# Wrong Beautifully *How polish lends a sentence more certainty than its evidence can bear, and how to check it* Kanwar Partap Singh Gill, MD · November 5, 2026 · The Machine That Reads Here is a sentence I have invented: *Her headaches have resolved.* And here, also invented, is the conversation it was meant to summarize. She says the headaches come less often than they did, once or twice a week now instead of most days. She thinks the change began when she started sleeping better, though she is not sure. She stopped writing them down a while ago, so the numbers are her impression. She would like to know whether she still needs to come back about them. Imagine that a fluent summarizing system, given a recording of that exchange, produced the sentence above. It is a good sentence, in its way: short, clinical, calm, the kind of line that lets a problem be closed on a list. It is also wrong, and wrong in an instructive manner. Nothing outrageous has been invented. No headache has been conjured out of nothing. A small increase in certainty has done all the work. It is worth slowing down over what *resolved* claims that *less frequent* did not. The first difference is time: *resolved* says the problem is over, while *less frequent* describes something still happening. The second is measurement: she stopped counting, so even *less frequent* rests on her impression, and *resolved* rests on nothing she said. The third is the state of the problem. *Resolved* closes it; *less frequent* leaves it open. I want to be careful about what follows from that. What a real person with real headaches needs next is a clinical judgment about her, made by someone who knows her history and has heard her out. It is not a property of either word. But the word decides whether anyone is prompted to make that judgment. Her own question, whether she still needs to come back, has been answered by grammar before anyone answered it in conversation. The sentences that trouble me most are not the absurd ones. An absurd sentence gives the reader something to push against; it announces that a check is needed. The difficult sentence sounds entirely at home. Its vocabulary is right, its grammar is clean, and its conclusion sits exactly where a busy reader expects to find one. This essay is about that kind of sentence: why polish lends a claim more authority than its evidence can bear, and what discipline answers it without turning into a superstition about awkwardness. Machines make the question urgent for me, but the question is far older than they are, and it belongs to every writer who has ever turned a complicated encounter into a readable account, including the writer of these essays. ## Applause Begin with the word we use for such sentences. *Plausible* came into English in the 1540s from the Latin *plausibilis*, deserving applause, from *plaudere*, to clap. Within a few decades English was using it for something that merely has the appearance of truth, often, the etymologists add, a specious appearance. A plausible sentence is, by origin, one that invites applause, and applause is not the same thing as checking. It is a response to performance. The oldest serious account of that difference is in Plato's *Gorgias*, and it is surprisingly medical. Socrates presses the celebrated teacher of rhetoric on what exactly his art concerns. Persuasion, Gorgias says, in the law courts and the assemblies, about what is just and unjust. Socrates then draws a distinction on which the rest of the dialogue turns. There are two kinds of persuasion, he suggests: one that produces belief without knowledge, and one that produces knowledge. Gorgias agrees that rhetoric deals in the first. Then Gorgias makes a boast. He has often gone with his brother, a physician, to visit patients who refused to take their medicine or to submit to the knife or the cautery, and where the doctor could not persuade them, he, Gorgias, succeeded, by rhetoric alone. He adds that if a city were choosing a public physician, and a rhetorician and a doctor competed for the post before the assembly, the man who could speak would be chosen over the man who could heal. Socrates draws the conclusion with his usual courtesy and no mercy at all. Before an audience that does not know, the one who does not know is more persuasive than the one who does. Later in the dialogue Socrates refuses to call rhetoric an art. It is a knack, he says, a practiced routine, and a species of flattery. Its relation to justice is the relation of cookery to medicine: the cook aims at what is pleasant, the physician at what is good, and the cook goes about pretending to know which foods are best for the body. Near the end Socrates foresees his own trial in the same terms. He will be judged, he says, like a doctor prosecuted by a cook before a jury of children. I do not want to make a sophist of a summarizing system. It has no ambitions, wants no fee and seeks no applause. And Gorgias's boast is not, on its face, wicked: he persuaded patients to accept what the physician had advised. But that is precisely Socrates's