Haute Lumière
Commerce · I.08 · MMXXVI · daylight
TEN CONCEPT BRIEFS · Chapter I.08 — When It Stalls
One page each. A reader who reads only these ten pages has the chapter.
BRIEF 1 — The Stall as an Ablation Study
The idea. While a programme is being pushed, you cannot tell what is being carried from what is being pushed. Your attention underwrites all of it at once, so all of it looks alive. Stop pushing and the question answers itself.
An ablation study is the experimental technique of removing one component to see what the system does without it. A stall removes the component that is hardest to remove deliberately — you — and it does it for free.
What it measures. Institutionalisation. Not how committed people say they are, not attendance at the steering group: which activities keep happening when nobody is making them happen. There is no other clean measurement of this, and a moving programme is structurally incapable of producing one.
Worked example. A pilot ran eleven activities: a monthly yield report, a supplier scorecard, a weekly stand-up, a shared dashboard, a training module, and six operational changes on the line. Four months into a stall, the yield report is still being produced automatically, three of the six line changes are still standard practice because the operators prefer them, and everything else has stopped. Seven of eleven gone; four of eleven alive. The stall has just told you which four things you actually built.
Why it matters. Because it converts the worst moment of a transition into the only reading of its kind — and because, as Brief 5 shows, that reading is worth a great deal of money when you come to restart.
You already know this because you have been away from a team for three weeks and come back to find out exactly which of your routines were theirs and which were only ever yours.
BRIEF 2 — The Survivor Ratio (σ)
The idea. One number for what the stall measured.
σ = activity still running with nobody pushing
─────────────────────────────────────────
activity running at the pauseHow to take it. From the activity list, not the calendar — a meeting that still happens is not an activity that still runs. Walk the list of everything that was running at the pause, and mark each one running or stopped. The denominator is stated, which means the reading is honest: σ is a ratio of countable things, not an impression.
When to take it. At month three. Not month one, because the first month is still coasting on the last push. Not month twelve, because by then the record has been tidied and people remember what they wish had survived.
How often. Once. A survivor ratio taken repeatedly becomes a target, and a target of this shape is very easy to perform for — keep a defunct meeting in the diary and σ rises without anything being true.
Worked example. Eleven activities, four still running, σ = 0.36. That single figure moves the restart window in a hostile room from under three years to well past seven. It took ninety minutes to measure.
Why it matters. It is the chapter's whole thesis in one number: the thing the stall gave you is the thing that buys the restart back.
BRIEF 3 — The Warm-Start Fraction W(t)
The idea. A paused programme is not one asset going stale at one rate. It is five assets decaying at five different rates.
| Component | Share of build cost | Still usable after a year |
|---|---|---|
| Discovery findings | 25% | 90% |
| Baseline comparability | 25% | 72% |
| Instrument documentation | 20% | 85% |
| Counterparty relationship | 15% | 80% |
| Team-specific know-how | 15% | 85% |
W(t) = Σ wᵢ · kᵢ^tOnly the last rate is measured rather than assumed: team know-how decays at your own voluntary turnover. The other four are assumptions, stated as assumptions in lib/verify/I_08.py, and every figure in the chapter moves when you replace them with your own.
The shape of it.
t (yr) W(t) lost
1 0.823 17.8%
2 0.681 31.9%
3 0.567 43.3%
5 0.400 60.0%The one figure to carry. Roughly a fifth of what a programme cost to build stops being an asset in the first twelve months it is not being pushed. The instantaneous rate at the moment of the pause is 19.9 percent a year.
Why it matters. Let us park it for now is a spending decision. It has never been quoted as one, and quoting it is a thirty-second intervention with a five-figure effect.
BRIEF 4 — The Carrying Cost of a Pause
The idea. A stalled programme costs money every month in two ways at once: the decay of what you built, and the return you are not collecting.
