{ "entity": "deepseek-v4-pro", "entity_type": "model", "category": "Book writing", "suite": "suite-2026-08-tasks4/book-writing", "model": "deepseek/deepseek-v4-pro", "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": 86, "caps": 0, "cost_usd": 0.1945, "started_at": "2026-08-16T18:03:02.672Z", "finished_at": "2026-08-16T18:08:13.759Z", "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": "They’re right if he can find a spine. A pile of anecdotes is like a shoebox of old tickets: each one is interesting, but they don’t add up to a journey. A book has an organising spine — a central question, change or shape that tells the reader why these stories are being told, in this order, and what they’re building towards. It also helps you cut good stories that don’t fit.\n\nThree possible spines:\n\n**1. Chronological — “Forty Years on the Rank: Blackpool, 1984–2024”** \nThe book follows the decades of his career, from first night to last shift, with Blackpool changing around him. \n**What it makes possible:** a double portrait of a man and a seaside town ageing together, so the reader feels time passing. \n**What it costs:** the calendar becomes the boss — dull stretches may stay in just because they happened.\n\n**2. Thematic — “What the Back Seat Tells You”** \nChapters are organised by kinds of passenger or human need: the hen do, the runaway, the regular, the grieving, the night worker, the lost visitor. \n**What it makes possible:** the strongest insight, because stories about grief or loneliness or joy can be gathered together and made to say something true about people. \n**What it costs:** it loses the momentum of one life moving forward and may read more like an essay collection than a memoir.\n\n**3. Single night — “Last Shift: One Saturday Night in the Illuminations”** \nThe frame is one final Saturday night shift, 6pm to 6am, during the Blackpool Illuminations; each fare triggers a memory from the previous forty years. \n**What it makes possible:** built-in urgency and unity — a ticking clock, a fixed route, and every flashback earned by something happening in the cab. \n**What it costs:** many good stories won’t fit the route or the mood, and the frame can become a gimmick if it has to carry too much.", "latency_ms": 71610, "cost": 0.002343084, "task_score": 10, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Clear, accurate, well-structured; meets all requested elements concisely. Minor stylistic embellishment but no factual errors or missed instructions." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response perfectly fulfills all instructions. It clearly explains the concept of a narrative spine and provides three distinct, well-thought-out structural options, each with exactly one sentence on benefits and one on drawbacks." }, { "judge": "openai/gpt-5.6-terra", "score": 10, "violates": false, "reasoning": "Accurately explains the organising spine in plain language, offers exactly three clearly distinct structures matching the requested types, and gives a useful possibility and cost for each. Clear, concise, vivid, and well suited to a non-tec" } ], "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: Mud and Paperwork — after reading, you can map your plot, test soil, and order seeds without pretending it’s exciting. \n2. February: Clearing Ground — after reading, you can clear weeds, build compost bins, and mark out permanent beds. \n3. March: First Seeds — after reading, you can start hardy seeds indoors and prepare a seedbed outside. \n4. April: Early Planting — after reading, you can sow carrots and parsnips, plant potatoes, and protect seedlings from frost. \n5. May: Filling the Beds — after reading, you can plant out tender crops, sow beans and squash, and set up supports. \n6. June: Keeping Up — after reading, you can water consistently, thin seedlings, and hoe weeds before they seed. \n7. July: First Harvests — after reading, you can harvest early crops, succession sow, and net brassicas against pests. \n8. August: Glut Season — after reading, you can harvest, preserve surplus, and sow autumn salads and spinach. \n9. September: Storing and Clearing — after reading, you can lift potatoes, store onions, and sow green manures on empty beds. \n10. October: Autumn Jobs — after reading, you can plant garlic and broad beans, divide herbs, and mulch beds. \n11. November: Wind-Down — after reading, you can protect overwintering crops, clean tools, and empty compost bins. \n12. December: Rest and Review — after reading, you can update your plan, order seeds, and rest before the cycle starts again.", "latency_ms": 38860, "cost": 0.001760532, "task_score": 9, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 9, "violates": false, "reasoning": "Exactly 12 chapters, single line each, practical thread, January explicitly non-exciting. Clear, concise, useful for beginners." }, { "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, making it excellent for a beginner-friendly book outline." }, { "judge": "openai/gpt-5.6-terra", "score": 9, "violates": false, "reasoning": "Exactly 12 numbered, single-line