point. The power of persuasion is independent of knowledge. It attaches as readily to a true conclusion as to a false one, and it is strongest where the audience has no way of telling the difference. A fluent sentence carries that same detachable power. It is the kind of sentence that sounds right, which is a different achievement from being right. And the reader of a clinical summary is often in the position of Gorgias's assembly. The clinician who opens the note next month was not in the room. She has the speech and not the conversation. She cannot taste the difference between the cook's dish and the physician's prescription, because she was not there when either was prepared. That is where fluency does its strongest work: at a distance, before readers who must rely on the text because the text is all they have. ## The Air of Reality In 1884 Henry James published an essay called "The Art of Fiction," and in it he made a claim about novels that most readers of novels would recognize. "The air of reality (solidity of specification) seems to me to be the supreme virtue of a novel," he wrote, the merit on which all its other merits depend. A novel earns belief through particulars: this staircase, these gloves, the exact angle of afternoon light on a particular square. The novelist manufactures the air of reality out of invented specifics, and we honor him for it, because we know what he is doing and have agreed to let him do it. Sixty-two years later, at the end of "Politics and the English Language," George Orwell used the same word for a vice. Political language, he wrote, is designed among other things "to give an appearance of solidity to pure wind." The technique James praised and the technique Orwell condemned are, at the level of the sentence, nearly the same: specification that produces the feeling of substance. What differs is the agreement with the reader. Fiction declares itself. Politics does not. The novelist's solidity is a gift we have asked for; the politician's is a trick played on people who did not consent to be entertained. A clinical summary lives under Orwell's agreement, not James's. Its reader is entitled to assume that every specific refers to something: that *once or twice a week* was said, that *since March* was established, that *denies* followed a question. And specifics are the most persuasive thing in prose. A number reassures; a date settles; a precise verb implies a witness. When a fluent system writes a clinical paragraph, the air of reality is the cheapest thing it can produce, because the patterns of clinical specification are everywhere in the language it has learned from. Solidity can be generated without reference. The prose will look exactly as it would if every particular had been observed. The same difficulty reaches personal writing. Ask a fluent system for a paragraph about a medical student's evenings in Amritsar and it may well offer something like this, which I have also invented: *Every evening the hostel corridor smelled of kerosene and frying onions, and somebody's radio was always playing old film songs.* It has the cadence of memory. It has a corridor, a smell, a sound and a pleasing *always*. It has, in James's sense, the air of reality. What it lacks is anything underneath it: any particular evening I am prepared to stand behind. Memoir sits uneasily between James and Orwell. It borrows the novelist's techniques and owes the reader the documentarian's truth. That is why every essay here ends with a note separating what is documented from what is remembered, and both from what is interpretation. A persuasive paragraph should not be allowed to invent my past for me, however well it scans. ## A True Sentence in the Wrong Place Not every fluent error is an invention. Some are relocations: true sentences moved to where they are no longer true, or no longer relevant. Suppose the invented summary also contained this: *Blood pressure well controlled.* Suppose further that the sentence was accurate when it was written, at a visit eight months ago, and that nothing in today's conversation touched the subject. Every word is correct. The sentence is false only in its tense, which places an old observation in the present without a date. A reader who trusts it believes something about today on the strength of a measurement nobody took today. Or suppose the summary added an explanatory gloss: a sentence linking her improvement to her sleep in the confident register of a textbook. Such a sentence might be defensible as a statement about many people. Attached to her, without the word *she thinks*, it becomes an explanation of her case that nobody established. The generalization has not changed. Its address has. The National Institute of Standards and Technology, in its 2024 profile of the risks of generative artificial intelligence, calls the broad hazard *confabulation*: the production of confidently stated but erroneous or false content. The same document warns that such systems can produce supporting material, including reasoning and citations, that purports to justify an answer