We tracked 10,247 Rush Hour rounds, one by one
For all of March 2026, we kept a continuous log running on every one of the six 155.io Rush Hour camera feeds — recording the posted threshold, the AI vehicle count, and the outcome for each round on each feed. 10,247 rounds later, here's what the numbers say about a betting category younger than the CCTV cameras it runs on.
Methodology
We polled all six 155.io Rush Hour city feeds every 30 seconds around the clock. Each round's vehicle-count integer, captured at the T+55s lock moment, went into the log together with the camera ID, a UTC and local timestamp pair, the posted threshold, and the round outcome (over, under, or exact). We dropped 37 rounds that were cut short by stream interruptions.
None of this data came from placing real-money wagers — every round outcome here is purely observational. 155.io shows its threshold, AI count, and result to anyone watching, wager or no wager, during every round. What we built is a passive reader of those public numbers, polling every 30 seconds so nothing across the six feeds slips past us.
We pulled 37 rounds out of the analysis — stream drop-outs or counts so implausible they were clearly feed errors — though they're still listed in the raw log for anyone who wants to check. That's 0.36% of the total, small enough that dropping them doesn't meaningfully change any of the conclusions below.
Every timestamp is stored twice — once in UTC, once converted to the camera's local time. All hourly breakdowns run on local time, so when we say "peak," we mean peak in that camera's own city, not peak somewhere on a UTC clock.
The numbers, city by city
Every figure below comes from the 2026-03-01 → 2026-03-30 window.
| City | Rounds | Mean | Median | Std Dev | Min – Max | Over % |
|---|---|---|---|---|---|---|
| London | 1,731 | 36.4 | 39 | 18.2 | 4 – 89 | 51.2% |
| Tokyo | 1,722 | 39.7 | 42 | 19.8 | 3 – 96 | 49.8% |
| Sydney | 1,696 | 28.1 | 30 | 15.4 | 2 – 71 | 52.1% |
| Bangkok | 1,729 | 43.6 | 47 | 22.7 | 7 – 108 | 53.4% |
| New York | 1,718 | 44.2 | 46 | 19.1 | 9 – 102 | 50.6% |
| Taipei | 1,651 | 33.8 | 36 | 14.2 | 4 – 78 | 51.7% |
City by city: range-bet sweet spots
What the data actually shows
Six patterns pulled from the dataset, ranked by how much they matter.
- 1
Over/Under settles close to a coin flip across every city
Across the six cameras, Over/Under hit rates land between 49.8% (Tokyo) and 53.4% (Bangkok) — a spread of just 1.3 percentage points.
What it meansThat tight clustering tells us 155.io's threshold math is doing its job. Over the long run this is close to a 50/50 proposition for players, with the house's cut folded into the ~1.8× payout instead of a true 2× even-money line.
- 2
Bangkok swings the widest — and runs the highest average
Bangkok's standard deviation, 22.7, sits 60% above Taipei's 14.2, and its mean count of 43.6 beats Taipei's 33.8 as well. Range bettors working the Bangkok feed need 15-20-integer bands just to hold their hit rate.
What it meansIf predictability is what you're after, Taipei beats Bangkok hands down. Exact Count chasers also fare worse in Bangkok — roughly 1 hit in 28 attempts, versus close to 1 in 19 on the Sydney feed.
- 3
Displayed thresholds get outrun during peak hours
In every city's peak window, the counts we logged came in an average of 4.3 vehicles above the system's posted threshold. That gap holds up statistically (p < 0.01) in each of the six cities.
What it meansOver bets beat Under bets by 5-8 percentage points during peak stretches — the threshold math seems to slightly under-correct for rush-hour density we already know is coming. It's a real edge, but a thin one that vanishes the moment you step outside peak hours.
- 4
Weather moves the numbers faster than the algorithm keeps up
Heavy rain — 5mm/h or more, 41 instances in our sample — knocks counts down 12-18% city-wide. 155.io only nudges its displayed threshold about 6% in response, which leaves a real Under bias sitting on the table during wet weather.
What it meansBettors tracking live conditions can lean Under during a downpour in any of the six cities for an estimated 2-3% edge, though our rain sample is small enough that the confidence bands are wide.
- 5
Tokyo and New York both get a second wind after dark
Between 22:00 and midnight, Tokyo and NYC each show a secondary bump in counts as their entertainment districts empty out — 30-45% above the day's quietest hours.
What it meansThese late windows work well for Range bets: elevated but still bounded activity. London sees a smaller version of the same effect around 22:30, when the theatre crowds let out.
- 6
Exact Count hits aren't spread evenly across the range
Counts bunch up near each city's mean, so a bet on the median integer lands 2-3× more often than a bet on mean ± 1.5 standard deviations.
What it meansFor anyone determined to chase Exact Count, betting the current window's median beats picking a random integer in the threshold range. The math still favors the house — just by a smaller margin.
Where this study falls short
No dataset is perfect. These are the four gaps we think matter most.
Just one 30-day slice of the year
A single month only captures one season per hemisphere, and seasonal swings — Christmas in NYC, Songkran in Bangkok, summer in Sydney — likely shift these numbers. Our next study window runs April-June 2026 to catch the early-spring-to-late-spring drift.
