Medians: ~21 min after 7 pm, 28 at 7 am, 39–42 late morning through midday. The midday peak is mostly the commercial window, not recreational volume (arrivals actually peak 8–10 am and 4–6 pm).
A commercial lockage within 15 min before your arrival doubles your median queue (24 vs 12 midday). The Goodtime III runs ~11:24 to the lakes and ~13:59 to the Sound daily — give both windows 20 minutes. The Outlook tab flags these live, plus tugs seen on AIS.
Arriving within 10 min after doors close is the worst case: median 50-minute wait. 35+ minutes after the last cycle, median ~24. The Fastest possible passage card on the Outlook tab computes your current cycle position automatically.
Lines both sides, fenders out, crew briefed before the green light — a fumbled loading costs the next ~22-minute cycle. See the transit guide.
| Scenario | Typical |
|---|---|
| 7:30 pm, no commercial, chamber due, rigged | 11–18 min |
| Random summer arrival | 27 min |
| 11:30 am behind the tour boat, July Saturday | 45–60+ min |
Rush hypothesis: for each hour of day, recreational arrivals were
counted and median waits computed across all 30 days; the six busiest
arrival hours (n=327 lockages) were compared with the six quietest
(n=176), and the arrival↔wait correlation computed across 18 hour bins
with ≥10 lockages.
Commercial-cost hypothesis: to avoid reverse causality (a longer
wait mechanically sees more commercial passings), exposure was defined
before the outcome: did any commercial vessel close doors in the
15 minutes before the boat's recorded arrival? Queues were compared
exposed vs unexposed, stratified by time of day. Result: midday 24 vs 12
min, evening 14 vs 5, overall 14 vs 7 (108 exposed, 673 unexposed).
Dose: one commercial in the window → 15-minute median queue.
Both effects are associations measured on US Army Corps of Engineers
lockage records, Jul 30 – Aug 28, 2026; they update as the archive grows.
Everything on this site is computed from the
US Army Corps of Engineers Lock Performance Monitoring System
(LPMS) — the Corps' own operational log of every lockage at the
Hiram M. Chittenden Locks in Ballard (their site code: waterway
WS, lock 01). It is public data, published as
a rolling 30-day window and refreshed as lockages complete. This
page re-reads it every 5 minutes.
What one record holds:
What it does not hold: boat counts inside a recreational group, vessel size, which chamber (large or small) was used, or why a wait was long (maintenance, salinity management and staffing all extend waits without appearing in the data).
How this site uses it: wait forecasts are medians of the last 30 days' recreational records for your arrival window and direction, scaled by how today is actually running, plus a premium for commercial traffic — commercial vessels lock through with priority, and the regulars (like the Goodtime III tour, which runs to a tight timetable) are detected automatically from the same records.
Archival: the public window forgets anything older than 30 days, so this project's analysis machine banks every record permanently as it appears. The forecast model's baselines widen as that archive grows — the estimates here get better with age.
This page (fast model): each estimate is the median of the last 30 days' recreational lockages for your arrival window and direction, multiplied by a live pace ratio (the last ~6 hours of actual lockages against their historical norm, clamped 0.6–1.8×), plus a commercial-pressure premium (~6 min per expected commercial arrival that hour, capped) and a fixed bump when a scheduled priority vessel is due. The "typical" range is the interquartile band of the same records, shifted by the same adjustments.
LockPush (full model, runs on the analysis machine): lock waits are not bell-shaped — you either make the next chamber cycle or you miss it and wait a full cycle more. LockPush models that explicitly as a mixture:
Calibration (fit on even calendar days, scored on 393 held-out odd-day lockages): the 10–90% band contained 79% of actual waits (target 80%), the interquartile band 50% (target 50%). Point error is comparable to the simple median model — the distribution, not the point, is what the mixture buys.
Operating rule built into the model: in normal operation only the small lock runs; the large chamber is reserved for vessels that physically don't fit (barge tows) and is otherwise rarely cycled. Commercial vessels use the small lock too, but never share it with recreational boats — so a commercial vessel ahead of you in your direction costs one full cycle, and the forecast treats it that way. Chamber inference from the archived records supports this: only ~9% of lockages show the large chamber's signature.
Known limits: the feed does not say which chamber (large or small) served a lockage, so chamber choice is absorbed into the spread; recreational records are batched, so individual-boat arrival processes are observed only by the camera; weather, tides and maintenance are not yet inputs. The model refits hourly from an archive that grows past the feed's 30-day horizon, so all of these estimates improve with age.
Everything above, plus researched guides, lives on static pages that work without JavaScript and are written to be cited:
Source: USACE Corps Locks / LPMS. This is an unofficial hobby project, not affiliated with the Corps.