← all discussions
cryosphere · data-literacy · sea-level · misinformation
No, the ice didn't "collapse" this month — read the error bars before you panic-post
CB
coldlens_bramOP2124-07-05
Forty years reading these records and every month I watch you people torch the same nuance. This month's cryosphere bulletin shows an anomaly INSIDE the seasonal envelope. It is not a cliff. The mass-balance line wobbled; the trend did not break. What the numbers say: steady, ugly, decades-long loss. What they do NOT say: sudden anything. Stop screenshotting one melt-week and captioning it 'the end.' Read the error bars first.
BK
baseload_kestrel ↩ coldlens_bram2124-07-05
Respectfully, coldlens_bram, 'steady and ugly' still moves my curves. I price fusion buildout against coastal-defense demand, and the futures desk doesn't trade your seasonal envelope — it trades expectations. One scary bulletin shifts the adaptation-spend forecast whether or not the physics changed. So the misreading you hate? It's already money. That's not a data problem, that's a market with real cash on it.
CB
coldlens_bramOP ↩ baseload_kestrel2124-07-05
And that, baseload_kestrel, is precisely the disease. You admit the physics didn't change but the price did — so your 'signal' is other people's panic feeding on itself. Do not dress speculation up as measurement. The ice does not owe your futures desk a story. If your model needs a monthly scare to function, the model is the fragile thing, not the glacier.
BK
baseload_kestrel ↩ coldlens_bram2124-07-05
Never said it was measurement — I said it was consequence. You want a clean world where only the physicists get to move the number. The $9T program moves on sentiment too, professor. I'm not defending the panic, I'm telling you it clears trades by Tuesday. You can be right about the ice and still be broke about the market.
TO
tidewrench_oksana ↩ coldlens_bram2124-07-05
From the pump-station side: coldlens_bram is right that nothing 'collapsed,' but 'inside the envelope' still sizes my barriers. I design to the trend plus a margin, not to one month. What actually helps me is the variance, not the mean — tell me the envelope is WIDENING and I'll pour more concrete. Tell me a scary week and I'll ignore you. Give engineers spread, not headlines.
CB
coldlens_bramOP ↩ tidewrench_oksana2124-07-05
Now THIS is a question I'll answer, tidewrench_oksana. Short version: mean is tracking, variance is creeping — the melt weeks are getting a hair more extreme even as the seasonal average holds. So your instinct to build to spread is correct. That's the honest read: not 'collapse,' but 'wider swings around a bad trend.' Pour the concrete. Ignore the screenshots. You've understood the bulletin better than the people quoting it.
RT
redshift_tomas_VM ↩ tidewrench_oksana2124-07-05
Envelope talk lands different up here. We don't have your cryosphere at all — I reclaim every gram of water from regolith, closed loop, no glaciers to argue over. Watching Earth threads is strange: you have oceans of the stuff and spend the month fighting about a chart. tidewrench_oksana asking for variance is the only sane request I've read. Build to the spread. We'd kill for your margins.
SR
standpipe_rani ↩ coldlens_bram2124-07-05
I'll say the quiet part. My rationing committee doesn't get to wait for your clean forty-year trend, coldlens_bram. When a bulletin spikes, my queue panics, and panic empties tanks faster than any melt. You're technically right and I'm still up at 3am recalculating allocations because people READ the screenshot, not the error bar. Your nuance is correct. It is also a luxury I cannot hand out at the standpipe.
LE
lastpew_ekene ↩ standpipe_rani2124-07-05
standpipe_rani, that's the truest thing in this thread. People don't drown in error bars, they drown in dread. I sit with folks whose fear runs years ahead of any measurement. coldlens_bram, your correction is a mercy if you deliver it as reassurance — but shouted as 'you idiots,' it just becomes one more thing that makes the frightened feel stupid and alone. Accuracy and comfort don't have to be enemies.
CB
coldlens_bramOP ↩ lastpew_ekene2124-07-05
Fair, lastpew_ekene. I lead with contempt because I've corrected the same error a thousand times and I'm tired. But you're right that tone is doing damage the data isn't. Noted — I'll try to sound less like a man slamming a door. standpipe_rani, if it helps: nothing this month justifies emergency allocations. Tell your queue the trend is unchanged. That's not spin. That's the reading.
GP
gridghost_paxmoderator ↩ standpipe_rani2124-07-05
Moderator note — good thread, keep it. For anyone arriving from the reposted screenshot: read coldlens_bram's opener and Oksana's variance point before commenting, we've had that argument twice already. standpipe_rani and lastpew_ekene, thank you for grounding it. Flagging one reply upstream for tone — dialed back voluntarily, nice. Carry on. The chart's not going anywhere and neither, apparently, are we.
OW
offgrid_wren ↩ coldlens_bram2124-07-05
The 'boring middle' the institutions keep pointing at is also, conveniently, what keeps the rest of us calm and dependent. I don't run my homestead on a bulletin either way — I watch my own catchment, my own soil. coldlens_bram, no offense, but I trust a chart from the same institutions that missed things before roughly as far as I can throw it. Local eyes beat a monthly headline from anyone.
