1 What a correction is here
That is the whole design, and it is what makes this safe to use. An entry is a lens the platform looks through, not an edit to your season. Applying one takes effect instantly, everywhere — the player, the Explorer, the tuning guide, the sail wind bands — including data summarised months ago. Removing one puts every number back to what it was, with nothing to rebuild.
The calibration lens explains why a whole season can turn on a single row like that.
The correction itself is linear — value × scale + offset — and that is not a simplification. Instrument calibration in practice is a zeroing and a span: a masthead a few degrees off, a paddlewheel reading 2 % slow, a heel sensor mounted upside down, which is a scale of −1 and is entirely legal.
2 Where it lives
Open Boats and take Calibration from the boat’s row, beside Polar and Targets. Same gating as those: admins and owners add and delete entries, everyone reads, and it is never billing-locked.
The page runs in three parts, top to bottom: the measuring tool, the form, and the record of what is in force. Measuring is the accurate way to reach a number, and the form is where you land either way.
3 Before you start: heading and boat speed
Nothing on a boat measures true wind. TWA, TWD and TWS are computed — from the masthead unit, certainly, but also from the boat’s heading and its speed through the water. So the order of work here is not a matter of taste. Correct the compass and the log first, and only then go near a wind channel: a log reading 2 % slow puts an error into every wind number the boat produces, and a wind correction measured through it is a number carrying somebody else’s mistake.
- Heading — compare against COG. Motor in flat water with no current and read the compass against course over ground. Do it on several headings rather than one: a compass is usually wrong by a different amount on different bearings, and one reading cannot tell a constant offset from that.
- Boat speed — compare against SOG. Same conditions, and take it across the speed range and at the heel angles you race at. A paddlewheel that reads true upright and slow at 20° of heel is entirely ordinary, and a single number will not describe it.
The Boat speed vs SOG row on the Analyze panel is exactly that second comparison, run over a window of your own sailing rather than over a measured mile. It stays informational and never becomes a recommendation, for the reason guided calibration gives: current flows straight into the ratio, and no amount of data separates a slow paddlewheel from a knot of tide. Read it as a prompt to go and check in flat water — not as the check itself.
4 Entering a correction by hand
Under Add a correction:
- Channel — start typing and pick from the list. This is the same channel registry the import mapper and the Explorer use.
- Effective from (UTC) — the moment the boat actually changed. See below.
- Offset — how much to add, in the channel’s own unit, which the field shows you.
- Scale — a multiplier. Leave it at 1 unless the instrument reads proportionally wrong. An instrument 2 % slow needs
1.02. - Note — optional, and worth it. “Masthead unit re-aligned after the rig came out” is the sentence you will want next season.
- Press Add correction.
Effective-from, which is the field people get wrong
A correction reaches forward only, from the instant you give it. Samples before it are untouched by that entry.
So the date to give is when the instrument started being wrong — or, more usefully, when it stopped: the Tuesday the rig came out, the morning the vane was re-seated. Almost always a date in the past, because you discover the error after the regatta. Future dates are allowed too, for a change you know is coming.
5 Angular channels have no scale
Pick a channel that carries an angle — wind direction, wind angle, heading — and the Scale field disappears rather than greying out, with the reason printed beside it.
An angle correction is a rotation. A scaled angle is not an angle: it wraps, so multiplying it produces a number that is no longer on the circle it came from. The offset rotates the channel and wraps properly — 359° plus 3 reads 2°.
Which channels are angular is the platform’s registry answering, not a guess made in the browser. A channel the registry does not carry keeps its scale field live, because naming an unknown channel is the server’s job.
6 Reading the timeline
Entries are grouped per channel and listed in the order the server consults them — newest effective-from first, and on a tie the entry that wins above the one it beats. The row order is the resolution order, so you never have to work out which applies.
| State | Means |
|---|---|
| In force | Corrects everything from its instant on. Today’s answer. |
| Scheduled | Its effective-from is in the future. It corrects nothing yet. |
| Earlier | Still correcting its own stretch of the season, up to the point where the next entry takes over. |
| Superseded | A later-created entry shares its exact instant and wins, so this one applies to nothing at all. It is listed and dimmed, never quietly dropped. |
An entry with offset 0 and scale 1 is legitimate, and the row says what it does: reads pass through uncorrected. That is not a mistake — it is how a crew stops an earlier correction applying to later sailing without deleting the history of why it existed.
Anything earlier than the bottom row of a channel reads uncorrected. A sample is corrected by the first row at or before its own time.
7 Deleting one
The × at the end of a row. The confirm makes the promise concrete: nothing recorded is lost, the samples were never changed, and every read of that channel goes back to the numbers it served before the entry existed.
