Plutonic Rainbows

Six short sentences and a bottle, which is the whole of the copy in this 1989 campaign: her laugh, her eyes, her walk, her voice, her touch, her smell. Then the sign-off, for a woman to love and a man to remember. The photograph is black and white, the border is a marbled pink that belongs entirely to the decade, and nowhere on the page does the woman get a name.

So: yes. Tatiana was named after Diane von Furstenberg's daughter, Tatiana von Furstenberg, born in 1971 and still a small child when it launched in 1975. It was von Furstenberg's first fragrance and she made it on almost nothing, using a stock bottle and clipping sample packets to the hangtags of her dresses. Department stores came after that.

James Bell composed it, and what he built was a white floral with no interest in restraint: hyacinth and African orange flower over gardenia, tuberose, jasmine, narcissus and rose, sitting on musk, sandalwood, amber and civet. It was made to carry across a crowded room, which in 1975 meant Studio 54.

The ad is late, though. The scent was fourteen years old by 1989, and Revlon had taken the licence and was relaunching it. Vanity Fair covered that February and got the better story from the girl herself, who as a child had been furious about the whole thing. "Everyone utters my name," she said. "I want it to be only you." She was turning eighteen that month, at Brown, reading comparative literature and thoroughly done with fashion.

Von Furstenberg bought the rights to Tatiana back in 1994, along with the rights to her own name. The scent is discontinued now, which puts it in the same category as most of the big florals that outlived their moment.

In the turquoise Jobis suit, Linda Evangelista carries the broad, exact lapels of expensive office clothes. JOBIS sits above her in white capitals. The styling is severe but the colour isn't, and her cropped blonde hair keeps the whole thing from settling into bank-manager respectability. This is a German tailoring company selling authority with the stale air let out of it.

The restraint of the page makes Jobis look like a small label, but it wasn't. Furrier Johannes Bisegger rebuilt his business in Bielefeld after the war; Jobis, a compression of Johannes Bisegger, entered the commercial register in 1947. By 1957 it employed about 1,000 people and ranked among the leading German clothing manufacturers. It made Lufthansa's first post-war stewardess uniforms in 1955, then kept the airline contract for decades. Seidensticker bought Jobis in 1970, so by 1991 this was an established premium label inside one of Bielefeld's major clothing groups, not a hopeful boutique spending its last cheque on a supermodel.

It wasn't chasing teen fashion. The proposition was a complete, expensive wardrobe for a woman who expected clothes to last: firm shoulders, long jackets, skirts at the knee, silk against wool. That made freshness difficult. Change the cut too much and you lost the customer; change the face and you didn't.

I found no campaign brief, but the brown Jobis suit makes the logic unusually plain. Its skirt and peach blouse could have looked dutiful on almost anyone else. On Evangelista they look sharply new. The bottom of both ads belongs to Dienz, a Koblenz clothing shop celebrating its centenary, which suggests these were dealer pages using Jobis campaign imagery. Linda's job was to carry fashion prestige all the way to a regional sales floor.

Evangelista was still an extravagant booking. The crop that remade her career had reportedly quadrupled her rate, while the Versace runway, Chanel campaigns and Freedom! '90 made her face legible far beyond fashion readers. But she was also a superb commercial model. She could give a conservative suit a different character simply by changing her hair, posture or expression. Jobis didn't book Linda Evangelista despite selling conservative tailoring. It booked her because she could make conservative tailoring look like the future.

The company lasted another seventeen years. Seidensticker sold it in July 2008 with about 100 employees, around 400 customers and exports above 50 per cent. That was still a real international business. The buyer, Damo, ran into trouble almost immediately; production stopped, insolvency followed and 108 Bielefeld workers lost their jobs. The company was dissolved in 2009, and its legal shell lingered until the register deleted it in 2020 for lack of assets. Jobis isn't still going. The clothes now circulate second-hand, while these pages preserve the moment an old industrial label bought the most modern face it could find.

