Weather on your rides: why yours is blank, and how to fix it

Nothing is looked up until the ride has a location, and no location is ever invented for you. We do not guess from the address you signed in from — that says where you signed in, not where you rode, and a ride placed in the wrong city would make a fault report wrong rather than vague. So a blank card is the usual starting point, and one of the three below ends it. Do the first one you can.

One — upload a route. On the Upload page (the Upload button in the header), the second panel is “Add a route or a phone recording”: a Sensor Logger .zip, or a .gpx from any app that writes one. You do not have to find the ride first, because the recording’s own timestamps are matched to the right ride for you. This is the best of the three, and the only one that can also tell you whether the wind was against you.

Two — pin the ride, or the whole trip. Open a ride whose card is blank (Timeline, then the ride) and it asks “What was the weather doing?” above a “Town or city…” box: type a place, press Find, pick from the list. The reading appears straight away. The trip page (Trips, then the trip) carries the same search and covers the whole trip in one action.

Three — say where the bike is kept. On your bike’s page (My Bikes in the sidebar, then your bike) there is a card headed “Where this bike is kept”. Say it once and every short ride with no track and no pin gets a reading, for as long as the bike lives there; “Change” and “Remove” sit beside it. Short has a number on it — 25 km, and the next section but one is about why. The weather card calls this the bike’s base, and that is the phrase used from here on.

Do them in that order if you can, because a ride that already has a reading is not looked up again on its own. Add a track to a ride already placed from the bike’s base and the old reading stays until you ask for a new one — the card has a “Look it up again” button for exactly that, and it re-reads from whatever the ride now has, so a track that arrived late takes over from a pin or from the base. In bulk, pressing “Change” beside the base on the bike’s page and picking the same town throws away every reading that came from the base and tells you how many will be looked up again.

The timing differs by method, which is worth knowing before you decide something is broken. A pin is looked up the moment you pick the place. The bike’s base — and a track that arrives after a reading was cleared — is looked up as you open the rides; nothing runs in the background. A trip fills in up to twelve of its legs per visit, and while it is still filling the box says so in its own footer: “Covers 3 of this trip’s 7 legs.”

“Apply to every ride in this trip” is ticked by default, both on a blank card and in the “Rode somewhere else?” override. Leave it ticked when the whole trip stayed around one area. Untick it when the legs ended somewhere else — pinning the town where you stopped for lunch would otherwise label the leg that finished 200 km further on.

One case none of the three fixes: a ride whose own clock is wrong. Some logs report dates in 1970. Nothing dated before 1990, or more than a day in the future, is looked up at all, because a real-looking reading for a day the ride did not happen on is worse than no reading. The card goes on offering you the search box, and a pin will not help.

Two words are used throughout. A trip is one journey; each ride inside it is a leg. And every figure here is metric, as the weather card itself is — it stays in °C, km/h and mm even if the rest of your pages are set to miles. The 25 km limit below is about 15.5 miles.

Inside one ride the card draws an hourly strip: one cell per hour with its time, a weather icon, a temperature bar and figure, and the millimetres printed on any hour that saw at least 0.1 mm. Where the temperature actually moved across the ride, the coldest and warmest hours are highlighted.

Those hours are your own local time — the same clock as the ride’s Start Time beside them, and the same clock as the ticks on the charts below. The weather service is asked in UTC, because that is how the hours are matched to the ride’s own window, but every moment is carried as a full instant and printed on the clock your browser is set to. Only a reading stored before this changed has nothing but the bare UTC clock face, and where that happens the hour is labelled “UTC” rather than quietly shifted by a guess.

Rain is given as a total in millimetres plus each spell named by the hour it started and the hour it stopped, never as an average. Timing is exactly what an average destroys: a shower at the end of a ride and one at the start say different things about a wet motorcycle.

Wind is given as the strongest hourly mean for the hours your ride covered, with the hour’s peak gust beside it — “up to 18.4 km/h gusting to 41.8 km/h” on one real trip, read on 2026-09-13. The pair is the point: a gust factor is a ratio, and a peak cannot be recovered from an average. Both figures are properties of the hour and not of the rider. The bike may have been parked for the worst of it.

