reconcile-tdbulisted

Interviews the user, then builds and runs a top-down/bottom-up new-product volume forecast that reconciles a trial-and-repeat model against a market-share model. Use this whenever someone wants to forecast first-year volume, units, or share for a new product or concept; mentions trial and repeat, repeat rate, share of choice, category penetration, MPCT (a monadic product concept test, the survey a concept is shown to one respondent group at a time), depth of repeat, or a volumetric concept test; asks to reconcile two forecasts or two modelling approaches that disagree; asks why a bottom-up estimate does not match a top-down one; or brings up the top-down bottom-up triangulation model from *The Art and Science of Forecasting* (chapters 20 and 27). Also use it for teaching or writing about reconciliation between estimation procedures, balancing variables, or survey overstatement correction. Trigger it even when the user has only partial data, the skill is built to interview for the missing inputs rather than re
KarpelesPublishing/the-art-and-science-of-forecasting · ★ 0 · AI & Automation · score 72

Install: claude install-skill KarpelesPublishing/the-art-and-science-of-forecasting

# Top-down / bottom-up volume reconciliation ## What this does Two accepted ways to forecast a new product's first-year volume usually disagree: - **Bottom-up (trial and repeat).** Start from how many people will try it and how often they will come back. Category penetration is given; market size falls out. - **Top-down (market share).** Start from the size of the category and take a share of it. Market size is given; penetration falls out. Each answer is defensible in isolation. When they disagree, at least one input is wrong, and the disagreement tells you which one. This skill runs both engines, measures the gap, reports what each engine implies about the inputs, and drives the two into agreement. `scripts/reconcile_model.py` is the author's top-down bottom-up triangulation model, the one described in *The Art and Science of Forecasting*, chapters 20 and 27. It began life as a spreadsheet and now lives here. Its `--selftest` reproduces the reference case the book quotes (3.55 and 3.53 million units, a 0.6 percent gap, two passes to reconcile). Every coefficient and curve constant in it is an illustrative starting value, from no particular source: generic numbers, set by judgment, that put an ordinary packaged-goods launch in the right ballpark. Tell the user this the first time a modelled number appears. If they have launch history, fit the trial and awareness relationships to it; if not, help them tune the constants against launches they know and record what cha