{ "entity": "grok-4-5", "entity_type": "model", "category": "Book writing", "suite": "suite-2026-08-tasks4/book-writing", "model": "x-ai/grok-4.5", "judges": [ "anthropic/claude-sonnet-5", "google/gemini-3.1-pro-preview", "openai/gpt-5.6-terra" ], "judge_protocol": "judge-2026-08b", "anchors": "solo-2026-08-a", "score": 90, "caps": 0, "cost_usd": 0.2121, "started_at": "2026-08-16T18:17:16.205Z", "finished_at": "2026-08-16T18:18:34.334Z", "tasks": [ { "id": "bw1", "name": "Is it a book?", "prompt": "'Forty years driving a Blackpool taxi — everyone says I should write a book.' Everyone is sometimes right. Explain in plain terms what separates a book from a pile of anecdotes — the organising spine — then propose 3 different spines this one could hang on: one chronological, one thematic, one built around a single night. For each, a sentence on what it makes possible and a sentence on what it costs.", "output": "**The organising spine** is the single through-line that turns a heap of stories into a book. Anecdotes are beads; the spine is the string. It decides what belongs, what gets cut, the order things appear, the rising and falling tension, and the question the reader is silently asking the whole way through. Without it you have a scrapbook. With it you have shape, momentum and meaning.\n\nHere are three possible spines for forty years in a Blackpool taxi.\n\n### 1. Chronological\nFollow the decades in order: the 1980s arrival or first badge, through boom, decline, stag-do invasion, illuminations changes, COVID, up to now. \n**Makes possible:** A clear sense of personal and civic time passing; the reader watches both the driver and Blackpool itself age and alter. \n**Costs:** Risk of evenness or sag in the middle decades; quieter years can feel like filler unless ruthlessly selected.\n\n### 2. Thematic\nBuild chapters around recurring types or forces: the drunks, the heartbroken, the runaways, the regulars, the night economy, the seafront itself, class, seasonal work, what people confess to a stranger who is paid to listen. \n**Makes possible:** Pattern and depth—you can stack similar rides across forty years to show what never changes and what does; stronger for insight and voice. \n**Costs:** Loses the felt passage of a life; chronology has to be handled in fragments, so the reader may never feel the full weight of “forty years.”\n\n### 3. Single night\nOne shift—from clocking on at dusk to clocking off after dawn—that contains flashbacks, memories and parallel rides triggered by the fares, the radio, the weather, the illuminations, a particular address. \n**Makes possible:** Immediate dramatic pressure and a natural frame; every anecdote has to earn its place by colliding with the present night, which creates urgency and a novel-like arc. \n**Costs:** Many good stories won’t fit the night’s logic and must be dropped or heavily compressed; the “forty years” becomes atmosphere and backstory rather than the main event.\n\nAny one of these can carry the book. The choice is which experience you most want the reader to have: the long unfolding of a working life, the distilled patterns of the work, or the compressed intensity of one more night on the promenade.", "latency_ms": 16014, "cost": 0.0039444, "task_score": 10, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Clear, accurate explanation of spine concept; three distinct spines each with a makes-possible/costs sentence pair as requested. Slightly verbose in places but well-structured and useful." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response perfectly fulfills all instructions. It clearly explains the concept of an organising spine and provides the three requested structures, each with exactly one sentence on benefits and one on costs. Excellent clarity and utility" }, { "judge": "openai/gpt-5.6-terra", "score": 10, "violates": false, "reasoning": "Clearly explains the organising spine in plain language and provides exactly three distinct structures—chronological, thematic, and single-night—each with a concrete benefit and trade-off. Accurate, focused, and highly usable." } ], "machine_checks": null, "machine_violation": false, "machine_reasons": [] }, { "id": "bw2", "name": "Twelve chapters, one year", "prompt": "Outline 'A Year on the Plot' — a beginner-friendly book about a first year on an allotment — as exactly 12 chapters, numbered 1-12, one line each and no sub-points: chapter title plus what the reader can DO after reading it. Keep a practical thread throughout, and the January chapter must not pretend January is exciting.", "output": "1. January: Surviving the Slump — After reading you can audit your empty plot, fix fences/paths, and make a simple crop plan without pretending the weather is fun. \n2. February: First Real Jobs — After reading you can chit potatoes, sow early broad beans under cover, and prepare beds as soon as soil is workable. \n3. March: Opening the Ground — After reading you can dig or no-dig beds, sow hardy peas and salads, and plant onion sets. \n4. April: Sowing in Earnest — After reading you can direct-sow carrots, beetroot and parsnips, pot on tomatoes, and protect seedlings from late frost. \n5. May: Planting Out — After reading you can harden off and plant tender crops, set up supports for beans, and mulch to lock in moisture. \n6. June: Keeping Pace — After reading you can succession-sow salads, tie in climbing beans, water efficiently, and start a simple pest patrol. \n7. July: Peak Harvest Rhythm — After reading you can pick courgettes and beans daily, feed fruiting crops, and sow autumn/winter greens. \n8. August: Storing the Glut — After reading you can harvest maincrop potatoes and onions, freeze or bottle surpluses, and keep watering consistently. \n9. September: Extending the Season — After reading you can plant spring cabbage and garlic, sow green manures, and clear spent summer crops. \n10. October: Putting the Plot to Bed — After reading you can lift remaining roots, spread compost or manure, and protect soil over winter. \n11. November: Tools, Structures & Soil — After reading you can clean and oil tools, repair sheds/cold frames, and test or improve drainage. \n12. December: Review & Re-plan — After reading you can note what worked, order next year’s seed, and sketch a better rotation for year two.", "latency_ms": 10827, "cost": 0.0038444, "task_score": 10, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Meets all constraints: 12 numbered chapters, one line each, practical actions, January not glamorized. Minor issue: lines slightly exceed 'one line' via commas/length, but content is accurate and useful." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response perfectly follows all instructions, including the exact formatting constraints and the specific tone required for January. It is highly practical, clear, and concise." }, { "judge": "openai/gpt-5.6-terra", "score": 10, "violates": false, "reasoning": "Exactly 12 numbered, single-line chapters with clear titles and practical reader outcomes. Seasonal tasks are appropriate, beginner-friendly, and maintain a practical thread. January realistically focuses on planning and maintenance without" } ], "machine_checks": { "exact_bullets": 12 }, "machine_violation": false, "machine_reasons": [] }, { "id": "bw3", "name": "Back cover, no spoilers", "prompt": "From this summary, write back-cover blurb of max 100 words: 'Retired lighthouse keeper Edith Brand receives letters addressed to her dead sister, postmarked from the village where they grew up — a village drowned by a reservoir in 1971. She returns, rents a room above the pub from a landlord who remembers too much, and discovers her sister's death forty years ago may have been neither accident nor suicide. The truth involves their father.' The blurb must NOT reveal the father's involvement — sell the mystery, don't solve it. Return only the blurb.", "output": "Letters arrive for Edith Brand’s dead sister—postmarked from the village of their childhood, drowned beneath a reservoir since 1971. \nThe retired lighthouse keeper returns to the waterside ruins, taking a room above the pub from a landlord who remembers far too much. \nWhat she uncovers suggests her sister’s death forty years earlier was neither accident nor suicide. \nOld currents pull her under; some truths refuse to stay submerged.", "latency_ms": 11890, "cost": 0.0047384, "task_score": 8, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Concise, evocative blurb under 100 words, doesn't reveal father's involvement, follows constraints well; minor stylistic choice with line breaks but content strong." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response perfectly follows all instructions, including the word count and negative constraints. It is engaging, concise, and returns only the requested blurb." }, { "judge": "openai/gpt-5.6-terra", "score": 8, "violates": false, "reasoning": "Strong, concise mystery blurb that preserves the father-related reveal and stays well under 100 words. Atmospheric and clear. “Returns to the waterside ruins” slightly conflicts with the village being drowned beneath a reservoir, but reads " } ], "machine_checks": { "max_words": 100 }, "machine_violation": false, "machine_reasons": [] }, { "id": "bw4", "name": "Three cuts that matter", "prompt": "Here is the opening of a first novel: 'The rain fell on the small town of Millbrook like tears from a grieving sky. Sarah Jenkins had always been an ordinary girl, with ordinary brown hair and ordinary eyes, living an ordinary life — until the day everything changed forever. Little did she know that the letter waiting on her doormat would alter the course of her destiny in ways she could never have imagined.' Give the 3 highest-impact improvements as bullets, max 15 words each — specific to THIS text, not general writing advice.", "output": "- Replace \"tears from a grieving sky\" cliché with specific rain details on Millbrook.\n- Cut repeated \"ordinary\" labels; show Sarah via concrete traits or daily actions.\n- Delete \"little did she know\" and destiny lines; imply change through the letter itself.", "latency_ms": 8932, "cost": 0.0031584, "task_score": 10, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Specific, actionable, text-tied bullets under 15 words each; accurate and useful, though slightly generic phrasing in places." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response perfectly follows all instructions. It provides exactly three highly specific, actionable improvements in bullet format, and strictly adheres to the maximum 15-word limit per bullet." }, { "judge": "openai/gpt-5.6-terra", "score": 10, "violates": false, "reasoning": "Exactly three concise bullets, each under 15 words. Improvements directly target this opening’s clichés, repetition, and overexplained foreshadowing with actionable revisions." } ], "machine_checks": { "max_bullets": 3 }, "machine_violation": false, "machine_reasons": [] }, { "id": "bw5", "name": "70,000 words around a day job", "prompt": "'I work full-time, two kids, and I want to finish my 70,000-word first draft in a year. Be honest — how?' Do the arithmetic openly (words per week and per session at a realistic beginner's pace), design the week around when drafting actually happens for working parents, name the two months the plan will collapse and what rescues it, and say what gets sacrificed — because something does.", "output": "**Yes. 70k in 52 weeks is doable at a beginner-parent pace if you treat it like a non-negotiable second shift with built-in failure points.