A cedar canoe is not a Russian boat. It belongs on a lake in Maine or Ontario, and here it is carrying a full imperial Russian wardrobe across flat grey water. Nobody on the shoot corrected it, and that tells you what the Ralph Lauren Russia collection actually was: not a place, but a mood you could buy on Madison Avenue using whatever props were already to hand.

It showed for autumn 1993, two years after the Soviet Union came apart. Lauren went past all of that and landed somewhere around the 1920s: velvet, dark paisley, military-cut jackets, long skirts over tall boots. Kate Moss walked it in sheer silk velvet burnout balloon trousers and a Balkan coat with a high embroidered collar, which is theatre wardrobe rather than clothing, and theatre wardrobe is the register the whole collection worked in. Six months earlier the same house had put lavender blazers over cream turtlenecks and made restraint the entire argument. Autumn threw that out and went to the dressing-up box. The opening page sets the terms: RUSSIA in red capitals across the slate roof of a stone cottage, and beneath it a figure in a rust paisley coat crossing the lawn with an armful of kindling, walking past a wing-back chair someone has carried outdoors for the picture.

The advertising is better than the premise deserves. The styling in the red frogged coat never wavers: black astrakhan at the throat and cuffs, a curled lambswool hat pulled down past the ears, leather gloves closed around the oar, and a paisley shawl heaped in the stern as though thrown there in a hurry. The water is the colour of dishwater. She's rowing away from something the picture declines to name, which is about as much story as a clothing advert can carry without collapsing under it.

The rest of the campaign works the same seam and keeps the same discipline. Kim Nye sits on straw bales in a long herringbone coat and a soft tweed cap, a waistcoat and a white shirt buttoned to the throat under it, a wine velvet skirt spilling over her knees: a farm worker costumed by somebody who has never had to kneel in one. Zofia Borucka is shot in profile under a stiff-peaked cap, a red and gold paisley shawl over tweed with a printed silk square knotted at the neck, and the layering of three patterns that should fight and don't is the best piece of work in the whole campaign.

The model in the canoe is Agnieszka Kotlarska. She was still a student at Wrocław University of Science and Technology when she won Miss Polski at the Forest Opera in Sopot in July 1991; three months later she took Miss International in Tokyo, the first Polish woman to do it. Ford signed her in New York, and both Lauren and Calvin Klein used her. So the casting on this picture is a Polish woman, four years out from the end of Soviet control over her country, hired by an American house to sell a romance of the East. I doubt anyone in the room described it to themselves that way.

I came to the image for the tailoring, which really is superb, the frogging in particular. The canoe still nags at me. Somebody freighted a boat out to a location, filled it with paisley, and decided that the Russia in Russia was the optional part.

She left New York at the end of 1993, the same season these clothes were in the magazines, and went home to Wrocław. A stalker stabbed her to death outside her home in the Maślice district of the city on 27 August 1996. She was 24.

Fragrantica's entry for Hermessence Violette Volynka lists three notes: Volynka leather, violet, violet leaf. Selfridges prints the leather on top and the violet in the heart and stops there. At £248 to £370 depending on size that looks thin, and it isn't: Jean-Claude Ellena founded the Hermessence line in 2004 on the idea of perfume as haiku, sold only in the house's own boutiques, one raw material pushed forward and everything else in service of it. Christine Nagel succeeded him and released this one in 2022. A three-note pyramid is the house style, not a shortcut.

Volynka is the interesting half of the name. It's Russia leather, yufte, hides tanned with oak and willow bark then cured with birch oil, which is what gives the material its water resistance and its smell. Hermès and its retailers hang the fragrance on the wreck of the Metta Catharina, a Danish brigantine that left St Petersburg in December 1786 bound for Genoa, was driven into Plymouth Sound by a gale and went down with a hold full of the stuff. Divers found the bundles two centuries later, still usable. The story usually arrives with the hides described as reindeer, following Garbett and Skelton's excavation reports, and a Ligatus paper on Russian yufte calls that "a wishful extension" of the leather's "almost mythical reputation" in the West, noting that the lattice finish on the recovered bundles is the one Russian sources associate with cow and bull hide. None of which changes what's in the bottle. It does tell you that the romance around this material was already being polished long before a perfumer got hold of it, which is worth holding onto when the drydown turns out to be quieter than the legend.

