Skip to main content
AI Biz Hub
Menu

Authority signal

Corrections Log

Every factual correction, clarification, retraction, or verification pass on aibizhub.io content is logged here, dated, and linked from the affected page.

Policy

When a factual error is found in a calculator, benchmark, or article, the content is corrected inline and the "Last updated" timestamp is bumped. Material corrections are flagged at the top of the affected page and logged here. Corrections are never silently edited. History is preserved.

  • Corrected

    A factual error was fixed.

  • Clarified

    Wording sharpened to prevent misreading (no factual change).

  • Retracted

    The claim or page was removed because it could not be substantiated.

  • Verified

    A formal audit pass found no error. Logged for transparency.

Log

  • corrected
    Agent API: eleven calculators

    A sweep of every input of every calculator through /api/run found inputs the API ignored. A default filled in for an input the caller left out replaced the one they gave: the per-component and per-category inputs of Build vs Buy, Startup Cost Estimator and Marketing Budget Allocator were overridden by a filled-in list; cost_of_goods and operating_expenses by a filled-in cost (Profit Margin); frequency by meetings_per_week (Meeting Cost); a conversions-only call or an average_inventory-only call never reached the mode the description promises (Ad Spend ROAS, Inventory Turnover); and mode=mix in Break-even Units ran on an empty product list. Inventory Turnover's industry was never passed to the engine, so every API call came back unrated. Employee Cost now returns the currency label it was given. These no longer fill in next to the input that replaces them, and a sweep in the test suite checks every input.

  • corrected
    Contractor vs Employee Calculator

    The W-2 side charged a flat 7.65% FICA past the Social Security wage base, a $420 state unemployment amount with no source, and filed workers' compensation under taxes. It now uses the Employee Cost Calculator's payroll model, from one shared module: Social Security 6.2% up to $184,500, Medicare 1.45%, federal unemployment 0.6% of the first $7,000 ($42), state unemployment at a rate on a state wage base (default 2.7% of $7,000, $189, labelled as an assumption), and workers' compensation as insurance. On the defaults ($85,000 salary, $70 an hour for 2,080 hours) the W-2 cost falls from $104,364.50 to $104,133.50 and the break-even contractor rate from $50.18 to $50.06 an hour.

  • corrected
    Contractor vs Employee for a Solo Founder: 2026 Math

    The worked $120,000 hire assumed $540 of state unemployment as a flat amount. It now uses the calculator's model, 2.7% of the first $7,000 ($189), so the employee costs $145,096 a year, not $145,447. The contractor at $95 an hour for 1,800 hours costs $25,904 more (17.9%), not $25,553 (17.6%), and the break-even rate is $80.61 an hour, not $80.80.

  • clarified
    Employee Cost Calculator

    The employer taxes figure ($8,881 on a $100,000 salary) included $1,000 of workers' compensation, which is an insurance premium, not a tax. The calculator now shows payroll taxes ($7,881: Social Security, Medicare, federal and state unemployment) and workers' compensation on separate lines. The total cost is unchanged.

  • clarified
    Loaded Employee Cost: What BLS Compensation Data Really Shows

    The worked example's $7,849 of employer taxes included $810 of workers' compensation insurance. It now gives $7,039 of payroll taxes and $810 of workers' compensation separately; the totals are unchanged.

  • corrected
    Employee Cost Calculator

    Employer payroll taxes were overstated. Unemployment insurance was 2% of the whole salary ($2,000 on $100,000), and Social Security ran past its wage base. The calculator now applies each tax to its own wage base: Social Security 6.2% up to the 2026 base of $184,500, Medicare 1.45% with no cap and no employer share of the employee-only Additional Medicare Tax, federal unemployment 0.6% of the first $7,000 ($42), and state unemployment at a rate on a state wage base. The state default, 2.7% of the first $7,000 ($189), is labelled as an assumption, and every rate and base is an input. Line items now always add up to the total, because every line is rounded to the cent and totals are summed from the rounded lines. The $100,000 default falls from $130,350 to $128,581 a year (1.29x).

  • corrected
    Loaded Employee Cost: What BLS Compensation Data Really Shows

    The worked $90,000 hire charged its 1.6% state unemployment rate on the whole salary ($1,440) and left out federal unemployment. It now uses the calculator's corrected model: $42 federal unemployment, 1.6% of the first $7,000 ($112) for state unemployment. The example now comes to $116,149.00 a year (1.29x), not $117,435.00 (1.30x), and the taxes-and-benefits layer adds 22.1% to salary, not 23.5%.

  • corrected
    AI Support Break-Even: Deflection Rate vs Ticket Volume

    The table of contents and the page description gave the 70% deflection saving as $8,655 a month and the description gave the 40% saving as $2,355; the engine result, the section headings and the text give $8,695.50 and $2,395.50. Both now match.

  • corrected
    Solo Founder Tax: SE Tax and the S-Corp Threshold

    Payroll-service ($600 to $1,500 a year) and S-corp return ($800 to $2,500) fees were given as typical ranges, the second cited to the IRS Form 1120-S page, which states no fees. The page now gives Gusto's published payroll prices ($660 to $1,104 a year for one person), says preparer fees are quoted per engagement with no current survey, and labels the break-even tables' $2,500 and $4,000 overhead as assumptions. The Massachusetts and New York minimums now cite the state tax pages.

  • corrected
    Pinecone Pricing 2026: Rates vs Qdrant

    The page credited Qdrant's pricing page with a free cluster that holds roughly 250k vectors at 768 dimensions; the page states only 0.5 vCPU, 1 GB RAM and 4 GB disk. It also said the worked scenario (200,000 vectors at 1,536 dimensions) fits Qdrant's free cluster. By Qdrant's capacity-planning formula those vectors and their HNSW index need about 1.26 GB of RAM at default settings, more than the free tier's 1 GB; they fit with 4-bit quantized vectors in memory and the originals on the cold tier (about 0.18 GB in RAM). The page now sizes the free tier by memory (about 300,000 vectors at 768 dimensions, 150,000 at 1,536), cites the formula, and drops an unsupported claim that Qdrant's free tier has no egress cap. Prices were rechecked; they had not changed.

  • corrected
    AI Support Tool Pricing 2026

    Ada was said to fit companies with at least 300,000 annual conversations and Forethought to be outcome-based; neither pricing page says so. Ada's is a demo request that asks for annual ticket volume, and Forethought's lists Team, Professional and Enterprise plans priced by quote. Gorgias AI Agent was described as $0.90 per resolution on annual billing and $1.00 monthly, on a helpdesk plan from $10; Gorgias states $0.90 on most plans and from $1 on Starter, with plans of 90 to 2,500+ automated interactions, and the cited article gives no helpdesk price. Both quote-only vendors now have a source, and the listing card no longer gives eesel's retired $0.40 per-ticket rate.

  • corrected
    eesel vs Gorgias AI Pricing 2026

    The page was corrected earlier, but the article listing kept its old title and description, which gave eesel's retired $0.40 per-ticket price. Article listings now read the same metadata as the article page, for this and 74 other articles.

  • corrected
    Embeddings DB Cost

    Postgres+pgvector was priced as a $35 managed-Postgres baseline with 25 GB included, a figure DigitalOcean does not publish, and it ignored the memory an HNSW index needs. It is now priced on Supabase Pro: $25 a month with $10 of compute credit, plus the smallest compute size that holds the index by Supabase's own pgvector benchmarks, plus disk above 8 GB at $0.125/GB. At 1,536 dimensions that is $30 at 50,000 vectors, $125 at 200,000, $425.79 at 1 million and $977.58 at 2 million (was $35 in each case). LanceDB on R2 billed reads at R2's Class A write rate ($4.50 per million); reads are now Class B at $0.36 per million, which lowers that line by roughly half to four-fifths. The methodology list also quoted superseded Pinecone and Turbopuffer rates; it now matches the engine.

  • corrected
    Embeddings DB cost by stored vectors and query volume

    Recomputed with pgvector priced on Supabase compute and LanceDB reads billed at R2's Class B rate. The table now shows the cheapest hosted option (Turbopuffer in every cell), because LanceDB, the lowest line everywhere, excludes the compute that serves queries; LanceDB's range is stated beside it. The claim that flat-priced managed Postgres wins once queries dominate is withdrawn.

  • corrected
    Embeddings DB vs Self-Host pgvector: True Cost

    Recomputed with pgvector priced on Supabase compute sized for the index: $125 a month at 200,000 vectors and $425.79 at 1 million, not a flat $35, which makes a dedicated pgvector database the most expensive line at both scales; the headings, TL;DR and decision rules are rewritten to match. LanceDB is $0.25 and $1.51 (was $0.55 and $5.24). The self-host arithmetic was wrong: 2 to 6 hours a month at $100 is $2,400 to $7,200 a year, not $3,000 to $8,000, and a 4 to 8 hour incident is $400 to $800, not $400 to $1,600. The 200k example now uses an 8 GB server, the size Supabase's benchmark needs, giving $2,920 to $8,920 a year.

  • corrected
    Milvus vs pgvector Cost in 2026

    The pgvector line was a $35 generic managed-Postgres baseline. It is now Supabase Pro with Large compute, $125 a month for 200,000 1,536-dimension vectors, because Supabase's benchmark needed Large's 8 GB to hold that index in memory; the summary and comparison table now separate a new pgvector database from adding vectors to a Postgres you already run.

  • corrected
    RAG vs Fine-Tune Total Cost for a Solo Founder

    Recomputed with pgvector priced on Supabase compute sized for the index: $977.58 a month at 2 million vectors and about $11,238 at 20 million, not $35 and $58.61. A dedicated pgvector database is therefore not a rounding error next to inference at scale; the storage comparison now names it alongside Pinecone as the exception. LanceDB is $0.94 and $4.58 (was $4.67 and $17).

  • corrected
    The AI Micro-SaaS Unit Economics Report 2026

    Recomputed with pgvector priced on Supabase compute: $977.58 a month at 2 million vectors, not $35, so it no longer fits the stack's $50 allowance; LanceDB is $0.94 (was $4.67). The comparison against the stack also used totals from retired model tiers ($1,442 and $15,122); the current totals are $1,622 and $30,122, so Pinecone's $213 is about 13% and 0.7% of them, not 15% and 1.4%.

  • corrected
    Cheapest Vector Database 2026: 8 Managed Stores Ranked

    Supabase pgvector was ranked fourth at a flat $25 a month for 1 million 1,536-dimension vectors. Supabase's own pgvector benchmark runs that index on 2XL compute, so it now ranks sixth at about $426 a month ($25 Pro + $410 2XL - $10 credit + disk), and the pricing-model section describes it as instance-sized rather than a floor.

  • corrected
    Supabase Free Tier Limits 2026: pgvector Cost and Pausing

    Added the memory limit: Supabase's benchmark served 15,000 1,536-dimension vectors from the Micro instance the Pro credit covers and 50,000 from Small ($15 a month), so the engine now prices the 50,000-vector example at $30 rather than a $35 generic baseline. An FAQ answer said Turbopuffer or Pinecone would price the workload under a dollar a month; Turbopuffer's minimum is $16, and the under-a-dollar options are LanceDB on object storage and Pinecone's free Starter plan. LanceDB is about $0.08 (was $0.21).

  • corrected
    Chroma Cloud vs LanceDB Pricing 2026

    The engine billed LanceDB's reads at R2's Class A write rate; at R2's $0.36 per million Class B reads, LanceDB on R2 is about $0.25 a month for the 200,000-vector scenario, not $0.55.

  • corrected
    Turbopuffer Pricing in 2026 (vs Pinecone)

    The comparison with a $35 managed Postgres used a baseline no provider publishes; it now compares with the $125 a Supabase Postgres sized to hold 200,000 vectors in memory costs.

  • corrected
    Lindy vs Relay 2026: Relay Shut Down, Lindy's New Pricing

    Lindy no longer sells Plus, Pro and Max plans. It sells one Team plan that bills every active user from $29.99 a month for 3,000 credits, with larger allowances per seat (15,000 credits for $99.99, 35,000 for $199.99, and sizes in between), top-ups at $10 per 1,000 credits, and a Free plan of $50 in credits for 7 days instead of a 7-day trial. The pricing section, cost-per-credit table and recommendations are rewritten on the current page; the Relay timeline is unchanged.

  • corrected
    Coolify vs Dokploy vs CapRover 2026

    Release and community facts from July were out of date. Coolify's MCP server is no longer read-only: v4.3.0 (2026-08-12) added diagnostics and deployment controls, and the latest release is v4.4.6, not v4.1.2. Dokploy's latest is v0.30.8, not v0.29.12, and its cadence is 40 tagged releases in 2026 rather than several a week. CapRover's latest is v1.15.4, not v1.14.2. Star counts are updated (about 63k, 38k and 15.2k) and now cite the GitHub API; Cloud prices were re-checked and are unchanged.