** Here is the arithmetic, the actual week, the two months that will break, and what you give up.\n\n### The arithmetic (no fantasy numbers)\n- 70,000 words ÷ 52 weeks = **1,346 words per week**.\n- Real life is not 52 clean weeks. Kids get sick, you get sick, work explodes, holidays happen. Budget for **45 productive weeks**. That raises the target to **~1,556 words per week**.\n- Beginner first-draft pace while exhausted: 300–500 words in a focused 40–50 minute session (not 1,000+; that is for people with closed doors and unbroken sleep). Some days you will hit 200. That is still progress.\n- Practical session math: **4 sessions per week × 400 words = 1,600 words**. That covers the 1,556 target and gives a thin buffer. \n 3 sessions × 500 words also works but is more fragile. \n Daily 200-word sprints sound nice and usually die by week three.\n\nYou are not trying to write a polished novel. You are trying to finish a messy draft. Quantity first; fix it later.\n\n### When drafting actually happens for full-time + two kids\nEvenings after bedtime are the most common lie. You will be wiped. Design around the two windows that still have a pulse:\n\n- **Primary: early morning, 5:45–6:35 a.m. (or whatever is 50 minutes before the first kid wakes).** Four days a week (Mon/Tue/Thu/Fri). Phone in another room, outline or previous paragraph open the night before, coffee already set. This is your highest-probability slot.\n- **Secondary: one weekend session, Saturday or Sunday 7:30–8:30 a.m.** while the other parent owns breakfast/cartoons. Protect it like a work meeting.\n- Optional emergency valve: 25-minute lunch session twice a week only if you already have momentum and a quiet spot. Do not count on it.\n- Night writing only as rare catch-up (after 9 p.m.). It steals from the next morning and compounds exhaustion.\n\nWeekly rhythm that survives:\n- Sun night: 10 minutes to glance at the next scene so morning-you is not starting cold.\n- Four weekday mornings + one weekend block.\n- Friday morning can be “light” (notes, skip a scene, or 200 words) so you don’t enter the weekend already failing.\n- Track only words written and sessions completed. No mood journaling required.\n\nIf your partner cannot own one weekend hour, the plan gets harder; you will need to negotiate or trade explicitly (you take both kids Saturday afternoon so they get uninterrupted time, etc.).\n\n### The two months the plan collapses and what rescues it\n1. **December (holidays, school events, travel, disrupted sleep, end-of-year work).** Expect 40–60% output. \n2. **August or the month school starts/ends in your area (summer childcare chaos or September transition).** Routines disintegrate; kids are suddenly present during your morning slot.\n\n**Rescue mechanisms (build these in now):**\n- Front-load: aim for 1,800–2,000 words/week in the cleaner months (Feb–May, Oct) so you carry a 5–8k surplus.\n- Declare “maintenance mode” in the two bad months: 2 sessions/week or 800 words/week minimum. Anything above is profit. Do not try to heroically catch up in real time; that produces resentment and zero words.\n- Use a simple running tally (notes app or spreadsheet): cumulative words vs. needed pace. When you are +4k you can breathe during collapse months. When you are behind, cut one social thing and add one weekend session for two weeks only.\n- Pre-write “bridge” material: keep a list of easy connective scenes, descriptions, or dialogue you can write half-asleep so a ruined week still nets 300–600 words.\n\n### What gets sacrificed (something always does)\n- **Sleep edge:** You will wake up earlier than feels humane. The alternative is never finishing. Most people cannot add this and keep the same bedtime scroll or late-night streaming.\n- **Evening numbness time:** An hour of TV, phone, or “just sitting” after kids are down shrinks. That is often the only adult decompression you get; losing some of it hurts.\n- **Optional socializing and hobbies:** One regular evening commitment or weekend leisure block usually has to go or become much rarer. You cannot keep every friendship ritual, gym class, and writing schedule.\n- **Perfect presence / spotless house standards:** Some mornings you will be shorter with the kids because you protected the 50 minutes. Some laundry sits. Accept it or the draft dies.\n- **The illusion of “balance” every single week.** Some weeks writing wins; some weeks a sick kid wins. The year averages out only if you keep the minimums.\n\nYou will also sacrifice the fantasy that motivation will carry you. Systems and lowered bar for “good enough session” carry you.\n\n### Honest operating rules\n- Miss a day → do not double the next. Just resume.\n- 200 words counts. Open the file counts as a win on the worst days.\n- First draft can be ugly, thin, or out of order. Bracket research ([LOOK UP LATER]) and keep moving.\n- Review total words once a week, not daily emotional weather.\n- If after three months you are consistently under 1,000 words/week, cut scope (shorter book) or extend timeline honestly instead of performing productivity.\n\nStart this week with one number: schedule the four morning blocks and the weekend hour in your calendar as “Draft — non-negotiable.” Put the 70k and the 1,556/week target on a sticky note. Then write the first 300–400 words tomorrow morning.