Editorial method

Editorial standard

Explanations are expected to connect the abstraction to a calculation, an implementation or system constraint, and the limits of the claim. A smooth paragraph is not evidence. A citation is not decoration. A benchmark is incomplete until its task, baseline, metric, assumptions, and date are visible.

Source hierarchy

1. Primary literature and standards. Papers, standards, specifications, official technical reports, and reproducible datasets carry the most weight for factual claims.
2. Authoritative synthesis. Textbooks and serious lecture notes provide stable framing and derivations.
3. Vendor material. Useful for describing a vendor's own architecture or claim, but labeled and checked against independent evidence where possible.
4. Secondary reporting. Used for context, not as the sole support for a technical conclusion when a primary source exists.

Review labels

Author-reviewed

The accountable author has checked the chapter for publication. It does not mean independent expert review.

Independent technical review

Shown only when a named or otherwise verifiable review record exists. No independent reviewer is invented or implied.

Source mapping

Each chapter has a registered source list, and rewritten chapters map their load-bearing claims to those sources inline. A reference supports only the bounded claim beside it; proximity does not transfer authority to unrelated claims.

What readers perceive as AI-shaped prose

These are editing cues, not proof of AI authorship. A human can write this way, an AI-assisted draft can avoid it, and no stylistic audit can establish who or what wrote a passage. The concern is the reader's experience and the work's evidentiary quality.

The signal is cumulative: statistically repeated transitions and sentence openings; interchangeable chapter skeletons; habitual negative-correction rhetoric; uniform cadence, length, and rhetorical beats; generic confidence without inspectable support; faux completeness and constant over-signposting; or simulated authority through fabricated precision, unnamed experts, synthetic quotations, and generic source notes.

No single cue is disqualifying. Useful textbook furniture may repeat, and a contrast may be the clearest sentence. Editing begins when the pattern overwhelms what the subject requires or when rhetoric exceeds evidence. The remedy is subject-driven structure, costly specificity that would break if the topic nouns were swapped, explicit limits, and traceable derivations, artifacts, measurements, or primary-source chains.

Dates and changing claims

datePublished records the publication date in the chapter manifest. dateModified changes only when the chapter or its publication record is materially revised; a routine build does not create a freshness date. Claims about companies, benchmarks, markets, or current systems must state their time boundary in the prose.

Automation and authorship

Generative tools were used during the 2026 rebuild for research routing, drafting and substantive rewriting, code assistance, copy editing, and quality-control automation. Steven Geller is the named author and remains accountable for every published claim. Tool output is not treated as a source, author review, or independent technical review.

No generated citation, credential, reviewer, experimental result, or claim is accepted merely because a system produced it.

Corrections

Corrections are separated from general release notes so a reader can see what was wrong, what changed, and which chapters are affected. See the errata process and publication changelog.