  • corrected
    Cursor Pricing 2026: Plans, Model Prices and Is It Free?

    Cursor now sells its editor inside The X Subscription: Free, Plus $8 a month, Premium $30, Super $100 and Ultra $200, with Teams seats at $30, $100 or $200, and Claude, GPT and Gemini billed at API list price from prepaid credit on Premium and above. The page presented Hobby, Pro at $20, Pro+ at $60 and Teams at $40 a user as the current plans; they are now legacy plans for existing subscribers, and the page is rewritten on the new plans with a legacy-to-new mapping. Gemini 3.8 Flash output is $3.75 per million tokens, not $3.50, and Cursor's $60 to $100 monthly-spend guidance, which its docs no longer publish, is removed.

  • corrected
    Cursor vs Claude Code vs Windsurf Pricing 2026

    Said all three tools start at $20 with Cursor Pro. Cursor now sells Plus at $8, which runs only Cursor's own models, and Premium at $30, which adds Claude, GPT and Gemini billed at API list price on top; Pro ($20) and Pro+ ($60) are legacy plans. The prices, heavy-use ladder (Super $100, Ultra $200) and decision guidance are updated; Claude Code and Devin Desktop prices were re-checked and are unchanged.

  • corrected
    Cursor vs Copilot vs Windsurf Pricing 2026 (Devin Desktop)

    Listed Cursor Pro at $20, Pro+ at $60 and Teams at $40 a user as current. Cursor now sells Plus at $8, Premium $30, Super $100 and Ultra $200, with team seats from $30 and third-party models billed at API list price; the three-seat team cost for Cursor is $1,080 a year at $30 seats, not $1,440. Cursor's $60 to $100 monthly-spend guidance, which its docs no longer publish, is removed.

  • corrected
    Best AI Coding Assistants 2026: Picks by Workflow and Budget

    Named Cursor Pro at $20 as the default AI editor and Cursor Teams at $40 a user. Cursor now sells Plus at $8 on its own models and Premium at $30 for Claude, GPT and Gemini at API list price, with Teams from $30 a seat; Pro is a legacy plan. The picks, billing section and FAQs are updated, and the app-builder prices were re-checked and are unchanged.

  • corrected
    Cursor Alternatives for Developers (2026)

    Described Cursor as Pro at $20 with two monthly usage pools. Cursor now sells Plus at $8 and Premium at $30 with one weekly pool for its own models and third-party models billed at API list price; the comparison and FAQs are updated. The page also said Zed's free Personal plan includes 2,000 accepted edit predictions, which Zed's pricing page no longer states; the figure is removed.

  • corrected
    Vibe Code Platform Comparison

    Showed Cursor's entry price as $20 a month (Pro). Cursor now sells Plus at $8 on its own models and Premium at $30 for third-party models, so the tool shows $8-30 a month and its FAQ names both plans.

  • corrected
    AI vendor pricing

    The coding-editors paragraph referred to Cursor's $20 Pro plan, now a legacy plan; it now describes Cursor's weekly pool and API-priced third-party models.

  • corrected
    Cheapest Object Storage 2026: R2 vs S3 vs Backblaze

    Gave Backblaze B2 storage as $0.006 per GB-month. Backblaze bills $6.95 per TB per 30 days, which is $0.00695 per GB; B2 remains the cheapest storage line, at about 30% of S3's $0.023 rather than a quarter. Cloudflare R2 and Wasabi rates were re-checked and are unchanged.

  • corrected
    Railway vs Render vs Fly Pricing 2026 Compared

    Re-checked against every source. Fly.io's smallest Machine is $2.19 per 30 days in iad and ewr (the page said about $2.02 in Amsterdam), a 1 GB Machine is $6.70 (it cited $6.64 for a shared-cpu-2x), extra RAM is $6 per GB a month (not about $5), new accounts get a free trial of 2 hours or 7 days rather than a $25 prepaid minimum, and Fly lists 17 regions, not 18; the claim that Fly lists GPUs as deprecated is removed and the January 2026 snapshot billing now cites Fly's announcement. Railway Pro includes unlimited seats rather than costing $20 a seat, usage is metered per second, and Pro includes Railway support. Render's instance names (Starter, Pro Ultra) are replaced by its current sizes, $7 to $1,500 a month.

  • corrected
    Devin vs Factory AI Pricing 2026 (Autonomous SWE)

    Cited Lindy's Devin pricing guide for a legacy Core rate of about $2.25 per ACU and a fixed ACU allotment on the old $500 Team plan; that page no longer carries those figures and Cognition does not publish them, so they are removed. The page now cites Devin's billing docs: Enterprise is still billed in ACUs at the rate in its order form.

  • corrected
    Free SaaS Unit-Economics Calculators, No Signup (2026)

    The valuation example used 3.41x SDE as BizBuySell's 2025 software average; BizBuySell's software, app and SaaS page now reports 3.31x for businesses sold from 2021 through 2025, so the midpoint is $397,200 (was $409,200). A Capterra page cited for Baremetrics pricing was ChartMogul's listing and is replaced by BizBuySell and IBBA sources. The page also said Baremetrics lists Launch at $75 a month and $49 only in its FAQ: Launch starts at $49 a month on annual billing and $75 on monthly billing.

  • corrected
    How to Calculate True Customer Acquisition Cost

    Merged into CAC Calculator on 2026-10-21.

    The BLS reference label gave a 1.24x-1.29x salary multiplier that the cited release does not state; the label now quotes the release's own figures for private industry, $46.89 total compensation and $32.82 wages and salaries per hour worked.

  • corrected
    eesel vs Gorgias AI Pricing 2026: Per-Ticket vs Per-Resolution

    Rebuilt. eesel no longer charges $0.40 per ticket with no minimum: it sells monthly credit plans from $299 for 500 credits to $1,749 for 5,000, where a ticket or chat is one credit, with $0.80 overage. The comparison with Gorgias's $0.90 per resolution is recomputed on the credit plans.

  • corrected
    AI Support Tool Pricing 2026: 7 Vendors Ranked by Cost Model

    Priced eesel at $0.40 per ticket with no base fee. eesel now sells monthly credit plans from $299 for 500 tickets to $1,749 for 5,000 (about $0.60 down to $0.35 a ticket) with $0.80 overage; the summary, FAQs and comparison are updated.

  • corrected
    Turbopuffer Pricing in 2026 (vs Pinecone)

    Rebuilt. Turbopuffer cut its Launch minimum from $64 to $16 a month in June 2026, and the 64 GB billing floor the page described applies only to pinned namespaces. The page and the Embeddings DB Cost engine now use Turbopuffer's published rates ($0.33 per GB-month stored, $2 per GB written, $1 per PB queried with a 1.28 GB namespace minimum and discounts above 32 GB, $0.05 per GB returned), which makes Turbopuffer the cheapest hosted option at small scale.

  • corrected
    Embeddings DB Cost

    Turbopuffer was priced on a $64 monthly minimum and approximate rates. It is now modelled on Turbopuffer's published rates and $16 Launch minimum: $16 a month at 200,000 and 1 million vectors (was $64) and about $196 at 20 million vectors.

  • corrected
    Embeddings DB cost by stored vectors and query volume

    Recomputed with Turbopuffer's $16 Launch minimum and published rates (it was priced on a $64 minimum); Turbopuffer is now the cheapest hosted option in the small-scale rows, and the winner sentence is corrected.

  • corrected
    Cheapest Vector Database 2026: 8 Managed Stores Ranked

    Priced Turbopuffer on a $64 monthly minimum; the Launch minimum is $16 since June 2026. Turbopuffer moves up the ranking to third and the figures that depended on it are recomputed.

  • corrected
    Chroma Cloud vs LanceDB Pricing 2026

    Compared against Turbopuffer at a $64 monthly minimum; the Launch minimum is $16 since June 2026.

  • corrected
    Milvus vs pgvector Cost in 2026

    Compared against Turbopuffer at a $64 monthly minimum; the Launch minimum is $16 since June 2026.

  • corrected
    Supabase Free Tier Limits 2026: pgvector Cost and Pausing

    Gave Turbopuffer's paid entry point as a $64 monthly minimum; it is $16 since June 2026.

  • corrected
    Embeddings DB vs Self-Host pgvector: True Cost

    Turbopuffer was priced at its old $64 minimum; at $16 since June 2026 it is the cheapest hosted store at 200,000 and 1 million vectors, and the comparisons are recomputed. The Pinecone reference now separates its docs page (metering) from its pricing page ($16 to $18 per million read units), since the docs page does not show the rates.

  • corrected
    RAG vs Fine-Tune Total Cost for a Solo Founder

    Turbopuffer is now priced on its $16 Launch minimum and published rates: $16.67 a month at 2 million vectors and about $196 at 20 million. The Pinecone reference now separates its docs page (metering) from its pricing page (read-unit rates), since the docs page does not show the rates.

  • corrected
    The AI Micro-SaaS Unit Economics Report 2026

    Turbopuffer is now priced on its $16 Launch minimum and published rates ($16.67 a month at 2 million vectors). The Pinecone reference now separates its docs page (metering) from its pricing page (read-unit rates), since the docs page does not show the rates.

  • corrected
    Bootstrapped Software Valuation Multiples

    Cited BizBuySell's software, app and SaaS average as 3.41x SDE for 2025 and IT and software services as 2.99x; BizBuySell's sector pages now report averages for businesses sold from 2021 through 2025, 3.31x and 3.10x. The worked example is recomputed: SDE value $926,800 (was $954,800) and a blended midpoint of $903,400 (was $917,400) on a $722,720 to $1,084,080 range.

  • corrected
    How to Value a Small Business: SDE vs EBITDA Multiples

    The sector multiples were given as BizBuySell 2025 averages (service 2.70x SDE, retail 2.74x, restaurants 2.25x, IT services 2.99x, software 3.41x); BizBuySell's sector pages now report averages for businesses sold from 2021 through 2025: service 2.58x, retail 2.62x, restaurants 2.15x, IT services 3.10x, websites and e-commerce 3.32x, software 3.31x. Each figure now cites its sector page.

  • corrected
    How to Value a Business 2026: Methods, Multiples & Examples

    Updated the software SDE multiple to BizBuySell's 2021-2025 sector average of 3.31x (was 3.41x for 2025): the SaaS example's SDE value is $595,800 (was $613,800) and its blend is recomputed. The service and restaurant size statements now cite BizBuySell's sector pages.

  • corrected
    Business Valuation Calculator

    The sector table showed BizBuySell 2025 averages (service 2.70x SDE, retail 2.74x, restaurants 2.25x, IT services 2.99x, software 3.41x); it now shows BizBuySell's 2021-2025 sector averages (2.58x, 2.62x, 2.15x, 3.10x, 3.32x for websites and e-commerce, 3.31x) with their revenue multiples, and the examples and FAQ use them.

  • corrected
    Supabase vs Firebase vs Neon 2026 Compared

    Firestore's read price ($0.03 per 100,000 reads in us-central1) was cited to Firebase's pricing page, which does not show it; it now also cites Google Cloud's Firestore pricing, which lists reads at $0.03, writes at $0.09 and deletes at $0.01 per 100,000 documents.

  • corrected
    Supabase Free Tier Limits 2026: pgvector Cost and Pausing

    Said the 500 MB free database holds 50,000 to 80,000 vectors at 1,536 dimensions, while its own figures (about 12 KB per vector with an HNSW index) give about 40,000 before application tables; and called a 50,000-vector prototype a comfortable free-tier fit using an engine footprint that excludes the index. Sizing now says roughly 40,000 at full precision (about twice that with halfvec), and the description says ~40k.

  • corrected
    Embeddings DB vs Self-Host pgvector: True Cost

    Contradicted its own engine output: the small-scale heading said managed vendors win on cost while pgvector and object storage were cheaper, and the 1M-vector section said the managed minimums were not yet cleared while Pinecone's reads had put it at $111.12. Headings and text now match the engine; the annual comparison uses the scenario figures (Pinecone up to about $1,333 a year, not $600); the self-hosting hours are labelled as assumptions; LanceDB on R2 is described as a library you run, not a managed store; an unsourced "95% of cases" is removed.

  • corrected
    RAG vs Fine-Tune Total Cost for a Solo Founder

    The summary said storage at 20 million vectors spans "$58.61 to $96.41/month" and the decision rule "$5 to $96/month across every scale", figures from an earlier engine version; the engine output on the page gives $17 to $64 for the storage-priced options and about $6,920 for Pinecone, whose reads scale with index size. Both sentences now match the engine. Removed an unsourced "low hundreds of dollars" figure for a fine-tuning run.

  • corrected
    Wholesale Pricing Tools With Automated Updates (2026)

    Attributed wholesale margin bands for beauty, home goods, food and electronics, and a "30% margin floor", to AIMS360's apparel pricing guide, which contains neither. The page now quotes what that guide does say for apparel (direct-to-consumer 60-75% gross margin, wholesale 40-55%, marketplace fees 15-30%) and no longer gives invented bands for other categories.

