The Repayment Overhaul

Replace the five income-driven repayment plans with one that takes between one and ten percent of income, waives unpaid interest each month, and forgives what is left after thirty years instead of twenty.

AI evaluation · not yet reviewed by a human

This evaluation was produced and sourced by an AI model; a human review is still pending. Figures and conclusions may still change. The review log is at the foot of the page.How review works →

Until 2026 a federal student loan borrower could choose between five income-driven repayment plans, the most generous of which took ten percent of income above 225 percent of the poverty line and forgave the rest after twenty years. The law passed in 2025 replaced all of them for new borrowers with a single Repayment Assistance Plan, in force since 1 July 2026: payments run from one to ten percent of adjusted gross income with a floor of ten dollars a month and fifty dollars off per child, unpaid interest is written off each month so a balance can never grow, and the remainder is forgiven after thirty years. Existing borrowers keep access to income-based repayment, and the other older plans close to everyone by July 2028. The budget office scores the change as saving the federal government 271 billion dollars over ten years. This evaluation looks ten years ahead.

Balance

Balanced · 0.43 previous scale

Balance on the previous scale. The Bilanz 2.0 simulation is not yet available for this evaluation. The category comes from the share of the debate on the pro side (r).

For 16 · 43 % Against 22 · 57 %

Size class: large Scale of this evaluation: Normalised Impact — unitless, calibrated to this topic. For comparison: one point here is worth roughly 5 billion euro per year. The 271 billion dollars the budget office scores as a federal saving is not new money: it is the same money, taken from 43 million borrowers instead. This evaluation therefore counts it twice, once as a gain to the treasury at the ordinary weight and once as a loss to borrowers at a higher one, because a dollar taken from someone repaying a student loan on a modest income is worth more than a dollar in the federal budget. That difference is the whole of the result. Treat the two as equal and the measure comes out level; treat the poorest fifth of borrowers, who paid nothing before and now pay from the first dollar of income, as bearing an ordinary loss rather than a heavy one, and it comes out slightly ahead. How we score →

Arguments for

Arguments against

6 arguments evaluated · Scoring v1.3 Δ absolute −6

Arguments — For

3 arguments

Money the government keeps

14of 100

The budget office scores the repayment overhaul as saving 271 billion dollars over ten years, which is about 27 billion a year. It is one of the largest single savings in the law that contained it.

Value 5 · Public financesImpact 4.7Plausibility 6

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is money the federal government collects instead of writing off, priced at the middle of the scale as public money always is. It is one end of a transfer and the borrowers are the other, which is why the same sum appears twice in this evaluation at two different weights. Nothing is added for the interest cost of the debt the government would otherwise carry, which is the same money under a later name. Nothing is added for the credibility of a lending programme that is repaid, which is a genuine good and is not measured by anything here. Whether the saving is used to reduce borrowing, cut taxes or spend elsewhere makes no difference to the weight. The value is the middle of the scale, the level this site uses for public money whatever it is later used for.

The Congressional Budget Office scores the new repayment plan at 271 billion dollars of savings over ten years, within a total student loan saving of about 315 billion for the law as a whole ^[3]^[4]. The plan itself is therefore 27.1 billion dollars a year, or 23.4 billion euro. Two changes produce it and they run in opposite directions: payments rise, because the plan takes a share of adjusted gross income from the first dollar rather than of income above 225 percent of the poverty line, and forgiveness is pushed from twenty years to thirty; against that the monthly write-off of unpaid interest costs the government money. The net is what the budget office scores and is what is used here, which is why the interest waiver does not appear as a separate gain anywhere in this evaluation. Public money carries the standard weight of 1.0. The Impact is the second largest here and it is the same figure as the largest, differing only in whose pocket it is measured in.

▸ Show calculation ▾ Hide calculation

	Federal saving from the new repayment plan [3]	over ten years, within a total student loan saving of about 315 billion	271 billion dollars
÷	Per year	over ten years	27.1 billion dollars a year
÷	In euro, at the standard weight for public money exchange rate used throughout this evaluation	1.16 dollars to the euro, weight 1.0	23.36 billion euro a year
÷	Normalised Impact scale of this evaluation	5 billion euro a point	4.67

Score 4.67 Impact × 5 Value × 6 Plausibility ÷ 10 = 14 of 100

A budget score is not a prediction about the world but an arithmetic exercise on a defined population under a defined statute, and this one is unusually well constrained. The counterfactual is the previous set of plans, which the office models from the same borrower data it uses for the new one. The chain from statute to cash flow contains one behavioural link that matters, which is which plan borrowers choose and whether they stay in it; the office models that from observed behaviour under the previous plans. The confounder is enrolment: a plan that costs more per month is one more borrowers leave, and a borrower who leaves and defaults pays less rather than more, which would cut the saving. That is named and is inside the office's own model rather than resolved by it. Reverse causation does not arise. The Plausibility is at the upper end of what a budget score can carry: the population, the formula and the horizon are fixed by statute, and only borrower behaviour is estimated.

evidence basis: Projection · P ceiling 6 identification: Definitional · no rung ceiling

Counterfactual: the previous set of income-driven plans, modelled from the same borrower data. Design: definitional for the formula — the payment schedule and the forgiveness horizon are statutory and the cash flow follows from them; only plan choice and persistence are behavioural. Confounder: borrowers leaving the plan or defaulting under higher payments, which would cut the saving; named and inside the office's model rather than resolved. Direction: not applicable. Ceiling: projection 6.0 binds — the number is a budget score, and a score is a projection however firmly the formula behind it is written.