Humidity is the peak humidity of those hours, and since this September it appears on the trip box as well as the ride card. It is not trivia: it is the one figure the damp check reads when deciding whether an isolation fault has the excuse its own advice offers.

Cloud cover exists for the picture and nothing else. Rain alone tells you only wet or dry, and a sun icon drawn from “it was not raining” would be claiming a sky nobody looked at — so cloud is fetched to choose the icon, and the clear-sky sun is only drawn below 30% cloud. Where there is no cloud reading the icon falls back to a neutral dry one. It feeds no range figure and no fault check. It does one other job, in the sun-soak section below.

On a trip, the box now says “No rain during the legs” where it used to say “No rain recorded”, and the change is a real limit rather than a rewording. The rain spells are drawn only from the hours the legs themselves covered, so a downpour during a two-hour coffee stop sat outside every window the box ever looked at. The box describes the hours you were riding, not the day.

When the weather service does not answer, the page loads without weather and nothing is kept — open it again later and it tries again. When a ride cannot be placed at all, nothing is substituted for the missing location: the card says so and offers you the search box instead.

A GPS track beats a pin, and a pin beats the bike’s base. The best one a ride has is the one used, and every card stamps its reading with which it was, in three words: GPS track, your pin, or bike’s base.

Where a ride has a track, the point used is the middle sample of the recorded positions — not the geographic midpoint of the route and not the halfway point in time. Only the place comes from there. The direction of travel behind the headwind figure is worked out separately, across the whole track.

A pin belongs to a ride, and one pin has covered a whole trip since the feature shipped in August 2026: that is what the tick box on the ride card does. What September 2026 added is the picker on the trip page itself, where the whole trip is the only sensible scope and there is no box to tick. Once a ride carries a pin, the card offers “Change place” to move it or take it away.

A pin is one place for a whole ride, which is the wrong shape for a ride that went somewhere. If a ride lasted more than an hour, the card lists its hours and you can name a place beside each one — and each hour is then answered by its own place, so the coldest hour of a ride that ran from the coast into the hills is genuinely the hour in the hills rather than a number the coast produced. You never type a time: the hours come from the ride itself and you only choose a place beside one of them.

A place you name stands for the hour it names and for at most one hour either side of it. Beyond that the hour is left unplaced, shown as a gap in the strip rather than filled in, and the card says how many of your hours it could describe. Leaving it as a gap is the point: carrying the last place forward indefinitely is not a refusal to guess, it is a guess — and it would confidently hand an inland hour to a town on the coast. Where a place does cover an hour you did not name yourself, the row says so.

None of this makes a ride into a route. There is no map and no line drawn between the places, because a line between two pins reads as a road, and the road between two towns is not the straight line between them — you may have gone inland and come back. For the same reason a ride placed this way still shows no headwind figure, and the card says so rather than leaving the space blank: two places an hour apart give a line between two towns, not the direction the motorcycle was pointing, and getting that direction wrong by enough turns a headwind into a tailwind. Only a recorded GPS track earns a direction of travel.

Naming several places is still your own account of where you were, exactly as one pin is. It is more detail, not better evidence: the card still says “your pins”, the reading is treated everywhere as a pin, and no number of remembered places becomes a measurement.

The bike’s base is one standing statement about the motorcycle rather than about any ride, changeable or removable whenever you like. It exists because the two stricter ways asked riders for something nobody was going to do a few hundred times: measured on 2026-09-13 across 13,677 rides on 167 motorcycles, 16 carried a GPS track and not one carried a pin. At roughly 82 rides per motorcycle, asking for a pin per ride asked for none.

The base is only ever considered for a ride with no track and no pin, and even then only for a recorded distance of 25.0 km or less. The test is literal: exactly 25.0 km passes, 25.1 km does not. The reason is geometry. The models divide the world into squares and give one answer per square, and the squares of the older of the two datasets are about 25 km across — so a ride no longer than 25 km cannot have ended more than 25 km from where it began, because the distance travelled is an upper bound on the distance moved, and the claim that the weather where the bike lives describes the weather the bike rode in is bounded by how far it could possibly have got. On a longer ride nothing bounds it. A ride with no recorded distance at all is refused too, because “we were not told how far” is a different claim from “short”.

