The Socratic Exponent

• Fidelity decay. Each generation transmits the practice imperfectly. If a fraction φ < 1 of the discipline survives per generation — in particular the withheld answer, the expensive move — the effective base shrinks generation over generation, and the squaring is replaced by something far slower or by outright decay.
• Graph overlap. The model assumes every new partner is fresh. Real social graphs are clustered, so a growing share of interchanges land on someone already reached. Overlap converts squaring into saturation.
• Dropout. Practitioners stop. The practice is effortful and unrewarded in most institutions, per Chapter 7, and the attrition term applies at every generation rather than once.
• Attention scarcity. The interchange requires sustained attention from a competent partner. Attention is finite, unequally distributed, and does not square. This is the term that carries the volume's central social objection: a method whose input is scarce senior attention may be captured by those who already have access to it.

x_{t+1} = ( φ · (1 − o_t) · (1 − d) · x_t )²

φ   transmission fidelity     o_t  overlap at generation t
d   dropout rate              attention cap: x_{t+1} ≤ A_t

Eq. 8.2 is still a toy model. Its value over Eq. 8.1 is that its parameters are measurable in a small pilot: fidelity by coding whether second-generation practitioners actually withhold answers, overlap from the interchange log, dropout from participation records, and the attention cap from the calendar.

Socratic probe

"If φ, o, d and A are unmeasured in your setting, what exactly does citing the exponent add beyond a feeling of momentum?"

For a reader who will carry one page out of this volume, the operative content is four questions and one artefact. Each maps to a chapter, and each has an answer that can be written down or shown to be missing.

• 1. Boundary. Where on our reward surface does the score rise while a constraint degrades? (Ch. 5, Step 1; Ch. 6, Stage 1.)
• 2. Comprehension. On which pre-declared held-out domains does this system's reasoning remain stable? (Ch. 5, Step 2.)
• 3. Override. Is the human override a variable inside the objective, and what protects the channel that sets it? (Ch. 6, Eq. 6.2 and its stated limit.)
• 4. Reasons. What fraction of our decisions carry a recorded reason and a named falsifier — and what fraction of revisions carry neither? (Ch. 7, procedures 4 and 6.)
• The artefact. An append-only decision ledger. Without it, none of the four questions can be answered after the fact, and the audit reduces to recollection.

That is the whole of what this volume asks an engineering organisation to do. It does not promise a beautiful world; the title names an intention, not a result. It proposes that the intention is only actionable if the reasons are written down where they can be attacked — by a colleague, by a regulator, or by the engineer who inherits the system after everyone who built it has moved on.