One thing the confirm adds that is easy to miss: deleting an entry does not necessarily leave the channel uncorrected. If an earlier entry for that channel remains, it applies again from its own effective-from. You are removing a layer, not clearing the stack.
8 Measuring one instead of guessing
Analyze this boat’s sailing reads a window of your own racing and tells you what it thinks the instruments are getting wrong.
The physics is worth one sentence, because it is what makes this work:
A constant angle error is the same on every beat, so it shows up as the boat pointing better on one tack than the other, every single time. Headers and lifts are much bigger, but they are not attached to a tack, so over many tacks they average out and the constant does not. Over one day of sailing, weather cannot fake a constant.
That constant is two things at once: where the masthead unit sits, and today’s wind shear — the breeze veers with height, so the vane at the masthead sits in slightly different wind from the mean your sails work in, and that turns the whole wind picture exactly as a crooked vane would. One window of sailing cannot separate them, and the same number corrects both. Guided calibration has the algebra, the magnitudes, and the two schools of thought about whether to correct shear at all.
The practical consequence lives here rather than there: analyse each sailing day on its own, and give each entry that day’s effective-from. The mounting half stays put until somebody touches the rig; the shear half belongs to the day. Keep that habit and your timeline becomes the boat’s shear log — the part that repeats is the mounting offset, and the movement around it is what the breeze was doing above the deck.
Running one
- Set From (UTC) and To (UTC), or use a quick pill — Today, Yesterday, Last 24 h, Last 48 h.
- Press Analyze.
Pick one sailing day with a proper beat in it. That is what this is built for, and 48 hours is the cap, because the analysis runs while you wait. A day of one-way sailing has nothing to compare and says so.
The analysis reads through the corrections already in your timeline. So after applying one, run the same window again: a good correction drives the next answer toward zero, which is the loop converging rather than a second opinion.
9 Reading what came back
The result opens with the window it read and how many tacks and gybes it found in it. Then cards, each naming the method that produced it:
| Method | What it measures |
|---|---|
| Tack symmetry | Your median close-hauled wind angle on port against starboard, over the steady stretches between tacks. Half the difference is the offset. |
| Wind-axis cross-check | The reported wind direction either side of each tack. Section 10. |
| Boat speed vs SOG | Speed through the water against speed over ground. Never a recommendation — any current along your track goes straight into the ratio, so a slow paddlewheel and a foul tide look identical. |
Every card carries its evidence: how many steady segments on each tack, how many tacks were used, and the spread — how much those segments disagreed, in the same unit as the number beside it. Small spread against a large finding means every stretch of sailing told the same story.
What each tack held
The tack-symmetry card also shows the two boards side by side: accepted time, median wind angle and median boat speed on each. The two wind angles differ by twice the correction above them — the arithmetic, in two rows.
The boat speed is there for one reason. Weather cannot fake a constant, but a crew can. A board sailed a shade lower and faster carries a wider wind angle on every beat of the day, which is exactly what a crooked vane looks like — and the fast board is the board sailed low. When the two speeds differ enough to matter, an amber note says so: some of the asymmetry may be the sailing rather than the vane, so confirm the crew held targets on both boards before you apply.
It is a caution, not a gate. The suggestion stays and the Apply button is unchanged — a boat can be both out of calibration and sailed unevenly, and only the crew knows which. If the boat carries no speed log, a plain line says that instead; an absent measurement is not a passed check.
The wind finding is one act over two channels. True wind angle and true wind direction get the same rotation and are written together, because calibration deliberately does not recompute derived channels. Half of that pair applied would leave your wind angle and wind direction disagreeing by the size of the correction — worse than doing nothing.
Apparent wind angle is deliberately absent, and the card says so, since it is the reader’s next question: the wind triangle does not carry a masthead rotation into apparent wind one for one.
Findings that did not clear their own bar are not hidden. They appear under Measured, but not recommended with the reason named, and offer Use this in the form below rather than a one-click apply — a weak answer you can see is worth more than a silence you cannot.
10 The cross-check, and why it has no button
The cross-check does not vote on the finding above it. It measures a different half of the error.
- Confirms the suggestion — the reported wind axis barely moves across a tack, which is exactly what a constant angle error looks like: it shifts both tacks alike, so it cancels here. That is the evidence the asymmetry really is a constant, and a constant is what an entry fixes.
- Something a single correction will not fix — the wind axis steps across the tack, and a constant cannot do that. Something that flips with the tack is in your data: leeway, current across the course, a vane error that changes with heel. Treat the suggestion above with care.
One useful property, worth knowing when the amber boat-speed note appears above: this check is largely immune to a board sailed low. The heading falls by the same amount the wind angle widens, so their sum — which is what wind direction is — does not move.
