I study why on-farm adaptation fails to offset the yield losses from extreme heat, using four major crops in Quebec over 1991–2022. Embedding a damage-control production technology in a model of phase-specific climate risk, I derive five predictions — defensive intensification, frontier disengagement, incomplete adaptation, cost escalation, and structural maladaptation — and test each using a thirty-two-year regional yield panel and three waves of a farm-level input survey. Heat exposure reduces yields for all four crops, with damage concentrated in the phenologically critical window for each: the sowing phase for cold crops, the growth and maturation phase for warm crops. Input responses are sharply heterogeneous: cold crops intensify pesticide and fertilizer use, while soybeans disengage from all variable-input markets as biological nitrogen demand collapses under reproductive heat stress. Unit production costs rise for all crops, reaching approximately $41 per tonne per additional degree-day above 25 °C for oats and $58 per tonne per additional degree-day above 30 °C for soybeans. The central result concerns barley: it is the only crop for which the phenological window driving observed pest pressure and input demand differs from the window determining yield damage. Consistent with this phase mismatch, barley alone exhibits a significant negative interaction between yield-relevant heat stress and pesticide expenditure, implying that defensive investment is calibrated to the wrong agronomic threat. This selective pattern — present only for the mismatched crop and absent for the three aligned crops — is strongly consistent with a mechanism of structural maladaptation that persists even under fully rational, immediate behavioral adjustment.

Governments overwhelmingly fight pesticide overuse with information-based instruments — national plans, quantitative targets, and integrated pest management (IPM) — and only rarely with taxes. Using a new, hand-validated database of all four instruments for 45 OECD and major agricultural economies (1990–2023), we show that the apparent success of the soft instruments is largely an artifact of selection. IPM — the most widely adopted pesticide policy in the world — carries the largest raw effect of any instrument, yet that effect collapses to zero once country-specific trends are allowed: adopters were already on declining pesticide paths before they adopted. Plans and quantitative targets have precisely-estimated null effects on the volume applied. Only the rarely-used tax shows a robust negative effect, though, resting on four adopters, its magnitude is suggestive. A circularity-free consumption footprint and a bilateral sourcing test find no large trade leakage. The robust lesson concerns instrument class: information-based policies do not measurably cut use, and their evaluation must confront endogenous adoption.

Pesticide-reduction targets in North America and Europe increasingly rely on price instruments, yet the literature offers little guidance on how the resulting revenues should be recycled. We build a computable general equilibrium model of Quebec agriculture in which conventional and organic farming enter as distinct technologies, with production parameters estimated from farm-level microdata. An ad valorem pesticide tax of 80% is required to meet Quebec's 12% reduction target, because the estimated pesticide demand elasticity is −0.16, five times lower than recent French estimates. At that rate, the revenue-recycling rule, not the tax itself, drives the environmental outcome: directing revenues to organic farmers expands organic acreage by 19.3%, against 3.55% when the government retains the revenue and 2.8% when households are compensated. A competitiveness-adjusted border tax further cuts import leakage by 5 to 8 percentage points. The binding policy choice in pesticide taxation is the design of revenue recycling, not the tax rate.

Mandatory agronomic prescriptions insert a licensed intermediary into the pesticide purchase decision, but their causal effect on farm-level demand is unknown. We exploit Quebec's 2018 prescription requirement as a natural experiment and find that the reform reallocated pesticide demand across active ingredients rather than reducing it in aggregate. Using annual compound-level data spanning 1992–2022 and combining difference-in-differences, slope-break, and synthetic control methods, we document three patterns. The prescription accelerated the exit of atrazine at roughly 39% per year faster than its pre-policy trend; it had no statistically detectable aggregate effect on neonicotinoids, the primary public target of the reform; and sales of diamide insecticides, an unregulated substitute class, rose sharply relative to controls. The substitution roughly offsets the gain from atrazine's accelerated exit, so the prescription's identifiable causal effect on within-panel risk-weighted aggregates is approximately zero.

In Quebec, organic-certification exits remained persistently high at 5–7% annually over 2021–2025, while entries fell sharply from 10.8% to 3.4%. By 2025, exits exceeded entries for the first time, contracting the certified stock and reversing a decade of expansion. Using enterprise-level data spanning all product categories, we document three life-cycle exit patterns: the exit hazard follows an inverted-U shape, peaking around the third year; more than 50% of exits occur within five years of certification; and enterprises certified in more product categories face lower exit risk. A calibrated dynamic model with heterogeneous time preferences and idiosyncratic pest shocks reproduces the hump-shaped hazard through the differential exit of impatient versus patient producers. Policy counterfactuals indicate that an 80% pesticide tax raises ten-year survival from 81.0% to 89.3%, while a 1% productivity penalty in organic production lowers it to 71.8%. Long-term retention is shaped by behavioural heterogeneity in time preferences and exposure to ecological (pest) risk.