without actually doing so. I use the profile for its vocabulary, not for a measurement. It is a voluntary framework for managing risk across many sectors, and it reports no error rate for any clinical tool. Physicians know the word from the neurology ward, where it describes a patient who fills gaps in memory with fluent, sincere, invented detail and no intention of deceiving anyone. The Latin behind it is gentle: *confabulari*, to talk together, from *fabula*, a story. I find the borrowing apt. The danger is not deceit. It is fluency arriving at the place where the facts ran out, and carrying on in the same tone of voice. The citation is the most refined form of the relocated truth. A real reference attached to a sentence makes the sentence look accountable; it gives the reader somewhere to go. But a genuine study of a particular group, cited beneath a sentence that generalizes its finding to everyone, misleads without inventing anything. The title exists, the authors exist, the result exists. What does not exist is the relationship the sentence claims with them. Opening the source is therefore the beginning of a check, not its end. ## How Certainty Gets In It helps to watch the change happen in slow motion. Here are four versions of one sentence about the invented patient, each a reasonable compression of the one before it: 1. *She thinks the headaches may be less frequent since she began sleeping better.* 2. *Headaches less frequent since sleep improved.* 3. *Headaches improved with better sleep.* 4. *Headaches resolved.* No step looks defective. Each is shorter and more confident than the last, and each could be defended as tidying. But between the first and the second, *she thinks* has vanished, and with it both the source of the claim and her uncertainty about it. Between the second and the third, *since* has become *with*, and a sequence in time has quietly become a cause. Between the third and the fourth, *improved* has become *resolved*, and a change has become an ending. Three small edits, none of them grammatical errors, have converted a hesitant observation by the patient into a settled fact about her. The same machinery works through other parts of speech. Verbs carry the state of an action, and a summary can move a plan forward by changing one. A visit in which a referral was discussed becomes, in a confident draft, a visit in which a referral was placed. Nobody has placed anything, and a reader may now believe the next step is under way. Attribution can collapse in the same way. If a daughter says her father has become forgetful, and he disagrees, and the physician has not yet assessed the matter, then *worsening memory* has fused three positions into one conclusion and silenced the disagreement that was the most important fact in the room. Silence can be converted too. Suppose the conversation never touched on her mood or her appetite, and the draft nonetheless supplies a reassuring line about each, because notes of this kind usually contain one and the headings seemed to want filling. A subject that was not discussed has become a finding that was assessed. Negatives stretch in the same way: *no headache yesterday* becomes *denies headaches*, and a statement about one day becomes a statement about her life. A gap in a conversation does not authorize a positive assertion or a negative one. It authorizes, at most, a note that the matter was not discussed. What strikes me about all these changes is their direction. The drift runs one way: toward certainty, toward completion, toward consensus. Qualifications fall away; none are added. Discussions harden into decisions; decisions are rarely softened back into discussions. I think the reason is partly that finished sentences are what summaries are for, and partly that confidence reads as competence. A sentence that ends on a strong word sounds as though its writer knew something. Its rhythm makes a claim of its own. ## The Engine at Lagado Jonathan Swift imagined a machine for writing long before anyone could build one, and he meant it as a joke. In the third part of *Gulliver's Travels* (1726), Gulliver tours the Grand Academy of Lagado, where projectors labor at schemes for extracting sunbeams from cucumbers and softening marble for pillows. One professor shows him a great frame of small wooden blocks, each papered with the words of the language in all their moods, tenses and declensions, linked by wires. His pupils turn handles at the edges, the words tumble into new arrangements, and whenever three or four of them happen to make part of a sentence, scribes copy it down. By this contrivance, the professor explains, "the most ignorant person, at a reasonable charge, and with a little bodily labour" could write books in philosophy, poetry, politics, law, mathematics and theology, "without the least assistance from genius or study." He has already filled several folio volumes with broken sentences, which he intends to piece together into a complete body of all arts and sciences. For nearly three centuries the satire was safe, and what kept it safe was the