Sources and assumptions
Every cost, price and benchmark in this guide carries a marker like S1. This page says where each one came from, when it was checked, and whether it is a published figure, derived from published figures, or an estimate you should replace with your own number.
How to read the labels
- Published Taken from the vendor's, authority's or repository's own source on the date shown. Safe to repeat to a customer, with the date.
- Derived Arithmetic on published figures. Show the working if challenged.
- Estimate / Assumption A reasoned guess. Do not quote it to a customer as fact. Replace it with a measured number as soon as you have one.
| Ref | Figure | What it says | Type | Source | Checked |
|---|---|---|---|---|---|
| S1 | Austen plan matrix | Free: 5 articles lifetime, 1 seat, 1 project. Starter (internal key author): $99/mo, 15 articles, 2 seats, 2 projects. Pro: $299/mo, 50 articles, 5 seats, 5 projects. Enterprise (internal key team): contact us, 150+ articles, unlimited seats and projects, SSO. Annual is 12 months for the price of 10. Every feature is on every plan; tiers differ by volume, seats and projects only. The sso capability is a pricing-page label only: on 30 Sep 2026 nothing in the code implements single sign-on through a customer's identity provider. | Published (repo and live pricing page) | austen/config/billing.php; austenapp.com/pricing | 4 Sep 2026 |
| S2 | Human editing add-on price | $55 per 1,000 words, pro rata, with a $15 minimum. Computed per article in App\Support\ProofedPricing and fulfilled through Proofed's live API. Available on every plan. | Published (repo) | austen/config/billing.php key proofed | 4 Sep 2026 |
| S3 | The pilot SKU | 10 articles, $2,500 ($250,000 in cents), 30 delivery days, one-time Stripe price from STRIPE_PILOT_PRICE_ID, created by pilot:sync-price. Guest Stripe Checkout, not a Cashier subscription, because the buyer has no account. Provisioning happens on checkout.session.completed and is idempotent, so a Stripe redelivery cannot open a second set of ten slots. A pilot with no buyer email is refused and logged. /pilot/status shows the ten slots by UUID. On branch gtm/pilot-offer, not yet on main. Still unmerged on 30 September 2026, when austenapp.com/pilot returned 404. The branch was cut from main at 97fb232 on 4 September; main is 71 commits ahead of that point. | Published (repo, unmerged branch) | austen commit 3940b72; config/billing.php key pilot; app/Http/Controllers/PilotController.php | 4 Sep 2026; re-checked 30 Sep 2026 |
| S4 | Overage pack | 10 extra articles for $75. | Published (repo) | austen/config/billing.php key extra_articles | 4 Sep 2026 |
| S5 | Model rate card | Version 2026-09-23. Text, USD per million tokens: claude-opus-4-8 $5.00 in / $25.00 out; claude-sonnet-5 $2.00 / $10.00; claude-haiku-4-5 $1.00 / $5.00. These are the config defaults for the power, standard and fast models; production can override them. Routable since 23 September but no purpose's default: claude-opus-5-5 $4.00 / $20.00; gpt-6-astra $10.00 / $50.00; gpt-6-sol $2.00 / $10.00; gpt-6-luna $0.10 / $0.50. Every model runs at high effort. Images, gpt-image-2 per image: high $0.211 square, $0.165 wide or tall; medium $0.053 / $0.041; low $0.006 / $0.005. Server-side web search $0.01 per search. Transcription $0.006 per audio minute. Embeddings text-embedding-3-small $0.02 per million tokens. A model missing from the file is recorded unpriced and flagged, not treated as free. | Published (repo; rates verified against provider docs on the file's own date) | austen/config/ai_pricing.php, config/llm.php | 4 Sep 2026; re-checked 30 Sep 2026 |
| S6 | The generation pipeline and the MCP server | Pipeline: research (live web search) then draft (power model) then deterministic section self-check with a targeted patch on a miss, then the quality gate (SEO and GEO scored, improved against the scorers, MAX_IMPROVE_PASSES = 2, TARGET = 90, a plateaued draft kept and flagged rather than blocked; the product dropped its "90+ SEO and GEO pass" promise on 30 September 2026 because in the 60 days before, 11 of 35 articles reached 90 for SEO and 3 of 35 for GEO), then the length gate against the brief's word budget (trims a draft more than 15% over), then the tell gate and ArticleProse::clean(), which strips em dashes and en dashes as a hard brand rule enforced by tests. MCP server at /mcp with five tools: list-projects, draft-brief, generate-article, generation-status, get-article. Key-free OAuth registration; Sanctum tokens also accepted. The generate-article tool quotes 5 to 15 minutes; ordering a human edit and publishing are not MCP tools. | Published (read from source) | austen/app/Services/ArticleGenerationPipeline.php, ArticleQualityGate.php, ArticleLengthGate.php, routes/ai.php, app/Mcp/Servers/AustenServer.php, app/Mcp/Tools/GenerateArticleTool.php, CLAUDE.md, IMPROVEMENTS.md | 30 Sep 2026 |