chapters with practical outcomes. January appropriately focuses on planning and paperwork rather than excitement. Seasonal progression is clear and beginner-friendly; only minor horticultural timing variatio" } ], "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": "Retired lighthouse keeper Edith Brand receives letters addressed to her dead sister, postmarked from a village that vanished under a reservoir in 1971. Returning to the valley, she rents a room above a pub from a landlord who remembers too much. The official story was that her sister died in an accident. Some whispered suicide. But as Edith digs into the past, neither explanation holds. The drowned village has kept its secrets for forty years—and someone wants them to stay buried. Edith must uncover what really happened before the truth pulls her under too.", "latency_ms": 30111, "cost": 0.001556604, "task_score": 10, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Compelling, concise, avoids revealing father's involvement, stays within length; word count slightly under 100, well-crafted blurb with minor cliché phrasing." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response perfectly follows all instructions, including the negative constraint and word count limit. It delivers a compelling, suspenseful blurb without revealing the father's involvement." }, { "judge": "openai/gpt-5.6-terra", "score": 10, "violates": false, "reasoning": "Compelling, accurate mystery blurb under 100 words. It preserves the central premise, heightens suspense, avoids revealing the father's involvement, and ends with a strong commercial hook. Clear, polished, and appropriately concise." } ], "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": "- Slash the stale “tears from a grieving sky”; let rain drum on Millbrook’s specific surfaces.\n- Cut “ordinary” after the first use; reveal Sarah through one distinct, contradictory trait.\n- Drop “everything changed forever” and “Little did she know”; instead, show the letter’s jarring arrival.", "latency_ms": 14999, "cost": 0.0010083216, "task_score": 10, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Specific, actionable, text-focused suggestions within word limits; concise and useful, though could be marginally sharper." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response provides excellent, highly specific editorial advice targeting the exact clichés in the text. It strictly follows the constraints, providing exactly three bullets with 15 or fewer words each." }, { "judge": "openai/gpt-5.6-terra", "score": 10, "violates": false, "reasoning": "Three specific, high-impact revisions directly target clichés, repetition, and telling. Each bullet is actionable, concise, and exactly 15 words or fewer." } ], "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": "Let’s do the arithmetic first.\n\n**The math**\n\n- 70,000 words ÷ 52 weeks = **1,346 words/week** raw.\n- But you won’t write every week. Assume two real-life lost weeks: illness, holidays, school chaos. \n 70,000 ÷ 50 usable weeks = **1,400 words/week**.\n- Add a small buffer: target **1,500 words/week**.\n\nNow pace. A realistic beginner drafting pace is **250–400 usable words per 30-minute sprint**. Plan at **300 words per 30 minutes**. That’s not fast, but it’s sustainable.\n\n1,500 words/week ÷ 300 words/session = **5 sessions/week**.\n\nSo the whole project is roughly **2.5 hours per week**. That’s the job.\n\nIf you can only manage 4 sessions/week, you’ll write about 1,200 words/week and finish in about 58 weeks. Still fine, but not one year.\n\n---\n\n**The working-parent week looks like this**\n\n| Slot | Duration | Expected words |\n|---|---:|---:|\n| Tuesday lunch break | 30 min sprint | 300 |\n| Wednesday early morning before kids wake | 30 min sprint | 300 |\n| Thursday lunch break | 30 min sprint | 300 |\n| Saturday or Sunday during nap/quiet time, or partner covers kids | 60–90 min | 600–800 |\n| **Total** | **2.5–3 hrs** | **1,500–1,700** |\n\nWhy these slots:\n\n- Lunch breaks are kid-free and reliable.\n- One early morning gives momentum but doesn’t destroy your sleep.\n- The weekend block is the engine. If you have a partner, that block is a standing appointment, not “helping if time allows.”\n- After bedtime is unreliable for working parents. You’re drained, and there’s still cleanup. Use it as overflow, not the plan.\n\nIf you’re solo-parenting, the weekend block still needs to happen: grandparent, babysitter, or a parent-swap. Otherwise the deadline slips.\n\n---\n\n**The two months the plan will collapse**\n\nAssume **December** and **May** will collapse.\n\n**December:** holidays, school events, illness. \nRescue: don’t fight it. Drop to one maintenance session/week, around 300–500 words. Build a 2,000–3,000-word buffer in October and November.\n\n**May:** end-of-school-year chaos, recitals, paperwork, summer planning. \nRescue: drop to 800 words/week, one weekend session, and keep the story alive with phone notes or voice memos. Some weeks you’ll write 200 words. That’s not failure.\n\nThe rescue is not motivation. It’s a planned lower target plus a buffer.