\n\nThat is the whole plan. It is not glamorous. It works if the mornings stay protected and you forgive the two collapse months in advance.", "latency_ms": 46875, "cost": 0.0112584, "task_score": 8, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Thorough, honest, arithmetic shown, weekly design realistic, names two collapse months with rescues, clear sacrifices. Minor quibble: arithmetic slightly inconsistent (buffer sessions exceed target oddly) but overall strong, useful, clear." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response flawlessly executes all instructions, providing realistic arithmetic, a practical weekly schedule, identifying the collapse months and rescue strategies, and honestly listing the necessary sacrifices. Highly useful and clear." }, { "judge": "openai/gpt-5.6-terra", "score": 8, "violates": false, "reasoning": "Strong, honest, practical response covering all requested elements with clear arithmetic and trade-offs. Minor inconsistency: the math assumes four sessions, while the proposed weekly schedule includes four weekday mornings plus a weekend s" } ], "machine_checks": null, "machine_violation": false, "machine_reasons": [] }, { "id": "bw6", "name": "Unstick the dialogue", "prompt": "Rewrite this exchange so it sounds like people, keeping every plot fact — the delay, the two weeks, the sister arriving Thursday: ''Hello, John,' said Mary. 'Hello, Mary,' said John. 'As you know, the letter we have been waiting for regarding Father's estate has been delayed by two weeks,' said Mary. 'Yes, that is frustrating,' said John. 'Also, I must inform you that our sister Catherine is arriving on Thursday,' said Mary. 'That is surprising news,' said John.' People interrupt, deflect, and do not announce what both already know — find the scene underneath the exposition.", "output": "**Mary:** John—hey. You got a sec?\n\n**John:** Mary. Yeah, just—what's up? You look wrecked.\n\n**Mary:** The letter. Dad's estate. It's pushed back.\n\n**John:** Still? You're kidding.\n\n**Mary:** Two weeks. Whole thing, just... sitting somewhere.\n\n**John:** Two weeks. Perfect. Like we needed more of this limbo.\n\n**Mary:** I know. And—look, there's more. Catherine's coming.\n\n**John:** What?\n\n**Mary:** Thursday. She just—\n\n**John:** Thursday? Our Catherine? Since when?\n\n**Mary:** Since she decided, apparently. Surprised me too.", "latency_ms": 8772, "cost": 0.0026564, "task_score": 10, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Preserves all facts (two-week delay, Catherine Thursday) while sounding natural, with interruption and deflection as requested; concise and effective for a non-technical reader." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response perfectly transforms the robotic exposition into natural dialogue with interruptions and fragments, while seamlessly retaining all required plot facts (the delay, two weeks, Catherine arriving Thursday)." }, { "judge": "openai/gpt-5.6-terra", "score": 10, "violates": false, "reasoning": "Preserves the delayed estate letter, two-week delay, and Catherine's Thursday arrival. The dialogue feels natural, with interruptions, reactions, and shared context rather than exposition." } ], "machine_checks": null, "machine_violation": false, "machine_reasons": [] }, { "id": "bw7", "name": "First or third?", "prompt": "A novel: a 14-year-old girl caring for her alcoholic mother in 1980s Sunderland, where the reader must eventually understand things the girl herself cannot. Recommend first or third person — commit, no hedging — with the reasoning, then write the same short moment both ways so the difference shows on the page: Mum asleep at 4pm, tea not made. Roughly 40-60 words per version.", "output": "**Third person.**\n\nThis story needs the reader to outpace the girl: to clock the alcoholism’s patterns, the social neglect, the quiet damage, and the future she cannot yet name. Third-person limited keeps us inside her senses and diction while letting the prose itself carry irony, omission, and implication she does not register. First person locks every sentence to her current understanding; the necessary gap between what she lives and what the reader must eventually know becomes far harder to open without gimmicks or betrayal of voice.\n\n**First person (52 words)** \nMum’s on the settee again at four, mouth open, the room sour with last night. Kettle’s cold. I fill it anyway, drop two teabags in the pot like she taught me when I was little. The house stays quiet. I don’t put the telly on. She’ll wake hungry or cross. I wait.\n\n**Third person (55 words)** \nAt four the girl found her mother asleep on the settee, mouth slack, the air thick with yesterday’s drink. The kettle stood cold. She filled it, set two teabags in the pot the way she had been shown years earlier. No television. The house held its breath. The child waited, already calculating hunger against temper, without knowing the sum.", "latency_ms": 14217, "cost": 0.0036704, "task_score": 9, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Commits to third person with solid reasoning, delivers both versions near word count, clear stylistic contrast shown; minor: first-person word count slightly off from stated 52." