Welcome to Off-Ramps! Today I’ll highlight four interesting pieces that I think you will enjoy reading. Please enjoy all of these on your morning commute, or save them for your weekend.
Let’s begin with a new occasional feature I’d like to introduce, Research Review, where I highlight a recent scientific paper with an interesting or counter-intuitive finding, and we stress-test it to see if it’s worth updating our previously-held beliefs (what a rationalist might call our priors).
Research Review: How Safe Are Two-Way Bike Lanes?
A two-way bike lane is just you think: two bike lanes, say a northbound and southbound one, that share a narrow strip of asphalt, with the barrier between them a mere line of paint. In the past, I have been both a user of bike lanes and a designer of them. In both capacities, I’ve proceeded on the basis of a strong prior that two-way bike lanes are unsafe. Riding at 30 km/h in one direction, constantly passing oncoming cyclists moving at the same speeds, just centimeters away, without a physical divider? It certainly feels dangerous, like each pass invites a collision. Surely my discomfort is a sign of real danger.
Enter a new paper in by Samuel Nello-Deakin, which suggests my instincts are wrong. Drawing on four years of crash data from Barcelona and matching it against actual ridership from hundreds of automated counters, Nello-Deakin finds that two-way lanes have slightly lower injury rates than one-way lanes, at 2.9 versus 3.5 per million bike-kilometres.
Barcelona has been busily converting its two-way lanes into one-ways, convinced they’re riskier. This paper suggests the city may be spending money to solve the wrong problem.
I’m disappointed to report that the the study isn’t airtight. It groups all one-way lanes together and all two-ways together, regardless of width, buffer, or intersection design. And geography confounds: many of Barcelona’s one-ways are concentrated in the dense Eixample district, with heavy car traffic and awkward angled corners, while two-ways are more common in quieter outer districts. Those contextual differences may explain much of the apparent safety gap.
There’s also a psychological confounder worth considering. I’m far from the only cyclist to feel discomfort in a two-way bike lane. That may make me more alert and cautious, reducing my own risk. Or that discomfort may deter less confident riders entirely, leaving behind a hardier group whose crash rates look lower on paper. The latter possibility is disquieting, which is that two-ways lanes are indeed safer, but only by screening out casual cyclists, who are just as deserving of accessible travel.
Still, the findings are useful for recalibrating what bike-lane designers, like I used to be, think we know. If you believed two-way lanes are inherently more dangerous, the evidence here should soften that conviction. The absolute difference, in Barcelona at least, is tiny: less than one additional crash per million kilometres. By comparison, the bigger risks lie elsewhere: at intersections, where turning cars and visibility failures account for most cyclist injuries, regardless of whether bikes approach from one or both directions. In other words, the direction of travel is not the lever that matters most for safety. I hope Barcelona’s policymakers bear this in mind: rather than converting existing bike lanes from two-way to one-way, they should aim to fix intersections, provide real separation from traffic, and build a connected network that gives cyclists safe and continuous routes.
Directionality, by contrast, is a sideshow.
The Canary in Tesla’s Showroom
I couldn’t go to work every day trying to train sales and delivery employees to sell EVs, knowing that the biggest detractor from the business was within our own company. Matthew LaBrot
Matthew LaBrot’s story caught my eye while researching The Tesla Doom Loop. (If you missed that essay, please go read it now for the big picture.) LaBrot’s firing is sad, but also interesting because it illustrates something I didn’t put into my longer piece: how Tesla’s culture prevents the feedback that might arrest its decline.
LaBrot spent five and a half years at Tesla, rising from assistant manager to head of North American sales training. He owned a Model Y and a Cybertruck. Then in June, he published a letter arguing that Musk was becoming a liability rather than an asset to the firm; that his presence was destroying demand. Though he’d written anonymously, Tesla nonetheless identified him and fired him within 48 hours of publication.
The letter itself matters less than what drove him to write it. LaBrot watched customer loyalty collapse from 73 percent to below 50 percent in nine months, a catastrophic defection rate. He saw Tesla lots fill with unsold inventory for the first time in the company’s history. Where the company once ran on months-long waiting lists, buyers can now drive away with almost any configuration the same day.
LaBrot’s letter did not prompt any introspection on the company’s part. Rather than asking why sales staff were losing confidence, the company leaned on Musk as the face of the company to an even greater extent than before, while also adding legal armour to ward off shareholder challenges. None of this addressed the problem LaBrot was pointing to: customers were leaving. Tesla chose to protect its leadership instead of its business.
The aftermath proves why more insiders don’t speak up. Despite his credentials, LaBrot went forty days without a single interview. That may be a bad job market, or it may be social contagion; I’m in no position to say. But it seems likely that at least some Tesla employees have taken on the lesson that raising alarms can mean professional exile.
LaBrot can take some consolation in his prescience. He forecast in June of this year that Q2 results would show “gigantic drops in sales, even potentially lower than Q1”. He was right. The data is in, and we now know that deliveries fell nearly 60,000 year-over-year, revenue slid 12% to $22.5 billion, and operating income plunged 42%. This isn’t happening because its cars stopped working, but because its brand has, and it’s taking punitive action against the employees who say so.
That’s the doom loop in action.