The limit is not a corner case. Measured on 2026-09-13 over those 13,677 rides: 12,388 carry a distance at all, and 10,318 of those are 25 km or shorter — 83% of the rides that can be measured, and three in four of all of them. The median ride is 8.0 km. The other 2,070 measured rides are longer than 25 km and are refused, and the 1,289 with no distance are refused as well. When a ride is refused the card says which of the two happened, with the number: this ride covered so many kilometres, past the 25 km inside which a base still describes the weather, or else that we cannot tell how far from home it went.

One short ride taken a long way from home is exactly the case this rule cannot catch, and there is a control for it. A card showing a base reading also offers “Rode somewhere else?”: say where that ride actually happened and it is used instead — for the whole trip, unless you untick the box. Your bike’s base is untouched either way and is still right for every other short ride it does. It is worth doing for more than tidiness: a pinned ride counts as a location the fault checks may read and a shared link may publish, and a base-placed one is neither.

Clearing 25 km buys the base one thing only — permission to describe your own ride page. Where you say the bike is kept is stored against the bike and shown only to you. It appears in no shared link and on no public profile, and any weather drawn from it is refused publication outright, so the most a reader of one of your links can learn from it is that some legs carried no weather they were allowed to see.

The search follows the language your browser is set to read in, so an English browser finds Rome and an Italian one finds Roma. If your town will not come up, try its local spelling, or a larger town nearby.

The pin does not have to be exact. It selects which square of the model the reading comes from, and those squares are 2 to 25 km across, so a pin a few streets out changes nothing. A few kilometres out can genuinely land in a different square, so pin the town you rode near rather than one two valleys away.

The town search is made by our server and never by your browser, and it goes to Open-Meteo’s own place service. That is a different thing from the town printed on a shared trip link, which is resolved from a list of towns kept on our own server — see the sharing section.

Wind and rain are the part your logs genuinely cannot see. A Zero records its own outside-air temperature on most rides and records nothing at all about what the air was doing, so this is new information rather than a second opinion on something the motorcycle already measured.

Two wind figures sit side by side. The first is the strongest hourly mean across the hours your ride covered. The second is the highest hourly peak gust — a separate figure asked for separately, because a peak cannot be recovered from an average. A wind range is only shown as a range when the spread across those hours is at least 8 km/h, and a temperature range only at 1.5 °C; anything narrower is model noise, and one figure is shown rather than implying a change the model cannot support.

The headwind is a third thing entirely: how much of the wind was blowing straight into you, worked out hour by hour and averaged across the ride. Because it is an average, a ride with one very windy hour and three calm ones reports a small number, which is correct and can still read as surprising. It is called a headwind when that average is more than 2 km/h against you and a tailwind when it is more than 2 km/h behind you; anything in between is a crosswind, and the card shows nothing at all rather than telling you the wind was sideways. On a trip the same rule is applied to the average of the legs’ own figures, and only when every leg had a direction of its own.

That is what a track is for. Working out how much of the wind was against you needs a direction of travel, and only a track carries one — a pin and a base are both stored without any direction, deliberately, because a pin says where a ride was and never which way it went. A track is necessary but not sufficient: we use the whole track to work out which way you were heading, so a loop or an out-and-back whose headings cancel out has no net direction, and some tracked rides carry no headwind figure either.

So read the absence carefully: no headwind figure never means there was no wind. It means no direction of travel was recorded, or the ride had no net direction, or the wind was within 2 km/h of sideways. The wind speed and the gust are shown either way, and where there is no direction the written ride summary is told so outright — no direction of travel recorded, so no headwind figure — rather than leaving the gap to be noticed.

“Modelled air” is the phrase used on the card and on the chart, and it means one thing: the outside temperature the weather service says it was, as opposed to what your bike’s own sensor read. The two are different measurements of different things and the site keeps them apart.

Your motorcycle’s own reading is the better record of what the pack sat in, and none of this replaces it: on 2026-08-10, 8,712 of 9,511 rides already carried a temperature the bike itself recorded — 92%. The modelled air is shown beside it and given to the written ride summary. It is never written over the bike’s own reading, never feeds the temperature-adjusted capacity figure, and never reaches the fault checks, which take only wet-or-dry from the weather.