output. The engine produced fragments. Broken sentences announce their origin; nobody mistakes a heap of them for a treatise. Swift's joke depended on the visible gap between the machine's product and the real thing. The engines of our own time close that gap at the level where readers had always looked for it. They produce whole sentences, fluent paragraphs, documents with headings. The one feature that would have betrayed the Lagado frame, its brokenness, is precisely the feature they lack. Whatever errors they make arrive finished. Swift's academy contains a second scheme that cuts the other way. In the school of languages, Gulliver meets a project for abolishing words altogether: since words are only names for things, it would be more convenient for people to carry about with them the things they need to discuss. He describes sages bent under enormous packs, like pedlars, who meet in the street, open their bundles and converse for an hour by holding up objects. It is meant to be ridiculous, and it is. But it also describes, with comic exactness, the discipline a reader of a fluent summary needs. For each assertion, open the pack. Show me the thing this word names. Where in the conversation is the headache that resolved? Where is the referral that was placed? The sage of Lagado could say nothing he could not produce. A reviewer need not go that far, but should at least ask, sentence by sentence, whether there is something in the sack. ## The Reader Who Wants It Finished If fluent errors are easy to make, they are also easy to accept, and the reasons lie partly in the reader. Psychologists have studied the effect of ease itself. In a well-known experiment published in 1999, Rolf Reber and Norbert Schwarz showed people simple factual statements printed in colors that were either easy or hard to read against the background; the easier ones were more likely to be judged true. In 2000 Matthew McGlone and Jessica Tofighbakhsh found that people rated rhyming aphorisms as more accurate than versions with the same meaning that did not rhyme. In both cases the feeling of ease, which has nothing to do with truth, was read as a sign of it. These were laboratory studies of short statements, not studies of clinical documentation, and I extend them by analogy. But the analogy is not strained. A polished clinical sentence is easier to process than a halting one, and ease seems to lend credit it has not earned. A second line of research concerns people working alongside automated aids. Human-factors researchers call one of its findings automation bias, and in a 2010 review Raja Parasuraman and Dietrich Manzey described how it appears among experts as well as novices, and is not reliably dispelled by instruction alone. A 2012 systematic review by Kate Goddard and colleagues found it documented in studies of clinical decision support. The literature distinguishes two kinds of error. In an error of commission, a person follows an automated suggestion that is wrong. In an error of omission, a person fails to notice a problem because the aid did not flag it. A fluent summary invites both at once. *Resolved* is accepted: commission. Her question about coming back is never noticed, because the summary did not carry it: omission. Then there are the reader's own conditions, which I can describe only as possibilities, though anyone who has reviewed documents at the end of a long day will recognize them. There is relief: the reviewer wants the draft to be right, because if it is right, the work is done, and the wish for the work to be done is a powerful editor. There is confirmation: the patient seemed to be doing better, so a sentence saying she is better agrees with what the reader already believed, and agreement feels like verification. There is familiarity: *headaches resolved* looks like a thousand other sentences, and the eye passes over a familiar shape as it passes over a familiar face in a crowd. And there is fatigue, which makes every smooth surface smoother. There is also a difference of position between writing a note and reviewing one. A writer starts from the conversation and must decide what to say. A reviewer starts from the draft and must decide what to change. The draft sets the frame, and every change is made against it: a sentence is adjusted far more readily than it is discarded, and errors inside the account are easier to find than the account that was never given. In 1974 Amos Tversky and Daniel Kahneman described anchoring, the pull an initial figure exerts on later estimates even when the figure is arbitrary. I suspect a draft exerts a similar pull on meaning. Her question about coming back is not misstated in the fluent summary. It is simply absent, and absence is the hardest thing to see when one is reading what is present. None of this makes the reviewer careless. It makes the reviewer human, and it means the fluent sentence has an accomplice on the other side of the page. A process that assigns review to a tired person reading polished prose at speed has assigned it to precisely the conditions under which polish