Field report estimating Alaska's 2026 commercial salmon harvest from in-season ADF&G blue-sheet data, with an interactive pink-run slider and a comparison to the Kalshi betting market.
● Live dataADF&G blue sheet · as of Jul 24, 2026 (stat wk 30, prelim.)Kalshi KXAKSALMON-26ALL
How big will Alaska's 2026 salmon haul be?
The number, the edge, and the bet: how many salmon Alaska catches in 2026, where that disagrees with the Kalshi market, and how to size it — built from the state's live in-season data. The evidence sits below the decision.
The trade — how a market inefficiency showed up in the state's own data
Starting July 22, 2026, I began tracking Alaska's in-season commercial salmon catch — the state's own “blue sheet” — against the Kalshi market KXAKSALMON-26ALL, which settles on the final statewide harvest. The market was pricing a big season: an ~84% chance of clearing 100M fish and an implied median near 123M, essentially just trusting ADF&G's preseason 125M forecast.
The data said otherwise. Sockeye and chum were already finishing below forecast — and the clincher was Prince William Sound, whose early-peaking hatchery pink run had collapsed to ~60,000 fish (normally millions by now), a failure ADF&G and the fleet went on to confirm. Reading the regions instead of the forecast put the honest number in the low-80s. Pink later ran hard — in weeks 32–33 and again in week 35 — lifting the model to ~92M, still well under the market's ~123M. That gap is the inefficiency.
Kalshi KXAKSALMON-26ALL — market snapshot · July 28, 2026
≥100M84¢≥110M73¢≥120M55¢≥125M47¢≥130M38¢≥140M24¢
Each price is the market's odds (in ¢ = %) that the final harvest clears that line. Implied median ≈ 123M. My model, reading the in-season data, lands at ~92M — putting ~13% on ≥100M. The live tracker below is how that thesis plays out through the season.
~85M fish
regional model · agnostic median · range 69–104M
The agnostic read is ~92M. Everything-but-pink is ~53M, modeled per-species: sockeye done (~38M), chum ~12M with its fall run still trickling, coho ~2.8M (forecast-anchored, its season now). Pink did the damage: it erupted 6M → 21M in weeks 32–33 (its normal peak), then delivered the real surprise — a second wind in week 35, past the peak, when Southeast and Westward each ran ~4M more. It's now done in week 36 at ~36M landed, and the total lands near ~92M. With 85M already banked and only a few million of catch still in play, I put ~13% on clearing 100M. The market says 74%. The cushion is real, and the season is nearly out of runway to erase it.
Where the estimate lands
How my estimate has moved
My number fell from a market-agnostic ~100M in late July to the low-80s as the regions reported in weak, then climbed back to ~92M as pink surged twice more. The market's price for “≥100M” barely moved the whole time. So the gap between us — the edge — is still wide, though it has narrowed from its peak.
Coral = my agnostic estimate (left axis). Slate = the market's implied chance of ≥100M — the price of the YES side I'm betting against (right axis). Labeled points are the reasons the number changed; the line keeps logging daily. My read fell ~15M while the market's ≥100M price held near 82–88¢ — the divergence is the thesis.
01
What I actually did
Four steps, plain version. No magic — just the state's numbers, a bit of history, and honest uncertainty.
Step 1
Read the blue sheet
Alaska publishes cumulative commercial catch by species all season. Right now it reads 40.5M fish landed — mostly sockeye, with pink barely begun.
Step 2
Ask “how far along?”
40.5M today isn't the final. I needed the % of each species' run typically done by late July — sockeye is nearly finished, pink is not.
Step 3
Check the history
I pulled the same blue sheet from past even years (2020/22/24) out of the web archive to see how each species really fills in by date.
Step 4
Project + stress-test
Scale each species to a full season, add them up, then wiggle the big unknown (pink) to see how much the answer really moves.
02
The haul so far vs. the forecast
Solid bar = caught so far. Faded bar = where I project it finishes. Hatched outline = the state's preseason forecast. Notice pink: almost nothing caught, huge forecast — that's the wildcard.