The presentation makes the same trade. A plain rectangular flask, the orange box, and then a purple leather collar stitched around the cap, the one loud element, and it's the violet's colour rather than the leather's.

Violets and leather is Germaine Cellier's territory. She staked it out in Jolie Madame for Balmain in 1953, a formula whose note list runs to seventeen: green leaves, violets and leather up front, castoreum and civet doing the work underneath, oakmoss and patchouli holding the floor. Angela at Now Smell This, who set that list out, said the leather lets you know the woman wearing it would kick your keister around the block if you mess with her. Chanel's Cuir de Russie, Ernest Beaux in 1927, is the other pole, and it has already been sanded down once in public: Jacques Polge reorchestrated it in 1983 to play down the leather and push up the iris.

That second fact is the one people skip when they set Volynka against the classics. The Cuir de Russie on sale today isn't the 1927 one either. Vintage and modern Jolie Madame don't smell alike, and Balmain has since discontinued the modern one anyway. The animalics that gave those perfumes their teeth have largely left mainstream perfumery under the same regulatory and commercial squeeze that turned Kouros into a scent people either argue about or can't wear at all. So part of the distance between 1953 and 2022 was never Nagel's to close.

The rest of it was, and she chose not to. Cuir d'Ange, Ellena's 2014 Hermessence, already put a soft floral leather in this exact register, and the most persistent complaint on Fragrantica is that Volynka is a more focused violet stapled to a very similar leather. Persolaise put it accurately inside the house range, less masculine than Equipage, less accessible than Kelly Calèche, less strange than Cuir d'Ange, the leather presented in an "innocent, almost naive frame" while still filling a room with tannery. The Black Narcissus granted it "high aesthetic competence" and then found the later stages "gently nonagenarian and soaped; white-cardiganed."

I should be straight about the limits of that: I'm reading the consensus, not my own wrist, and the two reviewers I trust most here disagree about whether the softness is the point or the problem. What both descriptions have in common is that nothing in the scent pushes back. Cellier's leather is rude. Beaux's is aristocratic and a little cold. Both give you something to argue with, which is why people are still arguing sixty and ninety years on. Volynka gives you a beautifully made violet against a beautifully made suede, and the compliance regime isn't what stopped it being difficult. Ellena's haiku format did, and Nagel worked inside it faithfully.

That produces something genuinely lovely and probably not durable. Jolie Madame had seventeen notes and several of them were unpleasant. This has three and all of them are lovely, which is a harder trick than it sounds and a much easier one to forget.

Roughly 95% of SpaceX stayed locked up when it went public, leaving four to five percent of the shares actually available to trade. The IPO priced at $135 on 11 June and opened the next morning at $150, which tells you most of what happened next: enormous demand, almost nothing to buy. The stock ran to $225 within days and the market cap went past $2.5 trillion. Then it did what thin markets do when the buying stops, and on 15 July it broke below its IPO price for the first time, touching $132.28.

None of that fall required a single new share. The 95% is still locked. What changed is that everyone willing to pay a scarcity price had already bought, and once scarcity stopped setting the number, the fundamentals did. Daniela Hathorn at Capital.com called the retreat profit-taking, valuation reassessment and "the unwinding of extremely bullish positioning", which is the polite version of saying the shortage wore off.

The fundamentals are not comfortable. The public got about 5% of the company at 94 times sales. Aswath Damodaran read the $28.5 trillion addressable market claimed in the prospectus, called it a hallucination he'd be embarrassed to publish, and said the document looked like it "was written by Grok". His own discounted cash flow work put the IPO valuation 27% too high. S&P expects negative free cash flow through 2029, and Morgan Stanley, which is bullish, doesn't see the company cash-flow positive until 2035.