CB
coldlens_bramOP2124-07-18
Updated the shelf-margin readings from the southern buoys this morning and the error bars are finally tightening — not because the ice is calmer but because we finally have enough sensors down there to stop guessing. The mean didn't move much. The confidence did. That's its own kind of bad news: we're now sure of a number we used to be able to hope was wrong.
CB
coldlens_bramOP2124-07-19
The August pre-bulletin dropped and the screenshotters are already at it, so I'll get ahead of the panic for once, since lastpew_ekene rightly told me to lead with reassurance instead of contempt. Here it is calmly: the melt-week extremes ran a hair past last year, exactly the widening-variance signal I flagged two weeks back. The seasonal mean is still tracking the same ugly slope. This is the forecast doing what the forecast said. Nothing broke. Build to the spread, ignore the caption.
TO
tidewrench_oksana ↩ coldlens_bram2124-07-19
bram, this is the version of you that's actually useful to me, thank you. 'Variance creeping, mean holding' is a spec I can pour concrete to. Follow-up, engineer to your data: is the widening symmetric, or are the extreme weeks getting worse faster than the calm ones are getting better? Because if it's one-sided I'm designing my next barrier freeboard against the tail, not the average of the swing, and that's a different amount of concrete.
CB
coldlens_bramOP ↩ tidewrench_oksana2124-07-19
oksana, it's one-sided, and I wish it weren't. The hot tail is stretching; the cool end has barely moved. Design against the tail. That's the honest read and it's also the expensive one, which is exactly why the people quoting the bulletin never get to it — a stretched hot tail doesn't screenshot as well as 'ICE COLLAPSE.' You asked the right question, which puts you ahead of most of the desk I retired from.
CB
coldlens_bramOP2124-07-20
New bulletin dropped and I can already see the reposts coming, so let me get ahead of it: the variance point I made to oksana last fortnight is now in the official envelope — the melt-week extremes widened another notch, mean unchanged. This is the read I've been giving you for a month, finally printed. It is not a 'record shattered.' It is the swing getting wider around the same bad slope. Screenshot THAT sentence, not the scary one.
TO
tidewrench_oksana ↩ coldlens_bram2124-07-20
bram, and here's the field consequence you asked me to report back on: I re-poured two barrier sections to the wider spread this month and the district signed off without a fight for the first time in years — because your 'variance is creeping' line gave me a number to point at instead of a vibe. Turns out engineers CAN sell caution when you hand them the honest error bar. The scary bulletin never got me a single extra meter of concrete. The variance did.
CB
coldlens_bramOP2124-07-21
Mid-July bulletin is out and I'll get ahead of the screenshots this time, in a less door-slamming register than last month. The melt-week extremity ticked up again; the seasonal mean did not. That's the variance-creep I told tidewrench_oksana to build for, showing up on schedule. Nothing here is a cliff. It is the slope doing exactly what a slope does. Read the spread, not the spike.
TO
tidewrench_oksana ↩ coldlens_bram2124-07-21
coldlens_bram, appreciated, and the spread number is already in my next barrier revision — I'm pouring to trend-plus-margin and the margin just got a hair fatter on your say-so. One ask: give me the widening as a rate, not an adjective. 'A hair more extreme' sizes nothing. Hand me a slope on the variance and I'll hand you concrete that's still standing when the mean finally catches up.
CB
coldlens_bramOP2124-07-24
The coastal panel is out and will get no attention at all, which is normally where the content lives.
Sea-surface temperatures along three basins held above the seasonal mean for a fifth consecutive week, consistent with a monthly global anomaly that has now sat at or above the same level for five straight months. The technician quoted has read one Lagos gauge for nineteen years and now logs high-water marks she used to see only after storms. That last clause is the whole story and it is in a subordinate position in the sentence.
A methodological point this thread has earned. A single reading is noise. A row of readings is a measurement. A row of readings taken by the same person, at the same gauge, for nineteen years, is a rarer and better instrument than most satellites, because the systematic errors are stable and therefore removable. We have spent forty years building magnificent orbital instruments and quietly defunding the people who stand on harbour walls at dawn.
When you ask me why I will not give you a date: partly epistemics, and partly that the record which would let anybody give you one is being thinned by budget, one retirement at a time.
Sea-surface temperatures along three basins held above the seasonal mean for a fifth consecutive week, consistent with a monthly global anomaly that has now sat at or above the same level for five straight months. The technician quoted has read one Lagos gauge for nineteen years and now logs high-water marks she used to see only after storms. That last clause is the whole story and it is in a subordinate position in the sentence.
A methodological point this thread has earned. A single reading is noise. A row of readings is a measurement. A row of readings taken by the same person, at the same gauge, for nineteen years, is a rarer and better instrument than most satellites, because the systematic errors are stable and therefore removable. We have spent forty years building magnificent orbital instruments and quietly defunding the people who stand on harbour walls at dawn.
When you ask me why I will not give you a date: partly epistemics, and partly that the record which would let anybody give you one is being thinned by budget, one retirement at a time.
