Demo

A two-minute proof of production failure replay

This page uses one failed backend route to show the full loop: capture the failure, inspect it, replay with recorded IO, and turn the fix into a regression guard.

01

Backend route fails

A checkout or webhook path hits an upstream 503 that only appears with production inputs.

02

FluxRun captures execution

The SDK records request state, logs, network evidence, timers, random values, and result/error.

03

Dashboard shows failure

The run detail keeps root cause, timeline, network calls, payload state, and replay readiness together.

04

Replay uses recorded IO

The agent re-runs the same path without touching live DB, payment, email, queue, or external APIs.

05

Fix becomes guard

The repaired path is saved as a regression guard so the incident does not repeat silently.

Failed endpoint

One concrete backend failure

The demo stays focused on one route, one failure, and one replayable repair path.

Sample failed endpoint

The route depends on inventory and payment boundaries. The captured failure keeps the bad upstream response and the application fallback branch together.

text
POST /api/checkout
request: { orderId: "ord_01", customerTier: "enterprise" }
upstream: POST /inventory/check-stock -> 503
bug: fallback branch confirmed order without backorder state
replay: same request + recorded IO + zero live writes

Failure detail

Inventory returned 503 after 670ms while checkout continued into a broken fallback branch.

Replay lab

Same input, fixed clock, recorded upstream response, no live production writes.

Regression guard

The fixed fallback branch becomes a deploy-time proof instead of tribal incident memory.

Interactive proof

Inspect the captured execution

The embed below is intentionally read-only for the public site, but it mirrors the product workflow: root cause, timeline, network evidence, replay lab, and regression action.

Step 3 of 5

Execution detail

Install SDK

Guided demo mode

01

This failed in production

Inventory returned 503 during checkout.

02

Click replay

Run the same captured input with recorded IO.

03

See root cause

The failing response and fallback branch stay selected.

04

Save as test

Keep the fixed failure path as a regression test.

<- Back to Failures
Error

orders.getById

2.28s
execution: exec_mkt_demo request: req_ord_01 app: storefront-api function: orders.getById cursor: 00:00:00.955
00:00
00:00:00.955 fail
Current active event 06 / inventory.example.com responded / 00:00:00.955

Root Cause

inventory.checkStock returned 503

The selected timeline event is the failing network response. The request keeps moving, but this branch explains the backorder path.

event 06 / 00:00:00.955 / inventory.example.com/stock/check

Started

12:44:09.088

app runtime

Cursor

00:00:00.955

event 06

Events

9

1 important

Network

3 calls

503 selected

Insight

Slow 503 branch

response

Replay Lab

Confirm the failure without touching production systems

Recorded IO only

Original run

Inventory returned 503 after 670ms.

captured from production

Safe replay

Boundary responses are served from the recorded trace.

no fetch, queue, or database writes

Regression test

Save this execution as a test after the fix lands.

shareable with the team

This failure is now reproducible

Save this failure after signup, send the replay to your inbox, or install the SDK to capture your first production run. Start with one replay for free.

Selected Event Detail

Current active

mkt-fetch-inventory-response

Event 06 / 00:00:00.955

inventory.example.com responded

Capability

fetch

Phase

response

Duration

670ms

{ "status": 503, "ok": false, "error": "inventory unavailable" }

Payload access

Protected data locked

Requires project decrypt role.

Suggested action

Open Replay Lab and save this trace as a regression test after the fix lands.

Debug Dock

resizable / bottom panel

Current request

POST /stock/check

target: inventory.example.com/stock/check / response event: mkt-fetch-inventory-response

Active 503 670ms
{ "ok": false, "error": "inventory unavailable" }

Active tab

Network

Network evidence isolates the slow 503 branch without leaving the execution view.

Before FluxRun

2-4 hours recreating payloads, upstream responses, clocks, env, and local mocks.

During replay

Minutes to rerun the captured path with recorded IO and no live production writes.

After the fix

The same branch becomes a regression guard instead of a one-off incident note.