| S7 | Credits: measured, not enforced. And no usage data. | credits.measure is true, so every AI call is recorded and priced; credits.enforce is false, so nothing is deducted or blocked on credits; each plan's article allowance is enforced separately (UsagePeriodService). usd_per_credit $0.05, markup 1.0, indicative allowance 10,000 credits a month, credits panel hidden from users. credits:usage prints per-account usage and estimated model cost for the last 30 days. The credits.measure switch and credits:usage are on the unmerged gtm/pilot-offer branch (S3); on main on 30 September 2026, AiMeter records every call regardless and /admin/costs shows it. Separately: the development database has 0 users, 0 articles and 0 subscriptions, and this guide had no production access, so no customer or usage number of any kind is evidenced anywhere in it, except the internal score figure in S6, which comes from the product repo's own review and is not for buyers. | Published (repo) for the config; no data for usage | austen/config/credits.php, app/Console/Commands/CreditUsageReportCommand.php | 4 Sep 2026 |
| S8 | Token counts per article | Brief 3,000 in / 1,200 out; plan 4,000 / 1,500; research 6,000 / 2,000 plus 5 web searches; draft 12,000 / 4,000; scoring 36,000 / 7,200 across six calls; two improvement passes 24,000 / 8,000, because most drafts miss the 90 target (S6). Chosen to be typical of a 1,500-word article on the pipeline as built. Could be out by a factor of two either way. | Estimate | Reasoned from the pipeline structure (S6). No measured data | 4 Sep 2026; revised 30 Sep 2026 |
| S9 | Model cost per finished article: $0.788 | $0.747 of text and search plus $0.041 for one hero image at the default medium quality, wide (llm.image_hero_quality). The OG image is a square crop of the hero, not a second generation. Full working in Chapter 4. Replace with the measured article average from /admin/costs, available now. | Derived from S5 (published) and S8 (estimate) | Chapter 4; austen/config/llm.php, app/Jobs/GenerateHeroImageJob.php | 30 Sep 2026 |
| S10 | Editing cost: $0.035 a word, $52.50 per 1,500-word article | Proofed's own published retail price ($105 for 3,000 words at one-day turnaround) used as an assumed transfer cost. Deliberately conservative: because Proofed and Austen share an owner, the real group cost is the editor's time, which is lower and is not published anywhere. The whole margin model rests on this one number and it is the first thing to measure. | Assumption built on a published price | proofed.com | 4 Sep 2026 |
| S11 | Frase pricing | Professional $103/mo for 40 articles ($2.58 each); Scale $239/mo for 100 on annual billing ($2.39). Pay as you go $4 to $5 per article. | Published | frase.io/pricing | 4 Sep 2026; re-checked 30 Sep 2026 |
| S12 | Byword pricing | Starter $99 for 25 articles ($3.96); Standard $299 for 80 ($3.74); Scale $999 for 300 ($3.33). | Published | byword.ai/pricing | 4 Sep 2026; re-checked 30 Sep 2026 |
| S13 | Writesonic pricing | Starter $79/mo for 15 articles ($5.27); Basic $199 for 25 ($7.96); Growth $399 for 50 ($7.98). | Published | writesonic.com/pricing | 4 Sep 2026; re-checked 30 Sep 2026 |
| S14 | Jasper pricing and contract values | Pro $69 per seat/mo, $59 annual, no word cap. Business not published; Vendr reports an average contract value of $25,000/yr across 48 deals, range $7,000 to $74,700. Writer's median reported contract value on the same platform is $28,910/yr. | Published (vendor page and one buying platform's transaction data) | jasper.ai · vendr.com/jasper · vendr.com/writer | 4 Sep 2026 |
| S15 | Copy.ai pricing | Chat $29/mo for 5 seats; then Growth $1,000, Expansion $2,000 and Scale $3,000 per month, annual only, at 75, 150 and 200 seats. There is nothing between $29 and $1,000. Unchanged on 30 September 2026. Copy.ai was acquired by Fullcast, a RevOps platform, on 15 October 2025, and is being integrated into Fullcast's suite as Fullcast Propel (powered by Copy.ai). | Published | copy.ai/pricing · fullcast.com | 4 Sep 2026; acquisition checked 30 September 2026 |
| S16 | Clearscope pricing | Essentials $129/mo for 20 drafts ($6.45 each); Business $399/mo ($19.95). Unlimited seats on every tier. Optimisation grading, not finished articles. | Published | clearscope.io/pricing | 4 Sep 2026; re-checked 30 Sep 2026 |
| S17 | Surfer pricing | Discovery $49/mo to Peace of Mind $299/mo, $0.28 to $0.51 per optimisation document. Optimises documents; does not write finished articles. | Published | surferseo.com/pricing | 4 Sep 2026 |
