AI Coding Expertise & Technical Debt Tracker

The tracker assigns default velocity one hundred ninety eight percent, skill fifty three, vulnerability twenty two point four per month and cognitive debt fifty. Skill follows one hundred minus AI adoption eighty five times point six five times junior share seventy five over sixty, plus twenty two for mentoring, giving fifty two point nine three seven five rounded fifty three. Native review time forty five adds an arbitrary fifteen-point skill bonus, producing sixty eight. Bars compare fifty three and sixty eight at four pixels per index point. These are assumed rubric values, not observed employee ability, cohort retention or a measured security incident rate. A chart of twenty four synthetic sprint positions is a drawn projection rather than an executed longitudinal study. The vulnerability heuristic multiplies AI adoption times point twenty two, junior ratio over fifty and thirty divided by review time, with fixed quality-gate and mentoring discounts. Increasing review time from ten to forty five changes its inverse-time factor from three to two thirds, lowering the caption twenty two point four to five point zero per month. Bars encode both at ten pixels per caption unit. Skill also jumps at strictly more than twenty minutes, while debt receives a separate point seventy five discount at strictly more than twenty five. These abrupt rules are not estimated causal effects of review practices. With skill sixty eight, debt computes sixty eight times one minus sixty eight over two hundred times point seventy five, rounding thirty four. Native range changes and JSON export preserve the source arithmetic and static canvas. The audit sandbox has three synthetic pull-request fixtures and a fixed list of intended bugs for each. Bars encode three fixtures and the one-correct-hit threshold at sixty pixels per count. Approving a fixture with known bugs increments audit count and reports a rubber-stamp warning, but does not change debt or vulnerability metrics. Requesting rework counts as accurate whenever at least one selected issue matches, even if other selections are false positives. Repeating an audit of the same fixture increments counts again without a uniqueness rule, so counts are interactions rather than independent reviewed PRs. Actual native approval and export record this behavior; the page sends no real pull request or production change. Original four-width tool and canvas drawing preservation and all bottom footer placement are local subset evidence only.