# Draft-position denominator study (working audit) ## Question and outcome Estimate the proportion of drafted quarterbacks belonging to FindMeMyQB's explicitly selected, 29-player franchise cohort. Current outcome membership is supplied by the downloadable `draft-selection-audit.csv` membership flags and the site's four-tier framework, updated November 15, 2026. Earlier frozen28 research artifacts remain historical records, not the current membership file. This is an editorial cohort, NOT a mechanically applied, neutral league-wide franchise-QB definition. In particular, Andrew Luck, Trevor Lawrence and Kirk Cousins are eye-test inclusions; explicit exclusions also exist. Numbers must be labeled accordingly. No 2026 on-field results are used to expand the membership; the project's intended accomplishment cutoff is the end of the 2025 season, including playoffs and awards. ## Denominator All regular annual NFL draft selections in 1998–2025 identified as quarterbacks in the contemporary/official NFL draft records, regardless of whether they made an NFL appearance. Do not condition on debut, roster retention, playing time, starts, or being active today. Each individual draft selection is one observation; resolve duplicate/re-draft cases explicitly if encountered. Supplemental drafts must be investigated separately before declaring all draft methods complete. College quarterbacks drafted for another NFL position need a recorded decision and sensitivity check when source positions conflict. The 29 includes Tony Romo (undrafted), leaving 28 drafted cohort members. Romo remains visible separately in the descriptive 29-player chart, but is excluded from drafted-player numerators and denominators. No UDFA success rate will be claimed without a complete, defined UDFA denominator. ## Follow-up and censoring Main mature-cohort slice: draft years 1998–2016, providing at least ten NFL seasons of potential follow-up through the 2025 season. This is NOT an equal-ten-season outcome test: older classes still had more elapsed career time. It reduces, but does not eliminate, follow-up differences. Separate recent years 2017–2025. An all-1998–2025 summary is only 'qualified by the 2025 cutoff', NOT the eventual career success probability. Recent nonmembers are not called busts/failures. Secondary ≥five-season sensitivity uses 1998–2021 and must also be labeled minimum, not equal, follow-up. A proper equal-ten-season analysis would independently classify every quarterback using only their first ten seasons and re-evaluate the successful numerator, rather than reuse all-career accolades. That is beyond this frozen-cohort comparison and requires a different outcome dataset. ## Bins and presentation Publish disjoint overall-pick bins: 1; 2–5; 6–10; 11–32; 33–64; 65–100; 101+. Also display actual draft rounds separately; pick32 was not always first round. Useful contrast: top10 versus after10, actual first round versus later rounds. Choose bins before reviewing results. For each provide selected-cohort count, total drafted QB count, percentage and preferably Wilson 95% intervals for sample-size visibility. No denominator of29 for these hit-rate estimates; 29 only describes cohort composition. ## Limits This is retrospective observational evidence, not causation. Earlier picks receive different opportunities/support and prospect quality is selected, not randomized. It does not measure the cost of trading up or establish that any particular prospect is worth a high selection. A selected group whose exceptions are editorial cannot be advertised as an objective validated prediction model. Advanced opportunity-adjusted/survival analyses would require separate data and assumptions. ## Source audit warning HOF 2025 table lists Alan Bowman/QB at pick118, while the Falcons and NFL Operations final list identify Billy Bowman Jr./DB. Automated position filtering of a single source is insufficient. Retain source discrepancies and corrected records rather than silently rewriting raw extracts. ## Scope is deliberately different from the site's debut-language universe The draft-choice question requires entry by draft selection year, not first NFL game. These denominators therefore cover regular drafts1998–2025, not all players whose first regular-season game occurred1998–2025. Some drafted players never debuted; they remain included. Pre1998 signees/draftees with later debuts are outside this draft-year sample. This is a separately labeled analysis, not a silent rewrite of the site's eligibility definition. All28 drafted members of the current29 fall within these draft years, so their numerator joins by exact year/overall pick. The mature subset has20 drafted selected members plus Romo as a separate undrafted selected member, hence21 cohort members descriptively but20 in the drafted numerator. ## Complete-source meaning The regular-draft denominator has335 records.173 records1998–2011 were crosschecked between HOF annual draft tables and the independently hosted nflverse/PFR-derived draft CSV, with primary team sources resolving position conversions.162 records2012–2025 were extracted from98 official NFL round pages and crosschecked against HOF (2013–2025), Ourlads (2012), and team primary sources for2025 conflicts. This is not two independently audited official records for every early row. The HOF2012 parsing gap and HOF2016 incomplete overall table are covered by full official NFL trackers; the2016 QB rows agree. ## Position/supplemental sensitivity Excluded college quarterbacks drafted for another NFL role: Steve Bellisari(2002,S), Brad Smith(2006,WR), Reggie McNeal(2006,WR), Isaiah Stanback(2007,WR), Julian Edelman(2009,WR), Tommy Mellott(2025,WR). All had picks beyond100, so including all six changes after-top10 denominator281→287 (12/287=4.18%) and mature after-top10 denominator200→205 (9/205=4.39%); the top10 results do not change. This alternative is not the preferred QB-at-selection policy. Ronald Curry stays because Raiders identify him as an initial QB who converted later, just as later conversions such as Logan Thomas remain. Terrelle Pryor2011 supplemental third round is recorded separately, excluded from standard overall-pick bins. If combining regular+supplemental by actual round, total336 (28/336=8.33%), later-round denominator252 (8/252=3.17%). Mature total232 (20/232=8.62%), later-round denominator180 (6/180=3.33%). Do not assign him an invented regular overall slot. Main headings must say regular NFL drafts to be precise. ## Statistical interval caution The stored Wilson intervals are binomial reference intervals, not uncertainty about the historical counts (which are enumerated) and not a correction for editorial selection, unequal careers, era effects, or confounding. Do not present them as validated predictive confidence bands. Prefer leading with counts and clearly described historical percentages.