Seven trade-offs the work keeps returning to, and the position taken on each.
Both ends of every axis represent legitimate engineering objectives, decision pressures or evidence lenses. Neither end is inherently better. A marker is not a score: it represents an owner-ratified declaration on a designer-selected five-step lattice. Where that declaration depends on a condition, the condition is named and the marker is drawn as a span rather than a point.
READING CONTRACT
These are my current positions. They are not skill ratings and they are not universal rules. They describe how I presently reason when these pressures conflict. A position may move when the evidence or the constraints change.
BASIS names where the reasoning comes from — professional practice, engineering reasoning, or both. These are kinds of grounding, not evidence grades, ranks or counts of strength.
PUBLIC SUPPORT records whether and how a qualifying public artifact supports a narrow claim. ASSERTS, SPECIFIES and DEMONSTRATES describe the relation to that claim, not strength, maturity or proficiency. Its absence records a limit of public verifiability, not a weakness in the position, the reasoning behind it or the professional practice it may draw from.
RELATED MATERIAL points to public work that helps explain a position. It does not support the position by itself. Each entry states the limit of what it establishes.
Declaration wording is owner-ratified. Lattice placement and anchor selection are design interpretations, not additional owner claims.
MATERIAL
01 · 02
Grounds the practice in physical supply conditions: what is held, and what the supply of it actually does rather than what it promises.
01 / MATERIAL
What deserves stock?
Toward consequence, conditionally.
CONSUMPTIONCONSEQUENCE
I do not decide that an item deserves stock from consumption history alone. I start with the consequence of its absence, then test that need against demand, lead time, substitutability, repairability and the credible ways of recovering availability.
A rarely consumed part can therefore deserve stronger protection than a frequently consumed one.
MOVES WHEN
The consequence of absence changes, a substitute becomes genuinely reliable, replenishment becomes sufficiently dependable, or another recovery mechanism can provide the same protection with less committed stock.
BASIS
Grounded in professional practice. Basis is a kind of grounding, not an evidence grade.
HERVÉ implements cycle counting, stock and replenishment. It does not establish this stock policy.
02 / MATERIAL
Which lead time do you plan against?
Toward observed.
NOMINALOBSERVED
A contractual or quoted lead time is a useful commitment, but it is not enough to describe operational exposure.
For planning, I trust the observed distribution of actual lead times more than a nominal number. The difference between the two is information, not noise.
MOVES WHEN
Observed performance becomes sufficiently stable and aligned with the contracted commitment that the nominal value becomes a credible description of the distribution rather than only a promise.
BASIS
Grounded in professional practice. Basis is a kind of grounding, not an evidence grade.
SYSTEM
03 · 04 · 07
Captures behaviour, resilience, optimisation and supportability — how the whole thing performs, survives and stays maintainable over time.
03 / SYSTEM
What should the system be able to survive?
Slightly toward resilience, conditionally.
EFFICIENCYRESILIENCE
I do not treat unused capacity, redundancy or protection as waste by default. Efficiency is valuable until the system becomes unable to absorb ordinary variability or recover from a credible disturbance.
The relevant question is not whether slack exists, but whether the cost of that slack is lower than the exposure it absorbs.
MOVES WHEN
The cost of protection is demonstrated to exceed the expected cost and consequence of disruption, and recovery remains sufficiently fast and predictable without it.
BASIS
Grounded in professional practice and engineering reasoning. Basis is a kind of grounding, not an evidence grade.
The public WORK entry describes an authored teaching simulation and its resilience trade-offs. Its source is not public, the model is not calibrated against a real plant, and its behaviour is not an independent result supporting this position.
04 / SYSTEM
Where is the real constraint?
Strongly toward system performance.
LOCAL PERFORMANCESYSTEM PERFORMANCE
A local improvement is provisional until its effect is visible at system level.
Increasing the performance of one station, team, stock point or algorithm does not establish improvement if the constraint simply moves, another part of the system absorbs the cost, or the final service does not improve.
MOVES WHEN
The position moves toward local performance when a local requirement is independently binding — for safety, quality, compliance or service — or when repeated evidence shows that local performance is a reliable proxy for system performance without displacing the constraint or transferring unacceptable cost or risk elsewhere.
BASIS
Grounded in professional practice and engineering reasoning. Basis is a kind of grounding, not an evidence grade.
The public WORK entry describes an authored teaching simulation in which a bottleneck moves when it is lifted. Its underlying source is not public; that behaviour illustrates the principle but is not an independent result or field measurement.
Omnion contains a documentary base and one sourced technique sheet with a test protocol. No physical trial has been run.
DECISION
05 · 06
Captures modelling, uncertainty, automation and authority: what a model is allowed to be, and who remains accountable for the decision it informs.
05 / DECISION
What happens when the model is wrong?
Toward uncertainty.
PREDICTIONUNCERTAINTY
A forecast, model or scenario is an input to a decision, not a record of the world.
I would rather preserve visible uncertainty than replace it with false precision. Prediction becomes useful when its error is measured, its limits travel with the result and it performs better than a simpler baseline.
MOVES WHEN
Repeated backtesting establishes stable and relevant error bounds and the model demonstrably improves the decision it supports. Even then, predicted and observed states remain different things.
BASIS
Grounded in engineering reasoning. Basis is a kind of grounding, not an evidence grade.
The project defines an explicit separation between assertions, evidence and epistemic state in its normative corpus. It does not evidence deployed demand planning, backtesting or a supply-chain forecasting method.
The project keeps heuristics labelled as heuristics. It does not evidence deployed demand planning, backtesting or a supply-chain forecasting method.
06 / DECISION
When does automation exceed its authority?
Toward human authority.
AUTOMATIONHUMAN AUTHORITY
I want automation to detect, calculate, compare, simulate and propose as far as those functions can be made reliable.
It must not silently acquire authority merely because it became technically capable of producing an answer.
The more consequential or irreversible the decision, the stronger the need for explicit authority, traceability and the ability to intervene.
MOVES WHEN
An automated decision is bounded, attributable, auditable, reversible where necessary, and demonstrably more reliable than the process it replaces. What may move is the boundary of delegated authority; authority itself must never become implicit.
BASIS
Grounded in engineering reasoning. Basis is a kind of grounding, not an evidence grade.
The public repository implements machinery that can detect a registered resource change while leaving correctness and authority outside that mechanism. It does not establish the reliability of automated decisions in general.
EXOMIND specifies that generative output and remote agents may not directly drive the motors. That boundary is designed, not yet physically demonstrated.