Maintain
Keep the current capability substantially as it is.
Chosen by the householdProvidence Nexus · Observatory study
Two builds, one engine
The same stations, read against two different realities. Evidence, calibration categories and the decision all differ, because what each build meets in reality differs.
Example · Offered to a buyer
Roof survey walkthrough
The Reading
Returned signals
- attention41marketRound 1
41 people saw the offer in two local buying groups. No questions the first week.
- conversation3marketRound 1
Three real conversations. All three asked what happens if the roof is fine — nobody wants to pay for good news.
- refusal1marketRound 2
One refusal with a reason: 'my surveyor already looks at the roof'. Problem clarity, not price.
- payment2marketRound 2
Two strangers paid £180 each after the promise was rewritten to 'how long this roof has left, in writing'.
- actual use2marketRound 3
Both reports were used in the price negotiation. One buyer got £2,400 off and sent the report to their solicitor.
- referral1marketRound 3
One referral to a colleague buying in the same street. Not yet booked.
What the evidence demonstrates
Recorded so far: attention, conversation, refusal, payment, actual use, referral. These are different kinds of evidence, not stronger versions of one another.
- attention — the offer was seen
- conversation — the subject is worth someone's time
- refusal — a specific reason the offer was declined
- payment — a stranger valued it enough to pay
- actual use — the thing delivered was used for its purpose
- referral — someone staked their own name on it
What it does not demonstrate
- attention — does not yet show that anyone wants it, or would pay
- conversation — does not yet show that the promise is clear enough to be bought
- refusal — does not yet show whether the same reason holds for everyone else
- payment — does not yet show that they would pay again, or that delivery holds
- actual use — does not yet show durability — one use is not a pattern
- referral — does not yet show a booking — a referral is not a customer
Attention was never the problem. The offer had to name the thing the buyer was actually afraid of before any conversation turned into payment. Compliments from friends told us nothing; the first refusal told us the most.
Enough evidence to decide
Sufficiency is relative to the decision in front of you. Evidence can be enough to justify another bounded test and still be nowhere near enough to conclude durability or transfer.
Enough evidence to decide
- Maintain — It worked in actual use 3 times (payment, actual use, referral). No unresolved confirmed strain blocks continuing, so keeping it substantially as it is rests on evidence. Repetition across further rounds is not required.
- Improve — It worked in actual use 3 times (payment, actual use, referral). Evidenced friction: refusal. If a change alters the promise, the person it is for, the process, the price or the delivery model, its effect is unknown — that is Run Another Test, not an integrated Improve.
- Run Another Test — Material uncertainty remains: whether the repair to refusal holds. A bounded next test could reduce it.
Not yet enough to conclude
- Partner — It worked in actual use 3 times (payment, actual use, referral). No confirmed evidence yet names a complementary capability, access point, role or control issue that partnership would supply.
- Retire — It worked in actual use 3 times (payment, actual use, referral). The evidence does not point to poor fit, weak value or excessive strain. The household may still judge it not worth continuing.
What would reduce the remaining uncertainty
- Whether the repair to 'refusal' actually worked, observed again.
These rules are Providence rules, applied deterministically. AI may explain them. It may not change them, and there are no confidence scores here.
What can be changed
One evidence-based change at a time. The categories below are the only ones available to this build, set by what it meets in reality.
Market
reach / exposure · problem clarity · audience fit · promise clarity · format · price · delivery · durability
Calibration · one change at a time
- Round 1 → 2: problem clarity. The promise changed from 'roof survey' to 'how long this roof has left, in writing'. One change only.
- Round 2 → 3: delivery. Report turnaround cut from five days to 48 hours; price held at £180 deliberately, so the change being tested stayed single.
Choose the smallest useful next move: run another bounded test, or carry what you have into Root Check.
You have decided which single thing changes next. The decision is yours, not the system's.
Gate · Evidence must exist before integration. Calibration may send the build back out instead.
Root Check
Regeneration
- Replenished: skill and confidence on the ladder, one new relationship with a local solicitor.
- Depleted: Saturday mornings, and roughly four evenings writing reports.
- Strain: one yes — report writing ran past 8pm twice in week three. Scope reduced: two surveys per Saturday became the hard ceiling.
Ownership & Control
Result
The report template, the photographs and the client list are ours. The buying groups are not — that reach belongs to someone else's platform and is the main control risk.
Documentation & Absence
Result
Absence test passed in part: Ada ran a full survey visit from the checklist while Jerome was away, but could not write the report without him. The report step is not yet transferable.
Answer honestly whether this build leaves the household more capable or more depleted, then record ownership and the absence test.
Both checks are recorded. Two strain flags require scope reduction before the build continues or grows.
