GFT / Principles
Few measures, clear ownership, source capture and evidence-gated scale. Each principle must survive contact with the operating floor.
Principle in practice
Pick a candidate. The knot ties only when the measure can guide a responsible action, with an owner, a boundary and a guardrail.
A rate with no accepted-output boundary can reward haste while defects, rework or fatigue pile up elsewhere.
Keep it out of the operating review as a standalone target.
Conceptual test. Candidates are examples, not client results.
Count units accepted at a named process boundary, in a stated period, with rework treated explicitly.
Enter it with its contract and its guardrails.
Conceptual test. Candidates are examples, not client results.
Water, energy or material use means more when attached to accepted output in the same period and boundary.
Enter it once the estimation rule is written down.
Conceptual test. Candidates are examples, not client results.
1 / Start with the decision
Build the decision backwards from action: the question, the owner, the observation, then the response. The display comes last.
The display comes first, so nobody owns the response. A wall of indicators can look thorough and still change no decision.
Frame an operating question with a boundary. Why did accepted output change on this line? is more useful than What does the factory dashboard show?
An owner needs enough authority to investigate and correct the cause. A measure with no owner becomes commentary, however precise it looks.
Choose the smallest reliable observation. A controlled manual record may answer the question before a sensor network or a model is justified.
Agree the review, the exception threshold and the escalation path before collection begins. If every value leads to one response, the measure adds little.
Design test. Remove any indicator whose absence would not change a review, an investigation or an operating decision. The twelve tiles are generic examples, not a client dashboard.
2 / Capture at source
Step 1 of 3 CapturedReviewedGoverned
Give the event a controlled identifier, a time boundary and a definition the operator can apply. Record a missing value as missing, never as zero.
Supervisors compare the record with the process, chase exceptions and settle definition questions. The test is sense, a stable unit and no contradiction.
A correction keeps the original entry, the revised value, the reason and the person responsible. A data repair then never passes for an operating change.
The first record sits close to the work. Its definitions and revision history let the team review, correct and trust the signal without hiding the earlier state.
State: A legible source record with known gaps and an accountable recorder.State: The team uses the record in review and knows how it fails.State: A governed record fit for comparison, learning and proportionate automation.
Illustrative record. Procedural schematic, not client data.
Pulled taut, the record becomes a load-bearing line.
3 / Keep technology proportionate
The right system is the smallest one that can support the decision reliably. Choose a route and read what it costs, what it risks and what evidence it needs.
The architecture stays subordinate to the operating evidence contract.
4 / Design for real constraints
A design that needs continuous connectivity, abundant specialist support or perfect measurement is not a finished operating design. Set the link and see what each constraint asks of it.
Point to a constraint to light the part of the link it acts on.
5 / Trace every claim
The reader should tell apart what a source reported, what arithmetic produced, what a planner assumed and what a confidential boundary protects. Tap a status and read the sentence again.
the on
Source, date, unit and qualification travel with the sentence.
It keeps the date, the unit and the qualification. A policy target stays a target, and a dated sector figure never becomes a current operating baseline by repetition.
Accepted units divided by units started, same boundary and period.
Inputs, formula and rounding are open to inspection. A derived result can be reproduced, but it is never relabelled as an observation from the floor.
The basis is stated beside the value, and the value is labelled illustrative.
An illustrative scenario demonstrates a method. It is not a client result, an industry benchmark or a forecast of what will happen.
The reader sees that a boundary exists. The protected content stays out.
Public method and public-source context can be examined while client identity, engagement identifiers and source deliverables stay unpublished.
Sample sentence for the demonstration only. It is not a GFT or client result.
Point to a part of the sentence, or pick one here
Last quarter fixes when the figure was measured. Without its period, a rate cannot be compared with any other rate.
The accepted-output rate is the measure. Its unit and formula travel with it: accepted units divided by units started.
Line A is where the figure was measured. Under Confidential the boundary is stated and the name of the line is withheld.
Improved is the claim itself. It holds only as far as the evidence status of the whole sentence allows.
6 / Prove before scale
A pilot is evidence only when its comparison and guardrails were agreed before the result arrived. Set both below. The gate reads the weaker one, because a gain that breaks a guardrail is not a gain.
ADAPT / 58Weakest input: guardrails
Scale: The gate clears. The comparison holds, guardrails stay intact and the team can run the new state reliably.
Adapt: The mechanism needs work. The signal is useful, but the operating condition, definition or implementation changes before another test.
Stop: The commitment is not justified. The use case fails its test, moves harm elsewhere or needs evidence the operation cannot yet produce.
Illustrative index, 0 to 100, not a benchmark. The zone edges at 33 and 66 are example settings for the demonstration. Agree real ones before the pilot, not after.
Protect the boundary
Method sheet, public pageMethod sheet, rule behind each line
Three rows stay barred in every view of the public page.This is a disclosure decision, not a claim that the underlying work was public.
Research register. These papers develop the measurement, utilisation and data-governance arguments; their web editions are in preparation and are not presented as published links.
Publish the method. Protect the engagement.
The company pages use public-source context and general programme method. They do not name the client, expose engagement identifiers or offer the confidential deliverables as downloads.
1 · Start with the decision. A measure earns its place when it answers an operating question and has an owner able to act.
Read against it: Utilisation Decides, edition in preparation.
2 · Capture at source. Definitions, validation and correction history belong to the first record, not to a later report.
Read against it: The Industrial Data Spine, edition in preparation.
3 · Keep technology proportionate. Proven tools come before bespoke systems, and history-dependent models wait for usable history.
Read against it: Measure Before You Model, edition in preparation.
4 · Design for real constraints. Power, connectivity, supervision and measurement error are design inputs, not footnotes.
No edition in this register yet.
5 · Trace every claim. Sourced context, derived arithmetic, planning assumptions and illustrative values stay visibly different.
No edition in this register yet.
6 · Prove before scale. A wider commitment follows a baseline comparison, guardrail checks and verified corrective action.
No edition in this register yet.
Missing references stay missing. They are not replaced by a plausible publisher, a reconstructed quotation or a confident sentence without a source.
Carry the principles into operation
Choose the kind of decision in front of you. The chosen thread runs on into the page that helps most.