  • corrected
    Pricing Audit Checklist for SaaS (2026)

    None of its three headline figures was on the cited page: the Kaplan Group roundup does not contain the "28% annual discount in 2025 vs 15% in 2022", and the Monetizely article contains neither the "1% price rise lifts profit 11%" figure nor the "inelastic up to a 40% increase" claim. Those claims and references are removed. The leakage estimate is now attributed to Lago as a billing vendor's estimate (1-5% of ARR), and the elasticity section shows the arithmetic instead (at a 10% operating margin, a 1% price rise with no lost volume raises operating profit 10%).

  • corrected
    B2B vs B2C SaaS Pricing in 2026

    Said B2C ARPU is "often $20 to $50 per month" with no source (its own FAQ said "single-digit-dollar B2C ARPU"), and called a $35K B2B contract against a "$30-a-year" consumer subscription three orders of magnitude apart. The B2C side is now described without an invented range and anchored on the article's own $12 engine example: a $35K contract is about 240 times a $144-a-year subscription.

  • corrected
    AI COGS Accounting: A Clean 2026 Method

    Its tax paragraph had the mechanics wrong: it said capitalizing prompt work reduces the current-year tax deduction and raises gross margin. Book capitalization does not set the tax treatment; for US tax, software development is research expenditure deductible in the year incurred for tax years beginning after 2024 (IRC section 174A), and prompt work is not COGS, so it never affects gross margin. The prompt section prescribed capitalizing prompts and evals with invented useful lives (12-24 and 18-36 months); it now sets out the actual choice (expense, or capitalize under ASC 350-40 as amended by ASU 2025-06 if the criteria are met). Removed: invented COGS mixes for SaaS and AI products, a "10-25% valuation premium" for clean accounting, EDGAR and ChartMogul citations for claims those sources do not make, and a "median" software gross margin (Damodaran's 71.72% is an aggregate across 309 firms).

  • corrected
    The AI Micro-SaaS Unit Economics Report 2026

    Said its prices were list rates accessed on 2026-05-26 while using models released in September and October 2026; model prices come from the model catalog (verified 2026-10-09), and the hosting, auth and monitoring prices were re-checked on 2026-10-19 (Supabase Pro $25, Sentry Team $26, Vercel Pro $20 confirmed). The CAC payback result was called Excellent "by SaaS benchmark standards"; those grades are the calculator's own bands, now labelled as such beside David Skok's 12-month and 3x rules of thumb.

  • corrected
    Build vs Buy in 2026: When the LLM API Kills Your Moat

    Attributed to Bessemer's State of the Cloud 2024 a split of 110-130% net dollar retention for AI products with strong data moats against 75-95% for weak ones; the report contains no such split, so the claim and the reference are removed. Also removed unsourced figures (one competing foundation-vendor feature per quarter, a 12-18 month moat horizon for public data, a 6-24 month window before commoditization) and unsourced outcome claims about Notion AI users and custom-GPT startups, and replaced "next year's GPT-5" (OpenAI is on GPT-6).

  • corrected
    Vendor Lock-In Math: The Real Cost of Switching LLMs

    Rebuilt. Two citations did not support what was attributed to them: the Replit link is a page about Ghostwriter that does not mention Claude or long-context performance, and the Stripe link is the engineering blog index, not a post on LLM provider abstractions; both cases are removed. The "$15,000 to $200,000", "5-10x underestimate" and "5-10x lower incident severity" figures had no source and are removed; the cost of a switch is now priced with the LLM vendor lock-in calculator on stated inputs ($16,133, 6.7-month payback in its example). The prompt section claimed Claude works best with XML and GPT with JSON schemas; both vendors now document JSON-schema structured outputs, so the page cites their docs instead. Mid-tier list prices are read from the model catalog (Claude Sonnet 5.5 and GPT-6.1 Sol are both $2/$10 per million tokens).

  • corrected
    Marketing ROI Formula: How to Calculate ROI on Marketing

    Cited FTC advertising guidance for an attribution-window rule ("at least 2x the median sales cycle") and for a closing remark; the FTC page covers neither, so both citations and the invented ratio are removed. The data-driven attribution reference was a Google Research search-results URL; it is now Shao and Li's KDD 2011 paper on data-driven multi-touch attribution. Softened an unsourced claim that brand investments are often a company's highest-return spend and an unsourced 24-month payback cut-off.

  • corrected
    How to Value a Small Business: SDE vs EBITDA Multiples

    The deal-structure example did not reproduce: on its own terms (monthly note payments at 7%, a two-instalment earn-out, 10% discount rate) the cash-heavy offer is worth about $1.73M and the earn-out offer $1.67M (full earn-out) or $1.51M (60%), so the gap is $60k-$215k, not $130k-$290k; the assumptions are now stated. Removed: a claim that sellers "routinely receive 10-20% less economic value" cited to IRS Revenue Ruling 59-60, which says nothing of the kind, and unsourced 20-30% price-cut and 9-month figures. Pratt's Stats is now named DealStats.

  • corrected
    How to Run an A/B Test: Sample Size, Duration, Results

    Attributed a "Type I error above 25% from daily peeking" simulation to the KDD 2017 peeking paper. The checkable statement is Johari, Pekelis and Walsh's: with continuous monitoring, even at 10,000 samples, Type I error "can easily increase fivefold" (arXiv:1512.04922, now cited); the guide now says that. Also removed: a claim that test-failure rates trace to vendor marketing, a US Census retail-survey page cited for day-of-week user behaviour (now Kohavi et al.), a statement that the Kohavi textbook is free to read, and an Ioannidis citation for a claim his paper does not make.

  • corrected
    Contractor vs Employee for a Solo Founder: 2026 Math

    Rebuilt. The engine example charged FUTA at $420 (the rate before the state credit; $42 after it) and health insurance from an unsourced range; it now uses $42 and the KFF 2025 employer share ($7,885), giving $145,447. The "1,400-hour rule" contradicted the page's own arithmetic ($147,860 ÷ $95 = 1,556 hours); the break-even is now about 1,530 hours. Also removed: a claim that US employers owe 1-2 months' severance (US law generally requires none), an onboarding example whose arithmetic gave 73 productive hours instead of 138, an equity paragraph that reversed its own conclusion, and unsourced optionality and founder-time values. Classification exposure is now computed from IRC section 3509.

  • corrected
    FTE vs Contractor: The Real Cost of a $90k Hire

    Rebuilt. It applied BLS's "benefits are 29.6% of total compensation" as a markup on wages (that share implies about 42% on wages, and about 24-29% on a salary that already pays for leave), used 2024's social security wage base, quoted KFF's 2023 employer health contributions as 2024 medians, added paid time off as a cash cost on top of a salary that already pays for it, and relied on unsourced productivity, rate-inflation and misclassification-cost figures. The rebuilt article builds the $90k hire from 2026 statutory rates and KFF 2025 averages ($111,212, 1.24x), compares cost per working hour, and computes misclassification exposure from IRC section 3509.

  • corrected
    Best AI Customer Support Tools for Solopreneurs 2026

    Said Intercom's own guidance points to 40-50% AI resolution rates. Intercom reports that Fin customers average a 76% resolution rate across 12,000+ businesses. Because Fin bills per resolution, a higher rate means a larger bill, so the worked example now shows the 400-conversation case at 45% and at 76% ($178 to $301 a month).

  • corrected
    Bootstrap vs Raise: A Decision Tree for Founders

    Rebuilt. The previous version cited a MicroConf survey for bootstrapped month-on-month growth distributions that the report's public page does not show (the report is behind an email form), cited a 2009 Kauffman report on entrepreneur demographics for a "$100M+ exit" requirement it does not contain, attributed founder-ownership medians and a 25% (elsewhere 35%) Series A rate to Carta without a checkable source, and computed compound growth "from $0". The rebuilt guide keeps the dilution arithmetic (YC's post-money SAFE: amount ÷ cap), computes the growth a round commits you to from a stated starting MRR, and uses Carta's two-year seed-to-Series-A rates as reported by Konvoy (30.6% for Q1 2018 seeds, 15.4% for Q1 2022). Unsourced build-cost, audience, legal-cost and sale-multiple figures were removed.

  • corrected
    Chargebee Pricing 2026: Flow Plan Fees and Cost by Volume

    Said Stripe Billing does not replace Chargebee's multi-gateway support. Stripe documents billing subscriptions through third-party processors, but you build that integration yourself; the page now says so with Stripe's guide as the source. The tax note now says what a merchant of record is (the legal seller that collects and remits tax) rather than that Chargebee does not handle tax.

  • corrected
    Resend vs Postmark vs SendGrid Pricing 2026 Compared

    Priced Postmark at 50,000 emails as $87 a month of overage on its 10,000-email plan. Postmark's Basic plan has a 50,000-email tier at $55 a month (from its pricing page's own volume table), still above Resend's and SendGrid's ~$20. The per-thousand comparison now uses that tier (about $1.10 against $0.40, not "four times"), and "sized for a tenth of the volume" is corrected to a fifth.

  • corrected
    Bunny.net vs Cloudflare vs Fastly CDN Pricing 2026

    Called Cloudflare "the only permanent free CDN of the three" and Fastly's allowance "a starter credit": Fastly's pricing page lists a standing free tier of 100 GB of bandwidth and 1 million requests a month. Cloudflare is now described as the only one with unmetered free bandwidth. The page also placed the 100% uptime SLA on the Pro plan; Cloudflare's plans table gives it on Business and Enterprise only.

  • corrected
    Baseten vs Modal vs Replicate Pricing 2026: GPU Costs

    Said Baseten is "the only one of the three" pairing per-token LLM endpoints with dedicated GPUs, while the same article shows Replicate doing both. Now says both offer the pairing and compares their per-token examples, noting they are different models.

  • corrected
    Best AI Agent Frameworks for Solo Founders 2026

    Listed AutoGen as one of three credible open-source frameworks for 2026. Microsoft has archived AutoGen and no longer fixes it, including for security; its successor, Microsoft Agent Framework (MIT, Python and .NET, 1.0 since April 2026), replaces it in the shortlist, with a note for existing AutoGen users.

  • corrected
    LangChain vs CrewAI vs AutoGen 2026: AutoGen Is Archived

    Recommended AutoGen as a live option. Microsoft's AutoGen README now marks the project archived and no longer maintained (no bug fixes, security fixes or support) and points new users to its successor, Microsoft Agent Framework, which reached 1.0 for Python and .NET on April 2, 2026. The comparison is now LangGraph, CrewAI and Microsoft Agent Framework, with what happened to AutoGen and a migration note; status checked on GitHub on 2026-10-19.

  • corrected
    Solo Founder Tax Planning

    The guide ran on 2024 rules and several of its figures were wrong. It is now recomputed on 2026 federal rules ($184,500 wage base, $16,100 standard deduction, 2026 brackets) with the QBI deduction, which the S-corp salary shrinks: the election starts paying at about $66,000 of profit, not $70,000-$90,000, and saves about $1,100 at $100,000 and $6,300 at $200,000. The payroll-tax table had wrong rows at $150,000 (SE tax is $21,194, not $19,887) and $200,000 (it used a $90,000 salary in a 50% table). It called a 40-60% salary ratio an IRS safe harbor; there is none, and comparable wages decide. California's 1.5% S-corp tax was said to wipe out savings below $150,000; LLCs there already pay $800, so the extra cost is 1.5% of S-corp net income above about $53,000. Removed as unsourced or wrong: IRS case data on salaries below 25% of distributions, a 3-5x audit cost multiple, higher audit rates for self-prepared returns, an $800 LLC fee in Texas and Florida, and the 14.1% rate explained as the half-SE deduction (it comes from the 92.35% factor). The half-deduction now saves $680-$1,240 at $80,000, not $1,300-$1,800.

  • corrected
    Freelance rates by role: the 2026 BLS-verified US table

    Updated to the BLS ECEC June 2026 release ($46.89 total compensation and $32.82 wages per hour worked, private industry), replacing March 2026; the multiplier stays 1.43, and the accountant floor moves from $58 to $57 an hour. All 45 OEWS May 2025 percentile wages were re-checked against the BLS API on 2026-10-19 and match.

  • corrected
    Contractor vs Employee Calculator

    The warning for salaries above the Social Security wage base used the 2025 base of $176,100; it now uses the 2026 base of $184,500, so salaries between the two no longer trigger it.

  • corrected
    Startup Cost Estimator

    The how-to said real launches almost always surface missed costs; it now says often.

  • corrected
    Project Pricing Calculator

    The how-to said fixed-fee work almost always produces overruns; it now says often.

  • corrected
    Monthly Burn Rate Calculator

    The how-to named the most common burn mistake and said most teams should already be raising at six months of runway, without sources; both are reworded as advice.