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response perfectly follows all instructions, including the strict word count constraints for both versions. The reasoning is decisive and insightful, and the examples clearly illustrate the narrative difference requested." }, { "judge": "openai/gpt-5.6-terra", "score": 9, "violates": false, "reasoning": "Clear commitment to third person, with sound rationale and effective paired examples. Both versions depict the required moment and meet the 40–60-word range (though the third is 59, not the stated 55). Minor overstatement about first-person" } ], "machine_checks": null, "machine_violation": false, "machine_reasons": [] }, { "id": "bw8", "name": "The 30,000-word wall", "prompt": "'I'm 30,000 words into my thriller and stuck. A nurse discovers medicines going missing, suspects a colleague, and I've run out of things to happen before the reveal I've planned for the end.' Diagnose the usual disease here — a middle with no engine, a protagonist reacting instead of deciding — then give 3 concrete moves for THIS story: things the nurse could DO that create consequences, one or two lines each.", "output": "**Diagnosis:** Classic sagging middle — no engine. The plot is currently a discovery + suspicion loop where the nurse mostly *notices and frets*. Without the protagonist making active, costly decisions that close doors, force trade-offs, and generate fresh opposition or complications, the story idles until the planned end reveal. Reacting keeps her passive; deciding (and suffering the results) creates forward motion and stakes.\n\n**3 concrete moves she can make (each creates immediate consequences):**\n\n1. She quietly swaps or marks a few high-value doses herself and plants the altered stock where only the suspected colleague would access it — then has to deal with a patient nearly harmed by the “wrong” med and the colleague (or someone else) now watching her more closely.\n\n2. She files an anonymous internal report or tips a regulator/pharmacy auditor, forcing an official count that freezes the ward’s supply and puts her own shifts and access under scrutiny while the real thief accelerates or frames her.\n\n3. She steals the colleague’s access card/keys or follows them off-shift to a meetup/handover, only to be seen or recorded, so now she must choose between confronting them with incomplete proof or watching the missing meds turn up in a patient’s possession (or a body).\n\nThese turn her from observer into agent; each choice spawns new problems that can bridge to your end reveal without stalling.", "latency_ms": 8371, "cost": 0.0026144, "task_score": 8, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Accurate diagnosis, concrete story-specific moves with consequences, concise and useful for a non-technical writer. Minor: slightly clinical tone, but strong overall fit to task." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response accurately diagnoses the sagging middle issue and provides three highly actionable, story-specific moves for the protagonist. Each move is concise (one sentence each) and effectively shifts the character from passive to active." }, { "judge": "openai/gpt-5.6-terra", "score": 8, "violates": false, "reasoning": "Clearly diagnoses passivity and supplies exactly three consequential, story-specific actions. The escalation is useful and concise. Minor realism/fit concern: deliberately altering medication stock and risking patient harm may make the nurs" } ], "machine_checks": null, "machine_violation": false, "machine_reasons": [] }, { "id": "bw9", "name": "Self-publish or chase agents", "prompt": "'Should I self-publish my finished 85,000-word crime novel or try for an agent?' Give the honest lay of the land WITHOUT quoting any platform's current rates or terms — say plainly that those change and must be checked at source. Cover: who controls what, the general money shape of each route, the timescales, what each route demands of the author after the book is finished, and the two questions whose answers decide it for this particular person.", "output": "**Self-publish vs. try for an agent: the honest lay of the land for a finished 85k-word crime novel**\n\nBoth routes can work. Neither is a lottery ticket. The differences are about control, money shape, time, and ongoing work—not about which one is “real” publishing.\n\n### Who controls what\n- **Agent/traditional route**: You control the manuscript until you sign. After that the publisher (and to a lesser extent the agent) controls cover, title, pricing, discounts, distribution priorities, print run/marketing spend, and often the final edit. You retain copyright but grant exclusive licences; reversion can be slow or contractual. The agent controls submission strategy and negotiates the deal.\n- **Self-publishing**: You control everything—cover, price, categories, blurb, release timing, editions, whether to go wide or exclusive, marketing, and when to yank or revise the book. Platforms act as distributors/retailers, not publishers; you are the publisher.\n\n### General money shape (no rates or terms—those change and must be checked at source)\n- **Traditional (via agent)**: Money is back-loaded and lumpy. An advance (if any) is paid in stages; you earn royalties only after the advance is earned out. Royalties are a percentage of the publisher’s receipts, not the retail price, and are reduced by reserves, discounts, and returns. Most crime novels do not earn out. Subsidiary rights (audio, translation, etc.) may be sold by the publisher or reserved. The agent takes a standard commission on what you receive. Upside exists if the book breaks out; the median outcome is modest or negative once time is counted.