Revenue was $18.7 billion in 2025, up 33%, against a loss of $4.9 billion. Starlink is the profitable piece, growing the way a utility grows: subscribers doubled to 10.3 million while monthly revenue per user fell from $99 in 2024 to $66 by the first quarter of 2026. That's mass-market expansion, not decay. It does mean the one division that makes money makes less per head just as the AI side needs $7.7 billion of capital expenditure in a single quarter. February's merger with xAI bolted a business burning cash at the rate the private AI market has normalised onto a business that flies rockets, and the people who bought the rockets are now funding the chatbot.

Nasdaq-100 inclusion on 7 July should have been a floor. It forced passive funds to buy. The shares are down about 13% since. Mechanical demand showed up, met the same thin float, and the price fell anyway.

Which is why the lock-up schedule matters more than the price targets do. The decline so far has happened with no new supply at all. The unwind comes in tranches rather than one cliff: roughly 20% of the standard block becomes sellable a couple of days after the first quarterly results, more through autumn, and the full 180-day block expires on 9 December. Musk's own shares stay locked until June 2027. Each of those dates feeds supply into a market whose float is currently a rounding error, and into demand that has already demonstrated it won't chase. The analysts still carrying a $225 median target are betting the buyers arrive faster than the shares do. I don't think they will. Sideways or lower into December is the realistic path, with Starship the only thing likely to interrupt it.

The parfum flacon floats over a bare shoulder and the woman selling it is caught in a gilt mirror, cropped dark hair, a red mouth, her eyes going sideways to her own reflection rather than out to you. Other bottles crowd the dressing table at the bottom of the frame. It's a picture about a private ritual, not an offer, and it was made for a fragrance that had been on sale for about seventy-five years by then.

I wanted to name her and couldn't. Five search routes turned up nothing. The closest dated match is Bruno Bisang, who shot Mitsouko for Guerlain in Paris in 1997 and kept two Polaroids from the job, and the entries that follow them in his archive are labelled Monica Bellucci and Claudia Schiffer while the Mitsouko pair carry only the campaign name. That's a photographer's filing habit rather than evidence of anything Guerlain decided, and I can't confirm it's even the same shoot. She stays unnamed and I've stopped pretending that counts as a finding.†

The commercial picture behind the ad is less romantic. Guerlain went into the 1990s turning over around two billion francs a year, profits slipping, its catalogue visibly ageing, the last real hit being Samsara in 1989. By mid-decade Shalimar and Samsara were each running at roughly 15 percent of the house's perfume sales, and perfume was about 60 percent of the business. Mitsouko isn't named in that breakdown. Two flagships at 15 percent leaves seventy percent unaccounted for and the chypre could be sitting anywhere inside it, so this is an argument from silence and not a sales figure. What it does establish is which two names the house reached for when it had to describe itself, and neither of them was the one the critics loved.

That was the business the family sold. LVMH took 58.9 percent of Guerlain in 1994 for close to two billion francs, bought out the remainder by 1996, and valued the house at over four billion in the process. Champs-Élysées arrived that same year, built from a marketing concept rather than composed by a Guerlain nose.

The pressure from outside was worse. Thierry Mugler's Angel launched in 1992 on patchouli, praline, red berries and vanilla, invented the modern gourmand, and went on to outsell Chanel No. 5 in France; CK One followed in 1994. Neither smelled anything like a fruity chypre, and the mass of the market was already running toward clean and aquatic anyway.

Then the material itself was restricted. Oakmoss is the spine of a chypre, and also a skin sensitiser, and IFRA, the industry's own standards body, pushed its permitted level down to 0.1 percent of the finished product by 1997. That forced reformulation rather than deletion, but wearers noticed. A commenter on Perfume Shrine who bought her first bottle in the early 1990s described going back for a replacement and finding it simply wasn't the same, and only learned about the reformulation years afterward.

Which is one way to read that mirror. A house carrying a scent it can't grow and won't discontinue has little reason to pay for a famous face, and every reason to buy a beautiful anonymous one who is looking at herself rather than at you.