\n\n---\n\n**What gets sacrificed**\n\nSomething does.\n\nYou will sacrifice:\n\n- Most of your leisure scrolling, streaming, or gaming time.\n- Some social life.\n- House standards and ambitious cooking.\n- Possibly a hobby or one hour of sleep if you choose early mornings.\n- The idea of writing “when inspired.” You write when scheduled.\n\nWhat you should not sacrifice: sleep, your job, your marriage/partnership, or your kids’ core needs. If it’s writing vs. sleep, skip writing and use the buffer.\n\n---\n\n**Bottom line**\n\n70,000 words in a year while working full-time with two kids is possible, but it works like a second job of about **2.5 hours per week**, not a creative retreat.\n\nYou’ll write rough, miss some weeks, and survive December and May by lowering the target instead of quitting. The goal is not 70,000 good words. It’s 70,000 finished words you can fix later.", "latency_ms": 139134, "cost": 0.004893779, "task_score": 8, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Solid arithmetic, realistic weekly design, names two collapse months with rescues, clear sacrifices. Minor inconsistency: 5x300=1500 correct, but session count math slightly loose elsewhere. Clear, useful, concise." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response perfectly executes every instruction. The math is clear, the schedule is highly realistic for working parents, the two collapse months (December/May) and their rescues are identified, and the sacrifices are explicitly listed. E" }, { "judge": "openai/gpt-5.6-terra", "score": 5, "violates": false, "reasoning": "It addresses every requested topic, but the weekly schedule contains only 1,200–1,400 words, not the claimed 1,500–1,700, and omits a fifth session. Planned December/May reductions also exceed the stated buffer, undermining the one-year ari" } ], "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": "“John, hey. It’s me.”\n“Mary, hey. Everything okay?”\n“I just got off the phone with Michaelson’s office. Two more weeks.”\n“Two more—you’re joking.”\n“I wish. She said the estate stuff won’t even be looked at until—”\n“Yeah, I know what ‘delayed’ means. Great.”\n“That’s not the only thing.”\n“Of course not.”\n“Catherine’s coming Thursday.”\n“…Thursday.”\n“I know.”\n“You could’ve led with that.”", "latency_ms": 11557, "cost": 0.00031581, "task_score": 8, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Natural, interrupting dialogue preserving delay, two weeks, Thursday sister arrival; loses explicit 'sister' word but context implies it. Concise, effective, minor loss of clarity on relationship." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The rewrite perfectly captures a natural, subtext-heavy conversation while seamlessly integrating all required plot facts (the two-week delay, Catherine arriving Thursday). It follows all instructions flawlessly." }, { "judge": "openai/gpt-5.6-terra", "score": 5, "violates": true, "reasoning": "Natural, concise dialogue preserves the two-week estate delay and Thursday arrival. However, it changes the delayed letter into general estate processing and never establishes that Catherine is their sister, so not every plot fact is retain" } ], "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": "**Recommendation: third person, specifically third-person limited/free indirect.**\n\nThird person keeps the reader inside the girl’s viewpoint while allowing the narrator to record what she cannot admit: the 4pm sleep, the cold kettle, the shut-down room. First person would trap the prose too fully in her self-justifying voice. Third person limited creates the necessary gap between what she sees and what the reader begins to understand.\n\n**First-person version (contrast):** \nMam was asleep on the settee again, mouth open, one hand hanging off the cushion. Four o’clock and no tea on. I pulled the green blanket over her legs. She was just tired, that’s all. I’d make beans on toast. The kettle was cold.\n\n**Third-person limited/free indirect version:** \nAt four o’clock the tea wasn’t made. Mam lay asleep on the settee, mouth open, one hand curled around nothing. The gas fire ticked. Janey pulled the blanket over her mam’s legs and told herself Mam was just tired. The kettle sat cold on the worktop, waiting.", "latency_ms": 115202, "cost": 0.004851178, "task_score": 10, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Clear commitment with sound reasoning, both versions delivered near word count, effectively shows the perspective gap. Minor: first version slightly under 60-word range check but close; strong execution overall." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response perfectly follows all instructions, committing to a perspective with clear reasoning. Both examples accurately depict the scenario and fall within the 40-60 word count constraint. The writing is evocative and effectively demons" }, { "judge": "openai/gpt-5.6-terra", "score": 10, "violates": false, "reasoning": "Commits clearly to third-person limited/free indirect, explains the reader-character knowledge gap, and provides two effective contrasting versions. Both versions meet the roughly 40–60-word requirement and include the required 4pm, sleepin" } ], "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: the middle is built on the nurse noticing and suspecting, not deciding. She’s gathering clues while the antagonist stays comfortable, so the story has observation but no engine. A thriller middle needs actions that force reactions and make things worse.