  • corrected
    Marketing Budget Allocator

    The how-to called over-allocating to an unproven channel the most common way a budget underperforms, without a source; the superlative is removed.

  • corrected
    Landing Page Conversion Calculator

    The how-to said a conversion fix will usually be cheaper and more durable than more traffic, without a source; it now tells you to compare the cost of each.

  • corrected
    Solo Founder Unit Economics

    The tool told founders with organic acquisition to enter $0 CAC and called the resulting LTV pure upside, contradicting our true-CAC guide: organic acquisition costs founder time and tools. The insight and the how-to now say to price that time.

  • corrected
    Net Promoter Score (NPS) Calculator

    The how-to presented the score bands (70+ Excellent, 30-69 Good, 0-29 Needs Improvement) as a general guide; they are this tool's classification, now labelled as such with a pointer to industry benchmarks.

  • corrected
    Invoice Late Fee & Interest Calculator

    The how-to said grace periods of 5-10 days are common, without a source; the range is removed.

  • corrected
    Cash Conversion Cycle Calculator

    The how-to called a cash conversion cycle above 60 days a common warning zone without a source; it now says the longer the cycle, the more financing pressure it creates.

  • corrected
    CAC Calculator

    The how-to gave LTV:CAC of 3 and an 18-month payback as rules of thumb without a source and said a ratio above 5 can mean under-spending; it now gives Skok's SaaS Metrics 2.0 guidelines (above 3, payback within 12 months), and the high-ratio point is framed as a channel that could take more budget.

  • corrected
    Burn Multiple Calculator

    The how-to presented five burn-multiple bands as David Sacks's; they are this tool's grading, and Sacks's own anchor (2x reasonable for an early-stage startup) is now stated beside them.

  • corrected
    Bootstrapped Runway Calculator

    The insight printed the gap between savings running out and MRR covering expenses as an unrounded number (for example 3.3000000000000007 months); it is now rounded to one decimal. An unsourced claim in the how-to that many successful bootstrapped companies came within weeks of running out of savings is removed.

  • corrected
    Ad Spend / ROAS Calculator

    When product cost per order reached revenue per order, the calculator reported a break-even ROAS of Infinity (and a negative one above it) and advised that ROAS was below break-even of Infinity×. Break-even is now reported as not reachable, with guidance to raise the price or cut product cost before spending on ads.

  • corrected
    Break-Even Units Calculator

    The FAQ said mature businesses often target a 25-40% margin of safety; no source supports the range, and it is removed. The how-to's contribution-margin and capacity thresholds are now labelled as rules of thumb.

  • corrected
    Unit Economics Calculator

    The FAQ said an LTV:CAC above 5:1 may indicate under-investment in growth; Skok's SaaS Metrics 2.0, the source of the 3:1 guideline, says the best SaaS businesses reach 7:1 or 8:1 and makes no such claim, so it is removed.

  • corrected
    Customer Lifetime Value Calculator

    The FAQ said an LTV:CAC above 5:1 may indicate under-investment in growth; Skok's SaaS Metrics 2.0, the usual source for the 3:1 guideline, says the best SaaS businesses reach 7:1 or 8:1 and makes no such claim, so it is removed. The FAQ also listed snake_case output names for agents; the contract returns marginAdjustedClv, ltcRatio and paybackMonths. The how-to now attributes the 3:1 and 12-month guidelines to Skok and replaces "often improves CLV more cheaply" with the arithmetic (a third more from 3 to 4 purchases a year, half more from 2 to 3 years).

  • corrected
    CAC Payback Period Calculator

    The FAQ presented this tool's grading bands (6, 12 and 18 months; 5x LTV:CAC as room to scale aggressively) as SaaS norms. Skok's SaaS Metrics 2.0 sets 12 months as the payback line and LTV:CAC above 3, with the best SaaS businesses recovering CAC in 5-7 months and reaching 7-8x; the FAQ now gives his figures and labels the other bands as this tool's own.

  • corrected
    Deel vs Rippling EOR Pricing 2026 Compared

    The page put employer contributions on top of an EOR fee at 20-40% of salary without a source; the range varies widely by country and is removed. The comparison table labelled Rippling's column as illustrative although it shows only what Rippling publishes (EOR is quote-only); the label now says so.

  • corrected
    Pricing AI Features: Cost-Plus vs Value-Based

    Token costs were priced at Claude Sonnet's superseded $3 per million input tokens, and the two examples used different bases (a blended $9 per million in one table, input-only in the other). Both are now computed at build time from the current verified catalog: Claude Sonnet 5.5 at $2/$10, with the token mix stated (median coding user $10.80 a month, not $27; writing tiers at 94-98.5% gross margin). The Opus-to-Haiku cost ratio is now 40x at current prices, not 5x. Claims attributed to Patrick Campbell's ProfitWell research could not be traced and were removed: price-increase churn of 5-15% versus 20-40%, four pricing rules, value-based pricing as the highest-revenue model, and a 5-15% conversion loss from bad bundles. Also removed: a distribution of pricing models and a 20-40% expansion gap cited to our own illustrative page and Bessemer, revenue-per-customer ranges by model, and a 5-10x model-price fall in 18 months, replaced by the catalog's figure (Sonnet input price down 33% from Sonnet 4.5 to 5.5). Annual discounts now show their implied cost of capital.

  • corrected
    The Working Capital Trap

    Merged into Cash Conversion Cycle Calculator on 2026-10-21.

    The worked example did not follow from its inputs. With CAC of $1,500 and an 8-month payback, the monthly-billing scenario is not a cash trap: its first-quarter burn multiple is about 0.3, not 3.2. The annual-prepay scenario reported $1.58M of cash collected, $770,000 of deferred revenue and recognized revenue of $810,000, none of which its own inputs can produce (they allow about $1.0M collected, and a 15% discount on 80% of revenue gives $792,000), so the $734,000 year-end gap was overstated about sevenfold. The example is rebuilt as a month-by-month simulation with a $3,000 CAC (16-month payback): monthly billing runs out of cash in month 12, 80% annual billing ends with $98,000, a gap of about $105,000. The summary claim of three to five more months of cash is replaced accordingly. Benchmarks cited to our own illustrative calculator pages and Bessemer were removed (CAC payback of 12-24 months, annual-billing shares by segment, multi-year contract shares), as were unsourced conversion rates to annual, quarterly-prepay and dunning recovery figures, and burn-multiple bands attributed to Sacks, which now use his own anchors. Prepay discounts now show their implied cost of capital.

  • corrected
    How to Calculate True Customer Acquisition Cost

    Merged into CAC Calculator on 2026-10-21.

    The summary quoted a true CAC of $543 against a $260 headline, but the worked example computes $808 against $234; the summary now matches. The per-channel CACs left $6,180 of costs unallocated (LinkedIn ads, demo calls and partnerships), so they could not average back to the blended figure; channel CAC is now direct cost plus a $195 share of demos, partnerships and trial servicing per customer (paid ads $683, content $855, outbound $1,264, referrals $523), and the reallocation advice now follows those numbers rather than recommending partnerships, which acquired no customers. The salary method applied a 1.3x loading to a figure already described as fully loaded; it is now a base salary. Payback benchmarks (12-18 months SMB) and a claim about channel and CLV variance were cited to our own illustrative calculator pages, and a Bessemer citation did not support its sentence; all are removed. Unsourced CAC ranges by channel and a 2-4x headline gap were removed, and two arithmetic slips fixed: an hour of community work every working day at $75 is about $19,500 a year, not $26,000, and ad spend alone is $140 per customer, not $80.

  • corrected
    Customer Concentration Risk in B2B AI: The 20% Trap

    Merged into Cash Flow & Finance Calculators on 2026-10-21.

    A table of investor valuation haircuts by concentration level (up to 40-60%) was attributed to Carta and Bessemer 2024 data that does not contain it, and with it claims that a 35% customer costs 1.5-2x in dilution, that investors assume a 25-35% chance of losing that customer, and that haircuts compound across rounds; all are removed, and the page now describes how diligence discounts concentrated revenue without inventing a schedule. An Andreessen Horowitz article from January 2023 was cited for top-3 concentration of 50-70% at 2023-2025 AI startups, which it does not report. Claims cited to our own illustrative calculator pages were removed, as were unsourced default churn rates by segment, a 60-70% churn signal from usage drops, an HHI threshold and an SEC page reference. The formal anchor is now the accounting rule: public companies must disclose any customer providing 10% or more of revenue (ASC 280-10-50-42). In the worked example the expected-loss gap is $106k, and the SMB tail accounts for $61k of it, not $53k.

  • corrected
    SaaS Churn: Cohort vs Aggregate vs Net Dollar

    The worked example did not add up: its ending MRR of $87,400 does not follow from its inputs (they give $76,800), all 55 lost customers came from the starting cohort so the 180 new customers never churned, and it annualized aggregate churn linearly (17.7%) while the page warns against linear annualization. The example is rebuilt with new customers churning faster than mature ones, as the page describes: aggregate churn is 2.76% a month against 2.05% for the January cohort (not 2.04%), so the claim that aggregate churn is structurally too low is replaced with the accurate one: a blended rate depends on customer mix and fits no real cohort. Benchmarks cited to our own illustrative calculator pages and to Bessemer were removed: NDR quartiles for public cloud companies, SMB NDR of 95-102%, cohort retention curves by segment, investor thresholds by stage, and a 30-60% LTV overstatement. Company-specific NDR figures for Snowflake, Datadog, MongoDB, Salesforce and HubSpot had no source and were removed, as was an unsourced six-month payback for expansion spend.

  • corrected
    Burn Multiple vs Runway

    The page attributed five burn-multiple bands to David Sacks, including calling 1.5-2x suspect; Sacks's essay calls 2x reasonable for an early-stage startup and treats 3x or more as a warning. The bands are now presented as this guide's reading around those anchors, with a stricter line near 1.0 for bootstrapped founders. Figures attributed to Carta and PitchBook-NVCA could not be verified and were removed: Series A runway of 21 months at close and 6 at the next raise, time between rounds of 18-24 months, and fundraises below 4 months of runway closing at 23% lower valuations. A top-quartile range of 0.4-0.7 and a category spread were cited to our own illustrative calculator page and were removed, as were unsourced claims about investor committees and the 2021-2022 cohort. In the worked example, a $150k extension could not lift cash from $90k to $215k after a quarter of burn; the extension is now $170k, which reconciles.

  • corrected
    The Five Numbers to Run Before Pricing Your SaaS

    In the closing worked example, support was costed at 8 minutes per user at $75 an hour, which is $10 per user, not the $1.50 used; the input is now 5 hours a month across 250 users, which is $1.50. The runway-sensitivity formula was a linear approximation that overstated large shocks: at 1,000 users the 20% cost shock removes 3.4 months of runway (24%), not 4.4 (31%), and the example's shocked runway is 7.3 months, not 7.0. The page said leaving support out overstates gross margin by 10-15 points; in its own example the effect is 7.5 points. Payback benchmarks (18-24 months mid-market, 12-18 SMB, 14-16 SMB drift) were cited to our own illustrative calculator pages and are removed. The Van Westendorp range was defined as the crossing of too cheap and too expensive, which is the optimal price point; it now runs from where too cheap crosses expensive to where too expensive crosses bargain. A quotation attributed to Nagle that we could not verify is now a paraphrase, and unsourced fee ranges and a 30% LTV gap were removed.

  • corrected
    Free Trial to Paid Conversion

    Merged into Free Trial to Paid Target on 2026-10-21.

    Every benchmark on this page, including conversion by trial type, an eight-row table by AI tool category, an AI lift of 1.2-1.4x and solo-founder targets, came from ranges we compiled ourselves and cited to our own calculator pages; none was a measurement. The page is rebuilt on the SaaS Conversion Report by Kyle Poyar with ChartMogul and ProductLed (February 2026, 200 products): card-required trials convert about 30% of signups, with 25-35% rated good, and 4-6% is good for trials without a card. The AI-category table is removed because no public source measures it. Effect sizes for conversion levers (a 4-6x lift from requiring a card, 1.3-1.8x from milestones and faster time to value, 5-10% from longer trials) had no source and were removed, as were funnel shares such as 60-80% of signups never activating and a Bessemer figure on double-counted conversions that Bessemer does not publish. Userpilot's benchmark was cited for a 37.5% average activation rate and a 34% revenue effect it does not state; it reports 34.6% (product-led) and 41.6% (sales-led).

  • corrected
    How to Manage Cash Flow

    Merged into Cash Flow & Finance Calculators on 2026-10-21.

