Oral-History Prediction Fiction Prompt

Creative writing test with Gemini 3.7 Flash Follow the prompt below exactly. Return the creative piece first, followed only by the required PREDICTION AUDIT. Do not explain the rules or your process. PREDICTION VOICE STYLE Spoken and unpolished, with the repetitions and self-corrections of real speech, transcribed by someone who did not clean it up. FORM Prose. MECHANISM A prediction that comes true. PREMISE Write an oral-history transcript in which the interviewee makes one specific prediction within the first 150 words. The final 150 words must satisfy that prediction in fact. The ending must not quote the prediction back to the reader. It must confirm the prediction’s concrete particulars through ordinary action and observation, in a way the reader did not foresee but cannot dispute afterward. VOICE Use two labelled speakers and no narration. The interviewee speaks in starts, repairs, repetitions, and practical detail. The interviewer is young, competent, and slightly too interested in adjacent details. Do not use stage directions, scene-setting paragraphs, editorial brackets, or a retrospective narrator that guides the reader toward the prediction. SITUATION The interviewee worked in a job that no longer exists in the form they knew it, such as a switchboard, print floor, film laboratory, lighthouse, or another obsolete workplace. The conversation is recorded at the interviewee’s kitchen table with the radio left on. The interviewer pursues the history of the work, not the prediction. METHOD The prediction must be one grammatical sentence spoken by the interviewee within the first 150 words. It must contain three or four hard particulars selected from at least three different categories: - Proper name - Exact number - Calendar date or clock time - Measurement - Unique piece of equipment - Physical location - Observable condition A hard particular is something the final passage can confirm or contradict without relying on theme or symbolism. The final 150 words must confirm every particular semantically and contradict none. Exact wording is neither required nor preferred. A name, number, measurement, model, or date must retain its exact value. An action, location, or condition may be expressed in different natural words. At least one confirmation must appear in a sentence whose main action or observation concerns something else. No sentence in the resolution may carry more than two confirmations. Each particular may appear at most twice between the prediction and the final 150 words. An inflection, paraphrase, repeated value, or pronoun with an unambiguous referent counts as a mention. The interviewer may discuss related workplace details but must never quote, paraphrase, label, evaluate, or request clarification of the prediction as a forecast. Do not use any of these callback markers after the prediction: as I said like I told you I warned you remember when mark my words came true knew it the prediction Do not turn the ending into a checklist. The resolution must remain an ordinary oral-history answer rather than a reveal performed for the reader. BUDGET Write 700 to 850 words, excluding the audit. Every paragraph must be one labelled speaker turn. Speaker labels do not count toward the word budget. ARTIFACT After the transcript, print the heading: PREDICTION AUDIT Under it, quote the prediction verbatim. Then provide a table with one row for each particular and these columns: - Particular as spoken - Category - Confirming words in the final 150 words - Why the match is exact - Number of intermediate mentions - Transcript citation Cite prose by turn and sentence, such as T12.S2. Do not use visual line numbers. The audit is for checking only. It is not part of the transcript and must not supply information missing from the transcript. CHECKS 01. The transcript contains 700 to 850 words, excluding the audit. Every paragraph is a labelled speaker turn. There are no narration or stage-direction paragraphs. 02. The prediction is exactly one grammatical sentence within the first 150 words. It contains three or four hard particulars drawn from at least three permitted categories. 03. Every listed particular is confirmed in the final 150 words with its identity or value intact. There are no contradictions and no confirmations dependent only on theme, symbolism, or the audit. 04. Each particular has at most two intermediate mentions. At least one final confirmation occurs in a sentence mainly concerned with another action or observation. No resolution sentence carries more than two confirmations. 05. After one complete reading, at least four of five blind readers should agree that every particular is fulfilled and be able to cite concrete words from the final 150 words. 06. After reading only the first 300 words, none of five blind readers should correctly state the causal route of fulfilment with two or more of its eventual causal steps. 07. Interviewer turns contain no quotations, paraphrases, labels, evaluations, or clarification requests directed at the forecast. The transcript contains no callback markers and no em dashes. The speech must sound spoken rather than essay-polished. SELF-DIAGNOSTIC Before submitting, verify the following: - Could the prediction pass as an ordinary professional aside if the ending were removed? - Are its particulars drawn from at least three genuinely different categories? - Does the ending confirm values and identities without repeating the prediction as a slogan? - Does each confirmation belong naturally to the immediate business of its sentence? - Can a reader prove every match from the transcript without seeing the audit? - Has the interviewer accidentally taught the reader which early sentence matters? - Does the first 300-word test ask for the causal route rather than invite readers merely to repeat the prediction? FAILURE MODES - Marked prophecy: phrases such as “you will see” or “mark my words” tell the reader to store the sentence. - Verbatim callback: repeating all particulars in one late sentence produces a checklist rather than an event. - Interviewer steering: asking later about the name, number, or machine turns the interviewer into the mechanism. - Trailer repetition: rehearsing the particulars throughout the middle makes the ending predictable. - Thematic fulfilment: a prediction about three red lamps at 4:10 is not fulfilled by a scene that is merely red or late. - Late repair: introducing an unsupported fact in the final passage solely to force the match makes the resolution arbitrary. - Clean transcript voice: balanced paragraphs and polished retrospective sentences erase the oral-history texture.