Caught so farMy projected finishADF&G forecast
The read: Sockeye (~37.5M caught, done) finished near 38M — under the 49.7M forecast. Chum is soft too. Those two alone knock ~15M off the state's headline number.
03
How 2026 is tracking vs. past years
Total fish landed as the season rolls on. Each grey line is a past even year; the thick coral dot is where 2026 sits today. The big vertical climbs are the pink run in August — the part of the story we haven't seen yet.
Even years only (pink salmon run on a strict 2-year cycle, so odd years aren't comparable). 2022 was a monster (153M); 2024 was weak (~103M). 2026's final almost entirely depends on which August it gets.
04
It all comes down to pink — try it yourself
Everything except pink is locked in at about 50M fish (ADF&G weekly data: sockeye ~37M, chum ~12M, coho ~2M). So the entire 2026 total is really one number: how many pink salmon show up. Drag the slider.
Projected 2026 total
at 26M pink salmon
Everything else (near-locked)48.0M
Pink (you're setting this) — regional model26.0M
2026 total74.0M
Which Kalshi “at least X” bets pay out
Why the range stays wide: a low pink count today can mean the run is genuinely small — or just late. We literally can't tell yet. That ambiguity is the whole reason I won't commit to a tight number until mid-August.
05
My estimate vs. the betting market
Kalshi lets people bet on “will the total be at least X million?” The teal line is what the market's prices imply; the solid coral line is my agnostic estimate (the regional model), and the dashed coral line is my lower bearish lean. The market sits far above both — it's pricing a normal pink run the regions aren't showing.
Read a point as “chance the final clears this line.” At 100M my agnostic read is now ~13% while the market says ~74% (my lean, 12%). The gap is enormous — the late pink surges lifted the tail off zero, but with 85M banked and only a few million still in play, ≥100M needs the remaining catch to more than double. At ~36M through week 36 pink caught up but stopped, still below past even years, with PWS collapsed. Real edge, but a thin, low-volume market, and my backtest over-projects weak years — so treat the size with humility.
The probability table — my two views, the blend, and what the market pays
Agnostic = the regional model: each pink super-region projected from its own pace + base rate, then summed with correlation → pink ~38M, non-pink ~53M (sockeye done, chum fall-run, coho forecast-anchored). My lean trusts the weak pace more. NO-edge vs agnostic = how many points cheaper the market prices “stays under this line” than even my neutral view. Pink surged twice — its normal peak in weeks 32–33 and a genuine second wind in week 35 (Southeast + Westward carried it, PWS still collapsed) — but with the run now over and 85M banked, my ≥100M tail sits at ~13%.
Bottom line: my agnostic estimate is ~92M and the market's is ~123M — pink erupted, faded, then surged a second time before finally stopping in week 36. I put ~13% on ≥100M — with 85M banked and pink done, reaching 100M needs the remaining catch to more than double. The market is still pricing the 125M forecast that came in 12–24% high in two of the last three even years. Size with humility — thin market, and the model has had to be corrected as it went — which is why the calculator defaults to ¼-Kelly.
06
How good is the state's forecast, really?
The 125M is ADF&G's preseason guess — and guesses have a track record. Each bar is the actual catch; the ▾ tick marks what they forecast. Gap between them = the miss. Three even years, by species, with 2026's forecast for comparison.
Actual catch (bar length)Preseason forecast (tick)2026 caught so far
The takeaway: sockeye forecasts are tight — within ~7% every year, so this year's soft sockeye is a real signal, not forecast noise. Pink is the opposite: dead-on in 2020 & 2022, then off by 42% in 2024. And the statewide total came in under forecast in two of the last three even years (−12%, −24%), dead-on the third. That history is exactly why I lean below the 125M.
How ADF&G builds the forecast
Sockeye
Most reliable
Sockeye return over several ages, so “sibling models” use this year's young returners to predict next year's older ones (plus spawner-to-return models), river by river. Bristol Bay dominates. Multiple age classes = forecastable.
Pink
The gamble
Pinks are all the same age (2 years) — no siblings to lean on, so the forecast hangs on juvenile-fish surveys (trawl counts of young pinks) and one year of ocean survival. Great when the ocean behaves, ugly when it doesn't.