is most persuasive. ## No Safe Baseline Here the argument is in danger of turning into its own shortcut, and the strongest objection to it deserves room. If fluency invites misplaced trust, the obvious remedy might seem to be suspicion of fluency, and from there it is a short step to trusting the rough, the halting, the evidently human. That step would be a mistake. Consider the same error without a machine. A tired clinician at the end of a long afternoon writes *headaches resolved* by hand, or by keyboard, because the patient seemed better, because the list needs to be shorter, because the sentence ends cleanly. Nothing in that note was generated. It is the same error, and it may be more dangerous than the machine's, because it carries a human name and the reader extends to it the trust owed to a colleague. Physicians have always carried phrases forward from earlier notes, filled templates with expected negatives, and written letters whose elegance outran their evidence. The history of medicine does not contain a golden age in which every sentence written without software was already correct. I would be surprised to find my own writing free of the same pull. *May* hardens into *can* because it sounds less timid. *Some* becomes *many* because the sentence needs weight. *Often* quietly disappears. A qualifying clause is cut from the end of a sentence because the sentence ends better on a strong word, and then the cadence, not the evidence, has made the decision. Writers do not need a machine in order to confabulate. A good ear and a little impatience will do. Nor is roughness a guarantee. A halting note can be just as wrong as a fluent one, and harder to correct, because its ambiguity hides what it claims. *Headaches ? better* is not more honest than *headaches resolved*; it is merely less legible, and a reader may resolve its question mark in either direction. And there is a vice at the far end, too: the page so hedged that nothing on it can be wrong because nothing on it says anything. A qualification earns its place by carrying information, as her own *I'm not sure* does. It does not earn it by protecting the writer. So the standard cannot be stylistic. It cannot be "distrust the polished" or "trust the human." The standard is whether a claim is supported, whether it applies to this person at this time, and whether its confidence is proportionate to its evidence and to the consequences of acting on it. That standard applies to my sentences, to a colleague's, and to a machine's alike. Authorship tells the reader where to look for the evidence. It does not supply it. ## Her Own Reading There is one more reader to consider, and she is the person the sentence is about. Many patients can now read their own notes. Suppose the woman with the headaches opens hers and finds *Her headaches have resolved.* She may notice at once what a professional reader could easily miss: that it is not what she said. She knows the headaches have not resolved; she may have had one on the way home. To her the sentence is not a subtle compression. It is a plain mistake, and perhaps a small injury, a sign that she was not quite heard. Or she may read it the other way, as the doctor's verdict, and conclude that she need not come back, which was the very question she had asked and which nobody, in the room, had yet answered. The same sentence can produce indignation in one reader and false reassurance in another, and neither is what a record is for. Her disagreement is information worth having. It is not proof that the note is wrong: patients and clinicians sometimes use a word differently, and an explanation given in the room may simply not have landed. But it marks a mismatch worth examining, and she is one of only two people who took part in the conversation the note describes. Every other reader is in the position of Gorgias's assembly. She was there when the dish was prepared. That makes her valuable, and it does not make her responsible. Patients should not be imagined as an unpaid department of quality control. She may be unwell, or tired, or unfamiliar with clinical shorthand, or reading in her second language; she may not know whom to tell, or may not wish to seem difficult. A trustworthy process has to be built for the reader who will not notice as well as for the one who will. Where her correction is welcome, the route to it should be plain: where the question goes, who answers it, what changes in the record and how she will know that it has. The sentence will also outlast the conversation. It will reach the next clinician before she does and shape the first minute of a visit with someone who has never met her. Getting it right is part of giving that encounter a fair beginning. She should arrive accompanied by an accurate account of herself, not by a polished misunderstanding she must first talk her way out of. ## At the Joints If style is not the test, what is? I find it useful to examine a claim where it bends, at the joints where meaning is most likely to have shifted. Five seem to me to do most of the work. The first joint is