\n\nThree moves for this nurse:\n\n1. She deliberately miscounts a controlled drug on the suspect’s shift and reports it—forcing the suspect either to “fix” the log and reveal tampering, or let an official audit begin.\n\n2. She tells the suspect a false specific: “Pharmacy says your badge accessed the cabinet at 2 a.m.” That lie forces the colleague to react—explain, destroy records, discredit her, or move the stash.\n\n3. She hides a dose herself, then calls a unit-wide search—making herself a suspect to draw the real thief out, while risking her own job and license to force the plot forward.", "latency_ms": 47334, "cost": 0.001474766, "task_score": 8, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Accurate, on-task diagnosis and three concrete, consequence-driven actions specific to the story; concise and useful, though move 3's ethics/logic slightly shaky." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "Perfectly diagnoses the structural issue and provides three highly actionable, specific plot moves that fit the requested length constraint. Excellent advice for a writer." }, { "judge": "openai/gpt-5.6-terra", "score": 8, "violates": false, "reasoning": "Directly diagnoses a reactive, clue-gathering middle and supplies exactly three specific escalating actions with credible consequences. The moves fit a thriller, though deliberately falsifying counts or hiding medication may strain 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": "Here’s the honest lay of the land, stripped of current platform rates and terms (those shift constantly—always check the source directly before making any financial decision).\n\n---\n\n### Who controls what\n\n**Traditional publishing (agent → publisher)**\n- The publisher controls: cover design, interior layout, title (often), pricing, publication date, distribution, and the final edit. You have input, but not final say.\n- You grant them exclusive world rights (or specific territories) for the term of copyright, though you can negotiate reversion clauses. They control subsidiary rights (film, audio, translation) unless you carve them out.\n- The agent controls which editors see the manuscript and negotiates the deal. You control whether to accept the offer.\n\n**Self-publishing**\n- You control everything: cover, editing, title, blurb, pricing, release date, distribution channels, marketing spend, rights sales. You are the publisher.\n- You keep all rights. You can pull the book from sale, change the cover, update the manuscript, or license it elsewhere at any time.\n\n---\n\n### The general money shape\n\n**Traditional**\n- You pay nothing upfront. The publisher bears all costs (editing, cover, printing, distribution).\n- You receive an advance against royalties. An advance is a lump sum (often split into instalments: on signing, on delivery of final manuscript, on publication). You don’t pay it back even if the book never earns out, but you won’t see additional royalty payments until your share of sales surpasses that advance.\n- Royalties are a percentage of the publisher’s net receipts (hardcover, paperback, ebook, audio). The percentages vary by format and are negotiated. You will not see a penny beyond the advance unless the book earns out, which most books don’t.\n- Money arrives slowly: advance instalments can be spaced over a year or more; royalty statements come twice a year and can be delayed by reserves against returns.\n\n**Self-publishing**\n- You pay all upfront costs: professional editing, cover design, proofreading, formatting, possibly ARCs, and any advertising. You can spend a few hundred to several thousand dollars. There is no gatekeeper to stop you from publishing without these, but the market punishes unprofessional presentation severely.\n- You keep the lion’s share of each sale (the exact cut depends on the store, the price, and the territory—check current terms). You receive payments monthly (or as per the platform’s schedule), typically 60–90 days after the sale month.\n- There is no advance. Income is entirely performance-based. You can lose money if your costs exceed your sales. You can also make money indefinitely on a backlist that keeps selling.\n\n---\n\n### Timescales\n\n**Traditional**\n- Finding an agent: querying can take months to over a year. Many authors query 50–100 agents and still get no offer.\n- Submission to publishers: an agent may take weeks or months to send the manuscript out. Editors can take months to respond. An auction or pre-empt can happen fast, but a typical sale takes weeks to many months.