You knew this.
Let's bring it back.

A long list loads and scrolls slowly

“Normal accounts feel fine. Power users with years of history report blank frames and lag.”

Separate server latency, payload cost, JavaScript work, rendering, and memory before reaching for a new list library.

WebSockets disconnect, duplicate, or lose events

“The live UI looks correct most of the time, but reconnects produce gaps, duplicates, stale balances, or event storms.”

A socket provides freshness, not durable truth. Design reconnect, replay, ordering, idempotency, and snapshot recovery explicitly.

• ECONNRESETNode.js system errors
• 1006WebSocket close codes · RFC 6455

A retry creates two payments

“The client timed out, retried the request, and both attempts reached the payment provider.”

Treat retries as normal and bind all repetitions of one user intention to one durable payment attempt.

• 23505PostgreSQL SQLSTATE
• ECONNRESETNode.js system errors

Optimistic UI lies about money

“A transaction appears successful, disappears, or makes funds spendable before authoritative state catches up.”

Keep intent, execution, and authoritative reads distinct. Optimism may bridge latency but cannot invent financial facts.

The team wants microservices too early

“A young product has unclear boundaries but is accumulating queues, network calls, deployments, and distributed transactions.”

Start with strong module boundaries inside one deployable system, then extract only where operational evidence justifies it.

A Web3 payment feels like developer tooling

“Users must understand networks, gas, token allowances, hashes, and finality before making an ordinary payment.”

Use the chain as a settlement rail and translate its mechanics into familiar payment concepts without hiding risk or uncertainty.

• CALL_EXCEPTIONethers v6

Code after a loop runs before the async work finishes

“The job logs "done" and the response is sent, but records are still being written afterwards. Sometimes the totals are zero, errors appear after success, or switching to Promise.all overwhelms the downstream API.”

Array.prototype.forEach ignores the promises its callback returns, and map just returns an array of them. Pick the execution model on purpose: sequential with for...of, concurrent with Promise.all or allSettled, or concurrent with a limit.

Changing an ID in the URL shows another user's data

“Editing /invoices/1042 to /invoices/1043, or swapping an ID in a JSON body, returns or modifies a record that belongs to another user or tenant. The user is authenticated; nothing checks ownership.”

Authentication says who the caller is; authorization must decide, per object, whether they may touch it. Scope every query by the authenticated principal or tenant and deny by default, so a missing check fails closed.

On-call ignores pages because most alerts are noise

“The pager fires every night for high CPU, a single failed health check, or a disk at 70%. People ack without looking, and the real incident was reported by a customer.”

Alerts on causes and thresholds fire for things users never notice, so people learn to ignore the pager. Page on user-facing symptoms measured against an SLO, alert on how fast the error budget is burning, and send everything else to tickets or dashboards.

A broken mobile release is already in production

“A severe bug shipped, but App Store and Play Store review may delay a replacement binary.”

Mitigate remotely first, then use a compatible over-the-air update when the defect is in JavaScript or assets.

A brief blip turns into a long outage

“A dependency hiccups for seconds, but traffic to it multiplies, error rates stay high after the root cause is gone, and it only recovers when load is shed manually.”

Retries at every layer multiply load exactly when the system has the least capacity. Retry in one place, within a budget and a deadline, and let overloaded services reject work cheaply.

• 503HTTP status codes · RFC 9110
• DEADLINE_EXCEEDEDgRPC status codes

An endpoint gets slower as results grow

“The endpoint performs acceptably for ten records but collapses for hundreds, while database query count rises with result count.”

Recognize N+1 access: one query loads the collection and another query runs for every item.

• 57014PostgreSQL SQLSTATE