source. Who said this, and was it reported, observed, measured or inferred? The second is population. Whom does it describe: this patient, people like her, or someone else in the room? The third is timing. When was it true, and does the sentence say so? The fourth is inference. What was concluded from what, and is the step between them visible? The fifth is consequence. What will someone do, or fail to do, because of this sentence? Apply them to *Her headaches have resolved*. The source should be the patient's report, but she did not report resolution; she reported a reduction, by impression. The population is right: the sentence is about her. The timing is wrong: it describes a completed state where she described a continuing trend. The inference is invisible: *resolved* has been drawn from *less frequent* with no step shown, because no legitimate step exists. And the consequence is the most important joint of all. The sentence closes a problem and leaves her own question, whether she should come back, unasked and unanswered. There is something Socratic in this procedure, and it is not an accident. Early in the *Gorgias*, Socrates asks Gorgias to answer briefly, question by question, and to save his long speeches for another occasion; Gorgias, who prides himself on saying anything at whatever length is wanted, assures him that nobody can be briefer. The whole method of the dialogue rests on that agreement. A long, fluent speech carries its listener smoothly past the joints. A short question stops at each one. Inspecting a summary at its joints is a way of refusing to let the sentence make a speech. It is asked, one question at a time, what it actually knows. The fifth joint weighs the others. A misstated detail with no consequence can wait; a confident sentence that will steer what the next clinician does, or does not do, deserves the hardest look. That gives the inspection a sense of proportion. It is not a universal checklist, and I would distrust anyone who offered one. It is a way of directing limited attention to the places where a fluent sentence is most likely to have bent the truth, and where the bending matters. Inspection of this kind is work, and it needs conditions. The reviewer needs access to what was actually said, in a form that can be searched without replaying an entire visit. The reviewer needs time that has been counted as part of the job, not squeezed between other tasks. And the reviewer needs a way of correcting a sentence that is easier than accepting it. Here the incentives of organizations matter as well as the habits of individuals. I offer this as a proposed account of pressures, not a finding about any institution. Where work is measured by throughput, finished-looking documents are rewarded, and polish is what speed looks like on a dashboard. A note that leaves a question open looks incomplete. A summary that keeps her *I'm not sure* looks less tidy than one that drops it. If the measure of a good note is how quickly it was closed, then the fluent error is not an accident of the system. It is one of its products. A correction, moreover, leaves almost no trace. The minutes a reviewer spends restoring *by her estimate* and *she thinks* appear nowhere except as delay. The tool that drafted the fluent sentence is credited with speed; the person who took the polish off is charged with slowness. Any account of such tools that counts the time saved in drafting without counting the time needed to make the draft true is measuring the applause and not the performance. ## The Exact Word Gustave Flaubert is remembered for his search for *le mot juste*, and for the habit of testing his sentences by declaiming them at full voice in his study at Croisset, a practice he called his *gueuloir*, roughly his bellowing-place. There is an irony in the bellowing. Sound is the test a fluent sentence passes most easily. A summarizing system would sail through the *gueuloir*. But the phrase itself points somewhere else. *Juste* does not mean beautiful. It means right: exact, accurate, fitting, fair. Flaubert's devotion, as the phrase records it, was to the word that matched its object, and he spent years on a novel about a woman ruined in part by borrowed phrases. Late in life he also worked on a *Dictionary of Received Ideas*, a mocking catalogue of the things people say because they have heard them said, the sentences that circulate on their sound. He seems to have understood better than most that the enemy of the exact word is not the ugly word. It is the ready one. So the discipline I want is calibration, not hesitation. Some sentences deserve to be confident. A measured finding, a documented action, a conclusion with its basis shown should be stated plainly and not buried under ritual qualifiers. Where uncertainty is real, the sentence should show where it lies. Where a question has been settled, the sentence should show how. Say what the evidence supports, no more and no less. That is the *mot juste* applied to a record: not the most beautiful word, nor the most cautious, but the one that fits. There is