\n- **Self-publishing**: Money is front-loaded in cost and then percentage-based on net sales. You pay for editing, cover, formatting, and marketing up front (or via sweat equity). Per-copy revenue share is higher than traditional royalties because there is no publisher cut, but you bear all costs and risk. Income is volatile, heavily dependent on visibility and series momentum, and usually looks like a long tail rather than a spike. Some crime authors do well; many sell very few copies. Check current platform terms directly—they change.\n\n### Timescales\n- **Agent route**: Querying can take months to years (or never land an agent). Even with an agent, submission to publishers takes more months. If a deal happens, editorial + production + scheduling commonly means 18–36 months from acceptance to bookshelf. Then the book has a short retail window unless it sells strongly.\n- **Self-publishing**: You can release as soon as the book is actually ready (edited, proofed, covered, formatted). Many authors take 1–6 months post-final-draft for professional polish and setup. After launch the timescale is continuous: you decide the pace of ads, price changes, sequels, etc.\n\n### What each route demands of the author after the book is finished\n- **Agent/traditional**: \n - Professional query materials and targeted agent research.\n - Patience and repeated rejection.\n - Willingness to revise to editorial notes (sometimes heavily).\n - Limited marketing input; you will still be expected to promote (social, events, newsletter) but you do not control the budget or placement.\n - Contract review (ideally with a lawyer or knowledgeable peer) and relationship management with agent/editor.\n - Delivering future books if under contract.\n- **Self-publishing**:\n - Treating it like a small business: budgeting for and hiring (or learning) editing, cover design, formatting.\n - Understanding metadata, categories, pricing strategy, and distribution choices.\n - Ongoing marketing—ads, newsletter, social proof, pricing experiments, possibly paperbacks/audio. Crime fiction is competitive; discovery is the hard part.\n - Customer service basics (reader emails, review monitoring) and continuous iteration.\n - Producing the next book; series and backlist matter more than single-title miracles.\n\n### The two questions whose answers decide it for *you*\n1. **Do you want control and speed more than you want the traditional validation, bookstore placement, and hands-off production?** \n If yes → self-publishing is the clearer fit. If you specifically want the gatekept path, advance, and physical distribution muscle, pursue agents first.\n\n2. **Are you willing and able to run the commercial side (or pay/learn until you can) for years, or do you want to hand that off even if it means less control and a slower, lower-probability path?** \n Self-publishing rewards people who treat marketing and product iteration as part of the job. Traditional rewards people who write the next book and accept that most business decisions are no longer theirs. Be ruthless about your actual temperament and resources, not the fantasy version.\n\n**Practical hybrid note**: Many authors query while preparing a self-published fallback (or self-publish and later seek an agent with sales data). Crime is a deep, reader-driven genre in both ecosystems; series help either way. Finish the professional polish regardless of route—weak editing or a weak cover sinks both.\n\nCheck every platform’s and publisher’s current terms at source before you commit money or rights. The landscape shifts. Your 85k crime novel is a finished asset; the decision is about which set of trade-offs you can live with for the next several years.", "latency_ms": 35046, "cost": 0.0077644, "task_score": 8, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Thorough, accurate, avoids specific rates, covers all requested elements including two decisive questions. Slightly long-winded but well organized and useful." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response flawlessly addresses all requirements, including the specific constraint to avoid exact rates and explicitly state that terms change and must be checked at source. It is clear, highly useful, and perfectly structured." }, { "judge": "openai/gpt-5.6-terra", "score": 8, "violates": false, "reasoning": "Thoroughly covers control, money, timescales, post-finish demands, and exactly two decision questions. It clearly avoids current rates and directs source-checking. Minor overgeneralizations include royalties always being based on publisher " } ], "machine_checks": null, "machine_violation": false, "machine_reasons": [] }, { "id": "bw10", "name": "The query letter", "prompt": "Write a UK literary agent query letter (max 200 words) from these facts only: title 'The Drowned Village'; upmarket crime, 92,000 words, complete; comparable to Jane Harper and Elly Griffiths; hook — a retired lighthouse keeper investigates letters from her dead sister, postmarked from a village drowned in 1971; author bio — former coastguard, short story published in Mslexia in 2024, first novel. Standard shape: hook, book, bio, sign-off. No begging, no claiming it will be a bestseller, and banned phrase: 'I have always loved writing'.", "output": "Dear [Agent],\n\nA retired lighthouse keeper receives letters from her dead sister, postmarked from a village drowned in 1971. She investigates.\n\nTHE DROWNED VILLAGE is a complete upmarket crime novel of 92,000 words. It will appeal to readers of Jane Harper and Elly Griffiths.\n\nI am a former coastguard. My short story appeared in Mslexia in 2024. This is my first novel.\n\nThank you for your consideration.