The Method
Diagnose. Deploy. Remember.
Installing operational memory is not a one-off project. It is a repeatable method: find where the memory leaks, deploy a governed loop that captures it, then let it compound. Same method, every site.
The machine proposes. The human authorises. The system remembers. The institution improves.
01 — Diagnose
Map where the memory leaks.
Every operation already has a memory — it is just scattered, informal, and walking out the door. The ApexIndex diagnostic finds exactly where it leaks across five dimensions before a single line of software is deployed.
Capture
Where evidence is created and where it never gets recorded. Verbal handovers, photos on personal phones, knowledge that walks out at shift end.
Structure
Whether what is captured can be read by anything other than the person who wrote it. Free text, inconsistent codes, no link between a fault and its asset.
Retrieval
How long it takes to answer "have we seen this before". If the last fix lives in an email thread from two years ago, the memory is effectively gone.
Governance
Who decides what becomes the record of truth. Without a review gate, bad entries compound as fast as good ones and the memory cannot be trusted.
Compounding
Whether the operation gets measurably better each cycle. A memory that is only ever written and never read is a cost, not an asset.
The ApexIndex diagnostic
A fixed-scope assessment scored against these five dimensions. You receive a ranked map of where memory is lost and what a governed loop would recover.
£2,500 · fixed scope
02 — Deploy
Stand up the governed loop.
The Sentinel pilot installs a single, closed loop in four weeks. Evidence in, structure proposed, human review, approved memory out, daily brief back to the floor. Five stages, one direction, no autonomous writes.
- 01
Evidence
Raw evidence arrives as it already exists — SAP exports, CMMS records, field logs, engineer notes, machine events. Nothing is reformatted by hand and no source is left out.
- 02
Structuring
Sentinel reads the evidence and proposes structure: the asset, the fault, the likely cause, the action. This is a draft. It is not yet memory and it is never treated as such.
- 03
Human review queue
Every proposed entry lands in a review queue. A competent human accepts, corrects, or rejects it. Nothing is written to the record of truth without that decision.
- 04
Approved memory
Approved entries become governed operational memory — structured, attributed, and auditable. The approval, the reviewer, and the original evidence are all retained.
- 05
Daily brief
Each morning the team receives a brief drawn only from approved memory: what changed, what repeated, what to check first. The loop closes and begins again.
Pilot: £8–15k / month · 4 weeks · one site, one loop, one record of truth.
03 — Remember
Let the memory compound.
Once the loop runs, the record of truth gets richer every cycle. The intelligence below is derived only from human-approved memory — never from raw, unreviewed evidence. That single rule is what makes the numbers defensible.
Recurring faults
The same fault is never new twice
Approved entries are matched across assets and time. When a pattern returns, it is flagged with the history of what was tried and what actually worked.
Health scores
Asset health you can defend
Condition scores are derived from the approved record, not from a dashboard guess. Every score traces back to the evidence and the human who signed it off.
MTTR
Faster, evidenced resolution
Mean time to repair falls because the next engineer starts from the last verified fix instead of from zero. The metric is measured against approved memory only.
Compounding is the whole point. A diagnostic tells you where you stand once. Governed memory raises the floor every month — and because it is approved, it holds up to audit.
04 — Why it holds
Governance is the product.
The model does not need to be clever. It needs to be reliable. And it can only be reliable if the thing it writes into is clean.
The agent can stay boring if the memory writes are clean.
The human authorises
No entry becomes memory without a competent person accepting it. The review gate is not a feature you can switch off — it is the load-bearing wall.
Every entry is attributable
Who approved it, when, and from what evidence. An auditable trail is what turns a pile of records into a record of truth.
No autonomous writes
Sentinel does not write to SAP or any system of record on its own. It proposes; your team disposes. Clean inputs in, trustworthy memory out.
The same method, every site
Diagnose, deploy, remember. Because the loop is identical everywhere, what is learned at one site can be governed and reused at the next.
Install operational memory.
Start with the diagnostic. We map where your memory leaks, then deploy the governed loop that recovers it — same method, your site.