Chum & coho
Rough
The state does forecast these (press releases just say “average”), but the numbers miss badly — chum was off −55% in 2020, coho runs high every year. Small totals, so it barely moves the headline.
Chinook
Not in the number
No statewide commercial king forecast — many stocks are of conservation concern and managed individually. The 125M is really the four species above.
07
How much does today's blue sheet actually tell us?
I pulled 15 years of archived blue sheets (9 with a clean end-of-season final) and re-scaled each to % of its own final. Every grey line is one past season's fill-in. Where the lines fan apart, the blue sheet is a weak predictor; where they converge, it's locked. The dashed line is where we stand today.
7 densely-tracked seasons shown; the bold coral line is the average across all 9. Y-axis = share of that year's final total already reported.
Late July — today
~50%of final is in
final = 1.7–3.5× today's total · weak
Mid-August
~75%of final is in
final = 1.1–2.3× · firming up
Mid-September
~90%of final is in
final = ~1.0–1.5× · locked
So, directly answering the question: today's blue sheet correlates only loosely with the final — across 9 years, late July had captured anywhere from 29% to 58% of the season, so a single reading now pins the total only to a roughly 2× band. It gets trustworthy fast: by mid-September the same reading is within ~10%. That's the honest reason I hold a wide range now and expect the tool to sharpen through August.
08
2026 vs. past even years, week by week
The reconstructed weekly harvest curve — every statistical week, per species — for 2026 against each recent even year (2020 / 2022 / 2024), built from ADF&G's in-season blue sheet and the ASMI weekly updates. Pink is the whole story: 2026 stalled far behind, then erupted in weeks 32–33 to close much of the gap, though it remains below all three past even years. Everything-but-pink is ~53M (sockeye ~38M done, chum ~12M with a fall run still trickling, coho ~2.8M just ramping).
Past years vs 2026 — statewide weekly cumulative (M fish), reconstructed from ADF&G/ASMI weekly updates. Pink is the one that decides it.
2020202220242026 (live)latest week — provisional*we are here (now)
The pink panel is the whole story: 2026 stalled far behind through July, then erupted in weeks 32–33 — 6M to 21M — closing most of the gap. At week 33, 2026 pink was 21.2M against 32.4M in 2024 (the weak year), ~43M in 2020, and 49.5M in 2022 at the same week. It caught up fast but remains below all three. Pink peaks weeks 32–34, so the run is mostly in but not over. Sockeye is landed and finishing low; chum's curve here undercounts its late fall run (the ASMI series ends ~wk36). Reconstructed finals validate against the official totals (2024 pink 37→40M, 2022 sockeye 74.3→74.8M). * The hollow endpoint is the current, still-accruing week — it fills in over the following days as fish tickets post, and updates daily from the live blue sheet.
Pink weekly velocity — the ignition signal. How much pink lands each week (M/wk). A run declares itself when this spikes, not when the cumulative creeps — so this is the single number to watch through August.
2020202220242026 (live)ignition (~2M/wk)
In all three past even years (2020, 2022, 2024) the weekly catch blew past ~2M/week by week 28–29 and peaked near 10–12M/week. Right now 2026 is running ~0.4M/week — roughly 5× below the ~2M/week ignition line that all three past even years cleared by week 28–29. It hasn't ignited, and it's now 1–2 weeks past when every prior run did. When a Friday sheet finally shows the weekly jump clear ~2M, the run has started; until then, pink is genuinely absent. * newest week provisional.
Daily by-region pink velocity — 2026 only, from the daily tracker (fills in each morning). The day-over-day change flags which region is waking up before the weekly number moves.
Δ = new pink landed that day (M fish), by ADF&G super-region. For pink these read as Central ≈ Prince William Sound (hatchery), Westward ≈ Area M + Kodiak, Southeast = wild seine. A region's daily Δ climbing toward ~0.3M/day (≈2M/week) is ignition. This is 2026-only — no daily record exists for past years.
Pink by region — where past even-year pink actually came from (cumulative M, parsed from the historic ASMI blue sheets 2020/2022/2024), and where 2026 stands. Each region is a separate run on its own clock.