Gate · Root Check and the ownership check must both be recorded before the horizon opens.
The Wider Horizon
Choose what happens now
Improve
Keep the capability and make a known operational improvement during normal use. This does not create another Providence test.
Partner
Keep the capability and pursue complementary external capability, because the evidence shows a missing piece is better accessed through partnership.
Run Another Test
The evidence is not sufficient for the decision the household wants to make, or a material change needs another bounded reality test. The cycle stays open and returns to First Contact with a new Evidence Round.
Retire
Stop the build, based on the evidence gathered.
Run Another Test does not close the cycle. It opens a new Evidence Round in this same Providence cycle and returns you to First Contact.
Expansion is not a Day-30 action. The record may note that this build looks worth reconsidering for expansion, and that question is answered at the 90-Day Horizon Review.
The household's decision
Ending
Completed — Integrated
In the household's own words
Maintain. The survey in its tested form — £180, 48-hour written report — keeps its place in the household exactly as it stands. We wanted to call the shorter report an improvement, but shortening it changes the delivery model, and nobody has bought the shorter version. An untested change is not an integrated improvement, so that becomes a bounded test in its own right rather than something we quietly fold in now.
What now belongs to the household
A tested survey checklist, a report format two people can follow, a paid client record, and a working relationship with a local solicitor.
Completed — Integrated means the household has decided the tested capability, asset, process or partnership deserves a continuing place in the household.
Next Horizon
- A shortened report format — a bounded test of its own, because it changes delivery. Parked, not started.
- A short 'roof age' guide that estate agents could hand out — parked, not started.
- Teaching the report-writing step to Ada so the build stops depending on one person.
Parked, not started. A new cycle begins only when the Decision Maker chooses one.
Choose one of the five and write it in your own words. If the change you have in mind alters the promise, the person it is for, the process, the price or the delivery model, that is Run Another Test, not Improve.
The decision is written. Maintain, Improve and Partner close the cycle as Completed — Integrated; Retire closes it as Completed — Retired; Run Another Test keeps the cycle open and returns you to First Contact.
Gate · A completed cycle requires real-world evidence and a final decision written in the household's own words — Maintain, Improve, Partner, Run Another Test or Retire. Expansion is a 90-Day Horizon Review question, not a Day-30 action.
Providence Record
Read the full recordExample · Internal capability
Invoicing without one person
The Reading
Returned signals
- knowledge captured14knowledgeRound 1
Fourteen steps written down. Two of them turned out to be 'and then I just know which supplier to call first'.
- process completed1capabilityRound 1
First full run performed against the written record by the original holder. 95 minutes.
- missing context discovered3knowledgeRound 2
Marie hit three points where the record assumed knowledge: which account, which reference format, when to escalate. All three repaired.
- another person performed the task1capabilityRound 2
Marie completed the run in 70 minutes with two questions asked. Not yet independent.
- absence test1capabilityRound 3
Four-day absence. Marie ran it alone, no questions, nothing missed. 55 minutes.
- errors / friction1capabilityRound 3
One friction: a late payer escalation was delayed two days because the record did not say who has authority to offer a payment plan.
What the evidence demonstrates
Recorded so far: knowledge captured, process completed, missing context discovered, another person performed the task, absence test, errors / friction. These are different kinds of evidence, not stronger versions of one another.
- knowledge captured — the steps now exist outside one head
- process completed — the process runs against the written record
- missing context discovered — a real gap in the record, found by use rather than review
- another person performed the task — the capability transferred at least once, with support
- absence test — the process survived the original holder being away
- errors / friction — where the process actually breaks
What it does not demonstrate
- knowledge captured — does not yet show that anyone else can act from them
- process completed — does not yet show that it runs when the original holder does not run it
- missing context discovered — does not yet show that the repair worked — that needs another run
- another person performed the task — does not yet show independence, or that it holds under absence
- absence test — does not yet show that it survives a longer absence or an unusual month
- errors / friction — does not yet show whether a fix removes the friction
The tacit steps — the ones that never get written because they feel like nothing — were the only steps that actually blocked another person. Writing it down was not the work; being absent was.
Enough evidence to decide
Sufficiency is relative to the decision in front of you. Evidence can be enough to justify another bounded test and still be nowhere near enough to conclude durability or transfer.
Enough evidence to decide
- Maintain — It worked in actual use 4 times (knowledge captured, process completed, another person performed the task, absence test). No unresolved confirmed strain blocks continuing, so keeping it substantially as it is rests on evidence. Repetition across further rounds is not required.
- Improve — It worked in actual use 4 times (knowledge captured, process completed, another person performed the task, absence test). Evidenced friction: missing context discovered, errors / friction. If a change alters the promise, the person it is for, the process, the price or the delivery model, its effect is unknown — that is Run Another Test, not an integrated Improve.
