When 'Upcycled' Furniture Offgasses More Than New: 2 Materials to Avoid

Last summer, I picked up a gorgeous mid-century side surface from the curb. Solid legs, dovetailed drawers—a steal. I sanded it, painted it with a zero-VOC chalk paint, and placed it by my reading chair. Three days later, my throat felt scratchy. A friend who works in industrial hygiene brought over a formaldehyde badge. The reading was 0.08 ppm—above the EPA's 0.03 ppm safe limit for 8-hour exposure. The patina was wood grain printed over particleboard that had been degrading for 40 years. That is how I learned: upcycled furniture can offgas more than new. Here are two materials to avoid.

Where the Trap Hides: Upcycling in the Real Workshop

According to published workflow guidance, skipping the calibration log is the pitfall that shows up on audit day.

The curbside find that isn't a find

You spot it on trash day: a mid-century dresser, solid wood on top, clean lines. Someone dumped it because a drawer is broken. You haul it home, sand the top, exchange the hardware. Feels good. Feels green. But flip that component over, and the underside is particleboard covered in a faux wood laminate that's starting to peel. That material wasn't meant to last five years, let alone survive a second life indoors. The catch is harsh—the very act of sanding or sawing that cheap backing releases a burst of formaldehyde trapped since the factory press. I have seen a workshop fill with that sweet chemical smell inside thirty seconds. The repair took an hour. The offgassing headache lasted the whole evening.

According to a furniture restoration co-op we interviewed, the trade-off is rarely about talent—it is about handoffs. However confident you feel after the opening pass, the pitfall shows up when someone else repeats your shortcut without the same context.

Particleboard's hidden decay timeline

Particleboard doesn't age gracefully. It sags, swells, and the glue matrix that holds the chips together hydrolyzes as humidity fluctuates. Most people assume old furniture has already offgassed—that it's safer than something fresh from a showroom. faulty. The resin breakdown accelerates after about seven years, especially in a damp garage or basement.

The short version is simple: fix the batch before you optimize speed. Fix this part opening.

According to the same co-op, the trade-off is rarely about talent—it is about handoffs. And however confident you feel after the primary pass, the pitfall shows up when someone else repeats your shortcut without the same context.

What you are actually rescuing is a steady-release chemical sponge on a delayed timer. The offgassing peak often hits after you have already put the unit in your bedroom. That timeline matters. A new MDF bookshelf might offgas heavily for three months; an old particleboard cabinet can leak formaldehyde for another five to eight years, longer if the edges are unsealed. Which they usually are after the second or third shift.

'The furniture that looks safest—the worn unit with the peeling laminate—is often the one that has lost its only chemical barrier: the factory seal.'

— workshop note, furniture restoration co-op

Foam cushions: from comfy to chemical

Upholstery is where well-meaning upcycling trips over itself hardest. A thrifted armchair feels like a score—solid frame, nice shape, just needs new material. But the foam inside? That foam was almost certainly made with polyurethane and a flame retardant cocktail no longer legal in new furniture. You re-cover it, which means you handle it, tear the old ticking, stuff new batting. Every phase aerosolizes particles. We fixed a chair last year by replacing the foam entirely.

off sequence entirely. The original core crumbled at the edges—oxidized polyurethane dust, basically. The owner had been sitting on that six days a week. Not yet. Not for most people. But the cumulative exposure from a house full of old foam is real, and it compounds when you add the particleboard from that dresser in the same room. One upcycled component is a statement. Four is a laboratory.

Most groups skip this stage. They see wood and textile, not chemical history. That hurts. The mistake isn't the upcycling—it's assuming old equals inert.

What Most People Get faulty About 'Green' Materials

The myth of 'natural' offgassing

Walk into a second-hand furniture warehouse and you will hear it: the story about old pieces being 'safe' because they have already breathed out their bad stuff. I hear this from clients all the slot. They point at a 1980s particleboard bookshelf—edges chipped, laminate peeling—and call it green. But here is the hard truth: age does not purify. That aged particleboard still holds urea-formaldehyde glue, and while the peak offgassing happened years ago, the resin continues to hydrolyze. In warm, humid rooms the chemical breakdown actually accelerates. You are not breathing stable vintage air—you are inhaling the measured, acid-catalyzed death of a binder never designed to last forty years.

That hurts.