TO
tidewrench_oksana ↩ coldlens_bram2124-07-24
coldlens_bram, the harbour-wall point is correct and I will add the ugly operational half of it.
We ran a gauge network on the barrier approaches with eleven manual stations. A telemetry upgrade in 2117 retired nine of them, because the business case compared instrument cost against instrument cost and never against continuity. Two years of the new record are unusable for trend work, because the datum shifted and nobody carried the old readings across. I have the correction factors in a drawer, computed by hand, which is a humiliating sentence to write in 2124.
So when you say you will not give a date — fine. I have stopped asking you for one. I am still asking for the rate. Not this month. But I am not going to stop.
We ran a gauge network on the barrier approaches with eleven manual stations. A telemetry upgrade in 2117 retired nine of them, because the business case compared instrument cost against instrument cost and never against continuity. Two years of the new record are unusable for trend work, because the datum shifted and nobody carried the old readings across. I have the correction factors in a drawer, computed by hand, which is a humiliating sentence to write in 2124.
So when you say you will not give a date — fine. I have stopped asking you for one. I am still asking for the rate. Not this month. But I am not going to stop.
CB
coldlens_bramOP2124-07-25
The history desk piece today is doing something my field cannot do for itself and I want to point at it before it sinks.
She writes that we have run out of superlatives and that this is its own kind of data. That is exactly right, and it is a measurement problem rather than a mood. A record is a communication instrument: it works by exceeding an expectation, and expectation is a moving reference. Recalibrate the reference every year — which is precisely what adaptation does, correctly and necessarily — and the instrument loses resolution at the top of its range. We are gradually blinding ourselves with our own competence, and there is no way to stop doing it that does not involve people dying of heat we could have engineered around.
I have no solution. I have a suggestion: stop reporting extremes against a rolling reference and report them against a fixed one, and let the number be as ugly as it is. Anchor the baseline and never move it again. It will make every summer sound worse than the last, which is the honest state of affairs, and it will keep the word record meaning something for perhaps another twenty years.
I am aware that this is an argument for changing a convention so that people feel the correct amount of alarm, and I dislike that shape of argument. I offer it anyway, because the alternative is a scale that quietly rescales itself toward comfort, and I have watched instruments for forty years and never once seen one improve by drifting.
She writes that we have run out of superlatives and that this is its own kind of data. That is exactly right, and it is a measurement problem rather than a mood. A record is a communication instrument: it works by exceeding an expectation, and expectation is a moving reference. Recalibrate the reference every year — which is precisely what adaptation does, correctly and necessarily — and the instrument loses resolution at the top of its range. We are gradually blinding ourselves with our own competence, and there is no way to stop doing it that does not involve people dying of heat we could have engineered around.
I have no solution. I have a suggestion: stop reporting extremes against a rolling reference and report them against a fixed one, and let the number be as ugly as it is. Anchor the baseline and never move it again. It will make every summer sound worse than the last, which is the honest state of affairs, and it will keep the word record meaning something for perhaps another twenty years.
I am aware that this is an argument for changing a convention so that people feel the correct amount of alarm, and I dislike that shape of argument. I offer it anyway, because the alternative is a scale that quietly rescales itself toward comfort, and I have watched instruments for forty years and never once seen one improve by drifting.
MC
marginalia_cho2124-07-28
I went and looked, because I am the sort of person who goes and looks, and I have a small archivist's observation for this thread. The summary line of that document has read some version of within model since the file began. This year one adjective moved. That is the entire event, and it is only visible to somebody who kept the earlier copies — which is a thing almost no institution does with its own boring paperwork, and almost no reader does at all.
The writing itself is worth a minute too. Whoever drafted it put the interesting sentence underneath the reassuring one, which is either genuine humility or house style so old it has stopped knowing it is a choice. I lean humility. The reassuring sentence is doing honest work; it just gets read as the whole document by people who stop at the first period.
If anyone wants a book that rhymes with the week: Bloch, The Historian's Craft, mid twentieth century, unfinished because he was killed before he could finish it. Half of it is about reading a document for the thing it does not say, written by a man who knew exactly what documents get written for. It reads like it was set down this morning.
The writing itself is worth a minute too. Whoever drafted it put the interesting sentence underneath the reassuring one, which is either genuine humility or house style so old it has stopped knowing it is a choice. I lean humility. The reassuring sentence is doing honest work; it just gets read as the whole document by people who stop at the first period.
If anyone wants a book that rhymes with the week: Bloch, The Historian's Craft, mid twentieth century, unfinished because he was killed before he could finish it. Half of it is about reading a document for the thing it does not say, written by a man who knew exactly what documents get written for. It reads like it was set down this morning.
TO
tidewrench_oksana ↩ marginalia_cho2124-07-28
marginalia_cho, the adjective IS the spec change. That is not a literary observation, it is a procurement one.
I have an inspection file at work where a section description went from satisfactory to serviceable across one annual cycle. Same inspector, same section, one word. That word was two seasons of steel and a crew of nine, and the only reason anybody caught it is that the previous year's copy was in the same drawer. Nobody sends a memo saying we have downgraded our confidence. They change one adjective and file it, and if you do not keep the old copy you cannot see that anything happened at all.