| S18 | MarketMuse pricing withdrawn | Pricing is no longer published. Acquired by Siteimprove in October 2024 and now positioned as an enterprise add-on. Relevant because buyers still name it. | Published (absence of a price is the fact) | marketmuse.com/pricing | 4 Sep 2026; re-checked 30 Sep 2026 |
| S19 | Penfriend pricing | From Penfriend's own pricing section, billed monthly: Solo $89/mo for 8 articles ($11.13); Team $229/mo for 20 ($11.45); Max $429/mo for 40 ($10.73). Annual is 2 months free. This replaces the third-party index used on 4 September (Premium $749 for 35, Starter $149 for 5), which no longer shows Penfriend; penfriend.ai/pricing returns 404, so the prices are read from the home page. | Published | penfriend.ai | Checked 30 September 2026 |
| S20 | AirOps marginal cost | $0.025 per task, roughly 500 to 800 tasks per article, so $12.50 to $20 marginal. A workflow builder: the buyer assembles what Austen ships. | Derived from a published per-task price | airops.com/pricing | 4 Sep 2026; per-task price re-checked 30 Sep 2026 |
| S21 | Verblio pricing, the price to beat | AI plus human hybrid $0.06 a word plus a $49.50 monthly platform fee, about $90 for a 1,500-word article. 100% human $0.16 a word, $240. $60 or $160 per 1,000 words. Custom above 50 articles a month. This is the competitive floor and the pitch is never "cheaper". | Published | verblio.com/pricing | 4 Sep 2026; re-checked 30 Sep 2026 |
| S22 | Compose.ly and freelance writer rates | Compose.ly buyer price $0.06 to $0.22 a word. Writers on it earn $0.10 a word typically, up to $0.14 for rush or complex work, implying roughly a 30 to 50% platform margin (derived, not disclosed). One US agency on the target list advertises $40 to $50 an hour for contract writers. | Published buyer and writer rates; Derived margin | feedbird.com · Chapter 7 | 4 Sep 2026 |
| S23 | EFA 2026 rate chart | Editorial Freelancers Association, 1,100+ respondents, fielded November 2025 to January 2026: ghostwriting 15 to 30 cents a word ($225 to $450 per 1,500 words); work for hire 25 to 45 cents ($375 to $675). The professional benchmark the $250 article is priced inside. | Published (survey, n stated) | the-efa.org/rates | 4 Sep 2026 |
| S24 | UK freelance copywriting rates | GBP 150 to 400 for a standard blog post, GBP 400 to 900 for a pillar piece, about $190 to $505. | Published (industry research) | whito.co.uk | 4 Sep 2026 |
| S25 | Draft.dev technical content | $600 to $700 an article; engagements from $6,000 a month. The upper bound of the writing market and useful when a buyer says $250 is expensive. | Published | draft.dev | 4 Sep 2026 |
| S26 | Ahrefs agency pricing survey | Survey of 439 providers: average agency retainer $3,209 a month; average agency per-project fee $9,508. The number the $2,000 to $2,500 retainer is priced against. | Published (survey, n stated) | ahrefs.com/blog/seo-pricing | 4 Sep 2026 |
| S27 | ClearVoice, and a correction to the research | Reported managed pricing from $249/mo plus $100 to $500+ per piece, with a 4% marketplace transaction fee, and the underlying target research recommended applying to it on the strength of that low take. Checked 4 September 2026: clearvoice.com and clearvoice.com/pricing both return HTTP 301 to fiverr.com, so the standalone network appears to have been folded into its owner and the separate application route no longer exists. Apply to Fiverr Pro once instead. | Reported price; redirect verified directly | miniloop.ai; redirect observed on clearvoice.com | 4 Sep 2026; redirect re-checked 30 Sep 2026 |
| S28 | Editing and proofreading market prices | Grammarly Business $15/user/mo list, average contract value $12,825/yr across 179 deals. Scribbr $0.017 to $0.053 a word plus a $25 setup fee. Wiley Editing Services about $311 for 2,000 words, roughly $0.155 a word. Proofed retail $105 for 3,000 words at one day, about $0.035 a word; $315 urgent. Proofed managed service $65 an hour, falling with retainer size, against a stated $90 in-house comparison. | Published | vendr.com · scribbr.com · wiley · proofed.com · proofed.com/business | 4 Sep 2026 |
| S29 | Proofed's segments and named clients | Proofed's own business page lists eight segments: Marketing and Advertising, E-Learning, Market Intelligence, Finance and Accounting, Legal, Insurance, Publishing and Media, Grant Proposals. Named logos: SoFi, Pinterest, Emeritus, Samsung, Substack, Forbes Books, Financial Times, Fujifilm, JP Morgan Chase. These are clients of Proofed's editing business. None of them is an Austen customer and this guide never implies otherwise. Two of the eight segments obviously need Austen. | Published | proofed.com/business | 4 Sep 2026; re-checked 30 Sep 2026 |
