This working draft addresses accuracy in radiocarbon and zircon U–Pb geochronology at thelevel of the individual sample — accuracy set by sample contamination and event attribution,not by the decay physics. The decay clocks themselves are taken as established and are notdisputed. Its central thesis is that the dominant accuracy failure shared by both chronometers has asingle structure: incorporation of foreign, older material biases a measured age towardgreater age (the radiocarbon reservoir / dead-carbon effect; zircon inheritance, antecrysts,and detrital recycling). The two methods differ only in detectability — in radiocarbon theold carbon is mixed atom-for-atom and is therefore homogeneous, inseparable, and hidden, andmust be corrected externally; in zircon the old material resides in discrete grains and soappears as separable age populations that can be excluded from the data itself. The paper formalises this as a two-axis decomposition (internal clock integrity vs. eventattribution), unifies the underlying physics through the Dodson closure / Fourier number, anddefines a single protocol — screen → classify → correct-where-characterisable → cross-check →report honest uncertainty — in which zircon populations are classified on age-independentgrounds (crystal position, texture, common-Pb content) rather than by selecting the youngest age. The protocol is demonstrated on real data: a leave-one-out reservoir correction on publishedradiocarbon pairs (RMSE 418→141 yr and 739→453 yr); the canonical Lava Creek Tuff zircondataset (Matthews et al., 2015), where classification by crystal position reproduces thepublished rim age (crystal faces, 626.5 ka) and isolates the older antecrystic cores (668.8 ka),consistent with the ~631 ka eruption age and an independent Marine Isotope Stage tie; andopen-access SHRIMP-RG and detrital-zircon data (common-Pb correction; youngest-grain maximumdepositional-age constraint). The work is explicit about the working range of each method (both the young and the extreme-oldends degrade), about the assumptions each correction depends on, and about one irreducibleepistemic asymmetry: in the deep-time interval, accuracy is established by consilience —concordance, cross-method agreement, astrochronology — not by direct comparison with writtenrecords. Every quantitative claim is tagged as directly confirmed [F], inferential [I], orassumption/open [A]. Companion to "Matrix-Informed Correction and Exclusion Protocol for Biogenic RadiocarbonDating" (doi:10.5281/zenodo.17718893), which this paper extends to a second, mechanisticallydifferent chronometer. Contents of this record: the paper as PDF and as a self-contained HTML version (with in-pagesearch); the LaTeX source with figures; and a lightweight reproducibility bundle (Pythonscripts) that regenerates the zircon analyses from the cited public data.