Controlling LLM Prose

Area VIII · Precision & Disambiguation

Hedging vs. Commitment

Calibrating certainty

Set the level of epistemic certainty deliberately rather than hedging by habit.

Definition & mechanics

Hedges (may, might, possibly, generally) and boosters (will, always, guaranteed) encode how certain a claim is. Technical prose should match confidence to evidence: hedge genuine uncertainty, commit to what is known, and avoid reflexive hedging that makes solid facts sound shaky.

A curiosity

Over-hedged docs erode trust as much as over-confident ones. If everything “may” happen, the reader cannot tell the load-bearing claim from the throwaway.

Examples — the same idea, several ways

One claim across the certainty dial—from bare speculation to absolute commitment. Match the rung to your evidence.
The claimThe cache reduces latency.
  • Speculative

    The cache might, in some cases, help reduce latency.

    Stacked hedges; appropriate only for an untested hunch.

  • Tentative

    The cache appears to reduce latency in most workloads.

    One hedge, scoped—honest about partial evidence.

  • Committed

    The cache reduces latency.

    Plain assertion; use when you have measured it.

  • Boosted / normative

    The cache MUST reduce latency below the 50 ms SLO under peak load.

    Requirement-strength commitment, testable and absolute.

Copy-ready prompt

Copyable Prompt
Apply the concept of Certainty Calibration (calibrate epistemic certainty to the evidence—state established facts plainly without hedging words, reserve hedges (may, might, can) for genuinely uncertain or conditional cases, and remove reflexive qualifiers that weaken solid claims) to the provided text and rewrite in place.

Apply the edits directly to the file as a clean, production-ready deliverable. Output ONLY the revised document text without embedding revision notes, changelogs, or meta-commentary in the file itself—put any explanations in your chat response only.