If you re-engineer mush, all you have is re-engineered mush.
Mush is work nobody can explain well enough for someone else to follow. It still gets done, because someone knows the exception, the workaround and the person to call. Change it without that knowledge and you rebuild the same mush in new clothes, at a higher price. This page is the guidance. The stories behind it are in the blogs listed at the end.
Piss poor planning provides piss poor performance
Philip Andreae recalls that, at Management Centre Europe between 1993 and 1995, a sergeant major opened with one line to a room of IT and marketing professionals: “You all are simply re-engineering mush.” Coming from a man of that rank, it was a wake-up call for Philip Andreae. The seven Ps, crude on purpose, take him back to his earlier years as crisis manager at NDC in Atlanta, and to why the crisis typically comes: nobody planned. This is what earns the charge:
- Changing what you haven’t documented.
- Removing the person who knows the workaround.
- Committing before requirements and dependencies have been reconciled.
- Adding people or money to make up time.
- Testing only the ordinary day.
- Starting without a way back.
Keep it simple
Prior planning provides progress. Four words, four Ps: the same reminder turned the right way round. Planning is not paperwork for its own sake. It is the moment you find out what the work really depends on, while finding out is still cheap. Keep it simple:
- Know how the work is done.
- Write it down.
- Ask the people who do it.
- Test it under strain.
- Keep a way back.
- Then change it.
Scale the record and the testing to the consequences of failure. A system that is becoming unsustainable cannot wait forever for perfect documentation, but it can wait long enough to learn what people depend on.
Know how the work gets done
Document the work as it actually operates, in enough detail for a successor to inspect it and for a team to trace the consequences of a change.
- Applications and interfaces. What each application does, who owns it, which versions are in use, and what systems, integrations or scheduled jobs depend on it.
- Data and data structures. Where the data comes from, what the fields and identifiers mean, how records relate, which rules validate or transform them, and where the data goes next.
- Work processes and procedures. The sequence, roles, handoffs, decision rules, exceptions and workarounds. Follow an ordinary case and one that went wrong. Find out who noticed, who could act, and who had to wait.
- Equipment, utilities and capacity. What the operation depends on, including load under normal and peak conditions. Note available capacity and what happens if a limit is reached or a supply fails.
- Human checkpoints. Who inspects, approves or stops the work, when they intervene, what evidence they use, and who takes over if the usual person is unavailable.
- Quality standards and adherence. The applicable standards, acceptance criteria and tolerances; who checks them, how and when, where results are kept, and how deviations are corrected and checked again.
Keep the record current. Ask someone who has never done the job to follow it through an actual case. Where they get stuck is where the record needs work.
Show me the transition
Know who you are, what you do, how you do it, and why it matters. Then describe the future you want—and plan how services and quality will continue while you build it.
Before approving a change to working mush, get answers to five questions:
- What must remain available, and who depends on it?
- What have the people doing the work shown that the documentation missed?
- What evidence shows the replacement can handle ordinary work, peak demand and failure?
- Who can stop the change, and what tested fallback will keep the service running?
- Who will own the operation and its record after the project team leaves?
Set an availability target that matches the service’s needs. Agree the measurement period, what counts as unavailable, and how interruptions during the transition will count. Five nines (99.999 percent), for example, permits about 5.26 minutes of unavailability per year when measured by time over continuous annual operation. Hold the change to the agreed target, including on the day of the switch. Google SRE availability table
Adapt, or replace
A market shifts. Volume grows. Costs must come down. Use the record to trace the repercussions first. Who gets more work? What might break? Which rule, team or customer would be affected? Who will address each consequence?
If the existing system can meet the new need with a limited adjustment, start there. When it can no longer meet the need, replacement may be the better choice. The same record tells you what the new system must carry forward, what must keep working during the transition, and what evidence will show the replacement is ready.
Practices that help
- Process mapping. Trace the work as people do it, including decisions, exceptions and handoffs, and walk the map with them. ASQ’s flowchart guide
- Value-stream mapping. Draw the current flow of work and information across teams before proposing a future state. Lean Enterprise Institute
- Formal change control. Record the baseline and assess a proposed change before approving it. NASA Systems Engineering Handbook
Leave a record someone can use
When you hand over, the next person should be able to read the record, check that the work still happens that way, and know who can help when it doesn’t. Leave them what you changed, why, and how you kept the service working in the meantime.
Stories
A documented incident: The FAA Ground Stop: Start With the Router Manuals. A possibility examined: Suppose: A Lot of Mush. Caused by What?
Then look ahead: Design the present as the past of the future.
Philip Andreae, in dialogue with ChatGPT and Claude. Philip originated the principle and supplied the 7 Ps, the 4 Ps and the Sergeant Major’s charge from his own recollection. ChatGPT drafted the earlier /mush/ text and the transition questions; Claude restructured the page as guidance on November 18, ’26. ChatGPT checked the cited guidance and links on November 18, ’26, and clarified the future objective, transition measurements and recollection attribution on November 19, ’26. Sources were gathered in the Swirl.