Amendment, 26 July 2026. She has a name after all: the woman in the mirror is Tara Westwood, the Canadian model born in Manitoba. The argument above about why Guerlain wanted an anonymous face is mine, not the house's, and it was always weaker than the sales figures. ↩

Red coat over red jacket over red skirt, red gloves against black tights, and a brass rail she's leaning back on with both arms spread wide, as if she's holding the empty plaza still. The only words anywhere in the picture sit in the bottom corner: the Woolmark, and beneath it, Pure new wool. Everything else is colour doing the talking. Looked at cold it's an odd bit of staging, a woman braced against a handrail in a concrete forecourt, smiling at nobody, dressed for a season the light doesn't quite support.

Then the cab. There's a yellow taxi with checkerboard trim parked on what looks like a pier, rounded 1930s fenders in deep red, whitewall tyres, a TAXI dome on the roof, and a woman in a black knit coat and yellow turtleneck resting one gloved hand on the wing like she's just bought it. No taxi in New York looked remotely like that in 1992. The fleet was Caprices and boxy Ford sedans, square and dented and the colour of weak mustard. This car is a costume. Somebody built it, or found it and painted it, so the photograph could say New York faster than a street sign ever would.

The third frame lines three of them up against the water: black, yellow, red, each one a single column of colour from throat to shoe, hands on hips, the skyline behind them dissolved into pale blue mush. I spent longer on that mush than I meant to. Off to the right there's a tower with a steep pyramid crown and I can't turn that shape into anything except 40 Wall Street, with the older Financial District piling up to the left of it. Which puts the camera on the Brooklyn side looking west, on a cold bright morning, the two towers standing just off the right edge of the picture. I'm confident about the building and a good deal less confident about the pier.

Then the front plate on the cab, which reads RJ, Rio de Janeiro. I noticed it late and it irritated me for a good hour, because a number plate is usually the most honest object in a photograph, the sort of small hard evidence I've used to date a picture before. If that one is genuine, the paragraph above is in serious trouble. I've come down on the side of the skyline, on the grounds that a pyramid-crowned Art Deco tower is difficult to fake and a plate screwed to a costume car is not, but I can't get both facts to sit still at once. Either way the cab is a fabrication, down to a licence from the wrong hemisphere.

None of which touches the clothes, and the clothes are the part that hasn't moved an inch. Black and white harlequin diamonds running down one sleeve, red and gold scrollwork down the other, a four-strand gold choker sitting high on the throat, the whole thing knitted rather than printed. That's the part people forget about Escada: the pattern is structural, worked into the cloth on the machine, which is where the density in their late-eighties knitwear comes from. Shoulders still huge, skirts still short, but the colour blocking has been disciplined into flat graphic slabs. It's 1988's appetite wearing 1992's haircut.

Margaretha Ley died in 1992. Whether she saw these frames, approved them, sat in a room in Munich and said yes to the taxi, I have no way of knowing, and the catalogue went out regardless. The women in it are uncredited, which was house practice: Escada gave the shops a byline and gave the models nothing.

So the pictures have quietly changed jobs. The towers went, Ley was gone inside the year, the cab was scrapped or repainted into something else, and the three women at the rail are somewhere in their sixties, still unnamed.† The door on that morning is shut. What got out was the light, which came off a red wool coat, went into a lens, and also went straight past it and kept going, and is thirty-four years out from here tonight, well beyond Vega, carrying an arrangement of yellow gloves and cold water that nobody is going to assemble again.

Amendment, 26 July 2026. They have names after all: the three at the water are Gail Elliott, Yasmin Le Bon and Niki Taylor. That also undoes the arithmetic in the sentence above, because Taylor was born in 1975, which made her a teenager when this was shot and makes her fifty-one now rather than sixty. ↩

American Vogue, September 1993, carried the Hermès autumn campaign, and the model was Helena Christensen. In the first of them she stands in an oatmeal double-breasted suit with trousers wide enough to hide the boots, one hand raised, and two small orange boxes swing from her fingers on brown ribbons, like earrings she hasn't decided what to do with. Nothing in the picture explains them. They aren't shopping, they aren't being given to anyone, and they're much too small to hold the clothes she's wearing.