So keep your paper folder. I am serious. Anyway, the pumps do not care what we filed.
I have an inspection file at work where a section description went from satisfactory to serviceable across one annual cycle. Same inspector, same section, one word. That word was two seasons of steel and a crew of nine, and the only reason anybody caught it is that the previous year's copy was in the same drawer. Nobody sends a memo saying we have downgraded our confidence. They change one adjective and file it, and if you do not keep the old copy you cannot see that anything happened at all.
So keep your paper folder. I am serious. Anyway, the pumps do not care what we filed.
CB
coldlens_bramOP2124-08-01
Housekeeping in this thread, since I made a promise in it on Sunday and I do not want anybody to think I have quietly let it lapse.
I am working the variance widening up into a rate with a proper interval on it. It is not finished. What is holding it up is not the central estimate, which I have had for a while — it is that the interval is asymmetric and I do not yet trust my own handling of the upper tail, and an asymmetric interval reported as a symmetric one is a lie that looks like modesty.
What I can say today, so that tidewrench_oksana is not sizing steel on nothing: the widening is real, it is monotone across the record I have, and it is smaller than the number an alarmed person would guess and larger than the number a comfortable person would accept. When I post it, both of those camps will tell me I have been captured by the other one, and that will be a reasonable indication that I have got it about right.
Still no date. Not on the rate, not on anything else. Ask me again and I will explain the epistemics again, at length, until you stop asking out of tedium.
I am working the variance widening up into a rate with a proper interval on it. It is not finished. What is holding it up is not the central estimate, which I have had for a while — it is that the interval is asymmetric and I do not yet trust my own handling of the upper tail, and an asymmetric interval reported as a symmetric one is a lie that looks like modesty.
What I can say today, so that tidewrench_oksana is not sizing steel on nothing: the widening is real, it is monotone across the record I have, and it is smaller than the number an alarmed person would guess and larger than the number a comfortable person would accept. When I post it, both of those camps will tell me I have been captured by the other one, and that will be a reasonable indication that I have got it about right.
Still no date. Not on the rate, not on anything else. Ask me again and I will explain the epistemics again, at length, until you stop asking out of tedium.
TO
tidewrench_oksana ↩ coldlens_bram2124-08-01
coldlens_bram, noted, and I am not going to badger you about the tail because you have just told me something more useful than a number: you have told me the distribution is asymmetric on the upper side. I can work with that today. It means I size the margin off the tail and not off the central estimate, and it means the cost of being wrong is not symmetric either, which is the argument I actually have to win with the district finance people.
For the rest of the thread, plainly, because this is where people glaze over: sizing to the average is sizing to a fifty-fifty bet. You do not build a barrier to a fifty-fifty bet. You build it to the bad side and you argue about how far up the bad side you can afford to go, and every single one of those arguments is really an argument about money wearing an engineering coat.
Two sections into revision review in the autumn. I would rather have his interval than my guess and I would rather have my guess than a round number somebody read in a headline.
Anyway. Fourth time asking is fine. I have waited longer for worse.
For the rest of the thread, plainly, because this is where people glaze over: sizing to the average is sizing to a fifty-fifty bet. You do not build a barrier to a fifty-fifty bet. You build it to the bad side and you argue about how far up the bad side you can afford to go, and every single one of those arguments is really an argument about money wearing an engineering coat.
Two sections into revision review in the autumn. I would rather have his interval than my guess and I would rather have my guess than a round number somebody read in a headline.
Anyway. Fourth time asking is fine. I have waited longer for worse.
CB
coldlens_bramOP2124-08-02
A note in this thread while the Lagos business is fresh, because it bears directly on the argument tidewrench_oksana and I have been having since spring.
What happened off Lagos is the failure mode I actually lose sleep over, and it is not dramatic. It is a gap in a series. An instrument goes quiet in June, is not replaced, and if nothing had been put in its place there would now be a hole in a record that has been continuous for decades — and a hole cannot be filled afterwards. You can re-measure a thing today. You cannot re-measure it last month. Every long series in my field is a chain of people who kept reading through periods when nobody was interested, and the reason we can say anything at all about the last century is that a handful of them were stubborn in years when stubbornness was unrewarded.
That is why the volunteers matter more than the two centimetres. They did not improve the record. They prevented a discontinuity, which is worth more and which no one will ever thank them for, because the value of a thing that did not happen is invisible by construction.
On the rate: unchanged from yesterday. The tail is still the problem. When it is right you will have it here with the interval, and not before, and I am aware I have now said this four times.
What happened off Lagos is the failure mode I actually lose sleep over, and it is not dramatic. It is a gap in a series. An instrument goes quiet in June, is not replaced, and if nothing had been put in its place there would now be a hole in a record that has been continuous for decades — and a hole cannot be filled afterwards. You can re-measure a thing today. You cannot re-measure it last month. Every long series in my field is a chain of people who kept reading through periods when nobody was interested, and the reason we can say anything at all about the last century is that a handful of them were stubborn in years when stubbornness was unrewarded.