There is deliberately no way to apply the cross-check. What it measures is the half of the error that flips with the tack, and no single offset or scale corrects that — offering the button would be offering something wrong by construction.
The upwash profile, directly under it
The cross-check reports one step for the whole window. The block under it reports the same step, resolved by wind speed — one row per 2-knot band, with tacks and gybes kept as separate profiles. Reading order is the argument: one number, then the curve that number was the average of.
What it is showing is your boat’s own upwash signature. The sail plan bends the air before it reaches the masthead, so the vane reads a wider angle than the boat is really sailing — on both tacks, which is why it lands as a step across the tack rather than as a constant. Light air bends more of it. Bigger in light air and smaller as the breeze builds is the classic shape.
Three things to read carefully:
- The step is the full step — the whole difference between the two tacks, which is twice the error on either one of them. Anything that ever corrects it carries half the step, never the step. The form below takes degrees, so this is worth getting right.
- A thin band is shown hollow, with its count and no number. Under three maneuvers a median is not a median. The row keeps its place rather than disappearing, because a gap would read as wind the boat never saw.
- Where the boat records backstay load, the tacks split in two at the window’s own median load, to test one prediction: that the step turns positive under load as mast bend twists the masthead unit to windward. The block prints the confound rather than burying it — load rises with breeze, so that split is substantially also a split by wind speed, and no split of sailed data separates the two. Each half carries its own median wind speed for exactly that reason.
Like the cross-check, the block itself applies nothing, and half of that is for the same reason: an error that varies with the wind has no single number to enter, and a per-band correction is not the linear kind the lens can carry. That is not a gap in the feature — it is why a correction table is a separate object rather than another row on the calibration timeline.
What the block does offer is Use as a correction table, which fills the correction-table editor on the same page from the bands measured here: half of each step, negated, tacks into upwind and gybes into downwind. Then it stops. Nothing is written and nothing is queued — what you have is a draft in your hands, to check, edit, save or throw away. Hollow bands hand over nothing, for the same reason they are hollow: you may type a number in there yourself, and the platform will not offer one.
The correction-table editor never saves as you type, which is the one place it differs from the tuning-target tables that look like it. A target table is the boat’s current state, so writing it back on every blur is right. A calibration entry is a dated row on a timeline, and an autosave here would append one entry every time a cell lost focus — a season of history nobody wrote, to be read through afterwards. So the panel is a draft until you press Save as a new correction, and that press appends exactly one entry, effective from the instant you give it. Entries are immutable: correcting a table means saving another one, or deleting the one you saved.
The two timelines are independent, and neither can produce the other’s finding. The linear entries above apply first, then the table. A table adds the same widening on both tacks, so it never produces a mounting-offset finding; a rotation adds the same constant to both tacks, so it never produces a step. That is why you can enter them in either order and why one does not have to be undone to work on the other.
Before you save one, read the upwash correction table. It is the arithmetic you are being handed — half the step and never the step, what happens between the bands and past the ends of them, and the eight limits it keeps on its own page.
11 Applying a suggestion
- Read the card, its notes and its evidence line.
- Check the cross-check verdict.
- Set Effective from (UTC). It defaults to the start of the window you analysed, not to now.
- Press Apply and confirm.
That default matters. “Now” would leave the very sailing that produced the evidence uncorrected, so re-analysing it would report the same offset for ever. The window’s start makes the loop close. And if you know the instrument changed earlier — the rig came out the Tuesday before — set it earlier: the correction reaches forward from whatever instant you give it.
After an apply, the result on screen is marked stale, because it was measured before the correction landed. Run the same window again to see what is left.
There is a habit worth borrowing from crews who calibrate by hand: enter a little less than the number you were given, and let more tacking confirm the rest before you go the whole way. On this platform that verification is already the workflow — apply, sail, re-analyse the same window the same way, and watch what is left shrink — and effective-from is how you stage it: the smaller entry dated to this window, a second one dated to the next day’s, with the timeline keeping the record of both.
If the wind pair fails halfway — one entry written, one refused — the page refetches the timeline first and then tells you exactly which channel landed and which did not, with the two ways out: add the missing one from the form, or delete the one that landed.
12 When it finds nothing
An empty answer is still an answer, and the page distinguishes four of them:
- Nothing worth correcting — your wind instruments look square, within what a day of sailing can separate from the weather. This is the good one.
- Not enough tacking to answer — a fixed error is only visible as the difference between the two tacks. Analyse a window with a proper beat in it.
- Not enough steady sailing to measure — there were tacks, but not enough settled close-hauled sailing between them.
- Measured, but not confidently enough to recommend — a number came out and did not clear the bar. The reason is named with the finding.
None of them is an error, and none of them is a reason to enter a number anyway. If you get the first one, you are done: the answer you wanted was that there is nothing to apply.