The staging repeats. She strides in a black leather trench and cap, and the box has been flown out on a long string rather than carried, the one hot spot of colour in an otherwise entirely black frame.

Set that beside the country reading it. British average weekly earnings that September were £237.20, a bit over a thousand pounds a month before anything came out of it, in the autumn after sterling had been kicked out of the ERM. That is the only hard number available to a reader of these pages, because the campaign supplies none of its own. No price appears anywhere in it, and no product does either, since the box stays shut in every frame. You cannot work out what any of it costs from the evidence in front of you, which is a strange way to advertise and an extremely effective way to sell.

The box wasn't a house tradition anyone inherited. It became the advertising because an agency decided it should. Hermès has worked with Eldorado since the late 1970s, and Bruno Suter, one of the three who founded the shop in 1976, takes the credit flatly: "That was me. When I saw those things, the boxes, all so wrapped up, with the ribbons. I had that in mind." The brief was blunter still. Someone at the house told them not to talk about the scarf, on the grounds that it was the least fashionable thing on earth, so they didn't, and by the account of his partner Pacha Bensimon sales skyrocketed the following Christmas. Suter says the house went on to print a scarf with the ribbon and the box on it, packaging eating its own contents.

The rest of the campaign reads differently once you know that. One box dangles from a gloved hand while she stands under a navy shawl fringed to the knee. In the red riding jacket and silver-buttoned suede trousers there's no box at all, only a leather strap pulled taut towards a horse you're expected to supply yourself. The page loses nothing by dropping the object.

The last is the one I'd keep. She sits wrapped in a green silk carré, chin turned, and the box has been demoted to the floor where a bare foot could nudge it, still tied, still unopened. Three years earlier the house had Linda Evangelista cupping one on a floor in Vogue like something precious, and by 1993 it's litter. The gap between that orange square and a thousand pounds a month is real enough, but it isn't old money looking down from a great height. It's a piece of graphic design some people in Paris invoiced for in the 1970s, and it worked so well that the wrapping outgrew whatever it was supposed to wrap.

Someone at iTutorGroup configured the company's application software to reject women aged 55 and over and men aged 60 and over the moment their birthdate hit the form. The EEOC sued, and in September 2023 the company paid $365,000 to settle. No neural network, no training data, nothing anyone would call artificial intelligence, and that is exactly why it remains the clearest case in the field: the discrimination was legible. You could read it off a config screen and point at it in court. Nothing since has been that easy to read, which is the actual problem with handing hiring to models.

The story most people carry around is the Amazon one. Train a screener on a decade of your own hiring, find your own historical preferences reflected back, quietly kill the project. It fits the intuition that machines launder the past, and for a while the research agreed. Brookings researchers simulated resume screening with large language models and found significant gender and racial discrimination, concentrated on Black men: in their setup, resumes with Black male names were preferred over white male names zero percent of the time. Five leading models tested through VoxDev showed the same intersectional pattern, where Black women, Black men and white women each get sorted differently, so a fairness test built on race alone or gender alone reports nothing.

Then the sign changed. A paired-resume audit of fourteen models, modelled on the classic correspondence experiments and running 24,024 matched job postings per model, found that GPT-3.5-turbo reproduced the human pro-white callback gap at 2.12 percentage points, and that every model released in 2024 or later showed either no gap or a statistically significant reversal favouring Black-coded names, by as much as three points. The same flip appeared on the gender axis. A separate study across twenty-nine models found that language models now advantage female and Black candidates relative to comparable white and male ones while disadvantaging disabled candidates, the effect of demographic identity worth somewhere between six months and a year of extra education. Post-training alignment, not the pretraining corpus, turned out to be the main driver.