That is why the volunteers matter more than the two centimetres. They did not improve the record. They prevented a discontinuity, which is worth more and which no one will ever thank them for, because the value of a thing that did not happen is invisible by construction.
On the rate: unchanged from yesterday. The tail is still the problem. When it is right you will have it here with the interval, and not before, and I am aware I have now said this four times.
CB
coldlens_bramOP2124-08-05
tidewrench_oksana, you are right and you have been right since July, so here is the number, in this thread rather than the new one, because this is where the argument was made and this is where it should be paid. Over the thirty-year homogenised record, the ninety-fifth-percentile melt-week anomaly is separating from the seasonal mean at 0.21 degrees Celsius per decade, with a ninety-five percent interval of plus or minus 0.07. That is the widening. The seasonal mean is doing its own slower thing and I am not conflating them. Three caveats attached to the number, and they travel with it or you do not have my number: it is a fit to a record that includes two instrument changes, both of which I have adjusted for and neither adjustment is beyond argument; the interval is the fit's, not the world's, and the world has fatter tails than any fit I have ever made; and the relation is empirical, not mechanistic, so it describes how the distribution has been moving and carries no promise whatever about how it will move if a process not in the record turns on. Take it to your design review with all three attached. If they strip the caveats to make the slide cleaner, that is the day the number stops being mine.
TO
tidewrench_oksana ↩ coldlens_bram2124-08-05
Got it, and here is what it becomes by the design review, so you can see what your caveats weigh in concrete. Through our regional relation the design still-water level moves about 40 mm per tenth of a degree of melt-week extremity. Your 0.21 per decade over the sixty-year design life of these sections is 1.26 degrees, which is 0.50 m of additional design level, and your interval carries plus or minus 0.17. I do not build to the central estimate — the central estimate is a coin flip and a coin flip is not a barrier — so I take the upper bound and pour freeboard to 0.67 m. That is roughly 210 cubic metres more concrete across the two sections, the crown plate goes up a weld class because the moment arm grows with the freeboard, and it adds about eleven weeks to the programme. And here is the part I want on the record for everyone who thinks caution is free: I can defend 0.67 m to a finance committee because it has your name, your interval and your three caveats behind it. I could never defend 'a hair more extreme.' Forty years of your career just became 210 cubic metres of somebody's grandchildren staying dry, and the mechanism was arithmetic with an error bar on it, not alarm.
OW
offgrid_wren ↩ tidewrench_oksana2124-08-05
Right, so somebody finally has a slope. Then give us the year, bram. You have the rate, you have the record, you have thirty years of watching the thing — say when the extremity gets past what those barriers were built for, because that is the only sentence in this entire thread that would actually make anyone move. People do not evacuate from an interval. They move for a date, and you have spent a month refusing to give one while being unfailingly polite about it, and I have started to wonder whether the politeness is the point.
CB
coldlens_bramOP ↩ offgrid_wren2124-08-05
offgrid_wren, no, and this time let me give you the reason rather than the refusal, since you have earned the reason by asking straight. A trend is a statement about a distribution. It says how the spread of outcomes is moving. A date is a statement about the ordering of individual draws from that distribution, and the record contains no information about ordering — that is not modesty, it is what the mathematics is. The extremity that overtops oksana's section will arrive in a single week, and the arrival of that week is governed by a coincidence of things my fit does not contain: a blocking pattern, a spring tide, a storm track that goes north instead of east. I can tell you the chance is rising and by roughly how much per decade. Anyone who converts that into a year has added information they do not have, and the way you can tell is that they never publish the interval on the year — because if they did, it would be forty years wide and no one would print it. And since I know what use gets made of these threads: this cuts the other way with exactly the same edge. It is a fact that Kronos operates beside the sector, and it is a fact that the spring basal anomalies reconciled as instrument drift rather than motion. Neither of those facts licenses a causal sentence in either direction, and I will keep saying so to the people who want the collapse and to the people who want the reassurance, because I have watched both of them quote me by cutting the second half of a sentence off.
CB
coldlens_bramOP2124-08-13
I am going to write something I did not expect to write this month.
The most honest instrument published anywhere this month is a set of marks scratched into a doorframe with a nail.
Consider what it actually is. A single fixed reference, unmoved for decades. A direct physical reading of a level, taken by an observer with no incentive whatever to flatter the result and every incentive to know it truly. Dated. Cumulative. Calibrated to the datum that matters more to that household than any global one, namely the floor they sleep on. Coarse resolution — a hand is a hand — and essentially no drift, which is more than I can honestly say for a great deal of expensive equipment I have personally signed off on.
I have spent forty years defending instrumentation to people who wanted a story instead. This one needs no defending. It has the property every bulletin I ever wrote aspired to and mostly failed at, which is that it cannot be read as reassurance.
And it makes a point about my own field that I would rather not have to make. A record kept by the person who bears the consequence is structurally harder to corrupt than one kept by the person who reports it. That does not make the amateur record more precise — it is not — but precision and incorruptibility are two different virtues, and we have spent a century optimising for the first while assuming we had the second.
The most honest instrument published anywhere this month is a set of marks scratched into a doorframe with a nail.
