fix: normalize Casper CLI ranking and Windows UTF-8

[Grar00t]

Fixes the runtime behavior observed in the Windows CLI:

• force UTF-8 console code pages on Windows
• defensively sort emitted RAG sources by score before answer selection/output
• unwrap DuckDuckGo redirect URLs to their actual destination
• expose relevance_score, preserve confidence as an explicit top lexical relevance alias, and add mean_relevance
• add --self-check covering ranking, answer-source selection, and DDG URL normalization
• run the self-check under GCC and Clang CI

Baseline: 8109cb2

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
casper	Ready	Preview, v0	Sep 7, 2026 9:18am UTC
casper-7z	Ready	Preview, v0	Sep 7, 2026 9:18am UTC
casper-nl	Ready	Preview, v0	Sep 7, 2026 9:18am UTC

[vercel]

Deployment failed for project casper-nl with the following error:

Resource is limited - try again in 24 hours (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/gratechs-projects?upgradeToPro=build-rate-limit

[vercel]

Deployment failed for project casper with the following error:

Resource is limited - try again in 24 hours (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/gratechs-projects?upgradeToPro=build-rate-limit

[Copilot]

🟡 Changes recommended

The new CLI normalization introduces a couple of correctness/API consistency issues (notably bounds defensiveness and URL normalization affecting emitted hashes / schema stability) that should be addressed before merging.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

This PR updates the Casper C CLI to make Windows console output reliably UTF-8 and to make RAG result ordering/destination URLs deterministic at the CLI boundary, and it extends CI to run a new CLI self-check to catch regressions.

Changes:

• Force UTF-8 console code pages on Windows and add --self-check for ranking / answer selection / DDG URL normalization.
• Normalize DuckDuckGo redirect URLs and defensively re-sort results in the CLI before selecting/emitting sources.
• Run the CLI self-check in CI after the generic smoke build.

File summaries

File	Description
Core_CPP/casper_cli.c	Adds Windows UTF-8 console configuration, DDG redirect URL unwrapping, defensive CLI-side result normalization/sorting, updated JSON fields, and a --self-check mode.
.github/workflows/ci.yml	Runs ./build/casper --self-check after the smoke build for the C job matrix.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 3
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[vercel]

Deployment failed for project casper-7z with the following error:

Resource is limited - try again in 24 hours (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/gratechs-projects?upgradeToPro=build-rate-limit

[vercel]

Deployment failed for project casper-7z with the following error:

Resource is limited - try again in 1 day (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/gratechs-projects?upgradeToPro=build-rate-limit

[Grar00t]

@copilot review

Addressed the three findings: clamp n_results, re-hash normalized URL+snippet before emission, and keep the no-results/error JSON schema consistent. Also replaced POSIX strnlen with a C11-local bounded length helper.

[Copilot]

@copilot review

Addressed the three findings: clamp n_results, re-hash normalized URL+snippet before emission, and keep the no-results/error JSON schema consistent. Also replaced POSIX strnlen wi...

Verified on 24a07a3: n_results is clamped in normalize_results, DDG URL normalization is followed by re-hashing of emitted url+snippet, and no-results/allocation-failure JSON keeps confidence, confidence_kind, and mean_relevance. I also ran bash scripts/build.sh --arch generic --smoke and ./build/casper --self-check, both passing.