On the ride page the modelled air is drawn on the bike’s temperature chart as a labelled dashed reference level, never as a second line: one level at the model’s warmest hour, or two — modelled air max and modelled air min — when the modelled temperature varied by at least 1.5 °C across the ride. It stays visible even when it sits a long way from what your bike reported, which is the only case anyone needs it in. A level rather than a curve because of the shape of the data on both sides: the model’s series is hourly, and 86.5% of rides last an hour or less — 13,573 rides with a believable duration, median 19 minutes, nine in ten under 78 minutes, measured 2026-09-13. A typical ride touches one or two hourly samples, and a curve drawn through two points beside per-second data from the bike would be invented detail.

The one figure on the weather card that involves both measurements is the sun-soak note. When your bike’s own average for a ride sits 8 °C or more above the warmest modelled hour of that ride, the card says so in words: it had been sitting in the sun, rather than the day being that hot. The worked example behind the rule is a bike averaging 38 °C against a modelled 22. The comparison is made against the warmest modelled hour, the most forgiving of the modelled figures for that window, and the 8 °C margin is wide because a false sun-soak note would discredit a reading that is usually right.

Since September 2026 the note only blames the sun when there was sun to sit in. Above 60% cloud the temperature gap is still reported and the explanation is withdrawn: the card says it was cloudy, so the sun is not the explanation, but the reading is still higher than the air around it. The bike may have come out of a warm garage instead. The gap is the measurement; the sun is a cause, and a cause has to be plausible.

Either way the note says the reading is inflated, not that it should be thrown away, and it is an inference rather than a measurement — the comparison is against a square of map 2 to 25 km across, not against a second sensor on the motorcycle. Two numbers are worth keeping apart here. The temperature-adjusted line on your capacity chart is built from the outside-air temperature recorded during a charge, while the sun-soak note is worked out on a ride — the same sensor, a different session — so the note itself does not move your capacity chart. And your state-of-health percentage is not temperature-adjusted at all, so a sun-soaked reading cannot move it by a point.

One caveat belongs on the note itself. It is raised whatever placed the ride, including the bike’s base, so on a short ride taken far from home the comparison can be against the weather somewhere you were not, and a genuinely hot day can be labelled as sun-soak. That is the other reason to use “Rode somewhere else?” on a ride you know was elsewhere.

Several useful-sounding things could be computed from a wind, a humidity and a temperature, and are deliberately absent. Each of these is a decision rather than a gap waiting to be filled, and each is the kind of claim that would be believed because it appeared beside real numbers.

Air density, and any drag, aerodynamic, Wh/km or range figure drawn from the weather. This one was designed, worked out and then refused. The physics is settled — four independent derivations of the same sample agree to five figures, giving 1.1476 kg/m³ for 31.5 °C at 40% humidity and 1010.6 hPa, about 6% below the standard-atmosphere value — and it dies on wiring rather than on arithmetic: there is no drag model on the ride side for a density to enter, so the number would have had nowhere truthful to land. One thing to unlearn while it stays unbuilt: humid air is less dense than dry air, not more. “Humid air is thicker” or “heavier” is the single most likely wrong sentence anybody would write from this, and it is wrong.

Anything about the road. Not that it was wet, not that it had dried out, nothing about grip and nothing about braking. The reading describes the air over a square of map kilometres across, an hour at a time. Whether the surface under you was wet depends on what fell before your ride, what the surface is made of, whether it is shaded, and whether anything had driven over it — none of which is in the answer, and none of which the site will pretend to.

Anything about tyres, and above all anything about tyre pressure. A Zero has no tyre-pressure sensor, so there is no reading to correct, nothing to compare a cold morning against, and no way to notice if you were wrong. A “your pressures will have dropped overnight” line would be arithmetic performed on a number nobody measured, printed beside numbers that were.

The modelled air as a cell or pack temperature. It is the air roughly two metres above the ground in the model’s world. A cell inside a case, behind bodywork, after an hour of discharge is somewhere else entirely, and reading one as the other is how a perfectly normal pack starts to look as though it overheated. Your bike’s own temperature channels are what the battery pages read, and they are the only thing that may answer that question.