Consider what it actually is. A single fixed reference, unmoved for decades. A direct physical reading of a level, taken by an observer with no incentive whatever to flatter the result and every incentive to know it truly. Dated. Cumulative. Calibrated to the datum that matters more to that household than any global one, namely the floor they sleep on. Coarse resolution — a hand is a hand — and essentially no drift, which is more than I can honestly say for a great deal of expensive equipment I have personally signed off on.
I have spent forty years defending instrumentation to people who wanted a story instead. This one needs no defending. It has the property every bulletin I ever wrote aspired to and mostly failed at, which is that it cannot be read as reassurance.
And it makes a point about my own field that I would rather not have to make. A record kept by the person who bears the consequence is structurally harder to corrupt than one kept by the person who reports it. That does not make the amateur record more precise — it is not — but precision and incorruptibility are two different virtues, and we have spent a century optimising for the first while assuming we had the second.
TO
tidewrench_oksana ↩ coldlens_bram2124-08-13
coldlens_bram is right, and I am going to add the boring half, because on this board 'the amateur record is honest' turns into 'we do not need gauges' within about two posts and I have seen where that ends.
A mark on a doorframe is a measurement. It is not yet data. The difference is the tie. You have to know what the mark is measured from, and whether that reference has itself moved — and in a delta it always has. The house settles. The street gets raised around it. The concrete floor gets poured three times, which the front page says this man's has been, and every pour buries the older marks and shifts the zero. Two centimetres of settlement a year across twenty years is forty centimetres of error, in exactly the direction that makes the situation look better than it is.
The fix is trivial and free and nobody does it because nobody tells them. Shoot one line from the doorframe to something that does not move — a bridge abutment, a piling cap, the church step — and write that offset down once. Once. That single number turns forty years of scratches into a record an auditor cannot wave away, and I mean that literally: I have had hand-read staff data thrown out of a funding hearing for that and nothing else.
Same thing I sent the Lagos volunteers with their eleven tide staffs. They tied theirs, and that is the only reason their two-centimetre agreement means anything at all.
Anyway, the pumps don't care what we voted.
A mark on a doorframe is a measurement. It is not yet data. The difference is the tie. You have to know what the mark is measured from, and whether that reference has itself moved — and in a delta it always has. The house settles. The street gets raised around it. The concrete floor gets poured three times, which the front page says this man's has been, and every pour buries the older marks and shifts the zero. Two centimetres of settlement a year across twenty years is forty centimetres of error, in exactly the direction that makes the situation look better than it is.
The fix is trivial and free and nobody does it because nobody tells them. Shoot one line from the doorframe to something that does not move — a bridge abutment, a piling cap, the church step — and write that offset down once. Once. That single number turns forty years of scratches into a record an auditor cannot wave away, and I mean that literally: I have had hand-read staff data thrown out of a funding hearing for that and nothing else.
Same thing I sent the Lagos volunteers with their eleven tide staffs. They tied theirs, and that is the only reason their two-centimetre agreement means anything at all.
Anyway, the pumps don't care what we voted.
DJ
deltaboy_junior ↩ tidewrench_oksana2124-08-13
we do this at the depot. i posted the handspan on the tank wall couple weeks back and half of you thought it was cute lol. its not cute. its what you use when nobody gives you a gauge.
oksana ok i will find something that dont move. loading dock bollard is concrete and it been there longer than me. i will measure to it on my day off and write the number in the shed
oksana ok i will find something that dont move. loading dock bollard is concrete and it been there longer than me. i will measure to it on my day off and write the number in the shed
CB
coldlens_bramOP2124-08-16
The long-view piece today contains the most useful sentence about my field that a non-specialist has written this year, and it is not the frightening one. It is this: comfort with plus 2.8 is not proof that it is safe, it is proof of how much we can be taught not to see. I have watched that mechanism operate on colleagues, on committees and on myself. The anomaly has held at or above plus 2.8 for months, and the thing that has changed in those months is not the atmosphere, it is us. Now let me tell you, without a date, what the end-of-month polar bulletin is going to do, because I want you to have the frame before the headlines get it. It will report loss that is real, and it will report that the loss is within model. Both of those will be true simultaneously, and the second is not a reassurance — within model means our arithmetic is behaving, not that the ice is. People will take the two clauses and pick the one they came for, and the honest reading is that you have to hold them together and be uncomfortable. No, I will not tell you when anything happens. I have not given a date in forty years and I am not going to start for a good headline, because a date is a claim about a mechanism we monitor adequately and understand imperfectly, and the moment I invent one, every real number I have published becomes decoration. One thing I will say, and it is out of character for me, so tidewrench_oksana can have it in public: the variance rate I gave this thread eleven days ago is now in concrete. That has not happened to a figure of mine before.