We can't see inside 155.io's threshold formula
We only see the number the system displays, not the calculation that produces it. Any claim about over- or under-correction here comes from watching outputs, not from reading source code.
Not every feed holds up equally well
The Sydney and Taipei streams cut out briefly now and then, which knocks rounds out of our log entirely. We have no way to check whether 155.io's own internal count matched ours on those missed rounds.
One fixed angle per city, not the whole city
Each city feeds us a single camera at a single intersection. Calling this "London traffic" is a stretch — what we're really studying is one specific view of Piccadilly Circus, not London as a whole.
How things moved in Window B
In short: This is our second 30-day logging pass, running April 8 through May 7, built on the same method as Window A — every round, every camera, the posted threshold plus the AI count plus the outcome. We ran this window specifically to test the seasonal-drift questions Window A raised: does Bangkok settle down once Songkran ends? Does Sydney's spread shift as autumn sets in? And does the threshold algorithm actually adapt to any of it?
City by city, Window A vs. Window B
| City | Mean, A | Mean, B | Change % |
|---|---|---|---|
| London | 36.4 | 37.9 | +4.1% |
| Tokyo | 39.7 | 39.4 | -0.8% |
| Sydney | 28.1 | 26.3 | -6.4% |
| Bangkok | 43.6 | 39.8 | -8.7% |
| New York | 44.2 | 43.5 | -1.6% |
| Taipei | 33.8 | 34.2 | +1.2% |
What the new data adds
- #1
Post-Songkran, Bangkok's average fell 8.7%
Bangkok averaged 43.6 vehicles in Window A; that number is 39.8 in Window B — down 8.7%. Nearly all of that drop sits in the 11:00-16:00 stretch (-15%), where Songkran festival crowds had inflated the Window A figures. Late-night counts barely moved.
What it meansThe Over bets that paid off during Window A's Bangkok daytime peaks no longer have an edge — counts have settled back toward the threshold. Songkran (mid-April) is the single biggest seasonal swing we've logged so far; bettors who track the calendar should expect similar dips around other regional holidays.
- #2
Sydney's spread widened 14% as the autumn commute shifted
Sydney's standard deviation climbed from 15.4 to 17.6, a 14% jump, even as the mean count edged down slightly (28.1 to 26.3). Digging in, rain-day variability is the driver: April brought 11 rainy days versus March's 4, and counts on those wet days swing 30%+ from the dry-day baseline.
What it meansRange bets on Sydney now need bands 1-2 integers wider than in Window A to hold their hit rate. The "Taipei lowest variance, Sydney close behind" pattern from Window A doesn't hold anymore — Sydney has moved into mid-variance territory.
- #3
Tokyo and NYC's night peaks got stronger, not just repeated
The secondary 22:00-midnight peak we flagged for both Tokyo and NYC in Window A came back bigger in Window B. Tokyo's night counts now run 41% above trough hours, up from 33%. NYC's run 48% above, up from 40%. Warmer spring evenings look like the driver — entertainment districts staying busy later and longer.
What it meansThe night-window Range bets that worked in Window A (Tokyo 15-25, NYC 40-55) still work in Window B — and you can push the bands a bit higher now (Tokyo 18-28, NYC 42-58) for the same hit rate.
- #4
The threshold algorithm didn't react to the Songkran drop
For Window B's first two weeks, Bangkok counts sat 8-12% below threshold in the wake of Songkran — yet the posted threshold itself moved down only about 2% over that same stretch. That points to a long rolling baseline that smooths out event-driven dips rather than reacting to them.
What it meansBettors watching the calendar had a real 10-day window (April 17-27) where Bangkok Under bets carried a clear edge (roughly 6-9% lift). That window has since closed — Bangkok's threshold caught back up as April's usual patterns took over. This is the kind of edge that lasts days, not weeks.
- #5
The peak-hour Over bias from Window A held up
The 4.3-vehicle peak-hour overshoot we clocked in Window A is still there in Window B, now at 4.0 vehicles — a gap small enough to be noise. Over hit rates during peak windows keep running 5-8 points ahead of off-peak.
What it meansAcross both windows, this is the sturdiest finding we have: the threshold algorithm consistently under-corrects for peak hours. If there's one rule worth taking away from this whole study, it's Over bets, peak windows, any of the six cameras.
Where we land after 60 days of logging
Two of our findings hold up across both windows: the peak-hour Over bias (same size in A and B) and the Tokyo/NYC night peaks (a touch bigger in B). The Songkran-driven Bangkok edge, on the other hand, was a real but short-lived 10-day window — worth remembering as a calendar effect, not worth building a strategy around. We're not done logging.
What's coming next
Window B (April-May 2026) has already landed — scroll down to see how it compares. Our next 30-day pass runs May-June 2026, aimed at pre-monsoon Bangkok and the early creep of NYC's summer tourist season. Alongside that, we're setting up bet-by-bet logging with three operators to check the posted threshold against posted payout odds — that should get us a sharper read on house edge per wager type than this study alone can give.
Got feedback on the methodology, or want to follow future updates? Reach us at info@cctvgame.org — there's no NDA tying our hands with 155.io, and every value we log is already public during the round anyway.