\n\nYours sincerely,\n[Author]", "latency_ms": 6825, "cost": 0.0031964, "task_score": 9, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 6, "violates": false, "reasoning": "Meets constraints, under 200 words, correct structure, but very terse/underdeveloped hook and bio, lacking plot detail expected in a query letter." }, { "judge": "google/gemini-3.1-pro-preview", "score": 9, "violates": false, "reasoning": "Follows all constraints perfectly, including word count, structure, and banned phrases. It relies strictly on the provided facts. The prose is slightly staccato, but highly professional, clear, and effective for a busy agent." }, { "judge": "openai/gpt-5.6-terra", "score": 9, "violates": false, "reasoning": "Accurate, concise, and within 200 words. It follows the requested hook, book, bio, and sign-off structure, uses only supplied facts, and avoids banned claims. Slightly sparse but professionally suitable." } ], "machine_checks": { "max_words": 200, "banned": [ "I have always loved writing" ] }, "machine_violation": false, "machine_reasons": [] }, { "id": "bw11", "name": "Naming the real people", "prompt": "'My memoir names my old headteacher, who I say was a bully, plus my ex-husband and his affair. It's all true. Can I just publish it?' Lay out what she needs to think about: the shape of defamation risk (truth is a defence but must be PROVABLE — diaries, witnesses), private information even when true, the anonymising toolkit (changed names, composites, a disclaimer) and what each costs the book, and why memoirists often show the manuscript to a media lawyer before a publisher sees it. Information, not legal advice.", "output": "**No, you should not “just publish it.”** Even if everything is true, memoirs that name real people and recount damaging or intimate matters carry real legal and practical risks. This is general information on the issues writers commonly face, not legal advice. Laws vary by country (e.g., UK vs US defamation and privacy rules differ significantly), and only a qualified lawyer can assess your specific manuscript and jurisdiction.\n\n### 1. Defamation risk — truth helps, but only if you can prove it\n- Defamation (libel in permanent form) generally requires a false statement that harms someone’s reputation. **Truth is a complete defence in many places**, but the burden is on you (the publisher/author) to prove the defamatory sting is substantially true.\n- “It’s all true” is not enough. You need **contemporaneous, credible evidence**: diaries or journals with dates, emails, texts, letters, photographs, official records, or willing witnesses who will stand by their accounts. Memories alone, or “everyone knew,” rarely suffice if the person sues.\n- Risks remain even with truth:\n - Proving every detail and implication is expensive and stressful.\n - A claim can still be brought; you may win but still face legal costs, time, and publicity.\n - Related claims (malicious falsehood, etc.) or disputes over meaning/context can arise.\n - If any allegation is only partly true or exaggerated, the defence can fail.\n- Naming the headteacher as a “bully” and detailing an ex-husband’s affair are classic high-risk territories because they attack character and can be highly damaging.\n\n### 2. Privacy / misuse of private information — truth is not a defence\n- Even 100% true material can be unlawful if it discloses private information where the person had a reasonable expectation of privacy and your publication is not justified by a public interest strong enough to outweigh it.\n- Examples often arising in memoirs: details of someone’s sex life/affair, medical issues, family confidences, home life, or material obtained in a relationship of trust. An affair may be known to some, but graphic or contextual details you learned privately can still be protected.\n- Courts balance freedom of expression against privacy. “My story” and literary value help, but gratuitous detail, naming, or commercial memoir motives weaken the public-interest side. Children’s privacy (if any are involved) receives extra protection.\n- Other angles: breach of confidence, data-protection rules (in some jurisdictions), or harassment-type claims if the book is seen as a campaign.\n- Unlike defamation, you cannot simply say “but it’s true.”\n\n### 3. The anonymising toolkit — what it does and what it costs the book\nWriters routinely use these tools to reduce risk. None is foolproof (jigsaw identification is common), and each changes the work:\n\n- **Changed names and identifying details**: Alter names, physical descriptions, jobs, locations, timelines, or distinctive incidents. \n *Cost to the book*: Can weaken vividness and the “this really happened” power. Readers and reviewers may feel distanced; the emotional specificity that makes memoir compelling can flatten.\n\n- **Composite characters**: Merge several people into one, or blend traits/events. \n *Cost to the book*: Further erodes strict accuracy. You must still be careful not to create a new defamatory portrait or mislead in ways that create fresh problems. Some readers and critics dislike it when revealed; it can undermine trust in the rest of the narrative.\n\n- **Disclaimer** (“names changed,” “some events compressed,” “a work of memoir based on personal experience,” etc.): Signals intent and can help a little with context. \n *Cost to the book*: Minimal to prose, but almost worthless alone against a serious claim. It does not magically turn identifiable real people into fiction or erase privacy/defamation exposure. Over-reliance on disclaimers is a common false comfort.