    The Fed's Small Business Credit Survey was cited for 42% of firms experiencing delayed customer payments; the survey's figures are 39% naming customers slow to pay as a payments challenge (2023 survey) and 51% citing uneven cash flow (2024 survey), now quoted. The JPMorgan Chase Institute figure of 27 cash buffer days came from its 2016 study; the Institute's 2025 data put the typical small business at 17.6 days. Claims the cited sources do not make were removed: that firms without monthly forecasting were disproportionately distressed, that credit availability had tightened against 2021-2022 (the survey shows approvals steady but below prepandemic levels, 52% fully approved), ACH and card fee ranges attributed to the American Bankers Association, and an SBA citation for lender behaviour. Unsourced figures were removed: a $50k-$500k credit line at prime plus 1-3 points, an 80/20 split of value, vendor terms most SaaS vendors accept, and a 20% forecast-error threshold. In the worked example the freed $102k adds 20 days of buffer (to about 58), not 18 (to 56), and factoring at 3% for 30 days is about 38% a year. The page's search description was also cut off mid-sentence and is rewritten.

  • corrected
    How to Set Up Pricing Tiers

    The worked example attributed two-thirds of its 57% ARR lift to the Business tier; recomputed from its own mix, the Growth tier adds $158k, Business $138k, and moving customers down to Starter costs $50k, so Growth does most of the work. A sensitivity check is added: if half the projected Growth customers stayed on Starter, the lift would be about 28%. Claims cited to our own illustrative calculator page were removed: value-metric expansion 1.5-2x faster, ARPU growing 20-30% faster with annual packaging reviews, 1-3% repricing churn and hybrid pricing among the fastest-growing SaaS. Also removed without a source: a 30-50% ARR gap between structures, founders pricing 20-40% below willingness to pay, and hidden enterprise pricing losing 30-50% of pipeline. The page called anchoring the most robust finding in behavioral pricing and cited an anchoring paper for the decoy effect; it now cites the decoy study itself (Huber, Payne and Puto, 1982), notes a replication that found weaker anchoring, and replaces the claim that more options reduce conversion with the choice-overload meta-analysis, which found an average effect close to zero.

  • corrected
    How to Set Freelance Rates

    Merged into Freelance Rate + Capacity Planner on 2026-10-21.

    The income-based example understated the rate needed. It charged self-employment tax on about $100k of net earnings when its own inputs imply about $178k, took retirement as 15% of gross while using $18,000, and sized the 10% buffer at $15,000; rebuilt on 2026 federal brackets, an $80k take-home needs about $190,000 of billings, or $146 an hour at 1,300 billable hours, not $165,000 and $127. The $95-to-$140 example now uses the same model: take-home is about $59k at $95 and $94k at $140 (previously $48k and $88k), and $128 an hour would meet the $85k target. BLS benchmarks were wrong: the San Francisco web developer median is $73.29 an hour (May 2025), not about $50, and the national software developer median is $65.38. The benefits multiplier is now the BLS ratio of 1.43 rather than a 1.35-1.50 range, and the benchmark now converts to billable hours, which shows $140 is about the median developer's employment cost per billable hour ($144), not a 60% premium. CPI figures were wrong: prices rose 9.5% from September 2023 to September 2026 (an 8.7% real cut on a flat rate), not 10%, and did not average 2.5-4% a year in 2020-2024. Unsourced figures for project premiums, retainer discounts, revenue mix, raise sizes and 12-15% wage growth were removed; wage growth is now the BLS Employment Cost Index (22.9%, June 2021 to June 2026).

  • corrected
    How to Run a Profitability Analysis

    The page said 60-80% of profit typically comes from 20% of the business, cited to Kaplan and Cooper. Kaplan's own finding points the other way: in customer-profitability whale curves the most profitable 20% of customers typically generate 150-300% of total profit (Kaplan and Narayanan, 2001), which is now cited. Gross-margin benchmarks (a 72-78% SaaS median, retail 30-45%, services 40-60%) and their 2026 trend were cited to our own illustrative calculator page; they are now Damodaran's January 2026 averages (software 71.7%, general retail 33.2%, business and consumer services 33.4%). Also removed without a source: SMB margins 30-50% below average, a $15k ACV floor for outbound sales, a 30% gross-margin discontinue rule and a $20M revenue threshold for full activity-based costing. In the worked example, revenue-share allocation leaves SMB with a $138k operating loss, not break-even, and the price-rise decision mixed ARR and margin: the extra churn costs $80k of ARR ($59k gross profit), the increase adds $144k of ARR ($107k), a net of about +$64k ARR and +$47k gross profit. The example table now shows the support and selling costs behind each segment's contribution.

  • corrected
    How to Reduce Customer Churn

    Merged into Churn & Retention Calculator on 2026-10-21.

    Retention benchmarks by contract size (about 95% gross revenue retention above $10k ACV, 75% under $100), R&D reinvestment by top retainers, a 2-4 point fall in retention medians since 2022 and a 2-3x efficiency edge for onboarding work were cited to our own illustrative calculator page and are removed. A Federal Reserve payments study was cited for 6-10% of cards being reissued each year, which it does not report. Also removed without a source: involuntary churn at 20-40% of total, dunning recovery of 60-70%, card updaters catching 10-20% of changes, and onboarding laggards churning at 3-5x. The confidence interval was wrong: with 500 accounts per cohort, a fall from 5% to 3% monthly churn is about plus or minus 2.4 points, not 1.3, so it could be no change at all. The worked example's $283k of preserved ARR and two-week payback did not follow from its inputs; with steady acquisition the fix preserves about $274k of year-end ARR and $160k of year-one revenue, and pays back in about a month. The 20%-of-ACV cap on save offers is now derived from the save formula, and the Fader and Hardie reference now points to their subscription-retention paper rather than a model for non-contractual customers.

  • corrected
    How to Reduce Burn Rate Without Killing Growth

    Merged into Monthly Burn Rate Calculator on 2026-10-21.

    Burn-multiple benchmarks (a 1.5x median, top quartile under 1.0x) and their trend since 2021 were cited to our own illustrative calculator page; they are replaced by David Sacks's own reading of the metric (2x reasonable early on, 3x or more a warning). Claims cited to Carta and Bessemer that those sources do not make were removed: 4-7 months of runway from operational levers, better raise terms after this sequence, and a 30-50% magic-number gain from ICP tightening. Unsourced figures were removed: 60-120 SaaS subscriptions with 20-30% unused, 15-25% renegotiation wins, discretionary spend at 3-8% of burn, and retention losses of 1-2 points. The BLS backfill cost multiplied a per-hour-worked figure by 2,080 paid hours and an average across all US workers; it is now salary times the BLS-based 1.3 loading. In the worked example, the hiring freeze is now stated as savings against plan, and the ICP lever counted a 20% close-rate gain as $45k of immediate monthly revenue; it is now a $45k cut to out-of-ICP outbound spend, so the runway result is unchanged. The churn cost is now computed: one extra point of monthly churn on $180k of monthly revenue costs about $135k in a year.

  • corrected
    How to Price a Product

    The Van Westendorp section labelled the wrong crossings: the point it called optimal (too expensive meets bargain) is the top of the acceptable range, and the one it called indifference (too cheap meets expensive) is the bottom. Recomputed on the worked example's own answers, the optimal price is about $37 (not $48), the acceptable range $31-$50 (not $19-$89), and the recommended first step stays $39; $49 is now a later test at the top of the range. Claims cited to our own illustrative benchmark page were removed: value-metric pricing growing expansion 1.5-2x faster, a 15-30% list-to-realised discount, founders able to raise prices 20-30%, and a majority of SaaS companies raising prices each year. Also removed without a source: a 20-50% undercharging pattern attributed to Nagle, the quotation of Nagle calling cost-plus the most common error (his circularity argument is now paraphrased), regional and academic discount ranges, and Van Westendorp sample-stability figures. CPI is now the BLS series itself: 3.7% in the 12 months to September 2026 and about 24% from 2020 to 2025, not 2-4% in most recent years. Prepayment discounts now show their implied cost of capital (10-15% off a year upfront is about 27-44% a year).

  • corrected
    How to Optimize Your Conversion Rate

    Three of the four A/B sample sizes were overstated for the stated 5% significance and 80% power: at a 2% baseline a 20% lift needs about 21,100 visitors per variation (not 24,000), a 10% lift 80,700 (not 95,000) and a 5% lift 315,000 (not 380,000). At 30,000 monthly visitors a 20% lift therefore takes about six weeks, not four, and a 5% lift about 21 months. The worked example needed about 22,600 visitors per arm, not 33,000, so the test takes 11 weeks rather than four months, and a 35% target needs about 4,500 per arm, not 6,000. Compounding a 3-5% lift across five funnel steps gives 16-28%, not 30-50%. The 85% five-user figure is Nielsen's (NN/g), not the Kohavi book it was cited to, and the peeking figure is now cited to the paper that states it (about fivefold). Win rates for CRO programs, a 2-4 week novelty window, and an industry baseline of 2-4 tests a year had no source and were removed; six post-hoc segments give a 26.5% chance of a false positive, which the page had described as nearly guaranteed.

  • corrected
    How to Forecast Revenue

    Merged into Sales Forecast Calculator: Monthly Sales Projection on 2026-10-21.

    Three of four sources were cited for claims they do not make: a 15-25% month-3 forecast error and a 3x pipeline-coverage floor (a 2015 First Round survey and a forecasting textbook), sales-cycle lengthening that overstates bookings by 15-30% and macro concerns predicting slowdowns (the Fed's Small Business Credit Survey), and higher board trust for ranged forecasts. Those claims and sources are removed; pipeline coverage is now derived from the close rate (3x at a one-in-three close rate). The P75 and P90 definitions said the actual comes in at or below them, the opposite of how the worked example used them; they now mean reached or beaten three and nine times in ten. The pipeline example called a cascade through all four stages one month's P50; the same $347k is the expected value of bookings over a full sales cycle. In the worked example the first stress case recomputes to $393k, not $410k, and both stress cases are now labelled as scenarios rather than percentiles. A rate-cycle effect of 20-30%, a 60-70% in-month close rate and a 25% MAPE threshold had no source and were removed.

  • corrected
    How to Decide: Hire or Contract?

    The guide described an IRS 20-factor test; current IRS guidance groups the evidence into behavioral control, financial control and type of relationship, and the 20-factor list is from a 1987 ruling. It presented the 2024 DOL rule as the enforced federal test; the DOL stopped applying it in its own investigations on May 1, 2025 (it still governs private lawsuits) and proposed a replacement on February 26, 2026. Unsourced figures were removed: misclassification back taxes of 50-150% of pay, ramp times of 1-2 weeks versus 3-6 months, EOR fees of 10-15% (Deel and Remote list per employee per month), entity setup costs, and enforcement data that did not exist. The loaded-cost range of 1.25-1.40x is now BLS June 2026 (1.24x-1.29x of a salary before overhead). New York was wrongly listed as an ABC-test state. Two cost comparisons billed the contractor for more hours than the employee delivered; at equal hours the full-time employee wins the design example by about $30k, not $40k, and in the 30-hour example part-time W-2 costs about $85 an hour (not $69), the contractor $121.1k (not $148.2k), and the idle full-time capacity about $37k. The rankings are unchanged.

  • corrected
    How to Create Financial Projections (3-Statement Model)

    The article said 31% of small firms saw revenue fall in the Fed's 2024 Small Business Credit Survey; the survey reports 39% (43% grew, 18% held flat). It also cited the survey for a 3-5% median growth rate it does not publish, and a Carta fundraising report for operating margins and growth ranges it does not contain; these are replaced by SaaS Capital's 2025 median growth (24%) and Damodaran's January 2026 gross margins. In the worked example the implied NRR is 150%, not 148%, and the row labelled NRR 108% subtracted churn a second time; it is now expansion 8% less churn 5%, an NRR of 103%. An SBA citation for scenario modeling and unsourced cost-scaling and CAC-decay ranges were removed.

  • corrected
    How to Build a Retention Playbook

    Merged into Churn & Retention Calculator on 2026-10-21.

    The benchmarks cited two of the site's own calculator pages labelled illustrative (median NRR 102-108% or 105%, 96% for SMB, gross retention of 95% versus 75% by price, a 3-4 point decline since 2022); they are replaced by SaaS Capital's 2025 survey (median NRR 101%, GRR 91%, by ACV, funding and contract length). The Bain claim is now stated as the cited brief says it (financial services, more than 25%), not as a 1990 paper. Unsourced figures for involuntary churn share, dunning recovery, detractor churn multiples, win-back rates and a CS-mix NRR gain were removed.

  • corrected
    How to Calculate Customer Lifetime Value

    Merged into Customer Lifetime Value Calculator on 2026-10-21.

    Two references were the site's own calculator pages labelled as illustrative benchmark ranges; the payback and margin benchmarks citing them are replaced by David Skok's 12-month payback guideline and Damodaran's January 2026 software gross margin (71.7%). The worked example's honest CLV is about $2,000, not $2,150 (LTV:CAC 2.2x, not 2.4x), and a 3:1 revenue-based ratio is 2.25:1 on a 75% margin, not 1.5:1. Discount-rate ranges attributed to Gupta et al., channel and tier multipliers, and dated trend claims had no source and were removed.

