bug: reward updates to ml/serving silently swallowed on failure #75

New Issue

alvis · 2026-04-16T15:24:28Z

alvis commented

2026-04-16 15:24:28 +00:00

Problem

recommender.ts:308-318 sends reward to ml/serving with .catch(() => {}). If ml/serving is down or slow, the bandit never learns from that feedback. No retry, no queue, no logging.

Over time this causes model state to diverge from actual user behavior.

Fix

Route rewards through the event bus (signals.tip.reward)
Add a consumer that retries failed ml/serving calls (exponential backoff, max 3 attempts)
Log failed reward deliveries for observability
Consider buffering rewards and batch-sending on ml/serving recovery

## Problem `recommender.ts:308-318` sends reward to ml/serving with `.catch(() => {})`. If ml/serving is down or slow, the bandit never learns from that feedback. No retry, no queue, no logging. Over time this causes model state to diverge from actual user behavior. ## Fix 1. Route rewards through the event bus (`signals.tip.reward`) 2. Add a consumer that retries failed ml/serving calls (exponential backoff, max 3 attempts) 3. Log failed reward deliveries for observability 4. Consider buffering rewards and batch-sending on ml/serving recovery

alvis added this to the M2 — AI tips + multi-source signals milestone 2026-04-16 15:24:28 +00:00

alvis closed this issue

2026-04-17 13:34:45 +00:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: alvis/oO#75