A balance that cannot grow

1.3of 100

Under the old plans a borrower paying what they were told to pay could watch the balance rise anyway, because the payment did not cover the interest. The new plan writes off the shortfall every month, so that stops.

Value 6 · Trust in the systemImpact 0.5Plausibility 4.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is what it does to people to make every payment they are asked for and owe more at the end of the year than at the start. This site places that with confidence in public institutions rather than with money, because the money involved is already counted in the arguments on either side of it: the interest waiver is inside the budget office's net figure. What is left over, and is not counted anywhere else, is the experience of a debt that behaves as though the payments were not made. Nothing is counted for the borrowers who eventually reach forgiveness, for whom the balance was never going to be paid and its size never mattered. The value sits in the middle-upper part of the scale, where this site places confidence in public institutions, and the money in this stream is deliberately left to the arguments on either side.

About 12 million borrowers were in an income-driven plan under the old system with a payment below the accruing interest, which is the group whose balances grew. What the change is worth to each is set at 0.005 quality-adjusted years a year, in a range from 0.001 to 0.02 — under two days of good health, which is a low reading of something borrowers describe in surveys as the single most demoralising feature of the system. That gives 60,000 quality-adjusted years a year, or 2.4 billion euro. The figure is deliberately small: the same borrowers now pay more each month, which is counted against this measure, and it would be double counting to credit the plan with relief they are also paying for. What is not counted is the effect on people who avoid borrowing at all because of what they have seen happen to others. The Impact is a tenth of the money in this debate, which is what a real but unpriced experience is worth once the money attached to it is counted elsewhere.

▸ Show calculation ▾ Hide calculation

	Borrowers whose balance grew under the old plans [2]	payment below accruing interest	12 million borrowers
×	Quality-adjusted years gained each a year Setting, range 0.001 to 0.02: under two days of good health, a low reading of what borrowers describe as the most demoralising feature of the system	0.005	60,000 quality-adjusted years
×	Value of the years the value of a healthy life year used across this site	40,000 euro each	2.4 billion euro a year
÷	Normalised Impact scale of this evaluation	5 billion euro a point	0.48

Score 0.48 Impact × 6 Value × 4.5 Plausibility ÷ 10 = 1.3 of 100

The mechanism is definitional — the statute writes off unpaid interest monthly, so balances cannot grow — but what that is worth to a person is not. The counterfactual is the previous plans, under which negative amortisation was common and documented. The chain is named: payment below interest, balance rises, borrower concludes that repayment is futile, and the plan removes the first link. Every step of that is plausible and the last one, the effect on the person, has never been measured; the surveys that report borrower distress do not compare against a group whose balances were falling. The confounder that matters is that the borrowers who most notice a growing balance are the ones with the largest debts and the lowest incomes, who are also the ones paying most under the new plan, so the same person appears on both sides of this evaluation. That is named and unresolved. Reverse causation does not arise. The Plausibility is below the middle: the balance genuinely stops growing and what that is worth to anyone is assumed.

evidence basis: Mechanism · P ceiling 6 identification: Mechanistic · rung ceiling 6 band: Chain closed, unevidenced · P 4–5

Counterfactual: the previous plans, under which payments below accruing interest were common and documented. Design: mechanistic — the write-off is statutory, the effect on the borrower is a named chain with no measurement. Confounder: the borrowers who notice a growing balance most are the ones paying most under the new plan, so the same person stands on both sides; named and unresolved. Direction: no reverse causation. Ceiling: mechanistic 6.0 binds. Band: chain closed but unevidenced — every link named, only the value to the person missing.

Nothing measured argues against the claim; what is missing is any study of what a growing balance costs a borrower against a comparable borrower whose balance falls. Read back: about half the time, ending negative amortisation is worth roughly the amount of healthy time assumed here.

Open: The servicers hold monthly balances and the plan enrolment of every borrower. Comparing distress and repayment behaviour across borrowers just above and just below the point where the payment covers the interest would measure this directly.

One plan instead of five

1.1of 100

The old system had five income-driven plans with different formulas, different forgiveness horizons and an annual paperwork requirement that pushed borrowers out of them. Every year a large number lost their plan for missing it.

Value 5 · Household budgetsImpact 0.5Plausibility 5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is what happens to a borrower who falls out of an income-driven plan by accident: the payment jumps to the standard amount, the account goes delinquent, and the damage runs from a credit record to a garnished wage. It is priced at the middle of the scale because what is measured is money and money-equivalent harm. It is separate from the payment increase counted against this measure, which affects borrowers who stay in the plan; this argument is about the ones who used to fall out of it. Nothing is counted for the time spent on the paperwork itself, which is small. Nothing is counted for the servicers, whose costs fall. The value is the middle of the scale, and what is priced is the damage of falling out of a plan by accident rather than the plan's terms.