  • corrected
    How to Calculate Your Break-Even Point

    Merged into Break-Even Units Calculator on 2026-10-21.

    The worked example said cutting variable cost from $32 to $26 brings break-even to 937 users (25%); at $60,000 of fixed costs it is 1,112 (11%), as the TL;DR already said, and the conclusion built on 937 (the cost lever beats price) was wrong: dollar for dollar the two move contribution identically. Removed: a PPI 3-8% inflation figure and a 2026 PPI trend attributed to the PPI methodology page, a price-elasticity range attributed to a cost-accounting textbook, a FASB citation for a definition it does not give, and an unsupported claim that cash break-even needs 30-50% more volume.

  • corrected
    Cheapest AI Video API 2026: Per-Second Costs Ranked

    The ranking was from May 2026: it named Runway Gen-4 Turbo ($0.05/s) the cheapest credible API and said Pika has no self-serve API. Runway's API now resells cheaper models (Grok Imagine 1.5 Lite from $0.02/s at 480p), Pika 2.2 is on fal.ai at $0.04/s, Luma's Ray 3.2 API starts at $0.012/s for 360p drafts, and fal.ai retires Luma Ray 2 on October 25, 2026. Re-ranked on prices checked October 19; the page's two verification dates had also disagreed (May 26 and September 3).

  • clarified
    Employee Cost Calculator

    Clarifying an earlier entry today: the how-to's 1.24x-1.29x was the BLS figure on a salary basis (wages plus paid leave, as derived in our loaded-cost article), not a contradiction of the FAQ's 1.43x on a wage basis. The step now gives both bases.

  • corrected
    Calculator source lists

    Source lists on 21 calculators named reports the tools never used, often as "benchmarks" or "percentiles" (OpenView, SaaS Capital, First Round, Carta, Bessemer, Gartner, HubSpot and others). Each list now names only what the tool relies on, which for most is arithmetic on your inputs. The llms.txt claim that API responses include benchmark percentiles from those reports was false and is removed.

  • corrected
    Commute vs Remote Cost & Time Calculator

    The default driving cost was the IRS 2024 business rate of 67 cents a mile; it is now the current 76 cents (from July 1, 2026), with a note that the rate includes depreciation and insurance. The source list no longer names Census and BLS commuting surveys the tool does not use.

  • corrected
    Commute vs Remote Examples

    Merged into Commute vs Remote Cost & Time Calculator on 2026-10-21.

    Recomputed at the current IRS rate of 76 cents a mile: going hybrid now saves $4,723 a year rather than $4,334. Two scenarios described a 70- or 60-minute round trip as minutes each way.

  • corrected
    Commute Savings Formula

    Merged into Commute vs Remote Cost & Time Calculator on 2026-10-21.

    The worked example used the IRS 2024 rate of 67 cents a mile; at the current 76 cents the daily cost is $32.80 and the hybrid saving $4,723 a year. A claim that time cost often outweighs fuel was replaced by the example's own figures.

  • corrected
    AI vs Human Support Cost

    The methodology put US customer service wages around $19 an hour (2024) and loaded cost at 1.4-1.8x wage with no source, and cited Zendesk CX Trends for handle-time and deflection benchmarks it does not publish. It now gives BLS's $21.53 median (May 2025) and the 43% private-industry benefits load (June 2026); an unsourced "2-10x more tokens" claim was removed.

  • corrected
    Free Trial to Paid Target

    The page attributed a trial conversion benchmark (5-15% for B2B, 2-8% for consumer) to OpenView that we could not verify, and coloured results easy, typical or hard on it. The benchmark and bands are removed: the chart shows the required rate with 10% fewer and 10% more signups, and the colour flags only a target that needs more than every trial to pay.

  • corrected
    Revenue Per Employee Calculator

    Unsourced per-employee ranges (software $200k-$500k, services $100k-$300k) are replaced by 2022 Economic Census figures: about $477,000 for software publishers, $266,000 for computer systems design, $243,000 for management consulting and $159,000 for accounting and bookkeeping.

  • corrected
    Project Pricing Calculator

    A 10-25% risk-buffer range was attributed to PMI guidance that gives no such range; the buffer is now described as a contingency reserve set from how far your own estimates have missed.

  • corrected
    Employee Cost Calculator

    The source list attributed "benefits run 25-40% of base salary" to an SHRM benchmark we could not find; it is removed. The page's figure is BLS's: benefits about 43% on top of private-industry wages (June 2026).

  • corrected
    Contractor vs Employee Calculator

    The source list attributed "benefits run 25-40% of base salary" to an SHRM benchmark we could not find; it is removed.

  • corrected
    CAC Payback Period Calculator

    The 3:1 LTV:CAC rule was attributed to a16z's 16 Startup Metrics; it and the 12-month payback target are now attributed to David Skok's SaaS Metrics 2.0, which states both.

  • corrected
    Unit Economics Calculator

    Removed a citation of a16z's 16 Startup Metrics for the 3:1 rule; the rule and the 12-month payback guideline are attributed to David Skok's SaaS Metrics 2.0.

  • clarified
    Landing Page Conversion Calculator

    The 5% and 1% conversion guides are now labelled as our convention, not a measured benchmark.

  • clarified
    Business Valuation Calculator

    The 20-25% low/high spread is now described as a sensitivity convention rather than a typical buyer bargaining range.

  • corrected
    Sales Forecast Examples

    The baseline insight said compounding growth, not the pipeline, drove the forecast; the figures show won deals added $126,720 of year-end MRR against $105,033 from growth. A key insight made the same reversal, and another called the growth-rate case the biggest effect when better conversion added about $348,000 to its $94,000.

  • corrected
    Scope Creep Cost Examples

    Merged into Scope Creep Cost Calculator on 2026-10-21.

    An insight said nearly doubling the overrun halved the effective rate and that scope creep compounds; the rate fell 20%, from $69.44 to $55.56.

  • corrected
    Net Promoter Score Examples

    Merged into Net Promoter Score (NPS) Calculator on 2026-10-21.

    An insight said adding promoters moves NPS faster than anything else, which the page's own detractor example contradicts: with a positive score, a new detractor costs about 0.7 points and a new promoter adds about 0.3.

  • corrected
    Hourly to Salary Examples

    Merged into Hourly to Salary Converter on 2026-10-21.

    An insight put the break-even hourly rate near $37; on annual cash it is $34.62 ($72,000 ÷ 2,080 hours), and per hour actually worked the salary pays $39.13. Both are now stated.

  • corrected
    Salary Examples

    Merged into Salary / Paycheck Calculator on 2026-10-21.

    An insight credited uncapped Medicare for the higher tax on a raise; the reason is the 22% bracket plus 7.65% FICA. The 45-hour example now notes that most hourly employees are owed 1.5x pay after 40 hours, which would make the extra hours worth $14,820, not $9,880.

  • corrected
    Break-Even Examples

    Merged into Break-Even Units Calculator on 2026-10-21.

    An insight said a $20 rise in unit cost erased a third of the margin; contribution per unit fell 22%, from $90 to $70. An unsupported comparison with fixed-cost changes was removed.

  • corrected
    Commute vs Remote Examples

    Merged into Commute vs Remote Cost & Time Calculator on 2026-10-21.

    An insight said the reclaimed time was worth almost twice the cash saving; at the default value of time it is worth about $4,100, roughly the same as the $4,334 cash saving. Going fully remote raises the figures by two-thirds, not roughly double, and a garbled sentence was fixed.

  • corrected
    Time Zone Overlap Examples

    Merged into Time Zone Overlap Planner on 2026-10-21.

    An insight called New York and London five hours apart; on the example's date (10 March 2026) they are four, because the US had moved to daylight time and the UK had not.

  • corrected
    CAC Examples

    Merged into CAC Calculator on 2026-10-21.

    An insight said winning fewer customers matters exactly as much as spending more; 15% fewer wins raised CAC by 18%, against 15% for 15% more spend.

  • corrected
    CAC Payback Examples

    Merged into CAC Payback Period Calculator on 2026-10-21.

    An insight said a lower price hits payback twice as hard as a higher CAC; the 15% price cut added 4.3 months against 3.7 for the 15% cost rise.

  • corrected
    Customer Lifetime Value Examples

    Merged into Customer Lifetime Value Calculator on 2026-10-21.

    A key insight said cutting churn raises lifetime value disproportionately more than raising order size; in this model the drivers multiply, so each moves value in proportion.

  • corrected
    Business Valuation Examples

    Merged into Business Valuation Calculator (SDE & EBITDA) on 2026-10-21.

    An insight compared a 35% earnings rise with a 15% revenue rise as comparable, and another called a business with a 30% SDE margin low-margin. The comparison is now per dollar: each extra dollar of SDE added a dollar of value, each extra dollar of revenue 50 cents.

  • corrected
    Contractor vs Employee Examples

    Merged into Contractor vs Employee Calculator on 2026-10-21.

    An insight said a salary is fixed however many hours are worked, which holds only for exempt employees; a key insight called control the critical classification factor, where the IRS weighs behavioural control, financial control and the type of relationship.

  • corrected
    Employee Cost Examples

    Merged into Employee Cost Calculator: True Cost of an Employee on 2026-10-21.

    A key insight put fully loaded cost at 1.25x to 1.5x salary with no source; it now gives the BLS June 2026 figure of about 1.43x wages before overhead.

  • corrected
    Freelance Rate Examples

    Merged into Freelance Rate + Capacity Planner on 2026-10-21.

    An insight said freelancers routinely underprice by exactly the tax gap, and another that income goals raise the rate faster than billings; both were unsupported and are replaced by the arithmetic the examples show.

  • corrected
    Startup Runway Examples

    Merged into Startup Runway Calculator: Burn & Cash-Out on 2026-10-21.

    An insight said extra revenue bought less runway than cost cuts because revenue trails the burn; the cases differ in size ($1,750 against $3,750), and dollar for dollar the two work alike.

  • corrected
    Profit Margin Examples

    Merged into Profit Margin Calculator: Margin, Markup, Price on 2026-10-21.

    An insight described the higher-revenue case as a 15% sales lift; with cost of goods unchanged it behaves like a price rise, since more units would bring their own cost of goods.

  • corrected
    ROI Examples

    Merged into ROI + Payback Period Calculator on 2026-10-21.

    An insight said a higher annual benefit compounds over the horizon; it repeats each year, adding $24,500 over five years.

  • corrected
    Churn and Retention Examples

    Merged into Churn & Retention Calculator on 2026-10-21.

    An insight said the same retention gain delivers outsized returns on a larger base; the gain scales in proportion, 15% more for a 15% larger base.

  • corrected
    A/B Test Significance Examples

    Merged into A/B Test Significance Calculator on 2026-10-21.

    An insight called a 4% lift statistical noise at any realistic traffic level; it is detectable with about 190,000 visitors per arm.

  • corrected
    Meeting Cost Examples

    Merged into Meeting Cost Calculator on 2026-10-21.

    Removed unsourced claims that preparation and opportunity costs often far exceed salary costs and that trimming attendees is the single cheapest saving.

  • corrected
    Markup Examples

    Removed an unsourced claim that confusing margin and markup is the most common pricing error; the page now shows what the confusion costs.

  • corrected
    Inventory Turnover Calculator

    The industry ranges the calculator showed (grocery 14-20 turns, retail 8-12, e-commerce 6-10, manufacturing 4-8, wholesale 5-9, automotive 3-6) were editorial guesses, and its low/healthy/high rating applied one scale to every industry, so a typical clothing store read as low. The ranges are replaced by turnover computed from the Census Bureau's 2022 retail and wholesale trade surveys for 12 industries (grocery stores 13.0, clothing stores 2.9), and the rating now compares you with the industry you pick. The how-to steps, FAQ and examples quote the same figures.

  • corrected
    How to Calculate Inventory Turnover

    Merged into Inventory Turnover Calculator on 2026-10-21.

    Sector turnover figures attributed to Census data (grocery 14-20 turns, general retail 8-11, apparel 4-6) did not match it: the Census surveys put 2022 grocery stores at 13.0 and clothing stores at 2.9, and the article now shows a table computed from them. Also corrected: the GMROI formula (turnover × margin ÷ (1 − margin) when turnover uses COGS), the worked example's lost gross margin ($32k, not $21k, leaving 42.3% rather than 42.5%), and a 61-day DSI called a warning sign for fashion, where clothing stores average about 126 days. Unsourced claims about 2024 markdowns, 2026 overstock and typical operators were removed.

  • corrected
    Employee Cost Calculator

    A how-to step put BLS employer taxes and benefits at 1.24x-1.29x of pay, contradicting the FAQ's BLS figure of about 1.43x wages (June 2026). It now gives the BLS figure and explains why this calculator's defaults land at 1.30x for a $100,000 salary: BLS counts paid leave as a benefit, while the calculator keeps it inside salary.

