The instruction issued "to be agreed"
An instruction issued on a "costs to be agreed" basis defers two things: the price and the time. Everybody chases the price. Almost nobody separately notifies the time, so the prolongation rides along inside the cost conversation with no notice, no dated start, and nothing tracking it as a delay event while it is happening.
That is the whole failure mode. It is quiet, it is common, and it feels safe at the time because the instruction is in writing.
Why does "to be agreed" feel safe when it is not?
Because it satisfies the instinct that gets drilled into everyone: get it in writing.
And the instruction is in writing. There is a document, it is dated, it is unambiguous that the work was instructed. On the entitlement question you are in good shape. That is exactly what makes the next part so easy to miss.
What "to be agreed" actually means is that two separate commercial questions have been deferred to a later date, and only one of them has an obvious owner.
| Deferred | Who chases it | When it gets resolved | What happens if nobody does |
|---|---|---|---|
| Cost | The QS, immediately and repeatedly | Valuation cycle, or at final account | It is visible in every application. It gets argued |
| Time | Nobody, by default | Usually never, until it becomes a claim | The programme impact has to be reconstructed from memory |
Read the bottom row. The time question has no natural owner, no recurring meeting where it surfaces, and no line on the valuation that goes red when it is missing. It is deferred by the same words and then simply forgotten.
What does this look like on a real package?
Here is a real one, sanitised.
A fit-out package, around £1.5m, two large amenity floors of roughly 3,200 m² inside a 31-storey building. JCT Design and Build, with a schedule of amendments attached, as almost all of them have.
A specification increase was instructed in writing, on a costs to be agreed basis. The work went ahead. The higher specification took longer and needed more labour, and the subcontractor claimed both the additional labour and the prolongation. The main contractor was then in the middle: holding a claim from below, needing to pass it up, against a client disputing it.
Notice what is not in that account. Nobody recorded the day the additional labour started. Nobody logged the programme impact as it happened. There was no separate notice on the time, because the instruction had already been issued and it felt like the paperwork was done.
How did that one end?
It resolved. The commercial relationship was good and the parties worked it out.
That is the honest answer and it is the most useful part of the story, because the reason it resolved was the relationship, not the record. Had the records been thinner, or the counterparty less reasonable, the same file would have had to stand on its own.
Goodwill is not a control. It works right up until the other side changes commercial director, gets bought, or has a bad year, and then a set of records created months earlier by people who did not know they were creating evidence has to carry the argument on its own.
Near misses are more useful than disasters, because nobody has to be blamed for the lesson to land.
Does the contract form change the risk?
Yes, and the difference is larger than most people assume, which is why "we're on JCT so we're fine" is an unsafe sentence.
Under standard JCT Design and Build, the delay notice is a notification duty rather than a condition precedent. Serving late is untidy but rarely fatal to the entitlement. Under NEC, the eight-week rule at clause 61.3 genuinely does bar a claim. And amended JCT frequently imports a condition precedent that the standard form does not contain.
On the package above, the contract carried a schedule of amendments and it is not known whether those amendments touched the notice provisions. That uncertainty is itself the point: a schedule of amendments is normal, nobody reads all of it under time pressure, and the question of whether your notice is a duty or a bar is answerable only by reading the actual document. The difference between the two forms is worth knowing before you need it.
What would have caught it?
Anything that turned "raised in a meeting" into "notice served on a date".
That is the entire remedy and it is unglamorous. Three specific things:
- Separate the two deferrals at the moment of instruction. When an instruction says costs to be agreed, that is the trigger to ask what the time impact is and to notify it separately. Not later. Then.
- Date the start of the impact. Not the date you noticed it, the date it began. That is the number that gets argued about eighteen months later and it is almost impossible to reconstruct from memory.
- Capture the labour as it happens. The evidence of additional resource is being generated daily on site, in messages and photographs, and it is usually already there in a form nobody can retrieve.
Why is this so hard to fix with discipline alone?
Because the person best placed to record the impact is on site, at the end of a long day, and the impact does not feel like an event when it happens. It feels like a slightly heavier week.
That is the honest reason procedure notes have never fixed it. You are asking someone to recognise, in the moment, that an ordinary difficult week is going to matter commercially in two years. Nobody does that reliably, and the ones who do are the ones who have already been burned.
Which is why the answer is capture rather than diligence. If the record assembles itself from what the team already produces, the day the labour increased is in the record whether or not anyone realised it mattered. The small omissions are what compound into disputes, and they compound quietly.
An instruction issued "to be agreed" is not a problem. It is a completely normal commercial mechanism. The problem is that it hands you two open questions and only one of them has anybody's name against it.