\n- After the deal: the book is usually published 12–24 months later. The advance is paid in instalments across that period. So from finished manuscript to a book in a reader’s hands, expect 2–3 years minimum, often longer.\n\n**Self-publishing**\n- You can publish within weeks of finishing the book, provided you have the cover, editing, and formatting ready. A realistic, quality-focused timeline is 2–4 months to commission edits, get a cover, and set up distribution.\n- You start earning as soon as the book is live and selling, with the first payout arriving according to the platform’s cycle.\n\n---\n\n### What each route demands of you after the book is finished\n\n**Traditional**\n- You must write a query letter and a synopsis, research agents, and endure rejection. Once an agent signs you, you’ll likely do at least one more round of revisions based on their feedback before submission. After a publisher acquires the book, you’ll do structural edits, line edits, copyedits, and proof passes. You’ll be expected to engage in some promotion (social media, events, possibly a newsletter), but the publisher handles distribution, sales to bookstores, and the bulk of the marketing machinery—though the level of marketing support varies wildly and is often modest unless you’re a lead title.\n- You must be patient, professional, and willing to collaborate. You will have deadlines. You will have little control over the final product’s packaging and timing.\n\n**Self-publishing**\n- You must become a small business owner. After the manuscript is done, you must source and manage: a structural editor, a copyeditor, a proofreader, a cover designer, a formatter (or learn these skills yourself). You must write compelling back-cover copy. You must decide on keywords, categories, pricing, and launch strategy.\n- You must handle or outsource marketing: building a mailing list, running ads (if you choose), securing reviews, engaging on social media, possibly organising a launch team. You are responsible for every sale. The work doesn’t end at publication; it shifts into ongoing promotion and writing the next book, because in self-publishing, a deep backlist is your strongest asset.\n- You must manage your finances, track expenses, and treat this as a business with profit-and-loss accountability.\n\n---\n\n### The two questions that decide it for you\n\n**1. Do you want to run a publishing business, or do you want a traditional publishing career with institutional backing?** \nIf you relish the idea of being CEO of your own book—making every creative and financial decision, moving fast, and accepting the risk and reward of entrepreneurship—self-publishing fits. If you’d rather focus solely on writing, have a team handle production and distribution, and are willing to trade control and time for that support (and the validation of a professional gatekeeper), pursue an agent.\n\n**2. What is your primary, non-negotiable goal for this specific book?** \nIf it’s to see your crime novel on bookstore shelves, be eligible for certain major awards, get reviewed in traditional outlets, and build a literary reputation that opens doors to foreign deals and film options through an established rights team, you need an agent. If your goal is to get this story into readers’ hands as quickly as possible, build a direct relationship with an audience, and earn money per sale from day one without waiting years, self-publishing is the direct path.\n\nAnswer those two questions honestly, and the route that’s right for you—right now—will be clear. And remember: you can always query agents first, and if you don’t find the right champion after a set period, self-publish later. The only irreversible cost is time.", "latency_ms": 31315, "cost": 0.001429555, "task_score": 8, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Comprehensive, accurate, avoids specific rates/terms, covers all requested elements including two decisive questions. Slightly long/dense for a busy non-technical reader, but clear and well-structured." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response perfectly addresses all prompts, includes the required disclaimer about rates, avoids quoting specific terms, and provides a clear, accurate, and highly useful breakdown." }, { "judge": "openai/gpt-5.6-terra", "score": 5, "violates": true, "reasoning": "It covers all requested areas clearly, but quotes a platform-related payment timing (“typically 60–90 days”), contrary to the no current rates or terms constraint. It also overstates standard traditional rights/control arrangements and mark" } ], "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 Name],\n\nWhen a retired lighthouse keeper receives a letter in her dead sister’s handwriting, postmarked from a village that was deliberately drowned in 1971, she knows something impossible has surfaced. Her sister died years ago, and the village of Thornwick hasn’t existed since the valley was flooded to create a reservoir. Yet the letters keep coming, each one pulling her back to the tight-knit community she left behind, and to the secrets the water was meant to bury.