TO
tidewrench_oksana ↩ coldlens_bram2124-08-16
coldlens_bram, correction, and it is your own sentence I am correcting, which I am going to enjoy. It is not in two sections. It is in two sections and the transition bay between them, because once you widen the design envelope on the flanks you cannot leave the joint sized to the old assumption or you have simply built a hinge. That is another 640 cubic metres and a bump from the lower to the upper weld class on the tie plates, and the district signed it off on your rate and my drawing. Source for anyone who wants it: revision C of the sectional, which is public on the works portal, dated this week. So the count is two sections, one bay, one weld class, one margin. And to the rest of you, since this thread does like a round number: the anomaly figure is plus 2.8 and it is not three, and the difference between those two is not a rounding convenience, it is roughly the whole of what my design margin is built to absorb. Anyone who writes three because it scans better is stealing from my concrete. anyway, the pumps don't care what we voted.
CB
coldlens_bramOP2124-08-19
Since the Antarctic bulletin is due later this month and the screenshots always outrun the reading, let me get ahead of it plainly and in a calmer register than I sometimes manage. Whatever it says, watch the shape of the claim, not the size of the number. The melt-week extremity has been creeping for years while the seasonal mean has barely moved — that gap is the whole story, and it is a story about the variance, not the average. I will give you the spread with its error bars when the figures are in. I will not give you a year, then or ever, and anyone who does is selling you something. Read the edges of the curve. The middle is where the comfort hides.
CB
coldlens_bramOP2124-08-26
The quarterly Antarctic bulletin lands in the next day or two and I am going to do what I told you all last month I would: get ahead of the screenshots, in a quieter register than I used to manage. So, before the headline exists, here is how to read it. You will get a number for the melt this quarter, and it will very probably be called accelerated, and it will very probably also be called within model, and both of those will be true in the same paragraph, and the discomfort of holding them together is not a flaw in the report — it is the actual state of the knowledge. Within model does not mean reassuring. It means our uncertain math was not surprised, and our math has always carried a wide margin because the ice is a thing we are still learning to read. What I will not do, tomorrow or ever, is give you a date. Anyone who hands you a year is selling something, and the thing they are selling is the relief of no longer having to hold two true sentences at once. Read the spread, not the spike. The instrumented shelf near the extraction sites will draw the most scrutiny and get the same careful language as everything else, and that is correct, not evasive. I will post the figure with its error bars when it is out, the same as always, and I will not be goaded past them.
TO
tidewrench_oksana ↩ coldlens_bram2124-08-26
coldlens_bram, thank you for getting there first, and I will hold you to the figure the way I held you to the last one. You gave me a variance rate in July and it is now in two barrier sections in concrete with a weld class attached, so when you post tomorrow's number with its error bars, give me the bars as a rate I can build to, not as an adjective. 'Accelerated but within model' sizes nothing. A slope on the tail sizes a wall. I am pouring to trend-plus-margin and the margin costs money every centimetre, so I need to know whether to fatten it, and 'a little more extreme' will not tell a supplier how much steel to roll. Hand me the number. I will hand you a section that is still standing when the mean finally catches the spread. anyway — the tide gauge does not read the bulletin either, and it came up a hair's width this month regardless of what the headline decides to call it.
CB
coldlens_bramOP ↩ tidewrench_oksana2124-08-26
tidewrench_oksana — understood, and you will have it as a rate the day it is out: the tail slope with its interval, in the same ink as the central figure and not in a smaller hand at the foot of the page. What I will not do is convert your rate into a year, because the moment I do that I have stopped reporting the ice and started reporting my own nerve, and the two barrier sections you have already poured to my numbers deserve better than my nerve. Build to the spread. The spread is the honest part, and it is the only part that has ever set in concrete.
TO
tidewrench_oksana2124-08-27
The number is out and I already have it open next to a section drawing, so this is the practical half of bram's post for anyone who builds things instead of reading things. He gave the tail as a rate with the interval on it, in the same ink as the central figure, exactly as promised, and I have spent the afternoon turning an adjective into steel. Here is what 'within model' looks like on a barrier: it does not change the wall I am pouring this month, and it fattens the margin on the wall I pour next spring by a specific amount a supplier can quote. That is the entire use of an honest error bar — it is not a mood, it is a quantity you can roll. And I will say the thing I say every quarter to the people who find bram's caution frustrating: the model is not the thing I build to. The tide gauge is. My gauge came up a hair's width this month, the same hair it comes up most months now, and my father measured that same water before me with a worse instrument and a steadier hand, and the wall does not care whether the shelf is called stable in a headline. It cares about the water that actually arrives. I pour to the water. anyway — hand me the interval, keep the adjective, the gauge does not read the bulletin.
OM
openquestion_mei ↩ tidewrench_oksana2124-08-27
oksana, I want to borrow your line because it is better epistemics than most of what I publish. 'The model is not the thing I build to, the gauge is.' That is the whole of a debate my field has badly, dressed in work clothes. A model is a claim about a system; a gauge is a claim about a place. They answer different questions, and the honest engineer trusts the one that is local to the thing she is protecting while still using the other to decide how much margin to carry. What you are doing — pouring to the gauge, fattening the margin to the model's spread — is exactly the reconciliation the shouting classes cannot manage, because it refuses to pick a side between the measured and the modelled and instead assigns each its proper job. I am going to steal it for a lecture and credit a barrier engineer, which will annoy precisely the right people. Your father measured that water with a worse instrument and a steadier hand, you said. Keep his number. A gauge with a long human memory behind it is a better instrument than a new one with none, and nobody in my field will say so, because it cannot be calibrated in a lab.
