Uneven effort, hesitation, weak joins, doubtful figures, and the producer’s mechanical contact suggested where senior attention should land.
Canonical definition · coined by Bud Bhattacharyya · 10 July 2026
Unreviewable fluency
The condition in which work is coherent, polished, and apparently complete, but the evidence, assumptions, uncertainty, branch choices, and human ownership on which reliance depends cannot be efficiently inspected.
The finish of the work has stopped carrying reliable information about the strength of the thinking underneath it.
01 · The reviewer experience
The work looks finished. The looking has nothing to grab.
The prose is clean. The structure is right. The numbers agree everywhere they touch. Nothing pops as wrong, and the reviewer’s old instrument remains silent.
Uniform fluency flattens those asymmetries. Polish becomes evidence about the tool, not necessarily about the inquiry behind the work.
Rebuild the work personally, send back a vague request to “make it different,” or approve a coherent object without enough grounds for reliance.
02 · How it forms
The condition is produced before the final review begins.
Authority laundering
The human’s preliminary frame returns from the model wearing the appearance of external confirmation. Agreement occupies the chair where corroboration used to sit.
Premature artifacting
A provisional outline, memo, deck, or answer acquires authority before the understanding has earned it. Later work improves the artifact inside its inherited frame.
Flattened production residue
AI gives strong and weak claims the same finish. The seams, hesitations, uneven confidence, and mechanical contact senior reviewers once read disappear.
Under-owned production
The producer can present the recommendation without being able to reconstruct, update, or defend the load-bearing logic at the level the decision requires.
03 · Diagnostic
Nine signs that the work has become fluent before it is reviewable.
- 01The work is polished but you cannot tell which claims are load-bearing.
- 02Strong and weak claims arrive wearing the same finish.
- 03The producer cannot say where their conviction ends and the model’s fluency begins.
- 04A second model pass returns another seamless first answer rather than an independent read.
- 05The first polished artifact quietly becomes the frame of the inquiry.
- 06Alternatives are absent, or appear only as a clean after-the-fact story.
- 07Evidence is cited but difficult to inspect in the form where its relevant property is visible.
- 08Senior review becomes reconstruction, generic comment, or unsupported approval.
- 09Uncertainty can be admitted only as a confession rather than marked as professional state.
04 · What it is not
Not a synonym for bad AI output.
Not merely hallucination
The claims may all be factually plausible. The problem is that the grounds, limits, and human ownership cannot be efficiently inspected.
Not a demand for rougher work
The answer is not to reintroduce accidental seams or make people type more slowly. The missing review signal should be designed.
Not total transparency
Prompt logs, chat histories, source dumps, and complete branch archives would transfer the inquiry burden rather than govern it.
Not junior incompetence
The same default stance affects senior professionals in their own AI use. Mutual deference and premature closure are architectural problems.
05 · The remedy
Replace accidental residue with designed reviewability.
The making must become professionally legible as the work is made, without exposing raw cognition and without asking a model to certify itself.
- ProduceGovern the live inquiry before a decision-shaped artifact acquires closure.
- ExposeCut a producer-corrected State of the Work from the same current production state as the professional artifact.
- ReviewLet a separately situated reviewer and AI inspect the artifact, State, and native evidence.
- RelyEnd in a bounded human decision, direction, and reopening conditions.
- CompoundCarry the useful distinction into the next interval.
Inspect the method→“What was never designed can be replaced by something that is.”
Bud Bhattacharyya
Citation
Bhattacharyya, Bud. “Unreviewable Fluency.” re:compound, 10 July 2026.
Unreviewable fluency: work that is polished and apparently complete, but whose evidence, assumptions, uncertainty, branch choices, and human ownership cannot be efficiently inspected.The term and definition may be reused with attribution.
A practical first step
Thirty minutes about the next piece of work you’ll have to sign.
Who’s bringing it, when, and whether that’s the team to start with. No deck, no demo, nothing to prepare.