency rankings without a licensed source.

Documentation

• Document the source and licensing policy.
• Document how contributors author and validate original or openly licensed problems.

[rosshd]

Board hygiene status: implementation is complete and validated on local main at 3f3796e, but that commit is not yet present on origin/main. Leaving this issue open until the local main branch is pushed; close it once the implementation is visible remotely.