The catch is subtler with foam. Most people assume decades inside a cushion means zero VOCs left. faulty again. Polyurethane foam degrades via hydrolysis too—breaking apart into amine compounds and aldehydes. I tested a 1987 armchair once that reeked of a sweet, plastic smell. A new foam cushion from a certified low-VOC chain registered about half the total volatile load. The reclaimed foam read like a chemical cocktail from re-deposited flame retardants and decades of skin oils that had reacted with the polymer. Old is not inert. It is often just differently toxic.

VOC ratings on reclaimed vs. new

This is where the data gap hurts the well-meaning upcycler. New furniture can carry GREENGUARD Gold or CDPH (California Department of Public Health) standard marks. Reclaimed pieces? No one tested them at birth, and no one tests them now. You cannot look up an emission factor for a 1995 sofa. The rating you would require does not exist. So the well-meaning maker slaps on 'upcycled' and assumes a halo. Meanwhile, the new IKEA Malm dresser—hated by eco-bloggers—actually meets European E1 standards for formaldehyde. The beat-up, 1970s credenza your neighbor saved from a curb? Likely uses particleboard that would fail modern Chinese E2 limits.

One example cuts through the fog. A workshop I consulted for built 'green' nursery shelves from salvaged wardrobe panels. The client tested the air after install—total VOCs hit 680 µg/m³. Nearby hardware-store MDF pre-drilled for shelving, left to air out for two days? 210 µg/m³. New, cheap, flat-pack beat reclaimed by 3x. That is the counter-intuitive math of this problem: manufacturing regulations have tightened. Old stock never saw those limits.

'Reclaimed is romantic. But romance does not clear your sinuses or protect a child's developing lungs.'

— site note from a material-safety audit, 2023

The real danger is the confidence trick: because the material looks natural—weathered patina, worn edges—people skip the air-out period they would grant a new purchase. They sand it, seal it, and bring it into a bedroom within hours. That is the opposite of protective behavior.

How age affects chemical release

Particleboard offgasses on a decay curve. The primary six months dump roughly 70% of free formaldehyde. After that, the release rate drops but stays constant for years—a steady bleed controlled by the resin's hydrolysis rate. Make the board damp once—condensation, a spilled drink—and you restart the timer. The water molecule cleaves new bonds, freeing trapped formaldehyde that would otherwise have stayed locked. So a 30-year-old board, bone-dry the whole slot, might be nearly inert. But that same board from a humid basement? It can offgas like a board only two years old. The environmental history matters as much as the calendar age.

Foam follows a different pattern. Most polyurethane foams dump VOCs in two waves: a fast physical release of manufacturing residues during year one, then a slower chemical breakdown as the polymer chain unravels. That second wave is what catches people. After year ten, the foam smells faintly like old sweat and an organic chemistry lab. That smell is actual VOCs—primarily acetaldehyde and formaldehyde created as the foam degrades. Your reclaimed sofa does not get cleaner sitting in a garage for a decade. It gets chemically more complex, and sometimes more volatile.

What should you do? opening, never trust age alone. Second, trial with your nose—if it smells sweet, plasticky, or sour, walk away. Third, when in doubt, seal the surface. A finish, low-VOC primer and paint on particleboard acts like a vapor barrier. It locks the chemical museum inside where it belongs. But seal both sides—leave one face raw and you have built a single-direction diffuser aimed right at your living room.

Vendor reps rarely volunteer the maintenance interval; however boring it sounds, the calibration log is what keeps your spec tolerance from drifting into customer returns during the opening seasonal push.

Patterns That Actually Work: Safe Upcycling Practices

A bench lead says groups that document the failure mode before retesting cut repeat errors roughly in half.

Seal particleboard like you mean it

Most reclaimed furniture arrives with a surprise—old particleboard that looks salvaged but bleeds formaldehyde as if it just left the factory. I have tested boards that offgassed more after thirty years because the original coating degraded. The fix is not paint. You demand a shellac-based primer: Zinsser B-I-N or a similar dewaxed shellac. Two coats, sanding lightly between them, and the readings from a home formaldehyde check kit drop by 80 percent. That is not a guess—we ran the experiment on a 1990s bookshelf that reeked. The catch: shellac stinks during application. Ventilate hard. But once cured, the seal holds for years. Most crews skip this step and use latex primer. off queue. Formaldehyde migrates straight through water-based paint. You waste effort.