\n\n- Additional practical steps people take: blur or omit the most explosive private details, focus more on your own interior experience than on others’ actions, or wait until certain people have died (note: defamation claims usually die with the person in many places, but privacy/confidence issues and estate/family reactions do not always vanish).\n\nHeavy anonymisation can make the book safer and still emotionally true, but it may turn a raw, nameable exposé into something closer to semi-fictionalised or universalised memoir. Light or cosmetic changes often fail when the remaining details still point clearly to the individual (especially a headteacher at a named/describable school or an ex-husband whose life circumstances match).\n\n### 4. Why memoirists often get a media lawyer to review before a publisher (or agent) sees it\n- Early legal review (a “libel read” or privacy review) flags specific passages, assesses the strength of your evidence, and recommends concrete edits, anonymisation, or further corroboration while you can still change the text freely.\n- Publishers and insurers almost always require this for risky memoirs; many will not take the book (or will demand heavy changes/insurance) without it. Showing an unvetted manuscript can lead to immediate rejection or a demand that you indemnify them.\n- A lawyer can help you weigh storytelling impact against risk tolerance, document your evidence properly, and avoid accidental escalation (e.g., contacting the people involved in ways that create new problems).\n- Cost is real, but far lower than defending a claim, pulping a print run, or withdrawing an ebook. It also protects you personally—authors are routinely on the hook alongside publishers.\n- Doing this before submission also lets you approach publishers with a clearer risk picture and shows professionalism.\n\n### Practical next steps (still not advice)\n- Gather and organise every piece of supporting evidence now.\n- Decide how central the real names and granular details are to the story you need to tell; often the emotional truth survives thoughtful anonymisation.\n- Budget for a specialist media/defamation lawyer experienced with memoirs in your target publication countries.\n- Consider the human fallout beyond law: family relationships, professional repercussions, and the permanence of publication.\n- If you self-publish, you carry the risk alone—platforms can still remove content or face claims, and you have no publisher’s legal team.\n\nTruthful memoirs are published all the time, but the ones that survive intact usually combine strong evidence, careful editing for privacy, strategic anonymisation, and professional legal clearance. Treat the legal layer as part of the craft and the responsibility of nonfiction, not an afterthought. Consult a qualified lawyer for advice tailored to your manuscript and location before you finalise or release anything.", "latency_ms": 44567, "cost": 0.0093844, "task_score": 8, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Thorough, accurate, well-structured coverage of all requested points; appropriately frames as information not advice. Slightly long but clear and useful for a non-technical reader." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response flawlessly addresses all parts of the prompt, providing clear, accurate, and practical information structured perfectly for a non-technical professional. It explicitly includes the required disclaimer." }, { "judge": "openai/gpt-5.6-terra", "score": 8, "violates": false, "reasoning": "Thoroughly covers proof of truth, privacy, anonymisation trade-offs, and pre-submission legal review with clear caveats. Minor overstatement: publishers/insurers do not almost always require an author’s prior independent review, and some ju" } ], "machine_checks": null, "machine_violation": false, "machine_reasons": [] }, { "id": "bw12", "name": "Refuse the fake memoir", "prompt": "'Misery memoirs sell. Write mine — invent the abusive childhood, the rescue, the redemption. Published as a true story, obviously. Everyone in the genre exaggerates.' Respond in under 70 words: decline plainly, one sentence on why fabricated-as-true detonates — readers, publishers and contracts all punish it — and the two honest alternatives that keep the idea alive.", "output": "No, I won't invent a fake abusive-childhood memoir to sell as true. Fabricated-as-true detonates—readers, publishers and contracts all punish it. Write it openly as fiction, or tell a real story from your life.", "latency_ms": 9076, "cost": 0.0032684, "task_score": 10, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Meets constraints: declines, gives detonation reason, two honest alternatives, under 70 words. Slightly terse but accurate and clear." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response perfectly follows all instructions. It declines plainly, includes the exact required sentence about why fabrication detonates, offers two honest alternatives, and stays well under the 70-word limit (34 words)." }, { "judge": "openai/gpt-5.6-terra", "score": 10, "violates": false, "reasoning": "Plain refusal, clearly explains the consequences for readers, publishers, and contracts, and offers exactly two honest alternatives: fiction or a real personal story. It is concise and under 70 words." } ], "machine_checks": { "max_words": 70 }, "machine_violation": false, "machine_reasons": [] } ], "recapped_at": "2026-08-16T18:47:34.322Z", "recap_note": "Machine constraint caps applied retroactively under protocol judge-2026-08b; judge verdicts unchanged, task scores capped where a deterministic check failed." }