And a dry hour is never reassurance about ice. The figure is a two-metre air temperature averaged over a square of map; bridge decks, shaded cuttings and valley floors sit several degrees below it. A positive number on this card has never been a statement that the road was not icy, and nothing on this page is riding advice.

Weather can only ever make a report stricter, never kinder. Rain never costs you anything. Dry weather can cost five points, and only when your logs already show an isolation fault — the fault is what costs the points, and the dry weather only removes the “it was just damp” explanation.

One check reads the weather, and it fires only when the weather across a window says dry and a fault our own catalogue blames on damp is actually present in that same window. There is no branch for wet anywhere in the fault checks. Rain can explain a fault; it can never soften, cancel or downgrade one.

Exactly one of the catalogue’s 36 fault types is marked as blaming damp: the chassis and high-voltage isolation fault. Your logs may write it as Chassis Isolation Event, Low Chassis Isolation, Chassis Isolation DANGER, or — most often, because these two are built from binary message types rather than from text — Chassis Isolation Fault or MBB BMS Isolation Fault. Its published advice says it often clears once the system dries out. That sentence is the excuse, and it is the whole worked example.

If your logs carry that fault across a window the weather says was dry, the excuse does not apply and the page says so. A note appears on the log’s diagnosis panel reading that the conditions during this window were dry, so damp or condensation does not explain this fault. The five points come off the site’s own rule-based score for that log — the number shown as “rule score 95/100” on the deep-analysis card, and handed to the AI as ground truth it is not allowed to contradict. Beside a cell-imbalance problem the same dry weather adds nothing at all: the imbalance is still reported, but imbalance is not a damp story, so the check does not fire.

That note is capped at the severity the catalogue already gives the faults present in the window, and never invents a worse one. Worth knowing, because it is the inference this page exists to prevent: the only damp-blaming fault type is ranked “worth a look”, so a damp note showing as critical means an unrelated critical fault is present in the same window — not that dry weather made the isolation fault critical.

What counts as dry is decided in a fixed order. First the reading has to be allowed to speak at all: only a GPS track or a pin you dropped count, so a reading drawn from the bike’s base answers “we do not know” however hard it rained there. Past that gate, any recorded rain makes the ride wet. With no rain, humidity decides — 85% or more at its peak is wet, 55% or less is dry, and anything strictly between, 70% say, answers that we do not know and fires nothing. A ride with no humidity reading at all answers we-do-not-know too, because absent humidity is not dryness. The gap between the two thresholds is what stops “we do not know” being rounded down to dry, which is how a ride would lose an alibi it never earned.

Across a window of several rides the asymmetry continues. One wet ride makes the window wet. Dry requires every ride carrying a reading to say dry, and a ride whose reading is inconclusive or base-placed casts no vote while still stopping the window being called dry. Rides with no reading at all sit outside the count entirely. One scope note for honesty: the automatic check reads up to five hundred of the window’s rides, so on a very long window it is a large sample rather than every last one. Either way a base-placed reading can withhold an answer and never supply one, which can only make a verdict milder.

The same question is asked in two places on the site. The one that runs automatically when a log is analysed looks at the fault types in the whole log rather than a top six, and at every ride on that motorcycle. The other asks over a date range you choose, looks only at the six highest-ranked fault types it is showing you, and counts only rides from your own uploads — plus any ride with no source file left to attribute. That second one has no screen on the site today, so what you see on a log’s page is the first of the two, and it is the one to trust.

One thing is code and one is instruction, and it is worth knowing which is which. The fault checks cannot see a base-placed reading at all — that is enforced. The AI that writes your ride summary can see it, is told plainly that it is where the motorcycle is kept, and is instructed to treat it as background and never to explain a fault with it. That second one is a rule we give the model, not a lock. The promise that survives both is the plain one: weather may only ever make a report stricter, and that is checked before every merge against a grid of seven bikes — from a clean one to one carrying a catalogued critical fault — each crossed with wet, dry and no reading. A bike with a 20 mV cell spread ridden in the rain comes back identical to the same bike with no weather attached.

