Two-expert linear pool forecast for five Fraser Summer-run sockeye stocks (Late Stuart, Stellako, Raft, Quesnel, Chilko), built by age class from brood-year spawner escapements. Fraser returns are dominated by age-1.2 (four-year-olds; 69-95% of each stock), with age-1.3 (five-year-olds) secondary. The dominant age-1.2 class is forecast by pooling, with equal weight, two estimators that bracket the density-dependence question: (1) a cycle-line geometric-mean return-per-spawner, which is density-independent and respects the Fraser's four-year cyclic dominance (2026 falls on the dominant parity-2 line: 2010-2014-2018-2022-2026); and (2) a linearized Ricker, ln(R/S) = a + b*S, which is density-dependent and discounts the large brood-2022 spawner loads for overcompensation. Age-1.3 is forecast by a sibling model applied to the observed 2025 age-1.2 return. Minor age-1.1 and age-2.2 components use geometric-mean productivity. Chilko carries an independent smolt-to-adult marine-survival estimator, reported as a cross-check but not pooled. Parameters are fit over the recent regime (broods >= 1998). No bias correction is applied. The five-stock total is 4.05M, reconciling to 94% of the DFO Summer management-unit p50 (4.305M, which additionally contains Harrison and miscellaneous N. Thompson stocks) - an independent bottom-up agreement with the agency's top-down anchor. A true retrospective (2021-2025, all parameters refit through the prior return year only) gives an overall MAPE of 41.4%. The model is most accurate in the analogous dominant-cycle year 2022 and least accurate in the anomalous high-marine-survival 2025 return.

Most Fraser sockeye come back at age four. For these five stocks, four-year-olds make up 69-95% of every run, with five-year-olds a distant second. So the forecast is really a question about two groups of fish, and we handle each separately. Four-year-olds come from eggs laid in 2022. We know exactly how many spawners were on the gravel that year, so the only question is how many offspring each one produces. We answer it two ways, then split the difference. The first way looks at cycle. Fraser sockeye run on a four-year rhythm (generally), and 2026 sits on a dominant year, the same line as 2022, 2018, 2014 and 2010. So we ask: on past dominant years, how many returning adults did each spawner produce? We apply that average to the 2022 spawners. The second way looks at crowding. When too many fish spawn at once, they dig up each other's nests and the young compete for the same food, so each spawner produces fewer surviving offspring. This is a Ricker curve. It matters here because 2022 had a lot of spawners in Quesnel and Chilko, so this method predicts a smaller return than the cycle method does. We don't know which is right, so we average them. That average is our four-year-old forecast. Averaging two reasonable methods is usually safer than betting on one. Four- and five-year-olds from the same parents come home a year apart. So the four-year-olds that returned in 2025 tell us roughly how many of their siblings will show up in 2026. We use the historical ratio between them. Note that three-year-olds and a minor five-year-old group are each under 1% of the run. We estimate them from long-run averages. They don't quite change the answer. Chilko is the only stock where someone counts the young fish leaving the lake. That gives us a second, independent way to forecast it: count the 49 million juveniles that left in 2024, multiply by the share that historically survive the ocean. It landed close to our answer, which gave us confidence. We tested it on 2021-2025 by pretending each year hadn't happened yet and forecasting it with only prior data. Average error was 41%. That sounds high, and salmon forecasting is genuinely hard. The useful detail: our best year was 2022 - the last dominant year, and the closest match to 2026. Our worst was 2025, when the ocean was unusually kind and every method underestimated the run. Fish returning in 2026 hit a poorer ocean, so we would not expect a repeat.