2020202220242026 (through wk28)
The regional read was more nuanced than the statewide flatline. The early Alaska Peninsula (Area M) opener showed up — 0.9M, between 2024 and 2022. The clear early warning was Prince William Sound: its hatchery pink normally flows by week 28 (4.0M in 2022, 1.1M in 2024) and 2026 had ~zero — a collapse that held. Southeast and Kodiak — which produce most of the pink — ignite in weeks 30–34, and they did: both ran hard through weeks 32–33 and again in week 35, carrying statewide pink to ~36M while PWS stayed down. Lines here run through week 28 (latest ASMI regional split). Parsed areas sum to the printed statewide within ~2% for pink/sockeye/chum.
ADF&G's 2026 forecast, by region — built stock-by-stock, and how those pink forecasts have fared
Region
2026 pink forecast
caught so far
the read
Southeast (wild)
19M
16.8M
ran hard twice, now done
Central / PWS (hatchery)
~14M
2.7M
collapsed — ADF&G-confirmed
Westward (Kodiak 12M + Peninsula ~9M)
~23M
16.4M
ran late, now done
Statewide pink
56M
35.8M
my model projects ~38M
Other species (statewide forecast): sockeye 49.7M (Bristol Bay ~33.5M of it) · chum 17.2M · coho 2.4M → 125.5M total. Each number is built up region-by-region — Southeast pink from juvenile-trawl surveys, PWS from hatchery-return models, Bristol Bay sockeye river-by-river.
How they fare vs. the actual catch: for pink, the state's forecasts run high. The 2025 statewide pink run came in 30% below forecast; 2024 pink was off 42%; PWS's PWSAC hatchery run was 56% below forecast in 2025; and the statewide total landed under forecast in two of the last three even years. So the 125M the market is trusting embeds a 56M pink forecast with a documented high-side miss — and 2026 pink is tracking well under that forecast (~38M) even after a strong weeks 32–33 run, with PWS's failure now confirmed by ADF&G and the fleet. Source: ADF&G report SP26-08 (Apr 2026) + in-season Cordova Times / ADF&G blue sheet.
Forecast vs. actual — the state's recent pink track record (from ADF&G's own season reviews)
Stock
Year
Forecast
Actual
Miss
Southeast pink
2025
29M
21.4M
−26%
Prince William Sound pink (run)
2025
66.7M
46.4M
−30%
↳ PWSAC hatchery pink
2025
28M
12.3M
−56%
↳ VFDA hatchery pink
2025
20M
26.1M
+30%
Statewide pink
2024
~66M
38M
−42%
The pattern is one-directional: nearly every recent pink forecast has landed below the number — Valdez/VFDA is the lone overshoot in the other direction. The statewide pink total missed low two years running (−42% in 2024, −30% in 2025). That high-side bias is exactly what's baked into the 56M pink / 125M total the market is trusting for 2026 — and PWS is already tracking to its worst miss yet. Older per-region forecasts (2020/2022) live in ADF&G's archived reports; this is the recent, most-relevant record.
In the news — the PWS pink collapse, confirmed by ADF&G and the fleet (not just our blue-sheet read)
“All we know is that they are not coming back as expected.”
— Todd Johnson, ADF&G Prince William Sound commercial seine manager
What the reporting shows (late July 2026):
PWS commercial pink catch: ~60,000 fish through July 23. The entire PWS commercial harvest (all species) had "crawled" to just 1.4M fish, with pink running ~80% behind 2024; aerial surveys of eastern PWS show minimal pink.
The seine fleet has been open and fishing since June 1 — not a closed fishery. "Fishing has occurred during openers, but catches remain dismal."
The hatcheries (PWSAC, VFDA) paused cost-recovery fishing to build brood stock — a move made because a return is weak, which further reduces the commercial harvest the market settles on.
Preseason, PWS pink was already one of the lowest hatchery forecasts on record (21.3M run) — and it's coming in below even that.
Industry is calling the recurring weak-pink situation a “huge disaster”; ADF&G attributes it to poor marine survival.
I back-tested the method honestly: stand at late July in each past year, project the final from the blue sheet using a completion curve built only from the other years, and measure the miss. Each bar is one year's error — right = over-projected, left = under-projected. Teal = odd years, coral = even years (2026 is even).