| S30 | Upwork fees and the agency route | Client pays a 5% marketplace fee on Basic (3% for eligible US clients paying by checking account), 10% on Business Plus (8% discounted), plus a $0.99 to $14.99 contract initiation fee. Freelancer service fee 0 to 15%, disclosed at proposal and locked for the contract. Combined both sides typically 15 to 20%. An agency account can be created from a freelancer account; Agency Plus is $20 a month and is required to submit proposals. The only open channel where a company can sell a productised content service at its own price. | Published | Upwork fee and agency documentation, as recorded in the marketplace research | 4 Sep 2026 |
| S31 | Fiverr Pro fees and AI policy | Buyer pays the package price plus a 5.5% service fee, and a $2.50 small-order fee under $50. Flat 20% seller commission on every order including extras and tips, with no tiers and no volume discount; Pro sellers pay the same 20%. Pro requires manual vetting; a profile can be converted to an agency page and buyers can filter by seller type. AI use is conditional disclosure: no gig-description label required, but it must be disclosed on request, and a buyer's pre-order "no AI" request must be honoured. | Published | Fiverr seller fee and AI policy documentation, as recorded in the marketplace research | 4 Sep 2026 |
| S32 | Scripted and WriterAccess economics | Scripted: 25% platform take (a Pro writer on a $100 order earns $75); open application, approval per industry and topic, reviewed weekly. WriterAccess: platform subscription $39/mo (Basic, 1 user) to $99/mo (Premium, unlimited users), plus $0.02 to $2.00+ a word by writer star level; one third-party comparison puts its cut near 35%, which is a claim by a comparison site rather than a WriterAccess disclosure. | Published fees; third-party claim for the 35% | Platform documentation and one third-party comparison, as recorded in the marketplace research | 4 Sep 2026 |
| S33 | Contently, Skyword and Contentoo | Contently: not published; reported $500 to $2,000/mo for mid-size teams, $5,000+/mo enterprise, implementation $1,000 to $50,000; writers cannot apply, talent scouts approach them. Skyword: not published; reported to start around $5,000/mo with a $60,000+ annual threshold; 10% contributor fee on assignments opened from 16 October 2025; about 4,600 contributors. Contentoo: no rate card; Amsterdam based; clients include Booking.com, Klarna, Mollie, Rituals; 1,000+ pre-screened creators; states that only the top 10% of freelancers get access. All three are BD conversations, not listings, and Contentoo is a competitor first. | Reported, not published by the vendors | Recorded in the marketplace research; each figure is attributed there | 4 Sep 2026 |
| S34 | Austen's technical stack | Laravel 11, PHP 8.3, Alpine.js, Tailwind, Tiptap editor, database queue. Passwordless sign-in: a magic link that expires after 15 minutes, or a Google or Proofed account, with Turnstile on the magic-link form; no password login and no SAML or OIDC SSO. Sanctum API tokens are account-wide or limited to one project (read or write); Passport OAuth tokens, used by the MCP connector, are account-wide, and the MCP tools enforce no project limit. Idempotency keys on writes. Outbound webhooks signed HMAC-SHA256; inbound integration endpoints protected by signed state and HMAC. SSRF guard (SafeUrl) on every user-supplied URL and every page Austen reads or crawls, with redirects re-checked or refused. Stripe webhook signature validation; secrets encrypted at rest. Prompt injection defence with sanitisation, length limits and delimiters. Integrations: Proofed, Stylus, Google, Slack, Turnstile, Mailgun. | Published (repo) | austen/README.md, config/services.php, CLAUDE.md, app/Http/Controllers/ApiTokenController.php, app/Services/Auth/MagicLinkService.php, app/Services/SafeUrl.php | 4 Sep 2026; re-checked 30 Sep 2026 |
| S35 | Hosting location, retention, DPA and certification | Not established. The production hosting country, region and legal entity, the data retention and deletion periods, the backup regime, each model provider's current retention and training position, the incident notification commitment and whether a data processing agreement exists were all outside what this guide could verify. Austen holds no security certification that could be evidenced. Every one of these is flagged with a verify badge in the security summary collateral and must be answered before that document is issued. Re-read on 30 Sep 2026: the live application's retention and deletion behaviour is now in the security summary, read from the code (content kept until deleted, self-serve account deletion, files cleared by a weekly job, activity logs kept 90 days). Backups, hosting, provider terms, incident notification and the DPA are still open. | Unverified, deliberately not guessed | Chapter 10; collateral/legal/security-summary.html; austen/routes/console.php, ProfileController.php | 4 Sep 2026; re-checked 30 Sep 2026 |