- Run Another Test — Material uncertainty remains: whether the repair to missing context discovered, errors / friction holds. A bounded next test could reduce it.
Not yet enough to conclude
- Partner — It worked in actual use 4 times (knowledge captured, process completed, another person performed the task, absence test). No confirmed evidence yet names a complementary capability, access point, role or control issue that partnership would supply.
- Retire — It worked in actual use 4 times (knowledge captured, process completed, another person performed the task, absence test). The evidence does not point to poor fit, weak value or excessive strain. The household may still judge it not worth continuing.
What would reduce the remaining uncertainty
- Whether the repair to 'missing context discovered' actually worked, observed again.
- Whether the repair to 'errors / friction' actually worked, observed again.
These rules are Providence rules, applied deterministically. AI may explain them. It may not change them, and there are no confidence scores here.
What can be changed
One evidence-based change at a time. The categories below are the only ones available to this build, set by what it meets in reality.
Knowledge / Memory
completeness · findability · understandability · usability · missing context · transferability · absence resilience
Capability / Operations
process clarity · hidden assumptions · repeatability · friction / errors · transferability · time burden · dependence on one person
Calibration · one change at a time
- Round 1 → 2: hidden assumptions. Two steps were tacit — 'I just know which supplier to call first'. The evidence showed unwritten knowledge, not an unclear problem, was blocking transfer. One change only.
- Round 2 → 3: transferability / usability. The record moved from prose to a numbered procedure with account and reference details in line, so a second person could perform it independently. Then the absence test was run.
Choose the smallest useful next move: run another bounded test, or carry what you have into Root Check.
You have decided which single thing changes next. The decision is yours, not the system's.
Gate · Evidence must exist before integration. Calibration may send the build back out instead.
Root Check
Regeneration
- Replenished: shared ownership of the money process, and noticeably less low-level anxiety about being away.
- Depleted: about three hours of Marie's time and two focused mornings writing.
- Strain: no. The boundary held — no evening work at any point.
Ownership & Control
Result
The procedure, the reference format and the escalation rules are entirely ours, stored where both authorised people can reach them. Nothing depends on an outside party.
Documentation & Absence
Result
Absence test passed. One gap remains: authority to offer a payment plan is undefined, so that decision still routes back to one person.
Answer honestly whether this build leaves the household more capable or more depleted, then record ownership and the absence test.
Both checks are recorded. Two strain flags require scope reduction before the build continues or grows.
Gate · Root Check and the ownership check must both be recorded before the horizon opens.
The Wider Horizon
Choose what happens now
Maintain
Keep the current capability substantially as it is.
Improve
Keep the capability and make a known operational improvement during normal use. This does not create another Providence test.
Chosen by the householdPartner
Keep the capability and pursue complementary external capability, because the evidence shows a missing piece is better accessed through partnership.
Run Another Test
The evidence is not sufficient for the decision the household wants to make, or a material change needs another bounded reality test. The cycle stays open and returns to First Contact with a new Evidence Round.
Retire
Stop the build, based on the evidence gathered.
Run Another Test does not close the cycle. It opens a new Evidence Round in this same Providence cycle and returns you to First Contact.
Expansion is not a Day-30 action. The record may note that this build looks worth reconsidering for expansion, and that question is answered at the 90-Day Horizon Review.
The household's decision
Ending
Completed — Integrated
In the household's own words
Improve. The run keeps its place: Marie performed it alone through a four-day absence with nothing missed. The one named friction — nobody knows who may authorise a payment plan — is a written rule we add during normal use. It changes no step of the process and needs no test of its own. What has not been shown is a longer absence or an unusual month, and we are not pretending otherwise.
What now belongs to the household
A tested monthly invoicing procedure, a second authorised person who has performed it alone, and a known gap that is now named rather than hidden.
Completed — Integrated means the household has decided the tested capability, asset, process or partnership deserves a continuing place in the household.
Next Horizon
- The same treatment for the supplier renewal calendar — parked.
- A written rule for payment-plan authority, which the friction exposed.
Parked, not started. A new cycle begins only when the Decision Maker chooses one.
Choose one of the five and write it in your own words. If the change you have in mind alters the promise, the person it is for, the process, the price or the delivery model, that is Run Another Test, not Improve.
The decision is written. Maintain, Improve and Partner close the cycle as Completed — Integrated; Retire closes it as Completed — Retired; Run Another Test keeps the cycle open and returns you to First Contact.
Gate · A completed cycle requires real-world evidence and a final decision written in the household's own words — Maintain, Improve, Partner, Run Another Test or Retire. Expansion is a 90-Day Horizon Review question, not a Day-30 action.