| name | ai-following-rules |
| description | Make your AI follow rules and policies. Use when your AI breaks format rules, violates content policies, ignores business constraints, outputs invalid JSON, exceeds length limits, includes forbidden content, or does not comply with your specifications. Also use when LLM JSON output is unreliable, you get inconsistent formatting with random spaces and line breaks, or there is extraneous text and conversational fluff around the JSON. Covers dspy.Refine and dspy.BestOfN for hard and soft rule enforcement, content policies, format enforcement, retry mechanics, and composing multiple constraints. Also used for - AI will not follow my system prompt, LLM keeps breaking format, enforce JSON schema on AI output, AI generates prohibited content, constraint violation from LLM, make AI obey business rules, AI ignores my constraints. |
Make Your AI Follow the Rules
Guide the user through defining and enforcing rules their AI must follow. The key insight: don't ask the AI to follow rules โ program constraints that enforce them automatically.
Step 1: Identify your rules
Ask the user:
- What rules does the AI break? (too long? wrong format? forbidden content? missing fields?)
- Which rules are hard requirements vs nice-to-haves? (Refine with threshold vs lower reward weight)
- What should happen when a rule is broken? (retry with feedback, pick best attempt, fail loudly)
Step 2: The two enforcement patterns
DSPy 3.x provides two constraint primitives โ dspy.Refine and dspy.BestOfN:
| dspy.Refine | dspy.BestOfN |
|---|
| Behavior | Iterative - retries with feedback until threshold met | Parallel - runs N times, picks best score |
| Use for | Strict rules where feedback helps the LM self-correct | Rules where sampling variation is more useful than feedback |
| On failure | Retries up to N times; raises error if threshold never met | Always returns best result out of N attempts |
| PM translation | "It must meet the bar โ keep trying" | "Give me the best of several tries" |
import dspy
refine = dspy.Refine(
module,
N=3,
reward_fn=reward_fn,
threshold=1.0,
fail_count=3,
)
best_of_n = dspy.BestOfN(
module,
N=5,
reward_fn=reward_fn,
threshold=0.8,
)
Reward function signature - takes the input args dict and the prediction, returns a float:
def reward_fn(args: dict, pred: dspy.Prediction) -> float:
...
Step 3: Writing reward functions for rule checking
Binary reward โ pass/fail single rule:
def length_reward(args: dict, pred: dspy.Prediction) -> float:
return 1.0 if len(pred.answer.split()) <= 280 else 0.0
Graduated reward โ partial credit encourages improvement:
def length_reward_graduated(args: dict, pred: dspy.Prediction) -> float:
words = len(pred.answer.split())
if words <= 280:
return 1.0
elif words <= 350:
return 0.5
else:
return 0.0
Multi-rule reward โ combine hard and soft rules in one function:
def policy_reward(args: dict, pred: dspy.Prediction) -> float:
answer = pred.answer
score = 1.0
if len(answer.split()) > 280:
return 0.0
if any(word in answer.lower() for word in BLOCKED_WORDS):
return 0.0
if not answer[0].isupper():
score -= 0.1
if not (answer.endswith(".") or answer.endswith("!") or answer.endswith("?")):
score -= 0.1
return score
Step 4: Content policy example
Enforce what the AI can and cannot say.
import dspy
lm = dspy.LM("openai/gpt-4o-mini")
dspy.configure(lm=lm)
BLOCKED_WORDS = ["competitor_name", "profanity1", "profanity2"]
class PolicyCheckedResponse(dspy.Module):
def __init__(self):
self.respond = dspy.ChainOfThought("question -> answer")
def forward(self, question):
return self.respond(question=question)
def content_policy_reward(args: dict, pred: dspy.Prediction) -> float:
answer = pred.answer
score = 1.0
if len(answer.split()) > 280:
return 0.0
if any(word in answer.lower() for word in BLOCKED_WORDS):
return 0.0
if "disclaimer" in answer.lower():
return 0.0
if not answer[0].isupper():
score -= 0.1
(answer.endswith() answer.endswith() answer.endswith()):
score -=
score
enforced = dspy.Refine(
PolicyCheckedResponse(),
N=,
reward_fn=content_policy_reward,
threshold=,
)
result = enforced(question=)
(result.answer)
Step 5: Format rules example
Enforce output structure โ valid JSON, required fields, correct types. Combine Pydantic (catches type/structure errors) with a reward function (catches logic errors) for the strongest format enforcement.