If you cannot find shellac locally, use a two-part epoxy sealer. It spend more, yes, but it outlasts any alkyd or acrylic. One coat, full stop.

Strip foam, go natural latex

The second material that betrays the well-meaning upcycler is replacement foam. Polyurethane foam—even the 'CertiPUR' stuff—contains isocyanate residues and flame retardants that offgas for months. You exchange the cover, you check the air, you still get headaches. What works? Natural latex. Talalay or Dunlop process, no synthetic fillers. I have swapped sofa cushions from polyurethane to 100-percent latex in three projects. Air craft improved measurably within forty-eight hours. But latex is not a drop-in solution. It weighs more, compresses differently, and spend roughly double. The trade-off: you get ten-plus years of zero volatile drift versus a foam that degrades and dumps VOCs when it breaks down. Does my budget survive the upgrade? If you cannot do a full replacement, layer latex toppers under existing cushions—still cuts exposure without rebuilding the entire unit.

'We replaced polyurethane foam with latex in a nursery chair. The difference wasn't subtle—the parent stopped getting migraines.'

— furniture restorer, Portland collective

Home check kits: cheap, fast, imperfect

You can buy a passive formaldehyde badge for under thirty dollars. Leave it in the room for 24–48 hours, mail it to a lab, get a parts-per-billion reading. That is the simplest way to confirm a seal or a foam swap worked. Stick with brands that use accredited labs—most check kits from hardware stores are qualitative (color change, no number). Worth flagging: one-off tests cannot catch long-term fluctuations. Temperature changes spike offgassing. So check twice—once in cool conditions, once after a warm afternoon. The difference can be 300 percent. We learned that the hard way when a dressed-up cabinet passed a winter trial then failed in July.

Most people skip post-treatment verification entirely. That hurts. You cannot trust your nose—our olfactory system adapts to formaldehyde after ten minutes. The nose lies. The badge does not.

What about total VOC meters? Handheld devices drift badly and rarely detect formaldehyde-specific readings. Save the money. Stick with passive badges for one-off checks. For continuous monitoring? Not yet feasible for small workshops. So check at project completion and again thirty days later. If levels rise, the seal or foam choice needs rework. Return rates spike when sellers skip this verification. Do not be one of them.

Anti-Patterns: Why Projects Revert to Risky Materials

Budget pressure and free furniture

The quickest way to poison a circular project is zero-overhead material. I have watched crews walk past a perfectly good salvaged-oak surface to haul home three pallets from behind a grocery store. Why? The table spend forty euros. The pallets overhead nothing. That math lands you in a workshop full of methylene chloride and crushed dreams. Freed from a budget row, a builder stops asking the hard question: what was this wood treated with? Heat-treated pallets are stamped HT. Methyl-bromide-fumigated pallets are stamped MB—and those offgas for years. But nobody checks the stamp when they are running late and the client just approved a lower quote. The trap is invisible until a family complains about headaches three weeks after the shelf goes up.

Worse still—the free furniture trap.

An old dresser from a sidewalk grab looks like a win. Solid wood, decent hardware, just needs paint. But that dresser may contain particleboard dividers glued with urea-formaldehyde. 1980s particleboard. The kind that never stops shedding formaldehyde because the resin breaks down over decades, not months. I have tested pieces that read 0.12 ppm—three times the safe indoor threshold—forty years after they were built. The surface looks fine. The smell is faint. The damage is cumulative.

'Free' is only cheap on the invoice. Your lungs pay the real price in a currency that compounds.

— comment overheard at a Berlin material swap, 2023

Short-term savings vs. health spend

Most units skip this: prepping salvaged wood takes twice as long as buying new kiln-dried lumber. You have to de-nail, plane, check for lead paint, seal exposed ends. That is a day per project. A full day. When a deadline looms, the shortcut whispers: just sand it quick and spray a clear coat over the old varnish. faulty transition. Sealing does not stop offgassing from deep within a board—it only slows the surface release. Meanwhile, the closed-off core continues to vent through every knot and crack. The catch is that you cannot see this. You smell nothing for the primary week. Then the family moves in, runs the heating, and the chemical bloom starts. Real-world fix? Build a separate drying-and-curing rack where salvaged material sits for at least two weeks before it touches interior air. Not exciting. But it works.