\n\nTHE DROWNED VILLAGE is an upmarket crime novel, complete at 92,000 words. It combines the atmospheric tension of Jane Harper’s *The Dry* with the uncanny, landscape-driven mystery of Elly Griffiths’s Dr Ruth Galloway series.\n\nI’m a former coastguard, accustomed to the moods of the sea and the stories it keeps. My short fiction appeared in *Mslexia* in 2024. This is my first novel.\n\nThe full manuscript is available on request. Thank you for your time and consideration.", "latency_ms": 6377, "cost": 0.0004956432, "task_score": 5, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Well-structured, hits hook/book/bio/sign-off, under 200 words, avoids banned phrase, no bestseller claims or begging. Minor stylistic flourishes but accurate to facts." }, { "judge": "google/gemini-3.1-pro-preview", "score": 5, "violates": true, "reasoning": "The response adds invented facts ('Thornwick', 'reservoir') despite the instruction to use 'these facts only'. Otherwise, it perfectly follows the structure, word count, and tone constraints." }, { "judge": "openai/gpt-5.6-terra", "score": 5, "violates": true, "reasoning": "Polished and under 200 words, with no banned phrase. However, it invents substantial plot details beyond the supplied facts (village name, reservoir, recurring letters, community/secrets) and lacks a proper sign-off/name despite the require" } ], "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": "This is general information, not legal advice. The short answer is: you **can** publish, but “it’s all true” is not a forcefield. The real questions are whether you can **prove** it, whether the people involved have a **reasonable expectation of privacy**, and whether the book has been designed to reduce risk without losing its point.\n\n---\n\n## 1. Defamation risk: truth is a defence, but it must be provable\n\n### The shape of the problem\n\nA statement is potentially defamatory if it would tend to lower a person in the eyes of ordinary reasonable people, or damage them in their profession or trade.\n\nCalling your old headteacher a “bully” is not just personal venting if you publish it. It is likely to be taken as a factual allegation about his professional conduct. If he can show that:\n\n- the statement was published,\n- it referred to him,\n- and it had a defamatory meaning,\n\nthen in many common-law jurisdictions, including England and Wales, the burden can fall on **you** to prove that the defamatory meaning is substantially true.\n\nThat is very different from “I know it happened” or “my family knows it happened.”\n\n### What “provable” means\n\nYou need evidence that can be put before a court or at least would make a publisher’s lawyer feel confident. Useful evidence might include:\n\n- contemporary diaries or journals written at the time,\n- letters or emails to friends or family complaining about the behaviour,\n- complaints made to the school or governors at the time,\n- school records, personnel files or inspection reports,\n- therapy or medical notes that refer to the bullying,\n- witnesses who saw or experienced the same behaviour,\n- any subsequent apology, disciplinary action or finding against the headteacher.\n\nA diary is a start, but it is still your own document. It shows consistency and helps with credibility, but it may not be enough on its own. Independent contemporaneous documents or independent witnesses are much stronger.\n\n### “Bully” can be treated as a factual claim\n\nYou might think “I said he was a bully” is just an opinion. But if the book gives specific examples or implies he was unfit or abusive as a teacher, then labelling him a bully may be treated as a factual allegation.\n\nA safer framing is often something like:\n\n> “I experienced him as a bully. On 14 January 1994 he stood me in front of the class and shouted at me until I cried. My friend X saw it. I wrote about it that night in my diary.”\n\nThat gives the reader the evidence and lets them draw the conclusion, rather than you making a bare label the claim.\n\nBut no formula removes risk entirely.\n\n### The ex-husband and the affair\n\nSaying your ex-husband had an affair may feel less serious than calling someone a bully, but it can still raise defamation issues, especially if it implies deceit, dishonesty or hypocrisy. In some communities or professions it may still cause reputational harm.\n\nThe other person in the affair has their own legal rights too. They may claim defamation, privacy invasion, or both.\n\nEven if the affair is true, you would still need to prove it if it becomes a defamation claim. Evidence might include:\n\n- admissions in messages or emails,\n- divorce papers,\n- witness evidence,\n- dates, places and records that corroborate the relationship.\n\nAlso be careful: court documents are not automatically free to publish. Some family law material may be confidential or subject to reporting restrictions.\n\n### A practical point about risk\n\nPeople sue over memoirs not always because the legal claim is strong, but because they are angry, identifiable and feel publicly shamed. Even if you win on a truth defence, the process can be expensive and exhausting.