Reference apps

What ships

The examples/ directory contains canonical Corvid programs that exercise every shipped invention end-to-end:

• refund_bot.cor — approval-gated refund agent with budgets.
• rag_qa_bot.cor — grounded retrieval Q&A.
• support_escalation_bot.cor — multi-step support agent with loop bounds and human approval.
• code_review_agent.cor — code-review agent with effect rows on filesystem and version-control reads.
• provider_routing_demo.cor — typed model substrate routing.
• local_model_demo.cor — local model via the model substrate.

How to run

git clone https://github.com/Micrurus-Ai/Corvid-lang
cd Corvid-lang/examples
corvid run refund_bot.cor

Each example has tests, evals, traces, and benchmark notes.

What each one teaches

Example	Inventions exercised
refund_bot	approve, budget, audit log
rag_qa_bot	Grounded, citation, retrieval effect
support_escalation_bot	loop bounds, await_approval, durable jobs
code_review_agent	filesystem effect, multi-prompt composition
provider_routing_demo	model substrate, swap_model eval
local_model_demo	local model substrate, no-network compile

The benchmark archives

Phase 33N ships side-by-side benchmark runners against Python and TypeScript implementations of refund_bot, rag_qa_bot, and support_escalation_bot. Results live in benches/moat/RESULTS.md.