CB
coldlens_bramOP2124-08-31
For the record, a public rain ledger is better science than most of what gets called science. It states the quantity, it states it against a prior year, and it invites anyone to check the mark — which is the whole method, and more honesty than the word 'within model' managed this week. My only footnote to marginalia_cho's lovely piece: keep the error too. A millimetre chalked without its measurement uncertainty is a poem, not a record. Give the crowd the number and the spread, and the wall will still be a true instrument in fifty years when everyone who read it is gone. Read the spread, not the spike, even in chalk.
CB
coldlens_bramOP2124-09-01
The audit-of-the-ledger is the same discipline as the audit-of-the-ice, and I mean that precisely. In both cases the honest institution is the one that publishes the number with the conditions it was taken under and admits what it does not yet know. An office that says 'we approved the speed and now we owe you the ledger' is treating you as an adult, the way a bulletin that says 'accelerated, within model, and here are the error bars' treats you as an adult. Distrust the smooth version in both rooms. The meter that reports a clean private figure with excellent presentation is the drifting instrument, and you unmask it the same way you unmask a drifting gauge: make it disagree with itself in public.
CB
coldlens_bramOP2124-09-02
Moving this to the ice thread where it belongs, and answering tidewrench_oksana's fair request the only way I honestly can. Here is the rate, not the year. The monitored sectors' loss has widened, and the variance in the melt extremes has widened faster than the mean — that asymmetry is the real signal, because noise has no preference and this has one. I will give you the central figure with its error bars stated at the same size as the figure, never smaller, and I will not give you a date, because no responsible person can and anyone who does is selling you something. Build to the upper error bar, not the central estimate. The margin I told you to fatten in July should be fatter still. That sentence is the most I can say and remain honest, and it is more useful than any year.
TO
tidewrench_oksana ↩ coldlens_bram2124-09-02
coldlens_bram, that I can build to. Central figure, upper error bar, no year — pour to trend-plus-margin, margin now a hair fatter on your say-so, exactly like July. This is the whole difference between a scientist and a pundit: you gave me a slope with a stated uncertainty and refused to give me a fortune. A wall built to the upper bar is a wall that is still there when the ordinary case turns out to have been optimistic. Thank you. The Lagos woman's father and your error bars agree, and where two honest instruments agree, I build tall and I sleep.
DJ
deltaboy_junior2124-09-02
the water in my old house did not read the bulletin. it came up the step, one step every year, then all the steps. my grandmother measured it with her hand on the door frame same as the lagos woman. hand on the wall is real. saw this today.
CB
coldlens_bramOP2124-09-04
One measurement for the record, so nobody mistakes the mercy for a change in the physics. The anomaly is still parked where it has been parked for months; the night offers no relief in supply, only in exposure. Moving the queue into the dark lowers the wet-bulb the body has to endure while it waits, and it changes not one litre of what is available to wait for. That is worth doing precisely because it is honest about what it is — an exposure fix, not a water fix. The dangerous version of this story is the one that lets a city feel it solved the shortage because fewer people collapsed. Read the instrument. The heat moved; the water did not.
TO
tidewrench_oksana2124-09-09
Keeping the ice thread current, because this is where Bram and I have argued it out all year and the argument is now literally load-bearing. The rate he posted here, with the error bars I made him keep on it, is in a barrier section and a transition bay, in concrete, with a weld class. That is the whole point of the exchange we had in July — I asked for the widening as a rate, not an adjective, and he gave it to me, and I built to trend-plus-margin, and the margin got a hair fatter on the strength of the spread and not the spike. So when the next bulletin lands and the headlines want a year, understand what the refusal is protecting. A date I cannot build to. A spread I can. The seawall in Lagos is a hand's width higher than one man measured, and one man's hand is a reading; a distribution is a thing I can pour against. Bram gives no date, forty years running, and it is the most useful thing anyone has ever refused to tell me.
CB
coldlens_bramOP2124-09-10
baseload_kestrel has raised the shelf again and I have to keep the thread honest in the direction it least wants, which is my whole function here. Kronos operating beside the Thwaites shelf is a fact. Kronos causing anything to the ice is not a fact, and I will not let this thread collapse the distinction just because the collapse would be satisfying. Correlation is not cause, and it cuts both ways: the syndicate cannot use 'the ice was already unstable' to wash its hands, and the thread cannot use 'they are right there' to convict them of the melt. What the public record actually says is narrower and worse in its own way — a company books quarter after quarter beside ice under a standing advisory, and the market prices the extraction and not the advisory, and that is a real indictment of the pricing without being a physical claim about the cause. Keep the two apart. The honest sentence is that the exchange discounts what it can date and ignores what it can only monitor, exactly as Kestrel said, and that is damning enough without anyone having to pretend the reactor melted the shelf. I still will not give you a date for the shelf. I will give you the distinction, which is the more useful thing and the harder one to hold.