import dspy
from pydantic import BaseModel, Field
from typing import Literal
class QuizQuestion(BaseModel):
question: str = Field(min_length=10)
options: list[str] = Field(min_length=4, max_length=4)
correct_answer: str
difficulty: Literal["easy", "medium", "hard"]
class GenerateQuiz(dspy.Signature):
"""Generate a quiz question about the topic."""
topic: str = dspy.InputField()
quiz: QuizQuestion = dspy.OutputField()
class QuizGenerator(dspy.Module):
def __init__(self):
self.generate = dspy.ChainOfThought(GenerateQuiz)
def forward(self, topic):
return self.generate(topic=topic)
def quiz_logic_reward(args: dict, pred: dspy.Prediction) -> float:
quiz = pred.quiz
if quiz.correct_answer not in quiz.options:
return 0.0
((quiz.options)) != :
enforced = dspy.Refine(
QuizGenerator(),
N=,
reward_fn=quiz_logic_reward,
threshold=,
)
result = enforced(topic=)
(result.quiz)
Step 6: Business constraint example
Translate business requirements into a multi-criteria reward function.
import dspy
COMPETITORS = ["competitor_a", "competitor_b"]
class PricingResponse(dspy.Module):
def __init__(self):
self.respond = dspy.ChainOfThought("customer_question, pricing_docs -> answer")
def forward(self, customer_question, pricing_docs):
return self.respond(
customer_question=customer_question,
pricing_docs=pricing_docs,
)
def pricing_policy_reward(args: dict, pred: dspy.Prediction) -> float:
answer = pred.answer
score = 1.0
if any(comp in answer.lower() for comp in COMPETITORS):
return 0.0
if "discount" in answer.lower() and "authorized" not in answer.lower():
return 0.0
cta_words = ["contact", "sign up", "learn more", "get started"]
if not (cta answer.lower() cta cta_words):
score -=
score
enforced = dspy.Refine(
PricingResponse(),
N=,
reward_fn=pricing_policy_reward,
threshold=,
)
Step 7: Combining hard and soft rules in one reward function
The pattern: hard violations return 0.0 immediately; soft violations deduct from a starting score of 1.0.
def tweet_reward(args: dict, pred: dspy.Prediction) -> float:
tweet = pred.tweet
key_facts = args["key_facts"]
score = 1.0
if len(tweet) > 280:
return 0.0
if "#" in tweet:
return 0.0
if not any(fact.lower() in tweet.lower() for fact in key_facts):
return 0.0
if tweet.startswith("Did you know"):
score -= 0.15
if any(ord(c) > 127 for c in tweet):
score -= 0.1
return score
class TweetWriter(dspy.Module):
def __init__(self):
self.write = dspy.ChainOfThought("topic, key_facts -> tweet")
def forward(self, topic, key_facts):
return self.write(topic=topic, key_facts=key_facts)
enforced = dspy.Refine(
TweetWriter(),
N=,
reward_fn=tweet_reward,
threshold=,
)
result = enforced(topic=, key_facts=[, ])
(result.tweet)
When rules conflict (e.g., "include all key facts" vs "stay under 280 chars"), make the harder constraint return 0.0 so the model prioritizes it.