Across all 9 years the method's median miss was ~9% — versus ~39% for ignoring the season and guessing the average. So in-season data genuinely helps. The misses cluster by parity: odd years are easy, even years hard. 2024 — the closest match to 2026 — is now a real test, rebuilt from the ASMI weekly reconstruction: standing at late July, the method would have projected ~91M and it finished 103.5M — a ~12% under-shoot, because pink and chum both surged after July.
vs. just guessing
9%vs 39% error
reading the blue sheet beats flying blind — validated
odd years
~5%error
big reliable pink runs → easy to call
even years (like 2026)
~14%error
smaller, erratic → the hard regime
The honest verdict: the method is validated — it beats guessing, and it's genuinely good for a typical year. But 2026 is an even year (the hard regime) with an unusually low pink start. The one truly comparable case is 2024, and it now has a real test: standing where we stand today, the method would have under-projected 2024 by ~12% — it finished higher than the July reading implied, because pink and chum ran late. That's a direct warning against over-trusting today's low number. So trust the direction (below the 125M forecast) more than the number: the honest band is ~69–104M with a real tail both ways, and it tightens as Southeast/Kodiak pink finishes running in August.
10
Strategy: how to spread a bankroll across the NO bets
This sizes a bankroll across the “NO” rungs using growth-optimal (Kelly) portfolio sizing — accounting for the fact that the rungs are nested and correlated, so it won't double-count. Everything flows from your view of the final total and the live prices — edit them and the plan recomputes. Every figure is a % of your bankroll, so it works at any stake.
My actual positions — live, against the model
avg entry = my size-weighted average fill price on that rung (pulled from my actual Kalshi fills). now = the current NO price — also the market's implied chance the bet wins. unrealized = mark-to-market vs my entry. if NO wins = return if the harvest finishes below that line (each contract pays 100¢). model win-odds = the regional model's chance NO wins. Blended across all three rungs my average entry is ~21.7¢; if the final lands under 100M — where all three win — that's about +360%. The three rungs are nested: under 100M wins all three, over 110M loses all three. Prices refresh daily from Kalshi's public API. Position sizes and dollar amounts are withheld — this shows percentage returns only. Returns are gross of fees; not advice — my own bets, against my own model.
Bet
NO price ¢
My P(under)
Edge
Stake %
Pays if win %
Your profit / loss by where the harvest actually lands
If you'd make just ONE bet — Kelly growth rate by rung (the peak is your goldilocks)
Why not stake the whole bankroll? Holding cash is the risk control — the fastest way to blow up here is to over-bet, because the NO rungs all lose together if pink erupts. At ¼-Kelly the math says stake ~24% and keep ~76% in cash. The spread across the rungs is the same recipe at any bankroll — only the scale changes. To deploy the whole bankroll, click Full above (that's the growth-max bet, but it swings hard: ~5% chance of losing more than half).
Read before betting: this is a decision-theory calculator, not a directive. The whole plan is only as good as your central estimate — it defaults to my agnostic 74; set it to 61 for my bearish lean and it bets harder on NO, or to 120+ (near the market) and the edge mostly vanishes. It defaults to ¼-Kelly because my estimate is uncertain (see the backtest above) — full Kelly on a shaky number is how you go broke. And it's a thin, low-volume market, so you may not get filled at these sizes. I can't place trades; the number you actually risk is yours.
The work behind the number — methodology · forecast track record · weekly & regional detail
✓
Sources & methodology
Every input to this model is public. Nothing here uses non-public, pre-release, or privileged data, and I hold no confidential relationship with ADF&G, any processor, or any hatchery. The edge is analysis of published data the market underweighted — not inside information. The estimate moves as the public blue sheet updates, which is the signature of public-data analysis, not a fixed number known in advance.
Live ladder prices from Kalshi's public, unauthenticated REST API. Read-only — no account or order data.
Provenance is machine-checkable: the tracker banks the raw blue sheet to disk with a UTC timestamp on every run, and projection_history.json logs each estimate with the public source it was derived from and the date it changed. This dashboard is itself a public, timestamped record of the derivation. This is not legal or investment advice.