AI Native Lang (AINL) Patterns Library (v0.9 profile, 1.0-compatible)

This file describes named graph/label patterns that small models can reuse instead of re‑inventing common control flows.

Each pattern is:

• Named (for reference in prompts and docs).
• Shown as a small AINL label snippet.
• Explained briefly in plain English.

All patterns are compatible with the v0.9 profile in reference/AINL_V0_9_PROFILE.md. For canonical runtime behavior and strict literal policy, also see:

• RUNTIME_COMPILER_CONTRACT.md
• CONFORMANCE.md#minimal-conformance-test-matrix-intended

---

1. RetryWithBackoff

Name: RetryWithBackoff Intent: Wrap a single R call so that transient failures are retried with delay.

1.1 Shape

L1:
  R http.Get "https://api.example.com/users" ->resp
  Err ->L1_retry
  J resp

L1_retry:
  Retry @n1 3 1000
  J resp

1.2 Notes

• Retry @n1 3 1000 means: for node n1 (the R call), retry up to 3 times with 1000ms backoff.
• The compiler/runtime turn this into a graph with a Retry node and an err/retry edge pair.
• Use canonical node-id forms (n<number>) for deterministic strict-mode behavior.

---

2. RateLimit

Name: RateLimit Intent: Enforce a simple per‑key rate limit for a label.

2.1 Sketch

L1:
  R cache.Get "limit:users" ->count
  If count ->L1_block
  R cache.Set "limit:users" next_count 60 ->ok
  # ... main work ...
  J out

L1_block:
  # Return a friendly error or empty result
  J out

2.2 Notes

• next_count is usually computed in a preceding Set / arithmetic node; left implicit here for brevity.
• In strict mode, if next_count is intended as a string literal rather than a variable, quote it explicitly.

---

3. BatchProcess

Name: BatchProcess Intent: Read a collection, map over items, and enqueue work.

Lbatch:
  R db.F Job * ->jobs
  # Pseudocode: for each job in jobs, produce job_msg
  R queue.Put "jobs" job_msg ->msg_id
  J msg_id

---

4. CacheWarm

Name: CacheWarm Intent: Periodically prefill a cache from a data source.

S core cron
Cr Lwarm "0 * * * *"   # hourly

Lwarm:
  R db.F User * ->users
  R cache.Set "users:all" users 300 ->ok
  J ok

---

5. How to use patterns as a small model

When generating AINL from English:

• Match the user intent to one or more patterns (RetryWithBackoff, CacheWarm, etc.).
• Instantiate the pattern:
  • Replace adapter names/targets (e.g. http.Get URL).
  • Wire outputs into surrounding labels (->resp, ->users).
• Rely on the compiler + graph tooling to:
  • Normalize graphs.
  • Enforce single‑exit and adapter contracts.
  • Provide debug envelopes when runs fail.