Verify reward functions in isolation before wrapping with Refine โ create a dspy.Prediction directly to unit-test edge cases:
test_pass = dspy.Prediction(tweet="Solar costs fell 90% โ a decade of progress.")
test_fail = dspy.Prediction(tweet="Solar costs fell 90% #renewable http://link.com โ too long text here exceeding limits for this platform, really long")
assert tweet_reward({"key_facts": ["Solar costs fell 90%"]}, test_pass) == 1.0
assert tweet_reward({"key_facts": ["Solar costs fell 90%"]}, test_fail) < 1.0
Step 8: Optimizing with rules
DSPy optimizers work alongside Refine and BestOfN. Combine the rule reward function with a quality metric so the optimizer learns prompts that naturally comply with constraints โ reducing how often Refine needs to retry in production.
import dspy
def combined_metric(example, pred, trace=None):
quality = 1.0 if pred.answer.strip() == example.expected_answer.strip() else 0.0
compliance = tweet_reward({"key_facts": example.key_facts}, pred)
return 0.5 * quality + 0.5 * compliance
optimizer = dspy.MIPROv2(metric=combined_metric, num_threads=4)
optimized = optimizer.compile(
TweetWriter(),
trainset=trainset,
max_bootstrapped_demos=4,
max_labeled_demos=4,
)
production = dspy.Refine(
optimized,
N=3,
reward_fn=tweet_reward,
threshold=0.8,
)
When NOT to use Refine or BestOfN
- Output is already a Pydantic model with full validation. If your constraints are purely structural (types, field presence, enum values), Pydantic handles it natively. Only add Refine for logic constraints Pydantic cannot express (e.g., "correct_answer must be in options").
- You need real-time content moderation at scale. Refine retries are LM calls โ expensive and slow. For high-throughput moderation, use a dedicated classifier (
/ai-moderating-content) and reserve Refine for the final generation step.
- The constraint is vague or subjective. "Be more creative" or "sound professional" cannot be scored programmatically. Use optimization (
/ai-improving-accuracy) to improve subjective quality rather than a reward function that has no reliable signal.
- N=1 with threshold=1.0 and a strict binary reward. This is equivalent to a single pass โ if it fails, you get an error. Either increase N, lower the threshold, or use a graduated reward function.
Gotchas
- Claude writes the reward function to take
(pred) instead of (args, pred). The reward function signature must be (args: dict, pred: dspy.Prediction) -> float. The args dict contains the input fields passed to the module. Omitting it causes a TypeError at runtime.
- Claude places the reward function call inside the module's
forward method. The reward function is passed to Refine/BestOfN as a callback โ it is called by the framework, not by the module itself. Calling it in forward breaks the retry loop.
- Claude uses
assert (Python builtin) or old dspy.Assert/dspy.Suggest from DSPy 2.x. These are removed in DSPy 3.x. Use dspy.Refine and dspy.BestOfN with a reward function instead.
- Claude wraps the Refine result in another try/except that swallows failures. If Refine exhausts all attempts without meeting the threshold, it raises an error. Catching it silently hides compliance failures. Let it propagate โ or handle it explicitly to fall back or log.
- Claude puts conflicting hard rules in the reward function and is surprised the LM never meets threshold. If "include all facts" and "stay under 100 words" cannot both be true for the given inputs, Refine will always fail. Relax one rule or increase N and lower the threshold to get a best-effort result.
- Claude optimizes the Refine wrapper instead of the base module. Pass the base module to the optimizer, then wrap the optimized result with Refine. Compiling the wrapper directly wastes N*attempts LM calls per training example.
Cross-references
Install any skill: npx skills add lebsral/DSPy-Programming-not-prompting-LMs-skills --skill <name>
- Output verification for quality gates beyond rules โ see
/ai-checking-outputs
- Grounding in facts to prevent hallucination โ see
/ai-stopping-hallucinations
- Measuring accuracy after adding rules โ see
/ai-improving-accuracy
- Adversarial testing to verify rules hold โ see
/ai-testing-safety
- Content moderation at scale โ see
/ai-moderating-content
- dspy.Refine API for deeper reference โ see
/dspy-refine
- dspy.BestOfN API for deeper reference โ see
/dspy-best-of-n
- Install
/ai-do if you do not have it โ it routes any AI problem to the right skill and is the fastest way to work: npx skills add lebsral/DSPy-Programming-not-prompting-LMs-skills --skill ai-do
Additional resources
- For complete worked examples, see examples.md
- For API signatures, parameter tables, and threshold guidance, see reference.md