# KDP physical-book price/page frontier: methods ## Question and scope How do binding, ink, paper and representative trim choice change the relationship between a KDP print book's billed pages, printing charge and tax-exclusive list price? Which modeled page counts can cover their printing charge at USD 9.99 but not USD 9.98? This is original deterministic scenario analysis, not a survey of books, observed author income, manufacturing costs incurred by Amazon, a historical fee-change study, or an optimal-price recommendation. The inputs are selected publicly stated Amazon.com KDP charges and limits retrieved on 2026-10-30. The retrieval date does not establish when each rule became effective. The analysis covers standard Amazon distribution only. ## Fixed design The local protocol was registered before derived scenario calculation. The 14 profiles comprise five paperback ink/paper choices at each of 6×9 inches (regular) and 7×10 inches (large), plus two hardcover choices at each trim. Paperback choices are black ink/white, black ink/cream, black ink/groundwood, standard color/white and premium color/white. Hardcover choices are black ink/white and premium color/white. Hardcover black/cream has the same documented charge and page range as black/white; it is not duplicated as another cost profile. These are deliberately chosen scenarios, not a representative sample or market shares. Exact trim-specific upper limits are 828 pages for paperback black/white and premium color, 776 for black/cream, 812 for groundwood, 600 for standard color, and 550 for hardcover. Paperbacks start at 24 pages except standard color, which starts at 72. Hardcover sources permit 75–550 pages. Our hardcover grid deliberately examines even counts 76–550; it does not declare 75-page manuscripts invalid. Paperback submission guidance explicitly rounds manuscript counts up to an even number when necessary. Page counts here are billed printed pages, including any applicable rounding, not manuscript words or text-only numbered pages. ## Cost and rate model For the applicable short-book band, printing cost is the fixed-only charge. Otherwise it is the long-band fixed charge plus billed pages multiplied by the per-page coefficient. All inputs are stored as exact decimal strings in `inputs.json`. The primary model rounds printing cost to the nearest cent with `ROUND_HALF_UP`. KDP's generic pricing page says costs and royalties are rounded to two decimal places but does not specify how exact ties are handled. We therefore repeat every output with `ROUND_HALF_EVEN`. This is a sensitivity exercise, not a claim about the production rounding implementation. The four changed output cells are listed in `calculation-summary.json`: two costs and their associated unrounded royalty values. No registered maximum page count, price floor, unlock interval, matched-paper savings, or boundary increment changes under that alternative tie rule. The modeled unrounded royalty is `rate × tax-exclusive list price − rounded printing cost`. The rate is 50% through USD 9.98 and 60% from USD 9.99. Display rounding is not used to turn a small negative amount into feasibility. Minimum prices are analytical cent-grid floors, not observed values accepted by a KDP user interface. No publishing or calculator submission occurred. For each possible rate band, reproduction computes the smallest cent-grid price covering printing plus the target royalty, respects the lower bound of that band, rejects prices above its upper bound, and selects the lowest remaining candidate. This avoids applying the 50% formula to prices that actually belong in the 60% band, or treating the 60% formula as valid below USD 9.99. ## Registered outputs * `page-capacities.csv`: 196 rows, covering 14 profiles × 7 prices × 2 targets. Prices are 7.99, 9.98, 9.99, 12.99, 14.99, 19.99 and 24.99 USD. Targets are 0 and 2 USD unrounded royalty. The maximum is selected from the eligible even-page grid. Blank maxima mean no feasible page count, never a zero-page book. Status distinguishes a physical page limit from a price/cost bound. * `price-floors.csv`: 140 rows, covering 14 profiles × 5 matched counts (100, 200, 300, 400 and 500 pages) × 2 targets. All selected counts fit all 14 profiles. * `threshold-unlocks.csv`: 14 rows giving the number and endpoints of even page counts feasible at 9.99 but not 9.98. The counts are hypothetical page-grid points, not numbers of books, authors or customers. * `groundwood-savings.csv`: 10 matched comparisons, covering 2 trims ×5 page counts. Savings percentages use the rounded total white/cream printing charge as denominator, not the variable per-page coefficient alone. * `band-boundaries.csv`: 10 adjacent even-count comparisons at fixed-only/linear boundaries. Cream and white are retained separately because their upper physical limits differ even though these boundary charges match. The USD 2 target is a deliberately fixed comparison, not a living-wage threshold or net profit. Additional author and publisher costs are outside the model. ## Source discrepancies preserved The detailed paperback and hardcover cost tables, Print Options, and trim-size guidance explicitly distinguish regular and large trim charges. The generic Print Book Pricing Page also contains a sentence claiming trim does not affect printing cost. We do not merge those conflicting statements. The explicit numeric cost schedules determine the model; the generic page is used for the independently corroborated royalty formula and its two-decimal statement only. The conflicting sentence remains documented in `inputs.json`. The paperback black long-band header starts at 110 pages while its fixed-only note includes 24–110. At 110, the two formulas do not agree. The primary model follows the explicit fixed-only note. `band-boundaries.csv` also provides the alternative linear formula at 110: 2.32 USD regular and 2.87 USD large, versus the fixed 2.30 and 2.84. The threshold-capacity and 100/200/300/400/500-page price results do not depend on 110-page treatment. ## Reproduce Python 3 standard library only is needed for public reproduction. In a clean folder, place `reproduce.py` and `inputs.json` together, then run: ```sh python3 reproduce.py --inputs inputs.json --output-dir derived ``` The command writes only the five derived CSVs and `calculation-summary.json`; it does not access a network, create an account, call an API or overwrite source inputs. CSVs are UTF-8 with LF line endings. To verify privately retained original HTML, additionally provide `--raw-dir /path/to/exact-originals --source-hashes source-hashes.json`. Raw-source verification is optional. Current official pages may have different hashes because they are dynamic; a changed hash alone does not prove the fee parameters changed. Raw HTML is not part of this public artifact pack. ## Interpretation limits A fixed page count across trims does not mean a fixed word count or identical layout. A larger page might accommodate more text, but no typesetting experiment was performed. Paper, color and binding are not interchangeable quality or accessibility choices. Fixed-only/linear discontinuities describe published billing rules, not inferred physical resource consumption. A nonnegative printing-only royalty does not pay for editing, cover design, advertising, proofs, shipping, returns, taxes or author labor. There is no demand curve, sales forecast, causal behavior claim, international comparison or uncertainty interval: the uncertainty concerns rule interpretation and implementation rather than sampling error.