Four Ways Metrics Fail
Every large organization knows the frustration: identify a metric to represent success, and watch it backfire. Examples abound, in every sector: sales teams hit their quotas while overall revenue shrinks, hospitals improve wait times while care suffers, schools chase rising test scores while actually learning falters.
This is Goodhart’s Law in action: once a measure becomes a target, it ceases being a good measure.
Goodhart’s Law is familiar to anyone who has seen metrics misused, even if they don’t know it by name. That’s why I appreciate Scott Garrabrant’s piece, almost a decade old now, that adds sharp precision to how and why Goodhart’s Law applies. As per Garrabrant, Goodhart’s Law contains several failure modes, each with its own mechanism. Anyone who is responsible for delivering on metrics in a large organization would do well to know them all.
As per Garrabrant, they are as follows.
Regressional Goodhart: Pump up a proxy, pump up its errors
When a metric contains a true goal, as well as random noise, choosing the highest values means also choosing the most noise. Examples might include a transit agency that puts a station where a survey of residents forecast high ridership… but then fails to achieve that ridership, because the survey polled lots of people who would never take the bus, but liked to imagine themselves as the sort of people who would.
Likewise, the highest-ridership route in any given transit system isn’t necessarily the most valuable to the network, once you consider coverage or transfers. Looking only at ridership elevates outliers as well as true performers.
Causal Goodhart: Don’t treat an effect as a cause
If everyone has an umbrella, that’s probably because it’s going to rain today, but you can’t make it rain by handing out umbrellas.
More-subtle versions of this error happen all the time. A neighbourhood with lots of vitality, like dense housing, safety, and cool destinations, often also has a transit line; but that doesn’t mean giving a transit line to a neighbourhood without those things means they will suddenly appear. Similarly, in boom economic times, there is lots of traffic on the road; but narrowing roads to make them crowded (say, by adding a bike lane) won’t generate a boom.
In both cases, intervention on the proxy misses the causal structure. If you want prosperity, fund the generators—access, reliability, land use—and not their outputs.
Extremal Goodhart: At the extremes, the old relationship stops holding
A design that balances flow and safety at moderate volumes can break when pushed to the edge. This is the error that makes a transit agency aim to always hit on-time arrival metrics. Your trains can indeed achieve near-perfect records on this score by generously padding the schedule and widening the on-time window. In this case, punctuality climbs, but passenger travel, as measured in door-to-door time, gets worse. The correlation ‘more punctual → better service’ was real in the middle of the range, but it collapses at the boundary.
Adversarial Goodhart: Make a measure into a game, and people will begin to play it
Once a number governs budgets or reputations, people adapt. Ride-hailing firms can flood an area with empty vehicles to meet availability targets; never mind that the streets in that area become congested. Transit agencies can meet accessibility checklists by adding ramps to each station, even if those ramps are difficult to reach: that’s technical compliance but functional exclusion, adhering to the letter of the law while ignoring its spirit.
Taken together, the taxonomy is diagnostic gold. It helps you look at each expression of Goodhart’s Law and ask: is this noise amplification, a causal confound, a boundary effect, or strategic behavior? The path to success depends on what kind of failure it is.
Joe Heath on ABUNDANCE
American progressives want a Swedish-style welfare state without doing any of the hard work that is involved in creating a state apparatus capable of delivering a Swedish welfare state.
Yes, it's another review of Klein and Thompson's Abundance. These are thick on the ground, but this one is worth reading, because Joseph Heath's Canadian perspective sidesteps American debates about the book, which have been done to death. (I hope to offer a similar breath of fresh air later this year when I finally give my own review.) While the debate in the USA considers whether abundance thinking betrays progressive ideals, Heath has something else to offer: government dysfunction in that country is so normalized that abundance thinking is a necessary prerequisite for delivering those ideals.
The centre of his review is a personal story, an 18-month IRS nightmare of hung-up calls, computer systems that couldn't delete redundant files, and agents who could see his problem, but not fix it, nor escalate it. Heath's point is that this is not normal: while the Canada Revenue Agency has its problems (as I can attest) it is nowhere near this bad… but Americans who have only ever dealt with the IRS don’t know this, and assume that their uniquely terrible public institutions are the norm rather than a global outlier.
This blindness creates American progressivism's core problem: wanting Nordic-style programs, but having only Kafka-esque institutions to hand, meaning that those programs can never be delivered.
What Heath offers is a useful reframing of administrative competence: not as ‘neoliberal surrender’, but ‘table stakes’. This can be done! By way of analogy, he describes how after Quebec separatists lost the 1995 referendum to exit Canada, they collectively decided that the reason Quebeckers were afraid to secede was because the province depended heavily on national subsidy to run its public institutions. This meant, in the separatists’ view, the only way to win was to create “winning conditions”: make the province wealthy, well-run, and effective by fixing provincial governance. Only once that was achieved would independence seem viable. Two decades of this has not (yet) resulted in Quebeckers being ready to leave Canada, but it has improved life in Quebec in a variety of ways.
Abundance, Heath suggests, can be the USA’s “winning conditions”, a setting-of-the- table for a robust Nordic-style welfare state. Will such a welfare state actually emerge? Perhaps, or perhaps not; but the only way it will is if abundance is pursued first. As the kids say, read the whole thing.