--- name: historical_accuracy_checking description: Check fiction and nonfiction drafts for period accuracy: hunt anachronisms in objects, food, institutions, and language, separate attested from available from common use, rank sources for historical claims, and fix errors with swaps that keep the scene's beat. Use when a draft is set in the past, when a historical claim needs verifying, when reviewing a manuscript, script, game, or exhibit text for period errors, or when a writer asks whether people in a given year would have had, eaten, worn, or said something. --- # Historical Accuracy Checking ## Purpose Find the places where a draft set in the past quietly borrows from the present, prove each one against a ranked source, and fix it in a way that keeps the writing alive. The failure this skill prevents is not the famous howler. It is the steady drip of small wrong details, a food that had not crossed the ocean yet, a word a century early, a routine object that was still a rich novelty, that teaches the reader a past that never existed. The core discipline is the three-date question. For any item ask whether it was attested by the scene's year, available to a person of this character's place and means, and common enough that nobody in the scene would remark on it. Most anachronisms pass the first test and fail the second or third. A fork existed in Venice in 1004. That fact does not put forks on ordinary tables in England in 1520. ## Workflow ### 1. Fix the frame before checking anything - Write down the scene's year or narrow range, place down to region or town, and the characters' wealth, class, and trade. Every check that follows is against that frame, not against the century in general or the continent in general. - Note what the draft is: strict historical fiction, popular fiction with a period setting, nonfiction, a game, or exhibit text. The frame sets how much deliberate anachronism is acceptable. The check itself does not change, the ruling on each finding does. ### 2. Run the first pass with the script - Run the watchlist script against the draft with the scene year. It flags a small set of verified hard dates and rare use cases. ``` python3 scripts/flag_check.py --year 1520 draft.txt python3 scripts/flag_check.py --self-test ``` - Treat the output as leads, not verdicts. The watchlist is small by design, every entry sourced in `references/watchlist-notes.md`, and a clean run proves nothing except that these terms are absent or in date. The manual passes below are the actual check. Worked result: the self-test passes all four cases. On a test passage set in 1520 ("coffee to the coffeehouse, tomato with a fork, 'Okay,' he said") the script returns three flags: coffeehouse as a hard error (first English coffeehouse, Oxford, 1650), okay as a hard error (first attested 1839), and fork as a rare use warning. The tomato draws no hard flag at 1520, because it passes the post-1492 floor, and the next pass is what catches it. ### 3. Work the checklist in category passes - Open `references/anachronism-checklist.md` and work one pass at a time: hard dates, diffusion and common use, material culture in the scene, language, then mindset. - Log every candidate flag as you go, with the exact line. Do not fix while hunting, or the fix-it brain stops seeing the next flag. - Give the diffusion pass real time. Ask of each surviving item who had it, where, at what price, and whether the scene's other characters would stare. A use within a generation of first arrival is a claim, and a claim needs a source. ### 4. Settle each flag against a ranked source - Open `references/source-ranking.md`. Rank the source before trusting the claim: period evidence and museum or archive records first, dictionary and reference works for dates and word senses, peer reviewed work for how common a thing actually was, and enthusiast material only as a lead to a better source. - Apply external criticism (is the source what it claims to be, who made it and when) and internal criticism (what did the maker want, does an independent source agree). Two pages copying one another count as one source. - Settle hard dates with one top rank source or two independent mid rank ones. Settle common use claims only with evidence about spread. One surviving rich example never proves routine use. - Record the ruling for each flag: confirmed error, rare use that must be remarked on in the scene, deliberate choice by the writer, or cleared, with the source and its rank next to the ruling. ### 5. Fix with swaps, not deletions - Replace the wrong item with a period item that does the same job in the scene: the alehouse for the coffeehouse, the knife for the fork, turnips for potatoes, assent in the scene's own words for okay. The beat, the action, and the rhythm survive. Only the borrowed modern object goes. - When a rare item is worth keeping, make the scene pay for it: let a character notice it, price it, or explain where it came from. A remarked on novelty is period colour. An unremarked one is an error. - When the error is structural, a habit or a piece of knowledge the period did not have, rebuild the beat on period means instead of swapping the noun. A character who cannot check the time behaves differently, and that difference is usually better writing. - Read the fixed passage aloud. If the fix flattens the voice, the swap is wrong, not the ruling. Find the period item with the same energy. ### 6. Deliver the check log with the draft - Return the draft with fixes made or marked, plus the log: each flag, its ruling, the source and rank behind the ruling, and any deliberate anachronisms the writer chose to keep, listed as choices. - Study `references/worked-checks.md` before your first full check. It works three passages end to end, a Tudor kitchen in 1520, a London street in 1905, and a Roman camp in the year 100, including two findings that clear, because a check that only ever finds errors is a check nobody trusts. ## Output Contract - The checked draft, with every change traceable to a log entry. - A check log with one line per flag: the line or item, the scene frame it was judged against, the ruling (error, rare, deliberate, cleared), the fix applied, and the source with its rank. - A short list of claims that could not be verified, stated as unverified, never silently kept as facts. - No new historical claims introduced by the fixes without the same ranking check the original flags received. ## Operating Rules - Every factual claim in a ruling must be verified against a source found for this check. Never invent a date, a first use, or an arrival year from memory. If you cannot verify it, the log says unverified and the draft takes the safer side. - Invention, first sale, and common use are different dates. Rule on the one the scene actually needs, and say which one you used. - Existence is not availability, and availability is not routine. Keep the three apart in every ruling. - Judge against the scene's place and class, not the widest place the item ever reached. Spread is uneven and the gap is the story. - A word's date is the date of the sense used in the scene. Check the sense in the dictionary, not just the headword. - Deliberate anachronism is the writer's right, in fiction especially. Your job is to make it a choice made with the evidence on the table, and to log it as a choice, never to veto it silently or to let it pass unnoticed. - Nonfiction gets no deliberate pass. In nonfiction every confirmed error is fixed or the claim is cut. - Do not modernise on the sly while fixing. A replacement item needs the same check as the item it replaces.