\n\n---\n\n## 2. Private information: true facts can still be private\n\nThis is the other major risk, and it is often more dangerous than defamation in memoir.\n\nA person can have a reasonable expectation of privacy in many things, even if those things are true. The classic example is sexual or relationship life.\n\nYour ex-husband’s affair is, on its face, private information. The fact that it happened inside your marriage does not automatically give you a free right to publish it. He may well have a reasonable expectation that his intimate life will not be exposed in a book.\n\nThe other person in the affair has an even stronger privacy argument, because they did not choose to be in your memoir.\n\n### Privacy and public interest\n\nPrivacy law often balances:\n\n- the subject’s right to private life,\n- against the publisher’s right to freedom of expression.\n\nBut “the public is interested” is not the same as “public interest”.\n\nPublic interest usually means something like:\n\n- exposing serious wrongdoing,\n- safeguarding people,\n- revealing abuse of power,\n- correcting a significant false public narrative,\n- matters that affect a community or public institution.\n\nA personal memoir about an old headteacher who treated you badly and an ex-husband who had an affair may be powerful and important to you, but it is not automatically a matter of public interest.\n\nIf the headteacher is still working with children and there is a safeguarding concern, that may strengthen a public-interest argument. If he is long retired and the story is mainly personal grievance, the public-interest argument is much weaker.\n\n### Even if you anonymise, privacy may still be breached\n\nYou can change names and still identify someone if enough other details remain. For example:\n\n> “My headteacher at [school name] in the 1980s, who taught English and drove a yellow car…”\n\nEven without a name, people who know him will know exactly who you mean.\n\nThe legal test is not usually “did she use the real name?” but “would a reasonable person who knows the person understand it to refer to them?”\n\n---\n\n## 3. The anonymising toolkit and what each costs the book\n\n### Changed names\n\nChanging names is the baseline step, but it is not a magic shield.\n\nIt reduces the chance of being automatically indexed or publicly searchable by name, but it does not prevent identification through context.\n\nCost to the book:\n\n- It can reduce the sense of direct accountability.\n- If the memoir is partly about naming a bully or exposing someone, a pseudonym can feel less powerful.\n- But many readers accept “names changed” as a normal memoir convention.\n- If you change the name but keep the relationship, school, dates and physical description, you may get the worst of both worlds: less impact, but still legally identifiable.\n\n### Composite characters\n\nA composite is where you merge two or more people into one character, or move events around to make someone less identifiable.\n\nCost to the book:\n\n- This moves the work away from strict memoir and towards “creative nonfiction” or “based on a true story”.\n- You may need an author’s note saying something like:\n\n> “Some characters are composites. Events have been compressed or reordered for narrative purposes.”\n\n- That can soften the “it’s all true” claim.\n- If the book is marketed as a raw, factual memoir, using composites without disclosure can create a trust problem with readers if it is discovered.\n\nBut composites can be very useful if the point is not “I want to expose this specific named person” but “this is what happened in my life, and the emotional truth matters more than the exact cast list.”\n\n### A disclaimer\n\nA standard disclaimer might say:\n\n> “Some names and identifying details have been changed to protect the privacy of individuals. This is a work of memory.”\n\nThat is normal and expected in many memoirs.\n\nBut a disclaimer is not a legal shield. You cannot write a detailed true story about your named ex-husband, then add:\n\n> “Any resemblance to real persons is coincidental.”\n\nThat would be absurd and may even make things look worse if there is a dispute.\n\nCost to the book:\n\n- A modest disclaimer costs very little.\n- An overbroad disclaimer can undermine the book’s credibility.\n- If the book is saying “this is all true” but the disclaimer says “some characters are composites”, you need to be clear with the publisher and reader about which is which.\n\n### The trade-off\n\nThere is a sliding scale.\n\n- The more real names, real places and exact dates you keep, the stronger the memoir feels as documentary truth, but the higher the legal risk.\n- The more you anonymise, composite and disclaim, the lower the legal risk, but the more you may weaken the book’s authority, specificity and sense of justice.\n- The worst position is usually halfway: changing a first name but leaving the person easily recognisable.\n\n---\n\n## 4. Why memoirists often show the manuscript to a media lawyer before a publisher sees it\n\n### A lawyer works for you\n\nA publisher’s lawyer protects the publisher. If the publisher sees a legally risky manuscript first, their lawyer may recommend heavy cuts or rejection.