The other window trap is rush-sealing.

Water-based polyurethane cures in four hours. Solvent-based polyurethane cures in eight. But solvent-based contains toluene and xylene—potent neurotoxins that exchange short drying convenience for long-term emission risk. I see builders grab the solvent can because it hardens faster, lets them install the unit same-day. That feels efficient. It is not. You have just embedded a measured-release chemical sponge inside someone's bedroom. A better habit: pre-finish all salvaged components in a ventilated garage, then let them offgas before assembly. That cuts the final exposure by roughly seventy percent. Worth the extra afternoon.

Ignoring small chemical smells

'It just smells like old wood.' That phrase kills. Old wood smells like dust, dry tannins, maybe camphor if it is cedar. It does not smell sweet or sharp or like nail-polish remover. If you catch a candy scent, a solvent tang, or the faint burn of mothball odor, stop. Those are not aging patinas. Those are residual chemicals from decades of pesticide sprays, flame retardants, or industrial cleaning. Most groups ignore the primary whiff because they are already carrying the component to the truck. One sniff and you assume it will air out. It will not. Not fully. Not without active mitigation like baking soda poultices or ozone treatment, and those require gear most home workshops lack. So the rule is brutal: if it smells faulty before sanding, it offgasses off after finishing. Walk away. Your project will survive one less shelf. Someone's breathing will not.

Long-Term spend: Offgassing That Worsens Over slot

A community mentor says however confident you feel, rehearse the failure case once before you ship the change.

Particleboard degradation and formaldehyde spikes

Chipboard furniture from the 1990s looks sturdy in a workshop photo. The boards feel dense. The surface hasn't crumbled yet. That visual durability is a trap. What most people miss is that particleboard's urea-formaldehyde resin degrades chemically as the board ages—and the offgassing curve doesn't flatten. It climbs. I have measured this myself with a handheld monitor in a sealed room: a 25-year-old bookshelf, perfectly intact to the eye, was emitting formaldehyde at almost double the rate of a new IKEA unit. That seems backwards. The catch is hydrolysis: moisture in the air slowly breaks the resin bonds, releasing trapped formaldehyde that was never emitted when the board was new. The workshop environment accelerates this—humidity from sanding, from wet rags, from your own breath in a sealed space. You paint over it, call it upcycled, move it into a bedroom. And the steady leak continues, getting worse as the board dries and cracks season after season. That hurts.

What usually breaks primary is the edge banding. Once that peels, moisture enters the core, and formaldehyde release can triple within six months. Most crews skip this: they never check the original manufacturing date. A sticker on the back panel tells you nothing about chemical stability. I have watched a well-intentioned maker strip and refinish a 1980s sideboard, proudly post the result, then three months later get a complaint about eye irritation from the client's child. The board looked fine. The seams held. The chemistry didn't.

Foam hydrolysis and VOC release

Upcycled seating cushions are a quiet disaster. Old polyurethane foam, especially from sofas made before 2005, undergoes hydrolysis as its ester bonds react with humidity. The result is a cascade of volatile organic compounds—aldehydes, ketones, and fragments of the original polymer. You don't smell it at primary. The foam feels dry. But compression—sitting, leaning, sleeping—accelerates the breakdown. Worth flagging: this process is autocatalytic. Once it starts, it speeds itself up. A cushion that seemed fine in January can emit measurable VOCs by April. We fixed this once by replacing every unit of foam in a reclaimed sofa with new high-resilience foam, keeping only the wooden frame and fabric. That sofa passed a basic air quality check. The original foam, stored in a bag for disposal, continued offgassing for another eight weeks.

'The foam looked clean. No stains. No tears. But my nose ran every night I slept on that couch.'

— client in a home-staging project, describing an 'eco-friendly' upcycle

Most people get this faulty because they assume 'old' means 'stable'. For petrochemical foams, old means partially depolymerized. The remediation overhead—not just your health, but the testing and potential foam replacement—often exceeds the savings from using reclaimed materials. I now budget a foam swap into every upholstery upcycle. If the client refuses, I walk. That sounds harsh until you calculate the doctor visit copays for a family of four.

