Typing a stage pays the typist nothing, so it gets dropped first. Stop storing the stage, store the facts, and derive it each time the pipeline is read.
Stages set by mood, in a hurry, on a Friday. Qualified means somebody felt good about it, and In conversation means a reply once arrived, months ago. At a thirty-person software company the pipeline is tidied the night before the meeting about it.
By hand today
“most of the time colleagues don’t bother to update records properly”
flavaz, on Hacker News, read 8 September 2026
Everybody knows this and nobody solves it, because the usual solution is asking people to be more disciplined. Discipline is not the problem. Typing a stage is a piece of admin that pays the typist nothing, so it is the first thing dropped in a busy week and the last thing done before a review.
The tools have tried to catch it from the outside. Pipedrive marks a deal red when it has not been “updated for longer than the defined rotting period”, and the same help page lists “Adding notes and files to a deal” as one of the things that resets the timer. So a dead deal with a note added to it looks alive again. The staleness check can be passed by typing.
The result is a pipeline nobody trusts and everybody quotes. It is read out on Monday, argued with for twenty minutes, and quietly discounted by whoever has to plan around it.
What it costs you each week
A hundred and thirty-five thousand dollars, the bottom of the salary band in one advert for the hire whose named duty is pipeline hygiene. Below that price, hygiene is a founder tidying rows on a Sunday evening.
Source: a job advert for a revenue operations role, OpenSesame, on Greenhouse, read 8 September 2026
That advert names the work plainly: stage criteria, ageing thresholds, next-step expectations and close-date discipline. Somebody is paid a salary of that size to make a database say true things, mostly by chasing colleagues.
Under fifty people, nobody is paid to do it, so measure it instead. Take twenty leads at random and ask, for each one, which stored fact puts it in the stage it is in. Not a memory, not a note. A field, a score, a booking, a reply. In most pipelines fewer than half the rows can answer.
The build
For Claude Code a stage is a fact rather than a feeling. Every stage is derived on read from something stored: a field, a score, a person, a reply. Nothing is archived without a reason, and no lead sits with no next step.
The move that fixes this is smaller than it sounds. Stop storing the stage. Store the facts, and work out the stage each time you read the pipeline.
Work in the folder that holds your CRM. Type:
Do not store a stage on any company. Derive it on read: fields filled
means enriched, a score above the floor means qualified, a named person
means people found, a lead on a list means in the pipeline, a human
reply means in conversation. Keep companies and people apart: companies
move through the funnel, people through the pipeline. Move one step at
a time and never backwards. Refuse to archive anything without a written
reason. List every lead with no next step drafted, and every company
whose derived stage disagrees with what somebody typed last month.
What comes back on the first run is uncomfortable and useful. A list of companies whose real stage is behind where the CRM said they were, and a list of leads that have no next message waiting. That second list is the honest measure of the pipeline.
Two structural rules keep it from drifting back. Companies and people are counted separately, because a company in the funnel and a person in the pipeline are different things that get confused into one number. And nurturing is a list, never a stage, so that the thing everyone parks deals in cannot inflate the pipeline.
The pattern is ArkOne’s own. docs/sales-pipeline.md lines 10 to 11 hold the split of companies in the account funnel from people in the lead pipeline, line 23 the one stage at a time rule, and line 137 the ruling that nurturing is a list and never a stage. docs/crm-module.md line 370 is where the stations are derived on read rather than stored, where an archive needs a reason, and where a human saying no stands until a person overturns it. Copy that shape and build your own. There is no demo to download.
Run it every Monday before anything else, and read the two lists first.
What it feeds
A forecast with reasons. A pipeline whose stages came from facts can be counted without anybody defending it, and the next chapter writes a number from it. That number is worth nothing if the stages behind it were typed, so this chapter is the one that has to hold.
Three numbers prove it. Leads with no next step drafted, which should reach zero and stay there. Companies archived with a written reason, against companies that simply disappeared. And stages that changed because a fact changed, against stages changed by somebody clicking.
What stays with a person
The no, and the close date. Only the person in the conversation knows they said no, and roughly when this might land.
A machine can see that a reply arrived. It cannot see that the reply meant no. When somebody tells you they are not buying, that is a fact only you can record, and it should outrank every hopeful signal on the record afterwards.
The close date is the same kind of fact. Nobody can derive it from a stage. It comes from a person saying they want this before the end of the quarter, or that budget opens in March. Written down as their words with a date, it is evidence. Guessed, it is the thing that makes forecasts fiction.
The tools you already pay for
Nothing new to buy.
CRM holds the funnel, the pipeline and the reasons.
The stop log from the sequences chapter moves the stations.
Your calendar carries a booking, which is a fact a stage can read.
A text file, hygiene.md, records what failed this week.
No hygiene product and no operations hire.
Stages that came from facts. Then Monday needs no defending.
Ask: If nobody touched your CRM this week, which of your stages would still be true?
This is chapter 9 of twelve, after booking the meeting; chapter 10 writes the number. The book, Building a real GTM engine with Claude Code and Codex, end to end, runs the whole engine in order, one chapter per stage. A free starter edition for firms under fifty people is coming through the 100k programme.