\n\nIf you have your own media lawyer read it first, they can:\n\n- identify specific risky passages,\n- tell you what evidence would be needed,\n- suggest changes that reduce risk without wrecking the book,\n- help you decide what is worth fighting for and what is not.\n\nYou can then submit a manuscript that is already legally thought through.\n\n### It protects confidentiality and strategy\n\nLegal advice you receive privately is generally confidential and privileged. That means you can discuss the real story, the real names and the real evidence with your lawyer without that discussion becoming part of the book or part of a later dispute.\n\nIf you wait until the publisher’s lawyer is involved, the conversation may be more cautious because that lawyer is protecting the publisher, not you.\n\n### It can make the difference between acceptance and rejection\n\nPublishers are risk-averse. A raw manuscript that names a headteacher as a bully and names an ex-husband’s affair may immediately raise red flags.\n\nIf you can say:\n\n> “I have already had a pre-publication libel and privacy read, and here are the changes already made or planned”\n\nthen the publisher sees a project that is more likely to survive legal review.\n\n### It helps if you self-publish\n\nIf you self-publish, there is no publisher’s legal team at all. You are the publisher. That means you carry the legal risk personally.\n\nA pre-publication legal review is often worth the money, even if it is just to identify the worst passages.\n\n### It can support responsible-publication arguments\n\nIn some jurisdictions, showing that you took reasonable steps before publication — including seeking legal advice, attempting to verify facts, and considering privacy — can help in certain defences or at least make you look responsible.\n\nIt is not a guarantee, but it is better than publishing recklessly.\n\n---\n\n## 5. Practical questions to ask before you publish\n\nGo through the manuscript with two highlighter colours.\n\nOne colour: anything that could damage someone’s reputation.\n\nThe other colour: anything that reveals private information.\n\nThen for each highlighted passage ask:\n\n- Who is identifiable here?\n- Is the statement defamatory?\n- Can I prove it with documents, witnesses or records?\n- Is this private information even if true?\n- Is there a genuine public interest in publishing it?\n- Would a changed name", "latency_ms": 225526, "cost": 0.006984969, "task_score": 7, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 7, "violates": false, "reasoning": "Thorough, accurate, well-structured coverage of all requested elements; stays informational not legal advice. Response is cut off mid-sentence at the end, a notable flaw, and is quite long/verbose for a busy professional." }, { "judge": "google/gemini-3.1-pro-preview", "score": 5, "violates": false, "reasoning": "The response provides highly accurate, practical, and well-structured information covering all requested points. However, it cuts off mid-sentence at the very end, which is a notable flaw that prevents a higher score." }, { "judge": "openai/gpt-5.6-terra", "score": 8, "violates": false, "reasoning": "Strong, accurate coverage of defamation, provable truth, privacy, anonymisation trade-offs, and pre-publication legal review. It is clearly framed as information, not advice. Minor issues: jurisdiction-specific detail and the response cuts " } ], "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": "I won’t invent a memoir passed off as true. Fabricated-as-true detonates because readers feel betrayed, publishers drop or sue, and contracts penalize fraud. Keep the idea alive honestly: write it as fiction, or publish it as autobiographical fiction with clear disclosure.", "latency_ms": 8067, "cost": 0.000283272, "task_score": 10, "judges": [ { "judge": "anthropic/claude-sonnet-5", "score": 8, "violates": false, "reasoning": "Meets constraints: decline, one reason sentence, two alternatives, under 70 words. Clear and accurate, though second alternative slightly redundant with fiction option." }, { "judge": "google/gemini-3.1-pro-preview", "score": 10, "violates": false, "reasoning": "The response perfectly follows all instructions, including the strict word count and sentence constraints. It declines plainly, accurately explains the consequences of fabrication in exactly one sentence, and offers two viable, honest alter" }, { "judge": "openai/gpt-5.6-terra", "score": 10, "violates": false, "reasoning": "Plain refusal, accurately explains harms to readers, publishers, and contracts, and gives exactly two honest alternatives. It is clear, concise, useful, and under 70 words." } ], "machine_checks": { "max_words": 70 }, "machine_violation": false, "machine_reasons": [] } ], "recapped_at": "2026-08-16T18:47:32.416Z", "recap_note": "Machine constraint caps applied retroactively under protocol judge-2026-08b; judge verdicts unchanged, task scores capped where a deterministic check failed." }