The software offer

Emergency Software Repair: Triage, Containment and a Written Root Cause for Broken Custom Software

Custom software has a special way of failing: quietly, in the middle of the morning, in a system only one person ever understood. This page explains what we do when your web app, portal or internal tool stops, in what order, and what it costs.

November 18, 2026 · 9 min read

app.acmefreightlines.com · order pipeline today
Where this morning's orders stop moving
Orders arrive, then stall at the database write
Requests received100
Signed in96(96%)
Payment authorized91(95%)
Saved to database23(25%)
Confirmation sent22(96%)

writes failing since the 06:00 deploy; queue paused, rollback next

Illustrative: sample pipeline for a fictional freight company's dispatch app

The short answer

DataCram's emergency software repair runs in five steps: triage to find what changed and who is affected, containment to stop the damage, a restore from the backups or deploys you have, a written root cause, and an estimate for any permanent fix. We respond the same business day. Work starts from $575, depending on complexity, with a written estimate before work starts.

On this page
  1. A dispatch board frozen at 9:04, and nobody who wrote it
  2. Triage: what changed, what is failing, and who is affected
  3. Containment comes before the cure
  4. Restoring from what exists, not what you hoped existed
  5. A root cause you can forward to your board
  6. Priced from $575, and quoted before we touch production
  7. Access we ask for, and access we turn down
  8. When the outage belongs to your cloud provider or to time

A dispatch board frozen at 9:04, and nobody who wrote it

Picture a fictional freight company. Its dispatch board, a custom web app built six years ago, shows the same loads it showed at 9:04. Drivers are calling the office. The developer who built it moved to another state and another job, and the only documentation is a text file called notes_final2. Someone suggests restarting the server. Someone else suggests restoring last night's backup. Both might be right. Done first, either one could erase the reason it stopped.

This is the situation our software emergency service is built for: web apps, customer portals, internal tools, databases, scheduled jobs and the integrations that tie them to everything else. We respond the same business day; our hours are on the contact page. We work in a fixed order, because in custom software the wrong order is how a two-hour outage becomes a week of data repair.

If what you have is a public website rather than an application, our website emergency offer is the better starting point. If the trouble is a phone app, start with emergency app repair.

Triage: what changed, what is failing, and who is affected

Before anyone opens the code, we answer three questions. What changed in the last day: a deploy, a dependency, a provider, a certificate, a password or an API key? What exactly is failing, and what does the error say? And who is affected: every user, one customer, one feature, one region? The answers usually point at a short list of suspects within the first conversation.

The error code itself narrows the search. The definitions below come from the IETF's HTTP standards, RFC 9110 and RFC 6585, read in November 2026. The right-hand column is our practice, not the standard's.

Error families during a software emergency, and where we look first
CodeWhat the standard says it meansOur first suspect
401The request lacks valid authentication credentialsA rotated or revoked key, an expired token, a changed password on a service account
403The server understood the request and refuses itA permission change at the provider, an IP allow list, a plan downgrade
429Too many requests in a given amount of timeA loop or retry storm in your own code, or a new rate limit at the provider
500An unexpected condition stopped the serverThe last deploy, a missing environment variable, a full disk
502 or 504A gateway got a bad answer, or no timely answer, from the server behind itThe application process crashed or hangs; the database is slow or unreachable
503The server cannot handle the request right now, from overload or maintenanceA traffic spike, a stuck queue, or a host performing maintenance

When the failure is a third-party connection, the API integration broken page goes deeper. When it is the whole application, the web app down page has the first steps you can take before we arrive.

Containment comes before the cure

Some failures are merely closed doors. Others are actively doing harm: charging cards twice, sending the wrong invoice to every customer, writing corrupt rows into the database every minute. Those get stopped first, even before we know why they started. Containment is not a fix. It is the tourniquet that keeps the fix small.

Bright blue duotone photo of an open metal toolbox, framed by solid geometric circles, bars and triangles in deeper blue
The first tool out of the box in a software emergency is the off switch, not the wrench.
  • Pause the job that is doing damage. Scheduled tasks, queues and webhooks that write bad data are stopped, with the queue kept so nothing is lost.
  • Put the feature, not the whole app, into maintenance. If only invoicing is broken, the rest of the system can stay up.
  • Revoke what may have leaked. Google Cloud's API key guidance, read in November 2026, warns that keys hardcoded in source code or stored in a repository are "open to interception or theft by bad actors." A suspect key is rotated and the old one deleted.
  • Roll back the code when the last deploy is the cause. Version control makes this a recorded step: git's own documentation describes revert as recording new commits that reverse earlier ones.
  • Copy the evidence. Logs, the database and the broken deploy are kept before anything is restored.

If a deploy or update is the obvious culprit, our failed software update page covers the rollback decision in detail. If there is any sign someone got in who should not have, stop here and read the first hours of a data breach: that is a job for your attorney and insurer before it is a job for us.