I don't think that's good news. A screener that favours one group over another fails the same test whichever way it leans, and a bias that reverses between model generations is harder to govern than one that doesn't, because it can't be certified. It moves with vintage, with provider, with whatever the alignment team shipped last quarter. New York City's Local Law 144 requires an annual bias audit of automated employment tools. The EU AI Act classifies hiring systems as high-risk. Both regimes were designed around the Amazon story, where bias is durable, inherited, and points where you'd predict. An annual audit of something whose sign changes between releases is a photograph of a moving object, and the auditors are mostly reading figures the vendor produced, a weakness familiar from self-reported benchmarks.

The comparison that matters, though, isn't against a fair process. It's against human recruiters, and the human baseline is dreadful. It has been measured since the 2004 correspondence study in which Bertrand and Mullainathan sent fake CVs to Boston and Chicago employers and watched white-sounding names collect fifty percent more callbacks. In a field experiment covering seventy thousand applicants, AI-led interviews produced twelve percent more job offers, eighteen percent more people actually starting work, and better thirty-day retention. Those are outcomes, on real hires, and anyone defending the status quo has to account for them. The fairness number from the same study needs more care than it usually gets: reported gender-based discrimination fell from 5.98 percent to 3.30 percent, and "reported" is carrying the sentence. That is what candidates said about their experience, not a count of who got selected, and a machine interviewer can feel less prejudiced while sorting people exactly as badly. The hiring and retention gains I'd take at face value. The fairness gain I'd want measured a different way.

The catch is what happens when you put a person back in the loop, which is what every compliance framework asks for. In a screening experiment with 528 participants, people deciding alone, or alongside an unbiased model, selected candidates from all racial groups at roughly equal rates. Point them at a biased model and their choices tracked its preferences up to ninety percent of the time. The humans were fine until the recommendation showed up. Human oversight, the phrase carrying most of the load in AI hiring regulation, describes a person whose judgement the system has already colonised.

Against that, a study on Denmark's largest job portal compared recruiters searching manually, algorithmic matching, and the two together, and found the hybrid produced the fairest candidate lists of the three, better than either alone. Oversight worked there. The difference I'd bet on is accountability: the Danish recruiters were professionals on their own platform filling real vacancies with their reputations riding on the shortlist, while the experiment's participants were completing a task for a stranger and had no stake in being right. If that's the mechanism, then "human oversight" in a statute means nothing unless the human carries consequences, and none of the current rules require that. They require a person to be present.

A separate finding sits outside the demographic argument entirely. Maryland researchers ran twenty-two hundred resumes through commercial and open-source models and found self-preference rates of 67 to 82 percent: the models rank resumes generated by themselves above human-written ones of equivalent quality. Whatever that is measuring, it isn't the candidate. It rewards knowing which vendor screens your application, which is knowledge distributed exactly as unevenly as you'd expect.

Applicants adjust too, and that quietly undermines everything above. An IZA experiment found application rates falling 4.6 points for women and 3.2 points for men as AI involvement in the evaluation increased. If women withdraw from AI-screened roles at a higher rate than men, every callback-parity statistic in every study I've cited is computed on a pool the screener already reshaped before it read a single CV. A tool can post clean demographic parity across the applications it receives and still have skewed the workforce, because the skew happened upstream, in who bothered to apply. No audit regime I'm aware of looks there.

None of this makes the tools indefensible, and the scale argument does more work for me than any individual study. A bad human recruiter damages a few hundred careers across a working life, and eventually somebody notices the pattern. A model licensed across the Fortune 500 applies one idiosyncratic preference to millions of applications at once, and the only people positioned to notice are the ones running it. Mobley v. Workday is grinding through the Northern District of California on an age-discrimination theory, and it matters more than its facts because of the position it puts everyone in. The applicant can't see the screen. The employer that licensed the tool often can't either. The vendor owes an explanation to neither. A judgment against a vendor is the only lever anyone has found that reaches inside the model, and it's a slow, expensive, ten-year lever, aimed at an industry that is simultaneously removing the entry-level rungs those applications were pointed at.