Questions & Answers

Only as far as the checks go, and the checks are published here including where they fail. What has actually been measured, from our own detection data rather than assumed:

• How many fish get from Bonneville to the Snake — measured. Of Snake-origin fish detected at Bonneville, 75.3% are later detected at Ice Harbor and 72.2% at Lower Granite (33,722 fish). That second figure sits inside DART's independently published range of 70–79%, which is a real check passing.
• How often we actually see a returning fish — measured. Of adults detected at Bonneville, the share later seen on a home-river array is 55% for the Upper Columbia, 34% Clearwater, 33% mid-Columbia, 31% Salmon, and 21% Grande Ronde/Imnaha. Four in five Grande Ronde fish are never seen again after Bonneville. Any estimate that ignores this under-counts the poorly-watched rivers — and until this week ours did.
• Dam counts — carried correctly. All 206 completed dam-years sit within 10% of the DART census, and a deploy is blocked if that fails. It proves we carry the official counts faithfully, not that our estimates are right.

Where it fails: our estimate of how the run divides between regions is roughly 13 percentage points from the best reference we can build, and a detection-corrected method built this week is closer at about 8. Neither is settled. The comparison chart on this page shows all of it.

No — and this is the weakest part of the site, so it is worth being blunt. Two separate problems. The split between Idaho basins is currently about 13 percentage points from Idaho Fish and Game's published figures year by year, with the Snake mainstem reading zero because it is absorbing the leftovers from a balancing step rather than being estimated. The level is separately low: measured against those same published numbers we have historically published roughly half the fish, because 35–45% of the run never resolves to a river and we leave it unattributed instead of spreading it around. Recent work narrowed the level to roughly 0.4–0.9× depending on the year. Practical advice: use the relative shape and the timing, treat any single river-year total as an estimate with real room in it, and check the river pages, which name their own evidence.

A lot, and mostly for reasons that have nothing to do with your river. At Bonneville the last seventeen years range from about 416,000 fish down to this year's pace. The year-to-year swings are largely shared: when we compare each river's year-over-year change, the median correlation between rivers is about 0.6 using our allocated estimates. Measured on raw tag counts instead it drops to about 0.16 — the first number shares a common census and overstates agreement, the second is dominated by detection noise and understates it. The truth sits between: call it roughly half the movement basin-wide, half river-specific. Our smolt-to-adult return index moves 2–3× between adjacent outmigration years, and DART's own published rate fell from 2.39% to 0.09% for the class returning now.

Less than most people expect, for the big swings. When this year's early one-ocean run collapsed, it collapsed uniformly — every region, hatchery and wild alike, all landing in the same narrow band of roughly a twentieth of last year's pace. No drainage was spared and no drainage was rescued by good local conditions. The first-order driver is what happened to that class in the ocean, not the creek it left from. The honest caveat: roughly 40–50% of the remaining year-to-year variation is river-specific, so home river still matters for the practical question of which water is worth your weekend — just not for whether it is a good year overall.

Both are counted, and separately. The dam counts distinguish clipped hatchery fish from wild ones, so the hatchery/wild split at Bonneville and Lower Granite is a census, not a model. Hatchery fish dominate the totals in most years, and they carry PIT tags at far higher rates than wild fish do — roughly 14 to 22 times higher — which is why we expand tagged fish by their own group's tagging rate instead of treating all tags alike. One caution we publish rather than hide: our hatchery/wild split at Lower Granite currently disagrees with the published figures and we have not fixed it, so read the origin split at Bonneville with more confidence than at Granite.

No. Those ranks are always to date: this year's count through today against every past year's count through the same date. Where the season finishes depends on the fall run still ahead. The shaded band on the timing charts is our estimate of the finish, built by taking the three most similar past seasons and following their remaining paths. It is a projection, it will move as September and October land, and we do not draw it at all for rivers whose run straddles New Year, because a calendar year cuts those runs in half.

The goal is one thing: estimate where the returning fish are going, and when. Everything else is scaffolding. Judged against that goal, here is the current scorecard. Working: the dam counts and their timing (those are censuses); the corridor picture of where the run sits between dams; travel times between checkpoints from tagged fish. Not working yet: the year-by-year split between Idaho rivers, which is off IDFG's published figures by about 13 percentage points a year. That is the core promise, and it is not being kept at the accuracy it should be. The error is not random — it tracks how much of the tagged sample is Clearwater fish in a given year, which points at incomplete correction for differing tagging rates between basins. That is now measured, gated, and open.

Separately, no version of this can tell you whether you will hook one: it estimates fish presence, never catchability. Flow, temperature, clarity, pressure and skill decide that.

Paul Van Valkenburg — one fisherman, working with AI assistance, on public data: DART dam counts, PTAGIS PIT-tag detections, USGS flow, and published agency run reconstructions. About 40,000 lines of Python turn fifteen-plus years of individual fish detections into these charts, and a validation suite scores every estimate against independent sources before anything is published. It is free, there is nothing to sign up for, and corrections are genuinely welcome.