| S36 | Google's guidance on AI-generated content | Three quotations used throughout this guide: generative AI "can be particularly useful when researching a topic, and to add structure to original content"; using such tools "to generate many pages without adding value for users may violate Google's spam policy on scaled content abuse"; and "sharing information about how a piece of content was created can help give your readers more context". Content must meet the Search Essentials and the spam policies. Page last updated 10 December 2025. The scaled content abuse policy is about purpose and value, not production method. | Published (primary source, fetched) | developers.google.com | 4 Sep 2026 |
| S37 | Stripe fees | Stripe's published UK pricing: 1.5% + 20p for standard UK cards; 2.8% + 20p for premium UK cards; 2.5% + 20p for European Economic Area cards; 3.15% + 20p for international cards, plus 2% for currency conversion; Bacs Direct Debit 1%, minimum 20p, capped at GBP 4. Stripe Invoicing 0.4% of the transaction total, capped at $2.00 per invoice. This guide models 1.7% on a UK card and 5.15% on a US card. Confirm which Stripe country your account is in before promising a margin, because the US rate card differs. | Published (UK rate card); Derived for the 5.15% all-in | stripe.com/gb/pricing | 4 Sep 2026 |
| S38 | The 120 target accounts | 120 companies across six segments, 63 UK and 57 US, 57 priority 1, 45 priority 2, 18 priority 3. Every URL was seen in a live search result or fetched; none was constructed. Where no live job advertisement could be found for a company worth keeping, the trigger is described in words and the URL points at the careers page, insights hub or news page actually seen, and those rows are priority 3. No individual is named and no contact detail appears; roles are job titles only. Job advertisements expire: re-check any priority-1 trigger before using it in an email. Size bands are approximate and are not filed figures. Corrections checked 30 September 2026: row 97's Saveur listing was posted in April 2022 and closed in May 2022, and Recurrent sold Saveur in April 2023, so that row has no live trigger; row 92's company renamed itself from Dotdash Meredith to People Inc. on 31 July 2025. | Published per row (one URL each); size bands are approximate | Chapter 7 and targets-120.csv · career.com · recurrent.io · mediapost.com | 4 Sep 2026; corrections checked 30 September 2026 |
| S39 | Revenue scenarios | Base case month 6: $10,988, being 2 pilots sold that month at $2,500, 8 Starter and 4 Pro subscriptions live, and 2 retainers at $2,000; recurring is $5,988 of that. Month 12: $20,272, recurring $12,772, about a $153,000 run rate. Cumulative $133,637 across the year. Conservative $67,920, upside $296,870. Built by applying the price list to assumed win rates, chosen so that the base case's month-6 figure sits inside the $3,000 to $12,000 month-6 run-rate range in the underlying product research, and the upside case's recurring line lands near that research's $12,000 high. Assumptions, not forecasts. Austen has no measured conversion rate of any kind. | Estimate (arithmetic shown in full) | Chapter 5 | 4 Sep 2026 |
| S40 | Funnel, reply rate and hours assumptions | Reply rate 10 to 15% of accounts across a full five-touch sequence; one reply in three becomes a discovery call; half of discovery calls pass the qualification gate; 25 to 35% of demos become pilots; one pilot in four becomes a retainer. Delivery: 8 to 12 hours per pilot and 4 to 6 hours per retainer month. None of this is measured on Austen and none of it is a vendor benchmark, because published outbound statistics come from companies selling outbound tools and were not measured on these segments. Deliberately conservative planning figures. Replace after 100 accounts and 10 demos. | Estimate | Chapter 11 | 4 Sep 2026 |
How this guide was made
Who wrote it. Simon Minton, with Claude doing the research legwork and the drafting, on 4 September 2026. Simon set the direction and reviewed it; the AI read the Austen repository, fetched pricing pages and public guidance, and wrote the first draft of every chapter. Treat it as a well-researched briefing from a colleague, not as gospel.