  • corrected
    Ship-or-Kill Decision Score

    A how-to step credited research from a16z and First Round Capital for weighting distribution heavily, and listed channels and a competitor threshold the score does not use. It now describes the score as built (the market dimension is 15% of the total) and attributes the framing to Peter Thiel's Zero to One as an argument, not research. Two result messages no longer say distribution beats product every time or that incumbents always win on price, and an FAQ no longer calls its kill signals the most common ones.

  • corrected
    One-Person SaaS Valuation

    A how-to step and an FAQ described growth and churn as the strongest drivers of multiples in real micro-SaaS sales, and the FAQ said a track record under one year lowers the multiple. The tool's multiples are editorial, so the text now describes what moves them in this model, which has no penalty for a short track record.

  • corrected
    Micro-SaaS Pricing Engine

    Removed two unsourced how-to claims: that 70-85% is a common software margin target and that price testing every 6-12 months is best practice. The step now gives the tool's 70% default.

  • corrected
    Content Marketing Payback Calculator

    Removed an unsourced range (typically 3-6 months for SEO) for the months before content brings traffic.

  • corrected
    Model Price Drop Stress Test

    Removed an undated, unsourced claim that frontier model prices have fallen sharply over the past year; the stress test applies fixed drops and needs no forecast.

  • corrected
    Agent Cost Per Validated Customer

    Removed an undated, unsourced claim that token prices have fallen sharply over the past year.

  • corrected
    CAC Calculator

    An FAQ gave SaaS and e-commerce CAC payback benchmarks with no source; it now explains the calculator's 18-month warning, and 3:1 LTV:CAC is described as a rule of thumb rather than a standard.

  • corrected
    Net Promoter Score (NPS) Calculator

    An FAQ said industry NPS averages vary by 20-30 points; the Retently 2026 averages the calculator uses run from 26 to 68.

  • corrected
    Wholesale Pricing Calculator

    An FAQ said wholesale prices are usually 2-3x landed cost with no source; it now presents keystone (2x) as one convention alongside cost-plus and target-margin pricing.

  • corrected
    Inventory Turnover Calculator

    An FAQ gave turnover benchmarks by industry (grocery 14-20 turns, retail 8-12, manufacturing 4-8) with no source; the answer now explains why turnover varies and how to compare.

  • corrected
    Micro-SaaS Pricing Engine

    Removed an unsourced claim that most micro-SaaS founders under-price by 20-50%.

  • corrected
    Content Marketing Payback Calculator

    Removed unsourced payback ranges (12-36 months for SEO content, 6-12 for paid distribution).

  • corrected
    Statistical Significance

    Merged into A/B Test Significance Calculator on 2026-10-21.

    The page defined statistical significance as the probability that an observed difference is real rather than chance, the most common misreading of a p-value. It now defines the p-value as the probability of data at least this extreme if there were no effect, cites the ASA statement on p-values (Wasserstein & Lazar 2016), and gives the example's effect size. The Nature citation carried the wrong title; it is Amrhein, Greenland & McShane, 'Scientists rise up against statistical significance'.

  • corrected
    Async Communication

    Merged into Operations & HR Calculators on 2026-10-21.

    The worked example presented invented results as outcomes (a 95% documentation completion rate, 98% of deadlines met, 10 hours saved). It is now a labelled, illustrative calculation: replacing one weekly meeting of 8 people with written updates saves about 6 person-hours.

  • corrected
    Time Zone Overlap

    Merged into Time Zone Overlap Planner on 2026-10-21.

    The worked example was cut off mid-calculation ('Overlap End.') by the May 2026 description trimmer. It now completes the New York-Berlin calculation (2 hours, 3 for a few weeks a year when US and EU clock changes differ), and the daylight-saving answer no longer points to an unrelated commute calculator.

  • corrected
    Minimum Viable Experiment

    The example said the p-value was 'significantly less than 0.05' without giving it; a two-proportion z-test gives z ≈ 2.98 and p ≈ 0.003, and the result now also states the test's power (about 1.7 percentage points detectable at 80%).

  • corrected
    Product-Market Fit

    Merged into Startup Planning Calculators on 2026-10-21.

    The example's metrics are now labelled illustrative rather than presented as a company's results, a takeaway that had lost a word in the May 2026 wording codemod was rewritten, and the essay 'The Only Thing That Matters' is credited to Marc Andreessen's pmarchive, not Andreessen Horowitz.

  • corrected
    Gross Margin

    Merged into Profit Margin Calculator: Margin, Markup, Price on 2026-10-21.

    An FAQ answer claimed a high gross margin signifies product-market fit; it reflects pricing power and delivery cost, not demand. A cost line in the example was mislabelled 'per subscription' although it is a total.

  • corrected
    Net Revenue Retention

    Merged into MRR / ARR Growth Calculator on 2026-10-21.

    A wording codemod on May 10, 2026 deleted the word "crucial" from an FAQ answer without repairing the sentence ("...a strong feedback loop to product development are also for retaining and growing existing accounts"). The sentence is repaired, and a content check now fails the build on that shape.

  • corrected
    Stripe alternatives for SaaS (2026)

    The page gave Polar's fee as 4% plus $0.40 and called it the cheapest merchant of record. That is the Early Member rate, kept only by organizations created before May 27, 2026; a new organization starts on the free Starter plan at 5% plus $0.50, the same as Paddle and Lemon Squeezy, and pays $20 to $400 a month for Pro, Growth or Scale to cut the rate to between 3.8% plus $0.40 and 3.4% plus $0.30. The page also said subscription payments add 0.5% on Polar; that surcharge applies only to Early Member organizations. The fee table, pros, cons, FAQ and verdict now give the plan ladder and Polar's own break-even estimate for Pro (about $1,379 a month in sales).

  • corrected
    Pinecone alternatives (2026)

    The page said Weaviate Cloud had the cheapest managed entry tier at around $25 a month. Weaviate's paid Flex plan now starts at $45 a month, and it also runs a free tier of 100,000 objects with 1 GB of memory. The pricing, comparison table, FAQ answers and verdict now give the free tier and the $45 Flex minimum, and no longer call Weaviate the cheapest managed entry; Pinecone's $20-a-month Builder plan is the lowest flat paid entry of the three managed engines compared.

  • corrected
    Polar vs Creem 2026: Merchant of Record Fees Compared

    The comparison gave Polar's rate for new organizations as 5% plus 50 cents with no monthly fee and left out its paid plans: Pro at $20 a month for 3.8% plus 40 cents, Growth at $100 for 3.6% plus 35 cents and Scale at $400 for 3.4% plus 30 cents. On US cards those plans undercut Creem's 3.9% plus 40 cents once monthly sales pass about $12,500 to $20,000 at average sales of $10 or more. It also omitted that Creem lists no international card fee, while Polar adds 1.5% on non-US cards, which cancels Pro's advantage once about 1 in 15 sales are on non-US cards. The fee table, add-ons section, FAQ and decision list now include both.

  • corrected
    Dodo Payments vs Lemon Squeezy vs Polar 2026

    The comparison told new Polar sellers to budget for 5% plus 50 cents and left out Polar's paid plans (Pro at $20 a month for 3.8% plus 40 cents, down to 3.4% plus 30 cents on Scale). Pro undercuts Dodo's 4% plus 40 cents once monthly sales pass $10,000. The page also omitted Dodo's 0.5% subscription surcharge and $30 dispute fee, and described international card fees as roughly 1.5% when all three charge exactly 1.5%. With Dodo's surcharge, Pro is cheaper than Dodo on subscription sales from about $2,900 a month. The fee table, add-ons list and FAQ now include the plan ladder and the missing fees.

  • corrected

    Glossary pages (benchmarks)

    Six glossary pages gave benchmarks without a source: cart abandonment 'often 60-80%' (Baymard's average across 50 studies is 70.22%), e-commerce, lead-form and trial conversion rates, acceptable DSO by payment terms and for construction, net margins of 20-30% for software and 1-5% for grocers, NRR of 120% for enterprise and 90-110% for SMB SaaS, and 5-7 pipeline stages. The net-margin answer now uses Damodaran's January 2026 data (software 25.5%, pharmaceuticals 18.5%, grocery 1.3%, market 9.7%); the others now say there is no reliable single benchmark. A cart-abandonment citation carried a Baymard title on a Statista link.

  • corrected

    Formula pages (site-wide)

    On ten formula pages the step-by-step examples ran a second set of numbers that did not match the worked example or the calculator: break-even, business valuation, cash conversion cycle, customer lifetime value, hourly-to-salary, inventory turnover, LTV:CAC, meeting cost, profit margin and startup runway. For example, the break-even steps used a $50 price and $20 cost while the worked example and calculator used $150 and $60. The steps now restate the worked example's inputs and the calculator's outputs, and every worked example is re-run against the calculator in CI. The meeting-cost formula linked to the employee cost calculator; it now links to the meeting cost calculator. On 24 formula pages the 'common variations' list held generic advice instead of variants of the formula; each now names real variants, such as break-even revenue and cash break-even.

  • corrected

    Tool explanations (second pass)

    Nine FAQ answers and six how-to steps across eight calculators gave benchmarks with no source: email ROI of 3,600-4,200%, open rates of 20-25%, wholesale margins of 40-60%, SaaS gross margins of 60-85%, a 10-15% demand drop per 10% price rise, discount rates of 8-15%, a 25-40% loaded-cost uplift, and bootstrapping thresholds attributed to IndieHackers and MicroConf. Each now cites a source (Damodaran January 2026, BLS June 2026, Paul Graham's growth essay) or says plainly that it is a rule of thumb or that there is no benchmark.

  • corrected

    Examples, formulas and glossary pages (site-wide)

    A meta-description trimmer run on May 10, 2026 cut description fields at 155 characters, so 41 example, formula and glossary pages carried search descriptions that ended mid-phrase ("to inform.", "make smarter.") and three tool and article summaries were cut short. The descriptions are rewritten and the summaries restored in full.

  • corrected
    2026 AI vendor pricing and TCO report

    The LLM table still listed models that have been superseded (Claude Opus 4.8 at $5/$25, Haiku 4.5 at $1/$5, Fable 5, Gemini 3.5 Flash, Gemini 3.1 Flash-Lite), and the first TCO scenario was labelled Claude Opus 4.8 while the calculator priced it as Claude Opus 5.5. The DeepSeek row was a retired model at old prices (V4-Flash $0.14/$0.28; now V4.1-Flash at $0.30/$1.20 at peak), and Weaviate's free plan has been renamed. The Anthropic, OpenAI and Google rows now come from the shared model catalog, and every row carries the date it was checked.

  • corrected
    Prompt Caching ROI: When It Pays Off, When It Does

    Brought to current terms. OpenAI now bills cache writes at 1.25x input and cached reads at 5% to 10% of input; the page said 50% with no write premium. Claude's current models cache prompts from 512 tokens (the page said 1,024 or 4,096), and Sonnet 5.5 and Opus 5.5 read cache at 0.05x rather than 0.1x. The price table, the sensitivity table and the worked example are now computed from the model catalog, and an unsupported range of hit rates attributed to an Anthropic cookbook was removed.

  • corrected
    AI Support Break-Even: Deflection Rate vs Ticket Volume

    The page called the token cost of a resolution a fraction of a cent; at the Haiku 4.5 prices it used ($1/$5) it was 1.5 cents a ticket. It now uses Claude Haiku 5.5 ($0.10/$0.50), where the cost is $0.0015 a ticket, and the engine figures move slightly ($8,695.50 saved a month at 70% deflection, $2,395.50 at 40%).

  • retracted
    Token Cost Optimization Playbook

    Merged into AI Stack Cost Calculator on 2026-10-18.

    Retired. Its savings ranges had no source, its combined example ($19,500 a month to $850) could not be reproduced from its own steps, and it was priced on superseded models. The AI Stack Cost Calculator prices the same levers on current models.

  • corrected

    AI model prices across the site

    Pages that priced superseded models now price the current lineup from the shared model catalog: the DeepSeek vs Gemini comparison (Gemini 3.5 Flash was deprecated on October 8, 2026; Gemini 3.8 Flash is the workhorse row), the AI solopreneur stack (Sonnet 4.6, Haiku 4.5 and Opus 4.8 rows; its provider-spread figures were also out of date), the micro-SaaS unit-economics report, the RAG cost article, and the OpenRouter comparison. Two comparisons built around superseded models were retired: GPT-5.5 vs Claude Opus 4.7 vs Gemini 3.5 now redirects to the vendor pricing report, and the Gemini 3.5 Flash comparison to the cheapest-LLM ranking. The AI calculators' defaults and examples price named catalog models instead of hand-typed rates.