Worked example — Chapter I.01's pilot: £180,000 to build, £62,000 a year of verified saving.
| Year | Decay | Foregone | Year total | Cumulative |
|---|---|---|---|---|
| 1 | £31,950 | £62,000 | £93,950 | £93,950 |
| 2 | £25,475 | £62,000 | £87,474 | £181,424 |
| 3 | £20,460 | £62,000 | £82,460 | £263,885 |
The headline. The cumulative carrying cost crosses the build cost at month twenty-four. A programme held still for two years has cost about exactly what it cost to build. On this pilot the run rate is roughly £8,000 a month.
What this is not. It is not an argument that stalling is always wrong. Sometimes the capital genuinely is not there and a freeze is the right call.
What it is. An argument that a stall should be decided rather than defaulted into, by somebody who has seen the monthly figure. Most stalls are not decisions at all; they are the absence of a renewal that nobody had to defend.
Why it matters. It is the only line in this chapter that reliably changes a conversation in the room where the stall is happening, because it converts silence into a number with a currency symbol in front of it.
BRIEF 5 — The Friction Premium (f) and the Restart Inequality
The idea. Restarting is not free even where W is high, because the second ask is priced differently from the first.
What f is. The share of a cold start you pay again purely because the programme is known to have stopped: re-litigated objections, a spent option, and a documented precedent that we tried this.
How to estimate it. Count the signatures on the original approval path. Mark which of those people are new. Mark which have the stall in their file. It is a judgement and it should be written as one — but it is a judgement about a room you can actually name.
The inequality. Restart beats beginning again clean while
W(t) > 1 + f − 1/(1 − σ)With no survivors, σ = 0, that reduces to the memorable form: W(t) > f.
The windows.
| f | The room | Window |
|---|---|---|
| 0.15 | Same sponsor, nothing on the record | 11.3 yr |
| 0.35 | New sponsor, stall on the file | 5.8 yr |
| 0.60 | Stall cited as evidence against | 2.7 yr |
Why it matters. The clock is not set by the programme. It is set by who will be in the room when you come back — which is why the second owner from Chapter I.01 was never a courtesy.
BRIEF 6 — Restarting at σ
The idea. The commonest way a good restart fails is that it asks for the original scope. Do not. Restart at the survivor set plus the smallest rebuild that makes it whole.
Why the arithmetic favours it. Restarting at σ rescopes both the rebuild and the ask by (1 − σ), which relaxes the threshold:
| σ | Window at f = 0.35 | Window at f = 0.60 |
|---|---|---|
| 0% | year 5.8 | year 2.7 |
| 10% | year 8.2 | year 3.8 |
| 25% | effectively no clock | year 7.5 |
| 40% | restart always wins | restart always wins |
A survivor ratio of a quarter takes the hostile-room window from under three years to over seven.
Why it also works politically. You are not asking a room that watched a facility stop to approve that facility again. You are asking it to formalise four things that are demonstrably still running and fund the small remainder. Those are different meetings and they have different outcomes.
The one-page restart proposal. Four figures, in this order: σ, current W, estimated f with the room named, and the crossing year. Then the scope. Then the pre-agreed terms, if you had the sense to write them into the facility.
Why it matters. It is the move that turns the stall's free measurement into the restart's actual mechanism, rather than leaving it as an interesting fact.
BRIEF 7 — The Dormancy Covenant
The idea. A clause in the facility that defines what happens when a programme stops drawing, so that stopping does not mean terminating.
The parts.
| Term | Setting |
|---|---|
| Trigger | Named in advance: sponsor departure, capital freeze, covenant action, σ below a floor |
| Effect | Converts to dormant; baseline stands; repayment pauses, never accelerates |
| Surviving obligations | Series runs · verifier retained · docs reviewed annually · second owner keeps named time |
| Reserve | Sized to fund those four — order of £7,500/yr on a £180,000 build |
| Sunset | Hard date, three years default; converts to ended unless extended by signature |
| Restart | Pre-agreed at the survivor set plus minimum rebuild, original economics, no re-approval |
The two clauses that carry the weight. Repayment pauses rather than accelerating — a facility that accelerates on dormancy will never be allowed to go dormant, so the programme is killed instead. And pre-agreed restart terms, which are worth more than the reserve, because they remove the signatures that generate f in the first place.
