Why does everyone report a project "green" until it's too late?

The step is a cliff, not a slope

Most organisations report project health as red, amber or green. Three states, easy to scan, and everyone knows what they mean.

The trouble is that project confidence isn't three-valued. It's continuous, and it moves in small increments — a fortnight where progress was slower than hoped, a dependency that keeps almost arriving, an estimate that's looking thin. None of those is a state change. All of them move the number.

A project manager in our research described what happens when you try to express that:

"Switching a ticket from green to yellow is such a big step change that it sets off all the alerts. There is a cost to calling out problems."

That's the mechanism in one sentence. Amber isn't a slightly more cautious green — it's a different regime. Amber means a recovery plan, a review, questions from people who don't normally ask questions, and possibly a mention in someone's report to the board. The gap between "I'm a bit less sure than I was" and "I would like to trigger all of that" is enormous.

So the team lead has three options. Report green, which is now slightly false. Report amber, which is disproportionate and expensive. Or wait a week.

Waiting is the rational choice, and here's the part that makes it self-reinforcing: it's usually right. Most weeks, the thing does resolve. The team lead who waited looks calm and competent; the one who escalated at the first wobble looks jumpy. The organisation is actively teaching everyone to wait.

Which works until the week it doesn't — and by then the same logic has been applied five weeks running, the project is genuinely in trouble, and moving to amber now means explaining why it was green last week.

The asymmetry underneath

Strip out the mechanics and there's a straightforward incentive problem.

The cost of amber is immediate, personal, and certain. You get the scrutiny. You have the awkward conversation. If the project recovers, you were the person who panicked.

The cost of green is deferred, shared, and uncertain. It lands weeks later, on the whole team and the organisation, and by then it's attributable to the situation rather than to your reporting.

Nobody is behaving badly. Faced with a certain, personal, immediate cost against a diffuse, deferred, uncertain one, people choose predictably.

The result is a system where not knowing is preferable to knowing an inconvenient truth. It's the oil light flashing on the dashboard of your car, and the response being to ignore it — or to remove the bulb. Which is a good deal less productive than acknowledging the warning and topping up the oil.

A developer summarised where that lands:

"Everyone wants to be green. Executive sponsor wants to look good."

Culture sets the size of the asymmetry

Every organisation has this problem. What varies is how bad it is, and the variable is what happens to the person who reports amber.

An engineering leader described a previous employer at the extreme end:

"A culture where everything was green. You'd get raked over the coals for saying something was red."

In that environment reporting is worthless in a specific way: it's not that the data is noisy, it's that the data is uniformly positive and therefore contains no information at all. The reports still get produced. The deadlines still slip.

Another leader named the diagnostic question, and it's a good one to ask about your own organisation:

"A big factor is the culture of the organisation. How comfortable are they with 'red'?"

The tell isn't whether people say they welcome bad news. It's what visibly happened the last three times someone delivered it.

What changes the arithmetic

Three things, and they work together.

A scale instead of buckets

On a one-to-five scale, moving from 4 to 3 is a small act. It doesn't assert that the project is at risk, doesn't trigger a review, and doesn't require justification. It says the person is somewhat less confident than last week — which is exactly the information you wanted and exactly what RAG cannot represent.

This matters because the useful signal is the slope, not the level. A team drifting 4.4 → 4.1 → 3.8 is telling you something well before any of those readings would have been amber.

Many people instead of one

RAG status is set by one person, who carries all the exposure. When everyone answers, no individual is the one who "called it amber." The reading is a property of the team, and nobody has to be brave.

Anonymity for the individual answer

Even in an aggregate, if people believe their own number is visible to their manager, the same asymmetry reappears at the individual level. Aggregating before anyone sees the data removes it — the argument in full is in why confidence votes should be anonymous.

Projects don't turn red. They fade.

The premise underneath RAG reporting is that a project has a state, and occasionally that state changes. That isn't how it works in practice.

What actually happens is that a team gradually loses confidence as things pile up. A test suite that got slower. An estimate that looked generous in January and doesn't now. A dependency that keeps almost arriving. A person who left. None of those is a state change, and no one of them is worth escalating — but each one moves the team's belief down a fraction, and they accumulate.

By the time the project is genuinely red, it has usually been drifting for two months. The red isn't the event. The red is the moment the drift became impossible to keep calling green.

Which is why the trend matters more than the reading. A team sitting at 3.6 that climbed there from 3.1 is recovering from something and probably needs nothing from you. A team at 3.6 that has come down from 4.5 over five weeks is in a completely different situation, and the two are indistinguishable from any single measurement — or from any traffic light, both of which would show amber.

The direction of travel, week over week, is usually more telling than the score itself.

How Genchi handles it

Genchi asks every team member for a one-to-five confidence rating at a cadence you choose. No single person owns the status. Small movements are expressible and cost nothing to give. Individual responses are aggregated before anyone sees them.

Critically, every check-in is kept, and the score is displayed as a line over time rather than as a current value. That's the part RAG cannot do at all: three consecutive weeks of small declines is a visible, unmistakable shape, where the same three weeks under a traffic light are three consecutive greens followed eventually by a surprise.

The colour-coding people use for scanning a portfolio still exists, because it's genuinely useful for that — but it's derived from a continuous underlying number with a recorded history, so the drift is visible long before the colour changes.