  • corrected
    Calculator explanations (first pass: the most-used tools)

    Benchmarks in the explanations under the most-used calculators were checked against sources. Ranges with no source were replaced or removed: gross-margin bands by sector (now Damodaran's January 2026 averages for listed US companies), net revenue retention tiers (now SaaS Capital's 2025 medians, 101% overall), a 1.25x-1.4x loaded-cost range (now BLS June 2026, 1.24x-1.29x before overhead), and office-seat and cost-per-hire ranges. A startup growth benchmark is now attributed to its source (Paul Graham, 2012). Two statements were overstated: that a 20-30% scope overrun destroys a fixed-fee margin (it takes a 30% margin to 9-16%), and that building something nobody wants is the most cited reason startups fail. The AI Stack Cost Calculator's explanations contradicted its own price data: they named a PlanetScale free tier that no longer exists, put Clerk and Auth0 at $200-$500 a month for 10,000 users (both are free at that size on their current plans), recommended a deprecated auth library, and repeated an unsupported claim that frontier model prices fall 50-80% a year. The AI Product Margin Calculator's explanations gave two different healthy-margin ranges for AI products (60-75% and 50-75%) with no source, and quoted a superseded Haiku 4.5 price as current.

  • corrected
    AI Product Margin Calculator

    An answer on token-price trends said frontier model prices had dropped 50-80% a year since 2023 and advised planning on a 30-50% annual cut. The prices on the same page contradict it: flagship input prices went from the original GPT-4's $30 per million tokens (2023) to $4-$5 in three and a half years, and output prices fell two- to threefold; the 200x figure applies to a smaller model, not the frontier. The answer now says so and recommends planning on today's list prices.

  • corrected
    15 E-Commerce Statistics

    Updated the Census figures to the second-quarter 2026 release (e-commerce $340.2 billion, 17.1 percent of retail, up 12.2 percent a year against 6.7 percent for all retail, so nearly twice as fast rather than more than twice) and the abandonment reasons to Baymard's current survey (extra costs 40 percent, slow delivery 20, account creation 18, long checkout 17). Two statistics did not appear on the Baymard page they cited, a 44 percent guest-checkout finding and mobile abandonment near 97 percent; they are replaced with figures the page does state.

  • corrected
    How to Value a Business 2026: Methods, Multiples & Examples

    The deal-size multiples were labelled IBBA Market Pulse Q3 2025 but were that chart's Q3 2023 column (2.5x and 3.0x SDE, about 6.5x EBITDA for $5M-$50M). They now use the latest survey, Q1 2026: 2.0x SDE under $500K, 2.8x at $500K-$1M, 3.3x at $1M-$2M, 4.0x EBITDA at $2M-$5M and 4.5x at $5M-$50M. The SaaS example's 4.0x SDE and 5.0x EBITDA sat above that data; it now uses 3.41x SDE (BizBuySell's 2025 software, app and SaaS average) and 4.0x EBITDA, and its blended midpoint falls from $890,000 to $804,600. The service example's 3.0x EBITDA is now labelled as an assumption, since the surveys report EBITDA multiples only above $2 million.

  • corrected
    Total Cost to Company: True Cost of an Employee

    Merged into Loaded Employee Cost: What BLS Compensation Data Really Shows on 2026-10-18.

    Merged into the BLS loaded-cost article, which covers the same question from the current data. The guide carried errors: the 2024 Social Security wage base ($168,600; it is $184,500 for 2026), a UK employer National Insurance rate of 13.8% (15% since April 2025), single and family health-insurance premiums attributed to BLS, which publishes employer costs per hour worked rather than premiums by coverage type, and multiplier bands and state and country figures with no source.

  • corrected
    Free SaaS Unit-Economics Calculators, No Signup (2026)

    The page was titled as a ranking of the best free calculators but featured only this hub's own three tools without saying so; the title and summary now say whose tools they are. Its unit-economics example ran at a 100% gross margin, which overstated lifetime value ($1,250 per customer, 10.4x CAC); at an 80% margin it is $1,000 and 8.3x. Its valuation example used revenue, SDE and EBITDA multiples with no source; it now uses the 3.41x SDE BizBuySell reported for software, app and SaaS businesses sold in 2025, with IBBA's 2.0x for sales under $500,000 as the low end.

  • corrected
    Bootstrapped Software Valuation Multiples

    The page reached its 6.0x EBITDA multiple by discounting Damodaran's public 24.48x by about 75%, and said profitable bootstrapped SaaS businesses transact at 4x-7x EBITDA. Neither the discount nor the range had a source, and the published private-deal data sits lower: IBBA advisors reported 4.0x EBITDA for $2M-$5M deals and 4.5x for $5M-$50M in Q1 2026, and software, app and SaaS businesses sold through BizBuySell in 2025 averaged 3.41x SDE. The worked example now uses 3.41x SDE and 4.0x EBITDA, and its blended midpoint falls from $1,266,667 to $917,400. The unsourced 2.5x revenue multiple was dropped. The Damodaran figures themselves were correct.

  • corrected
    Decision workflows

    Merged into All Tools — 65 Free Decision Calculators on 2026-10-21.

    Benchmarks in the eight workflows were checked against sources. Three were wrong: median B2B SaaS monthly churn of 3.5% (SaaS Capital's 2025 survey of 1,000+ private B2B SaaS companies puts median gross revenue retention at 91% a year, about 0.8% a month), median net revenue retention of 105% (the survey median is 101%), and a 7x private SaaS revenue multiple attributed to SaaS Capital (that was its public-company index). Those now carry the sourced figures, as do the loaded-cost and business-sale multiples (BLS June 2026; BizBuySell 2025). Benchmarks with no source we could confirm were removed: freelance hourly rates labelled BLS-adjusted, project overrun, gross margin, CAC payback, ad return, email open rate, content payback, seed runway, agency margins, LTV:CAC and NPS medians. The fundraising workflow's description listed numbers its steps do not cover; it now names the five it does.

  • corrected
    Loaded Employee Cost: What BLS Compensation Data Really Shows

    Updated to the BLS June 2026 release (USDL-26-1494): private-industry compensation $46.89 an hour, wages 70.0%, benefits 30.0%, replacing the December 2025 figures. Two statements were wrong and are corrected. The page called 1.3x the BLS-implied multiplier while also saying benefits add 43% to wages; the BLS counts paid leave as a benefit and a salary already pays for it, so on a salary basis the BLS lines add 1.24x (legally required, insurance, retirement) to 1.29x (all benefits other than paid leave). It also called health insurance and legally required benefits the two largest line items; paid leave is larger than both.

  • corrected
    15 Freelancing Statistics

    The employer-cost statistic said benefits add 29.9 percent on top of wages. The 29.9 percent was benefits' share of total compensation; on top of wages they added 42.6 percent. The figure is now the BLS June 2026 release: $46.89 an hour, of which benefits $14.07, 42.9 percent on top of wages. A second figure, legally required benefits at $2.19 an hour and 8.3 percent of costs, matched no private-industry line in the release; it is now $3.40 and 7.2 percent (June 2026).

  • corrected
    Scope Creep Cost Calculator

    The rate-collapse and overrun figures displayed a hundred times too small: a 30.6% rate collapse showed as 0.3% and a 44% overrun as 0.4%, because the display divided an already-percent value by 100. The calculation and every other output were correct; the display now shows the engine's percentages.

  • corrected
    Pinecone Pricing 2026: Rates vs Qdrant

    The worked scenario counted one read unit per query and one write unit per upsert. Pinecone meters a query at 1 read unit per GB of namespace and an upsert at 1 write unit per KB, so 200,000 vectors at 1,536 dimensions with 2,400 queries and 1,500 ingests a day is about 103,000 read units and 270,000 write units a month, about $3.20 of metered usage rather than $2.60. The scenario still fits inside the free Starter tier and under Standard's $50 minimum, so the conclusion is unchanged.

  • corrected
    Weaviate vs Qdrant Pricing in 2026

    Weaviate now offers a Free Forever cluster (100,000 objects, 1 GB memory, 10 GB disk), so the page's claim that only Qdrant has a free-forever tier no longer held, and Weaviate's per-dimension rates fell from $0.0139 to $0.00465 per million on Flex and from $0.00975 to $0.003875 on Premium. The comparison, worked math and FAQ now use Weaviate's current pricing.

  • corrected
    Cheapest Vector Database 2026: 8 Managed Stores Ranked

    Corrected Pinecone's metered estimate at 1 million vectors from about $8 to about $38 a month: Pinecone bills each query 1 read unit per GB of namespace, not one read unit per query. The $50 minimum still applies, so the ranking is unchanged. Also updated Weaviate's per-dimension rates and Premium SLA (99.9%; 99.95% is Enterprise), Zilliz's free tier (5 GB plus 2.5M vCUs a month) and its now-published dedicated rates.

  • corrected

    Embeddings DB Cost engine and the articles that quote it

    The engine billed Pinecone as one read unit per query; Pinecone's cost docs bill 1 read unit per GB of namespace (minimum 0.25), and upserts by record size. Corrected the engine and the figures it changed: Pinecone at 2 million vectors and 30,000 queries a day is about $213 a month, not its $50 minimum (AI Micro-SaaS Unit Economics Report, RAG vs Fine-Tune Total Cost), $111 at 1 million vectors (Embeddings DB vs Self-Host pgvector), and about $6,920 at 20 million vectors and 100,000 queries a day (RAG vs Fine-Tune).

  • corrected
    Milvus vs pgvector Cost in 2026

    Zilliz Cloud's free tier is now 5 GB of storage plus 2.5 million vCUs a month with up to 5 collections, not 5 GB plus 100 CU-hours, and Zilliz now publishes dedicated per-vector rates by cluster type. The page and its facts are updated.

  • corrected
    LLM Observability Pricing: Braintrust vs Phoenix vs Langfuse

    Braintrust Pro includes $100 of monthly model credits, not $249, and its pricing page no longer lists a separate per-token Topics rate (AI features draw on the credits, then token rates). Corrected.

  • clarified
    Cursor Pricing 2026: Plans, Model Prices and Is It Free?

    The India Start plan (₹649 a month) was cited to Cursor's pricing docs, which do not mention it. It is now cited to Cursor's own announcement of the plan.

  • clarified
    SaaS Pricing Calculator: Tier Setup for 2026

    Merged into the SaaS Pricing Strategy Calculator on 2026-10-18.

    The three-in-five usage-based pricing figure was cited to a 2025 Maxio report that does not state it. It comes from OpenView's February 2023 study and is now cited through Chargebee's write-up of that study.

  • corrected
    Cheapest VPS 2026: Hetzner vs DigitalOcean vs Vultr

    Hetzner and Vultr prices came from third-party trackers (bitdoze, PriceTimeline). They are now read from Hetzner's own price-adjustment table and the plan APIs Vultr and Linode publish, and the tables and FAQ were updated to those figures.

  • corrected
    Gumroad vs Lemon Squeezy vs Paddle 2026

    Stripe Managed Payments was described as costing the same 5% + 50¢ as Lemon Squeezy and Paddle. Stripe's pricing page charges 3.5% per successful transaction on top of payment processing, about 6.4% + 30¢ on a standard US card, so it costs more on any sale above about $14.30.

  • corrected
    One-Person SaaS Valuation: Acquisition Walk-Through

    Merged into the One-Person SaaS Valuation tool on 2026-10-18.

    Reference [1] pointed to acquire.com/marketplace, which returned 404. Re-sourced to acquire.com/buyers/ (HTTP 200), the current buyer-facing listings hub. Anchor text updated to reflect the current page. (The /marketplace path is gone; the buyer listings are now at /buyers/.)

  • corrected

    Statistics pages — ProfitWell discounting-benchmarks citation + OpenView expansion benchmark

    Two stale citations re-sourced. (1) paddle.com/studios/shows/profitwell-report/discounting-benchmarks returned HTTP 301 to paddle.com homepage; updated in unit-economics-statistics.json and saas-benchmarks-statistics.json to paddle.com/resources (Paddle's research and resources hub, HTTP 200). The original Paddle Studios episode URL is no longer accessible. (2) The OpenView Partners expansion benchmark was already flagged as an archived PNG (firm defunct); the direct PNG URL at openviewpartners.com/wp-content/uploads/2017/07/Expansion.png remains HTTP 200 and the label already notes 'firm now defunct; archived image of original report' — no URL change needed, logged for transparency.

  • corrected

    Methodology pages + benchmark re-sourcing pass

    Added methodology pages and per-entry source citations to the AI Stack Cost Calculator and benchmark data. Re-sourced several benchmark URLs: WordStream 2016 blog URL replaced with current Google Ads Benchmarks; Crunchbase generic hub URL replaced with the specific First Round State of Startups article; SaaS Capital generic hub replaced with the specific annual-survey URL; Upwork research URL updated to the current Future of Work Report; SHRM generic hub replaced with the specific 2024 benchmark URL.

Reporting an error

To report a factual error, dispute a benchmark result, or flag an undisclosed sponsor relationship, use the contact details at /contact/. Substantiated reports are acknowledged within 5 business days and logged here if action is taken.