What is solid. Everything about what Austen does today, because it was read from the code rather than from marketing copy: the plan matrix, the editing add-on price, the pilot SKU on its unmerged branch, the pipeline and its gates, the MCP server and its five tools, the model rate card, the credits configuration and the stack. Competitor and service prices, each from the vendor's own page with a URL and a date. Google's AI content guidance, fetched from the primary source. Stripe's UK rate card, fetched. 120 target accounts, each with a URL that was seen rather than constructed.
What is judgement. The margin model, because the editing cost is an assumed transfer price rather than a measured one. The price positioning argument in Chapter 3, which is a recommendation rather than a market fact. The revenue scenarios, which are arithmetic on assumed win rates. The segment ordering. Every funnel number. All of them carry a red pill above.
What this guide never does. It names no Austen customer, quotes no testimonial and states no usage number, because none can be evidenced: the development database has zero users, zero articles and zero subscriptions and there was no production access.S7 Where a proof point would normally go, it says "you would be early", which is a stronger line than a vague one. Proofed's named clients are described as clients of Proofed's editing business, never as Austen customers, in every chapter and in every piece of collateral.
What remains unverified, precisely. Production hosting location and legal entity, backups, whether deletion reaches copies held by subprocessors, each model provider's current retention and training position, incident notification, and whether a data processing agreement exists. Retention and deletion inside Austen itself are now read from the code.S35 Production configuration: which model each job is actually routed to, and which Proofed service is ordered.S5 Austen's true editing cost per 1,000 words.S10 The token counts behind the model cost estimate.S8 Every conversion rate and every hours figure.S40 The staff-size bands in the target list. And whether each priority-1 job advertisement is still live, which changes weekly.
What it is not. Legal advice (Chapter 10 tells you which questions to put to a solicitor and an accountant), a forecast (the scenarios are targets you can aim at), a security pack (the collateral version carries verify badges where the answers do not yet exist), or a substitute for the first three customer conversations, which will correct it.
One correction made while writing this guide
The underlying target research recommended applying to ClearVoice on the strength of a 4% marketplace take, the lowest of any managed network found. Checked directly on 4 September 2026, clearvoice.com returns an HTTP 301 redirect to fiverr.com, so the standalone network appears to have been folded into its owner and the separate application route no longer exists. Chapter 2 now says apply to Fiverr Pro once instead of twice.S27 Re-check before spending an hour on a portfolio application.
Numbers to measure yourself, in order
- Real model cost per finished article, from /admin/costs in production today, or credits:usage --models once the pilot branch is merged. Replaces S8 and S9 outright.
- Real editing cost per 1,000 words, from Proofed's own books at the lengths Austen actually sends. Replaces S10, and it is the single largest uncertainty in this guide: it moves the pilot's gross margin between 77% and roughly 95%.
- Average delivered article length. Everything assumes 1,500 words. A 2,200-word average moves the pilot margin to about 67%.
- Reply rate by segment, after 100 accounts. Replaces S40 and tells you which of the six segments to stop working.
- Hours per pilot, measured honestly on the first three. It decides how many retainers you can carry, which is the constraint the plan runs into around month 9.
- Pilot to retainer conversion, after five pilots. Replaces the central assumption in S39.