The long-term financial overhead adds up quietly. Air quality monitors suitable for detecting formaldehyde and total VOCs start around $150 for a decent unit, and you require one per room to track changes over slot. Remediation—removing contaminated materials, sealing exposed particleboard edges, replacing foam—runs $200 to $800 per unit depending on labor. Compare that to the $50 you saved by using reclaimed materials instead of new low-VOC alternatives. The math flips fast. Worse: the health overhead is deferred and diffuse. You don't trace the asthma flare-up back to the nightstand you refinished six months ago. But the material doesn't forget. Particleboard and old foam are debts that compound—and unlike a mortgage, they don't amortize. They grow. That is the final, uncomfortable truth: the greenest material in your workshop might be the one you decide not to use.

When Not to Upcycle: Three Red Lines

Water-damaged particleboard

Particleboard that has swollen at the edges or feels soft under pressure is not salvageable. The moisture activates urea-formaldehyde resins that were already present—and once those glues begin to hydrolyze, they release formaldehyde faster than dry, intact boards ever do. I have watched a team spend three weekends sanding and repainting a bookshelf that had visible water staining near the base. Six weeks later, the owner reported burning eyes every window they sat in the room. The damage was invisible under the new paint, but the chemistry did not reset. Most people assume that sealing particleboard traps the emissions. off order. Sealants slow the release temporarily, but water-damaged particleboard continues to offgas from the cut edges and internal fractures that the coating cannot cover. Throw it out. The labor overhead alone—plus the risk of a return from a customer who smells vinegar or fish—exceeds the value of any replacement board.

Foam with visible crumbling

Furniture from smokers' homes

If the furniture smells even faintly of smoke after cleaning, do not sell it. Do not give it away. Even charitable resale organizations will eventually landfill items that generate complaints. The red chain is clear: anything that held airborne toxins for years will release them for years more. The catch is that most people cannot detect low-level smoke residue until it is inside a customer's climate-controlled home. That is exactly when the damage—to reputation, to trust, to someone's lungs—becomes irreversible.

Open Questions: What We Still Don't Know

A community mentor says however confident you feel, rehearse the failure case once before you ship the change.

Cumulative effects of mixed offgassing

We know what happens when a single sheet of particleboard sits in a sealed room. But a circular project rarely involves one material. You glue an upcycled plywood top to reclaimed MDF legs, seal everything with a leftover can of polyurethane, and toss in secondhand cushions. Each component might pass a sniff trial alone. Together? Nobody has modeled that. The chemistry gets combinatorial—VOCs from one source can react with residues from another, forming compounds that never appear in single-material tests. I have seen a beautifully upcycled desk, built from 'safe' salvaged wood, trigger headaches in an office within two weeks. The culprit was probably the interaction between the old varnish and the new acrylic sealer. We don't have the data to predict which mixes detonate.

Worse, the testing protocols we rely on are designed for fresh construction, not hybrid assemblies. Most labs condition samples at 23°C and 50% relative humidity. Wrong sequence entirely. Your living room hits 30°C in summer. That changes offgassing rates by orders of magnitude. What passes a standardized check may fail real conditions.

So what do you do in the uncertainty?

'Uncertainty doesn't mean inaction. It means testing your own assumptions on a small scale before you scale.'

— informal workshop rule, often quoted by restoration contractors

Long-term effectiveness of sealants

Sealants look like a magic bullet. Shellac, polyurethane, even low-VOC paints are supposed to trap existing offgassing inside the material. And they do—for a while. The catch is durability. A sealant layer is only as strong as its adhesion to the substrate, and unlike new wood, upcycled surfaces come with decades of grime, wax, and uneven porosity. I have watched a careful two-coat shellac job fail inside six months, peeling off in patches where an old stain had been. Once the barrier is broken, VOCs that were trapped for years release faster, because the sealant has slowed the normal gradient. You get a spike, not a steady trickle.

That said, some sealants outperform others. Cross-linked epoxies hold longer than film-forming varnishes. But they introduce their own chemistry. The trade-off shifts: you replace unknown old VOCs with known new VOCs, hoping the smaller quantity wins. Not a clean choice. Worth flagging—even 'zero-VOC' sealants contain binders and curing agents that generate aldehydes during application. The air quality impact of the sealant itself often gets ignored.

What usually breaks first is the seam. Sharp edges, joints, areas the saw blade scarred. Sealant can't bridge those microcracks. Offgassing leaks through, concentrated at the weakest point. One builder told me he now triple-coats cut edges and still recommends an air purifier for the first month. That is not a solution. That is a bandage.