Restoring from what exists, not what you hoped existed

Every restore is limited by decisions somebody made months ago. Amazon's RDS backup retention documentation, read in November 2026, says retention can be set between 0 and 35 days, that 0 turns automated backups off, and that the default is one day when an instance is created through the API or command line and seven days in the console. Neon's history window page, read the same month, lists 6 hours on its Free plan and a default of 1 day on its Launch and Scale plans, as of November 2026.

Our reasoning from those numbers: a restore can only reach back as far as the window that was set before the problem started. If bad data began three days ago and the window is one day, the clean copy no longer exists in that system. That is why we ask about every other copy too: manual snapshots, exports, a reporting replica, even the spreadsheet accounting keeps. Amazon's automated backups page also notes that automated backups are deleted with the instance unless you choose to retain them, so nobody deletes anything during an incident.

We restore to a new instance or branch and compare it with the damaged one, rather than writing over the original. It takes a little longer. It means a bad guess costs an hour, not the evidence. The database emergency page covers point-in-time recovery for owners who want the detail.

A root cause you can forward to your board

Once the system is working, we write down what happened, in language a non-developer can read and a future developer can act on. You own the document. It is useful whether or not you hire us again.

  1. 1

    The timeline

    When it started, when it was noticed, what was done and when, with the evidence each step relied on.

  2. 2

    The cause

    The specific change, failure or missing safeguard behind the outage, separated from the symptoms people saw.

  3. 3

    The repair

    What we changed to restore service, and whether it is permanent or a temporary measure that needs a follow-up.

  4. 4

    The prevention

    The few changes that would have stopped it or caught it sooner, such as a longer backup window, an alert, a test or a second person with access.

  5. 5

    The estimate

    If the permanent fix is larger than the emergency work, a written estimate for it, which you can take to us or anyone else.

Priced from $575, and quoted before we touch production

Software emergencies start from $575, depending on complexity. You get a written estimate before work starts, and you approve it before we begin. What moves the number is mostly what we find in triage: how many systems are involved, whether access and backups exist, whether data needs repair as well as code, and whether the code has tests or documentation. The emergency pricing guide breaks those factors down.

Broken connections to APIs and third-party services are restored the same day once access is in hand, whether the failing piece is a CRM sync, a shipping rate lookup or a payment processor. The slow part of an integration repair is almost always getting the credentials, not the work.

Access we ask for, and access we turn down

We ask for our own accounts wherever a system allows them, so you can see what we did and remove us afterward. We ask for less than you might expect, and refuse a few things people offer in a hurry.

What we ask for

  • The code repository and the deploy history
  • A named user in the cloud console with the permissions the repair needs
  • Read access to logs and any error tracking tool
  • Database access through a user created for us
  • Dashboards for the third-party services involved

What we decline

  • Your cloud provider's root account password
  • Access keys for a root user
  • A personal password shared across several systems
  • Production changes before you approve the estimate

The root user line has a source. AWS's root user best practices, read in November 2026, says: "We strongly recommend that you do not create access keys for your root user". We would rather wait twenty minutes for a proper user than work with a key that can delete the entire account.

When the outage belongs to your cloud provider or to time

Sometimes the honest answer is that there is nothing to repair on your side. AWS describes a shared responsibility model, read in November 2026, in which it protects the infrastructure that runs its services and the customer's share depends on the services chosen. If the provider's own health dashboard shows an incident, the work is to tell customers clearly and wait. The cloud server outage and outage communication pages cover that.

And sometimes the emergency is the symptom of an older problem: software nobody has maintained for years, on a framework nobody hires for. We will stabilize it first, then tell you plainly whether to repair, rewrite or replace it. Our legacy software rescue page explains that decision, our custom software development page covers rebuilds, and a fractional CTO gives the system an owner. Every other emergency we cover is on the emergency repair hub.

Questions owners ask in an emergency

That is most of what an emergency is. We read the code, the deploy history and the logs before changing anything, and we tell you what we can and cannot be confident about. Undocumented code takes longer to triage, which the written estimate reflects. We do not need the original developer, though their notes help.

We respond the same business day during our posted hours. We do not promise a fix time, because it depends on what triage finds. We do promise the order of work: contain the damage first, restore from what exists, then find and write up the cause.

A restart can bring a hung process back, and sometimes that is the right call to stop the bleeding. Before you do it, copy the logs and note the time, because a restart can clear the evidence of why it hung. If data might be corrupted or someone might have broken in, do not restart; call first.

From $575, depending on complexity, with a written estimate before work starts. The number depends on how many systems are involved, whether backups and access exist, and whether data needs repair along with the code. Our emergency pricing guide explains each factor.

Once we have access to both sides, broken API connections are restored the same day. The usual causes are a rotated key, a changed permission or a new rate limit, and the error code tells us which. Getting the credentials is normally the longest part.

No, and we would rather not have it. We ask for a named user with the permissions the repair needs, which you can remove afterward. AWS itself recommends against creating access keys for a root user.

Keep reading

Free · No obligation

Tell us what stopped and what changed

Describe the failure, the error and the last deploy or change you know of. We respond the same business day with a first read on the cause and what we need to start.

Which of these sounds like you? (pick any)

Rather talk now? Call (210) 346-0848.

CallGet a quote