Everything except your bike’s own thermometer comes from Open-Meteo, and none of it is a reading from a weather station near you. Rides from 2017 onwards use a live weather record — successive real model runs, each started from measurements, joined end to end. Older rides use a historical reconstruction: consistent with itself, but not a record of what anybody near you observed. The card always says which one answered.

Which one is asked is decided by the ride’s date, and it is a date rule rather than a fallback: if the live record fails on a ride it covers, the answer is that we do not know, not a quiet substitution of the coarser reconstruction. A trip takes the note of its weakest-placed leg, so a reconstruction cannot be laundered into an observation by a better leg sitting beside it.

What leaves this site is a point per place you named — one for most rides, and at most one per distinct square of map when you have said where you were hour by hour — each rounded to about a kilometre and sent from our server. Never your browser, never your name, never your route, never anything that identifies the motorcycle. Open-Meteo sees a square of map and a date. The rounding is also why two places three kilometres apart cost one lookup and not two, and why a day’s riding in one area costs a single lookup however many rides it held.

Seven things are asked for and no more: air temperature, humidity, rain, wind speed, wind direction, peak gust and cloud cover, hour by hour. The list is deliberately short, because every extra thing we ask for is bandwidth on a free service we are a guest of — no key and no account is used, as Open-Meteo asks for none for non-commercial use. The gust is the newest of the seven, which is why a reading looked up before September 2026 shows no gust: absent means absent, never stale.

Once a ride has a reading, it keeps it. Nothing expires off your pages. The two ways a reading is deliberately thrown away are both yours: taking a pin off a ride, and changing or removing the bike’s base, which discards the readings that came from it and tells you how many.

A shared trip link does name a town, and it comes from your route rather than from the weather. The link publishes the route you chose to publish — if you asked for the ends to be trimmed, they are trimmed before anything is published — and the name is then read off the ends that survive, against a list of towns kept on our own server, so no coordinate of yours goes to anybody else to be named. A route whose start you trimmed is named from where it visibly begins. Your shared links never carry a place name or coordinates from the weather itself.

Weather on a shared trip is off until you tick it, like every other thing a link can carry, and the choice is fixed when the link is made. On a link already in someone’s hands you can change the profile listing and the two route-privacy settings, and those two may only ever be tightened. So showing weather on a trip you have already shared means publishing a new link, with a new address.

Even with it ticked, the link does not simply publish what your own page shows. It works the trip’s weather out again for publication, and two filters run first. Legs placed from the bike’s base are dropped — only a track or a pin may be published. And legs whose reading was looked up after the link was created are dropped too, because a reading arrives the first time somebody opens a ride and, without that bound, a link shared with no weather could grow some by itself. A reading with no date on it is refused as unprovable rather than assumed to be old.

A reader gets the sky and its timing: lowest and highest temperature with the hour each happened, the strongest wind and the peak gust, the humidity, the rain spells with their start, end and millimetres, and — only when every contributing leg was tracked — the headwind. Alongside them come the coverage counts, how the surviving legs were placed, whether they count as evidence, and which dataset answered. The sun-soak note is not published at all: it quotes your bike’s own sensor, which is governed by the metrics toggle rather than by the weather one.

Because those filters run before the summary is worked out, a link can occasionally show more in one respect than your own page does. A trip with one tracked leg and three placed from the base reads on your page as base-placed with no headwind, since every leg carrying weather contributes there; published, the three are dropped first, so the link reports the surviving leg’s placing and its headwind. It still discloses less overall — narrower in what it reveals, not narrower in every respect.

Partial coverage is printed rather than hidden: a trip whose legs do not all carry publishable weather says so with both numbers, and the larger of the two counts every leg, including the ones dropped for being base-placed or looked up late. If every leg is dropped the link publishes no weather at all, rather than a coverage line reading zero.

Two edges are worth knowing. A shared bike recap never publishes weather in any form, and the share settings report the weather choice as not applicable for a recap rather than as on or off. And the written AI paragraph kept on a public trip page is not held to the publish-time bound the weather block is: if you press refresh on it, that paragraph can mention conditions looked up after you handed the link out. It still carries no place name and no coordinates. If that matters to you, leave it as it was published.