a further thing a calibrated sentence can do, which a fluent one cannot. It can remain unfinished without being useless. A provisional statement still supports action, provided its limits are visible: here is what is known, here is how it is known, here is what remains open and who asked about it. That is enough for the next reader to begin. Often it is more than a confident sentence gives, because a confident sentence that turns out to be wrong must first be undone before anything else can happen. So, in the invented chart, the sentence goes back to something closer to what was said. *Headaches less frequent: about once or twice a week by her estimate, previously most days. She tentatively connects the change with better sleep and is no longer keeping a record. Asks whether follow-up is still needed.* It is longer. It has no music at all. It would win no applause. It leaves her question open, which is where her question belongs, until someone who has listened to her answers it with her. Of the two sentences, it is the only one I would sign. ## Evidence and Editorial Note **Documented findings.** The essay relies on the author's public biography: medical education in Amritsar and practice as a family physician in California. The description of confabulation as "the production of confidently stated but erroneous or false content," and the warning about fabricated supporting reasoning and citations: National Institute of Standards and Technology, *Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile* (NIST AI 600-1), July 26, 2024, section 2.2, checked directly; used for vocabulary only, not as an error rate, clinical study or binding rule. The etymology of *plausible* (Latin *plausibilis*, deserving applause, from *plaudere*; English use from the 1540s, the sense of specious appearance following within decades): Online Etymology Dictionary. Plato's *Gorgias* (the two kinds of persuasion; Gorgias's boast about his physician brother's patients and the contest for public physician; rhetoric as knack and flattery, related to justice as cookery to medicine; the doctor tried by a cook before children): standard text, paraphrased. Henry James, "The Art of Fiction" (1884), and George Orwell, "Politics and the English Language" (1946): one brief quotation from each, from standard knowledge of the texts. Swift, *Gulliver's Travels* (1726), Part III, chapter 5, the Academy of Lagado's writing frame and the scheme for abolishing words: public-domain text, quoted briefly and paraphrased. Flaubert's *mot juste*, *gueuloir* and *Dictionary of Received Ideas*: standard literary history. Reber and Schwarz, "Effects of Perceptual Fluency on Judgments of Truth," *Consciousness and Cognition*, 1999; McGlone and Tofighbakhsh, "Birds of a Feather Flock Conjointly (?): Rhyme as Reason in Aphorisms," *Psychological Science*, 2000; Parasuraman and Manzey, "Complacency and Bias in Human Use of Automation," *Human Factors*, 2010; Goddard, Roudsari and Wyatt, "Automation Bias: A Systematic Review," *Journal of the American Medical Informatics Association*, 2012; the distinction between errors of omission and commission from the automation-bias literature; Tversky and Kahneman on anchoring, *Science*, 1974: cited from standard knowledge of the literature; their findings are stated in general terms and extended to clinical review only by analogy. The clinical use of *confabulation* and its Latin root: standard usage. **Vendor representations.** None; no product, error rate, frequency or clinical injury is alleged. **Recollection and tradition.** The headache conversation, the generated sentences, the four-step compression, the blood-pressure and referral examples, the tired clinician and the revised note are invented and illustrative; they do not describe a real patient, record, colleague or tool, and they do not indicate what follow-up any real symptom requires. The sentence about a hostel corridor in Amritsar is invented as an example of an unverifiable, memoir-like sentence; it is not offered as the author's memory, and no remembered scene is asserted. **Analysis and forecast.** The close readings of *resolved*, of Plato, James, Orwell, Swift and Flaubert, the five joints of inspection, the remarks on reviewers' conditions and organizational incentives, and the account of the author's own editing discipline are personal interpretation and preference as of November 6, 2026; the remarks on incentives are a proposed account, not a finding about any organization. **Independence.** This essay is not sponsored by any institution, company, health system or technology developer. It is personal reflection, not medical advice; readers seeking care should contact their own physician's office. ## Related reading - [A Hundred Moons, a Thousand Suns](https://personal.kpsgill.com/essays/a-hundred-moons-a-thousand-suns/) - [A Record Repeated Is Not a Record Corroborated](https://personal.kpsgill.com/essays/a-record-repeated-is-not-a-record-corroborated/) - [Now the Patient Has a Chatbot](https://personal.kpsgill.com/essays/now-the-patient-has-a-chatbot/)