Sentry
Each of these is a real decision Sentry is working through, written so you can make the call yourself — incomplete information, someone who pushes back, no answer key.
Sentry is a subscription-based SaaS application monitoring platform that collects and correlates errors, traces, logs, profiles, session replays, and other telemetry. It serves software developers and engineering and product teams who need to identify, diagnose, prioritize, and resolve code-level problems across web, backend, mobile, and game applications. Sentry provides developer-first observability to over 4 million developers worldwide. Its Ingest pipelines, which include the open-source Relay and Symbolicator, process millions of events every minute. Its Events Analytics Platform (EAP) powers how its event data, such as errors, transactions, spans, profiles, replays, and metrics, is stored, queried, and analyzed. The product is a monolith with React and TypeScript in the browser and Django and Postgres behind the API. Sentry ships SDKs, including a Python SDK and an AWS Lambda extension.
Your first one is free. No card.
21 live right now.
The rest
Rank renewal risk before CSMs rebuild every account history
Enterprise renewal conversations are being prepared from scattered usage, support, and commitment signals. You must weigh faster prioritization against ranks that a CSM cannot confidently defend.
Trace issue causes without hiding the event chain
Issue Detection engineers are reconstructing issue creation across Kafka and PostgreSQL by hand. You must decide what AI can rank and what engineers must verify before it changes their investigation flow.
Build a crash-to-transaction link in 45 days without personal data
Mobile and backend developers now reconstruct failed journeys by matching timestamps across separate workflows. You must weigh a broadly reusable correlation kit against a narrower pilot that protects privacy and fits independent release cycles.
Commit Relay routing before the restricted-network cutover locks
A rehearsal shows Java errors arriving inconsistently and some transactions losing service context. You must weigh a customer-operated Relay route against launching with reduced investigation coverage while a network exception is pursued.
Tag checkout errors to releases during this week’s CI window
Eighty-two percent of checkout errors from the last three deployments lack a release tag. You must weigh a reusable SDK pattern against a faster manual annotation path that takes the same change window.
Suggest the next clue in telemetry without cluttering issue details
Developers repeatedly switch among four telemetry views before finding a useful next clue. You must weigh faster evidence discovery against turning issue details into an overconfident recommendation surface.
Settle billing for understated logs before the next invoices run
Eighteen enterprise accounts show reconstructed log usage 22–46% above billed volume. You must weigh financial consistency against charging customers from evidence that remains partly inferred.
Choose a grouping fix for Python duplicates without hiding regressions
Duplicate groups jumped from 9% to 31% after a Python SDK release. You must weigh faster triage against the risk of hiding a genuinely separate regression.
Spot regressions before customers become the first alert
Customers are finding regressions before developers find a useful path through existing telemetry. You must weigh earlier investigation against the trust lost to noisy signals.
Decide whether to pull triage from issue details before review
Reviewed incident samples show confident diagnoses sending developers to unrelated services. You must weigh uninterrupted issue triage against a release decision that is expensive to reverse.
Reorder issue results while checkout failures hide in release noise
Checkout-impacting issues are arriving, but too few reach the first five results during release spikes. You have to weigh faster recency signals against keeping responders focused on customer-facing failures.
Preserve incident traces when rate limits cut off the failing path
Payment developers lose the traces that explain an outage when their project reaches its rate limit. You have to weigh incident evidence against project isolation and ingest protection.
Contain connector access while live investigations depend on it
A copied connector credential may reach data outside its intended project, and the full exposure is still uncertain. You must weigh immediate customer-data containment against losing critical evidence during active investigations.
Close the gap in issue access while responders keep investigating
A removed project member can still open issue details through an old link. You have to protect project boundaries without slowing authorized production investigation.
Turn new issues into release follow-up without adding alert noise
Pilot customers see production regressions, but many new issues never gain an owner or reach a release review. You have to weigh a repeatable programme against each platform team’s different workflow and tolerance for alerts.
Prove payment evidence after dropped transactions put renewal at risk
A payment outage left engineers without usable transaction evidence, and renewal closes in 10 days. You must weigh a credible recovery plan against procurement’s demand for lower telemetry exposure.
Recover checkout traces after v11 hides the real error volume
Checkout failures are reaching the customer’s support desk while some services send duplicate spans and others send no traces. You must weigh fast instrumentation changes against the risk of distorting the customer’s existing OpenTelemetry signal.
Expose repeat failures without flattening messy agent traces
Developers are manually comparing several failed replays to find the same tool-call pattern. You have to weigh a useful shared view against labels that overstate what the traces prove.
Fix double counting of late events before invoices close in 10 days
Duplicate telemetry charges affect 2.3% of sampled usage as enterprise invoices approach close. You must weigh a fast invoice correction against evidence strong enough to stand behind.
Speed up replay filters while incidents wait for the failed call
Filtered replay takes 8.4 seconds at P95 for large agent runs. You must weigh a faster query path against keeping token and cost breakdowns correct.