Topic 04 · The Haystack Syndrome
More numbers. No answer.
Velocity, coverage, utilisation, deploy frequency, cloud spend. The wall is full. The questions that matter — is Throughput rising, where is the constraint, should we start this — still go unanswered. Goldratt called that the haystack: buried in data, starved of information.
After Critical Chain. Six sections. The three measures from Flow come back as a filter.
01
Data is not information
Data is the raw count. Information is the answer to a question that helps you raise Throughput, cut Inventory, or protect the constraint.
- Velocity
- Deploy count
- Coverage
- Utilisation
- PR size
Information
Does this change Throughput at the constraint?
- Sprint velocity
- Story points
- Coverage %
- Utilisation
- Commits
- Bug count
- Queue at review
- Outcomes used
- Lead time
- WIP on the board
- Cloud spend
- Deploy frequency
Outcomes used answers the question. The dim tiles can be measured. They are not information.
02
Three measures, as a filter
Prefer what increases Throughput. Cutting Inventory or Operating Expense matters when it protects that, not instead of it.
Throughput
Outcomes the system delivers through software that is in production and used.
Inventory
Time and money stuck in work not yet delivering: open branches, half-done features, knowledge in one head.
Operating Expense
What it costs to turn ideas into Throughput. Salaries, cloud, tools, meetings, switching.
Traditional view
More capacity.
03
What the system has to answer
A useful board answers five questions. Everything else is optional, and most of it trains the wrong behaviour.
- Where is the constraint? Where does work wait?
- Is Throughput rising, flat, or falling?
- Is Inventory — unfinished work, lead time — under control?
- Are we subordinating, or maximising local busyness?
- This tool, hire, or process: what does it do to T, I and OE?
- Velocity
- Points
- Coverage
- Utilisation
- Commits
- PR count
- Story throughput
- Meeting hours
Eight numbers, none of which say whether the system got better. Adding a ninth will not fix that.
04
Sift
For any number on the wall: does it say something about the constraint? If we improve it, does Throughput rise, or do we just make more Inventory?
Team velocity
Queue time at review
Utilisation heatmap
Outcomes used per week
Coverage percentage
Cycle time to production
0/6 match the filter
Cycle time to production, queue time at the constraint, and the rate of adopted outcomes usually survive. Coverage targets, velocity charts and utilisation heatmaps usually do not.
05
When the metric is the constraint
A lot of software constraints are not a server or a specialist. They are a rule about a number.
“80% coverage before merge.”
A gate on every change, whether or not defects are the constraint.
“Every engineer fully utilised.”
Non-constraints stay busy and start more work. Inventory grows.
“Velocity must not drop.”
The number is defended. Starting work becomes safer than finishing it.
“Story points go to the board.”
The report becomes the work. Local points outrank shipped value.
06
One metric you will stop defending
List what leadership watches. Ask, of each, whether improving it ships more valuable software. Then replace one of them.
- The top measures the team or the board actually looks at.
- For each: if this number improves, does Throughput rise?
- The current constraint, and one measure of it. Queue, wait, or buffer.
- A Throughput proxy: outcomes that reached production and were used.
- Stop emphasising the numbers that reward local optima.
- The next tool, hire or process: a short T / I / OE line before it is approved.
This week
0/4
Cheat sheet
- Haystack
- Plenty of data. Almost no answer to the question that would change a decision.
- Information
- The answer to a specific question about Throughput or the constraint.
- The quote
- Tell me how you measure me and I will tell you how I will behave.
- Order
- Increase Throughput first. Cutting Inventory or Operating Expense is secondary if it does not protect T.
- Local optima
- Velocity, utilisation and coverage targets usually create Inventory.
- The system
- Where is the constraint? Is Throughput rising? Is Inventory under control? Is this change going to move T?
Check yourself
5 questions. No timer.