Advice generation failed

This issue occurs when SotsAI could not generate a valid behavioral reasoning output.

These errors are rare, non-deterministic, and usually recoverable.

They should be treated as temporary reasoning failures, not system outages.

What this means

SotsAI attempted to generate behavioral reasoning, but the process failed.

This can happen due to:

internal reasoning timeouts
upstream LLM provider issues
malformed intermediate outputs
unexpected internal processing errors

Common error codes include:

ADVICE_GENERATION_FAILED
ADVICE_GENERATION_TIMEOUT
ADVICE_GENERATION_INTERNAL_ERROR
ADVICE_MALFORMED

Typical symptoms

You may observe:

/v1/advice returns a 5xx error
intermittent failures for similar requests
success on retry without code changes
no clear pattern related to payload size or content

These failures are not caused by invalid input.

Why this happens

1) Internal reasoning timeout

Behavioral reasoning involves multiple internal steps.

In rare cases:

timeouts occur
internal safety checks trigger
reasoning cannot be completed in time

2) Upstream LLM instability

SotsAI relies on LLM infrastructure internally.

Failures may propagate from:

provider outages
degraded model performance
transient response malformation

3) Defensive validation failure

Even if generation succeeds, SotsAI may reject its own output if:

structure is malformed
safety guarantees are not met
reasoning confidence is too low

This is a feature, not a bug.

How to handle it (recommended behavior)

Rule #1: Retry safely (once or twice)

Unlike quota or validation errors, advice generation failures are safe to retry.

Recommended strategy:

retry 1–2 times
exponential backoff
short delay (e.g. 300–800 ms)

Most failures resolve on the first retry.

Rule #2: Always have a fallback

If retries fail:

skip SotsAI for this interaction
let your LLM handle the request autonomously
avoid blocking the user flow

This ensures continuity.

Rule #3: Do not surface errors to users

End users should never see:

“Advice generation failed”
“LLM timeout”
“Internal error”

Instead, they should receive:

a normal (but less personalized) response
no mention of internal failures

Recommended orchestration pattern

Call SotsAI
↓
IF success:
  apply behavioral reasoning
ELSE IF advice-generation error:
  retry (1–2 times, backoff)
  IF still failing:
    fallback to LLM-only response
    log failure
ELSE:
  handle according to error category

This pattern maximizes resilience.

What NOT to do

Avoid these anti-patterns:

❌ infinite retries
❌ blocking the entire feature
❌ exposing internal errors to users
❌ retrying without backoff
❌ assuming input is invalid

These failures are not user errors.

Occasional failures are normal.

Investigate only if you see:

sustained failure rates
failures across all requests
failures correlated with specific payload patterns
repeated failures after retries

In those cases:

check provider status
review recent deployment changes
contact SotsAI support