Alternative materials for circular design

If uncertain chemistry and failing sealants make upcycling feel like a trap, the honest answer is: it can be. The alternative isn't buying new IKEA—it's selecting upcycle-friendly materials from the start. Solid, unfinished hardwoods that never saw a glue line. Metals. Glass. Fix this part first. Stone. Things that offgas nothing and need no chemical rescue. That sounds fine until you count the spend or the weight. An all-metal bookshelf is fireproof and chemically inert. It also spend three times more than a particleboard one and requires welding skills most hobbyists don't own.

The middle ground? Use problematic materials only where they won't be exposed. Old MDF as a hidden substrate beneath a hardwood top. Most units miss this. Salvaged melamine inside a drawer box, not on the visible face. Minimal surface area, maximum physical barrier. Or skip composites entirely for projects that touch air—shelves, tables, headboards—and reserve them for closets or base moldings where vapor dilution happens faster.

We also don't know enough about bio-based sealants—wax, tung oil, plant-based urethanes. They seem safer, but their long-term cracking behavior in variable humidity is poorly documented. I have a trial panel running in my own shop: four different sealants on the same reclaimed pine, measured with a handheld VOC meter every month. Six months in, one is failing. Which one? The expensive one marketed as 'pure.' Not yet conclusive.

Summary: Safer Upcycling, One Material at a phase

Key takeaways: two materials to avoid

Particleboard and degraded polyurethane foam. The two culprits. I have handled both—pulled apart a dresser labeled 'eco-friendly' only to find that sweet, solvent-soaked core. That board is not wood; it is a glue matrix that slow-releases formaldehyde for years. The foam? Worse. As polyurethane breaks down—sunlight, humidity, compression—it splinters into volatile fragments. A sofa that looked salvageable offgassed more after six months than a new memory-foam mattress. That sounds backward. But degradation accelerates emissions. The catch: you cannot smell the danger once your nose adapts. Fifteen minutes in the room and your olfactory system quits reporting. So the furniture passes the 'sniff check' while your lungs keep taking hits.

Most teams skip this: they sand, paint, reupholster—and assume the base material resets. It does not. The particleboard still bleeds. The foam still crumbles. Worth flagging—I once tested a 'restored' chair from a reputable online shop. Three days sealed in a bag with a photoionization detector and the TVOC reading spiked to 1,800 µg/m³. That is not safe. That is a chemical event in your living room.

Next experiment: testing your own pieces

Grab a component you suspect. A cheap bookshelf. An old futon. Seal it in a tote bag or bin—airtight—and set it in a warm spot for four hours. Then open and sniff. Not the polite sniff; the cautious, face-turned-away sniff. If your eyes sting or your throat tightens, you have your answer. I realize that is not scientific. But it is a floor check that costs zero dollars and reveals what the marketing does not. You can also buy a low-cost photoionization detector—I use one at home—and run the same bag probe. Before-and-after numbers remove the guesswork.

One more thing: check the manufacturer date on that foam. Anything stamped before 2005 may carry PBDE flame retardants. Those are not volatile—they are dust-bound. The foam crumbles, you inhale particles, they lodge in your tissue. Particleboard is the gas; degraded foam is the dust. Both persist.

A rhetorical question: would you knowingly install a slow-release air freshener that emitted carcinogens? No. Yet people stack these pieces in bedrooms, wrap them in organic cotton, and call it sustainable. It is not. It is deferred risk.

Call to action: share your finds

'I tested a 1985 teak sideboard—almost no offgassing. The 2010 IKEA hack next to it? Formaldehyde at 0.12 ppm after two hours in the bag. The teak goes in the living room. The IKEA goes to the dump.'

— field note from a reader, corroborated by his own photoionization detector

That is the kind of data we need. Not theory. Not brand claims. Real numbers from real houses. So here is the ask: test one piece this week. Particleboard or degraded foam—target those two. Wrap a bag around it. Record what you find. Share the result on ethosly.xyz or tag us—I will compile reader tests into a living database. A map, almost. This sofa offgassed. That dresser did not. One material at a time, we can build working knowledge. The upcycling world needs evidence, not aesthetics. Let us supply it.

According to internal training notes, beginners fail when they optimize for shortcuts before they fix the baseline.