When to write it. Now, into the template, for every facility, whether or not you expect to use it. It costs nothing to include and it is unavailable to include later — a dormancy clause negotiated during a stall is negotiated by people who are already arguing.
Why it matters. It is the cheapest clause in the instrument catalogue and the only one that pays out precisely when nobody is paying attention.
BRIEF 8 — The Unbroken Series as Audit Evidence
The idea. Keeping the measurement running through a pause is not sentimentality. It is the evidence that a capitalised asset is not impaired.
The mechanism. Where the programme capitalised an asset, dormancy raises an impairment question at the next reporting date, and impairment is a real charge against a real number. An unbroken series lets you show the unit's performance through the dormancy instead of asserting it.
The arithmetic. Keeping the series running and the verifier retained changes baseline comparability from 72 percent a year to about 97, and the counterparty relationship from 80 to about 95.
t = 1 yr W plain 0.823 W dormant 0.908 preserved £15,300
t = 3 yr W plain 0.567 W dormant 0.754 preserved £33,600
reserve, three years at £7,500 £22,500
asset preserved £33,600 1.5xAnd the part that is larger than the 1.5×. A reserve of £7,500 a year that protects a capitalised balance against a write-down is not an insurance product. It is the cheapest audit evidence available, and that is the conversation to have with your auditors — service potential and useful economic life, which they discuss every year anyway.
Why it matters. Because a report that has been automated is cheaper to leave running than to switch off and rebuild, which makes this the single highest-return intervention in the whole chapter: do not switch off the report.
BRIEF 9 — The Baseline Cliff
The idea. W assumes the baseline decays. Sometimes it does not decay at all. It goes to zero in one step.
When. Whenever the underlying process is re-engineered during the pause — a new ERP, a line rebuild, a reorganisation that redraws the cost centres, a change in how the metric is captured. The old baseline is then not a worse comparison. It is not a comparison, because the two sides measure different things.
What it does to the arithmetic. Baseline comparability is a quarter of the warm-start fraction, so the cliff removes up to 0.25 at once.
t = 2 yr W 0.681 -> W after the cliff 0.551At year two after a cliff, W sits barely above a new sponsor's friction premium and well below a hostile one. With no survivors, the inequality fails.
What to do. Declare the programme ended, in writing, and begin the successor clean. This is the condition under which this method loses, and it is stated here rather than discovered later, in the tradition of Ostrom's design principles — the value of a rule is in specifying where it stops holding.
The consolation, which is real. The discovery findings and the instrument documentation survive a cliff untouched. A clean successor inherits both, so it starts well ahead of where the first one started.
Why it matters. Because without this named threshold, the method is a machine for restarting things forever.
BRIEF 10 — The Appreciative Bias Against Ending
The idea. Discovery orients you to what is working. A stalled programme will therefore always present a survivor set, and a survivor set will always look like a reason to restart. That is the method's own blind spot and it is worth naming plainly.
The literature. Barry Staw's 1976 study of escalating commitment and Arkes and Blumer's 1985 work on sunk cost describe people who kept going because they had already gone. Appreciative Inquiry applied to a stall, without an inequality, is an unusually efficient version of exactly that: it supplies evidence of life on demand.
The guardrail. The restart inequality is not decoration. It is what makes the appreciation safe, because it can return no — and a method that cannot return no is not measuring anything.
The structural fix. Make ending a real status with a real ritual. Where nothing can be ended, everything is dormant, the register fills with fiction, and within two years nobody reads it. Dormancy is only a real status if ending is also a real status.
What a good close contains. What it cost. What it returned while it ran. Its final survivor ratio. One paragraph on which layer proved load-bearing. These become the most-cited internal documents in an organisation, because they are the only place anybody wrote down what actually holds.
Why it matters. A person who has run something that stalled should be the person given the next programme — because of the reading they now have. That is only possible where stopping is speakable, and stopping is only speakable where ending is honourable.
All figures in these briefs are computed in lib/verify/I_08.py and sourced in the chapter's Works Cited. Four of the five decay rates are declared there as assumptions to be replaced with the reader's own measurements.