Google Maps tells you what traffic is doing right now.
It does not tell you when to put your shoes on.
Every Angeleno knows the routine: refresh Maps, do mental subtraction, guess.
This does the arithmetic instead. Tell it where you're going and when you need
to be there, and it picks the latest departure that still gets you
there on time, counts down to it, and keeps re-checking traffic in
the background until it tells you to go.
UCLA to LAX Terminal 7, arriving 6:45 PM. Relaxed says
24 minutes of driving. Important says 61.
Same route, same target — the mode decides whether you leave at 6:15 or
5:38.
3
risk profiles side by side, so you choose your margin per trip instead
of guessing one buffer for everything.
37 min
the spread between Relaxed and Important on that one real trip. That is
the decision the tool exists to make.
9
Routes API calls per poll cycle: three sample times × three traffic
models, because sampling only "now" is the wrong question.
1 file
A single self-contained HTML page. Paste in your API key and open it.
↓ where the three modes come from
Not invented. Google already publishes the range.
The three profiles are not a heuristic layered on top of Google. They are
Google's own trafficModel parameter, queried three times:
Relaxed = OPTIMISTIC
Assumes conditions are better than typical. The latest leave-by time.
For lunch with a friend who is late anyway.
Usual = BEST_GUESS
Google's realistic point estimate from live and historical data.
Important = PESSIMISTIC
Assumes worse than typical. The earliest leave-by time, with margin
built in. For flights and interviews.
That spread is exactly the "typically 22–50 min" range the consumer
Maps app shows you. The difference is that Maps hands you a range to eyeball
and the developer API hands you three numbers you can compute with. The tool
is not predicting traffic better than Google — it is
turning Google's own uncertainty into a decision.
Three models, three modes.
trafficModel is silently ignored unless
routingPreference is TRAFFIC_AWARE_OPTIMAL. No
error, no warning — you just get the same number back for all three
models and it looks like real data.
The gotcha that cost an embarrassing amount of time
↓ the design decision
Sampling the future, not the present
The naive build polls "how long is the drive right now" every few minutes.
It works, and it answers the wrong question. If it is 5:00 PM and you plan to
leave at 6:00, current conditions tell you about the 405 at 5:00 — and the
405 at 5:00 and the 405 at 6:00 are close to different freeways.
So instead the tool samples inside your likely departure window.
On startup it fires one anchor probe for a departure 20 minutes before your
target, uses that baseline to compute a realistic departure window, and then
each cycle samples two times within that window plus one "now" sample
for live conditions. Three samples × three traffic models = nine calls per
cycle.
Anchor, then window, then samples inside the window.
↓ using it
Open a tab while you're getting ready
It is a single-page app you leave open in a browser tab. It shows the
countdown, updates the recommendation as traffic shifts, and when the moment
arrives it fires a browser notification and an on-screen
alert so you do not have to keep looking at it.
↓ what it costs and what it can't do
You bring your own API key
This is the real friction and it is worth stating plainly. The tool needs a
Google Maps API key with four APIs enabled — Routes, Maps JavaScript, Places
(the classic one, not "New"), and Directions — pasted into the HTML file. It
is not a hosted service you can just click into, and at nine calls per poll
cycle a long monitoring session is not free if you exceed Google's free tier.
It also inherits every limitation of its source. If Google's pessimistic
estimate is wrong about a crash that has not happened yet, so is this. The
tool improves the decision, not the underlying forecast.
Finish · what it is built on
Where the numbers come from
Routing
Google Routes API, queried three times per sample with different trafficModel values
Required setting
routingPreference: TRAFFIC_AWARE_OPTIMAL, without which the traffic model is ignored
Sampling
One anchor probe, then two in-window samples plus one live sample per cycle
Alerting
Browser notification plus an on-screen alert at the leave-by moment
Build
One self-contained HTML file. No server, no build step, no framework
Setup
Your own Google Maps API key, pasted into the file