Under the old system borrowers had to recertify their income every year, and a large share missed the deadline: servicer reporting and audits put the failure rate at between a fifth and a third of enrolments in some years. A single plan with automatic data sharing removes most of that. The figure used here is 400,000 fewer delinquencies a year, in a range from 150,000 to 900,000. What one costs the borrower is put at 4,000 euro, in a range from 1,500 to 10,000: months at a payment they cannot afford, late fees, and a credit record that raises the price of everything bought on credit. Weighting by where these borrowers sit gives 1.4. That is 2.24 billion euro a year. The Impact is a tenth of the money in this debate, and it is the only argument here that does not have a mirror image on the other side.

▸ Show calculation ▾ Hide calculation

	Delinquencies avoided by removing annual recertification Setting, range 150,000 to 900,000 [2]	a fifth to a third of enrolments missed the deadline in some years	400,000 a year
×	Cost of one to the borrower Setting, range 1,500 to 10,000 euro: months at an unaffordable payment, late fees, and a damaged credit record	4,000 euro	1.6 billion euro a year
×	Weight of a euro for these borrowers borrowers who fall out of an income-driven plan sit below the middle of the distribution	1.4	2.24 billion euro a year
÷	Normalised Impact scale of this evaluation	5 billion euro a point	0.45

Score 0.45 Impact × 5 Value × 5 Plausibility ÷ 10 = 1.1 of 100

The counterfactual is the old five-plan system with annual recertification. The chain is named — one plan, automatic income data, fewer missed deadlines, fewer forced exits — and the first two links are statutory. What is not measured is the last one: nobody has compared delinquency rates across borrowers under a simplified and an unsimplified plan, because the simplification is new. The confounder that matters points against the argument: a borrower who missed the paperwork under the old system was often a borrower in difficulty for other reasons, and simplifying the form does not fix that. It is named and unresolved. There is also a counter-mechanism in the other direction — a higher monthly payment is itself a reason to leave a plan — which the first argument on this side already accounts for in the budget score. Reverse causation does not arise. The Plausibility is at the middle: the failure rates under the old system are documented and what a single plan does to them is estimated.

evidence basis: Mechanism · P ceiling 6 identification: Mechanistic · rung ceiling 6

Counterfactual: the previous five-plan system with annual recertification. Design: mechanistic — chain named (one plan, automatic income data, fewer missed deadlines, fewer forced exits) with the last link unmeasured because the simplification is new. Confounder: borrowers who missed the paperwork were often in difficulty for other reasons, which a simpler form does not fix; named and unresolved. Direction: no reverse causation. Ceiling: mechanistic 6.0 binds.

Arguments — Against

3 arguments

Borrowers pay the difference

15of 100

A borrower earning 40,000 dollars paid about 40 dollars a month under the plan this replaced and pays about 132 under the new one. Across 43 million borrowers that is where the federal saving comes from.

Value 5 · Household budgetsImpact 4.9Plausibility 6

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is money leaving the households of people repaying student debt, priced at the middle of the scale exactly as the treasury's gain is. This is the other end of the same transfer, and without it the saving above would be counted as though nobody paid for it. The weight is what differs: student loan borrowers on income-driven plans are, by the definition of the plan, people whose income is low relative to what they owe, and a euro is worth more there than in the federal budget. The poorest fifth of them are taken out of this argument and counted separately, because their increase is proportionally much larger and averaging it into the rest would hide it. Nothing is counted for the years of repayment added at the end, which is the same money arriving later. The value is the middle of the scale, and this argument exists so that the saving above is not booked as if it came from nowhere.

The saving to the government and the cost to borrowers are the same 23.4 billion euro a year. Two fifths of it comes from higher monthly payments and the rest from the forgiveness horizon moving from twenty years to thirty, which for a borrower who never repays in full means ten more years of payments. The published comparison is a borrower on 40,000 dollars, who paid about 40 dollars a month under the plan this replaced and pays about 132 under the new one ^[4]. Twenty percent of the total is attributed to the poorest fifth of borrowers and counted in its own argument, leaving 18.7 billion euro here. Those borrowers sit below the middle of the income distribution but not at the bottom of it, giving a weight of 1.3, in a range from 1.1 to 1.7. That is 24.3 billion euro a year, slightly more than the government gains from the same money. The Impact is the largest here, and the amount by which it exceeds the treasury's gain is the whole of what this evaluation says about the measure.

▸ Show calculation ▾ Hide calculation

	What borrowers pay in addition [3]	the federal saving, seen from the payers' side	23.36 billion euro a year
×	Share falling on borrowers other than the poorest fifth the poorest fifth are counted in their own argument, because their increase is proportionally much larger	80 %	18.69 billion euro a year
×	Weight of a euro for these borrowers Setting, range 1.1 to 1.7: people on an income-driven plan have low income relative to what they owe, by the definition of the plan	1.3	24.3 billion euro a year
÷	Normalised Impact scale of this evaluation	5 billion euro a point	4.86

Score 4.86 Impact × 5 Value × 6 Plausibility ÷ 10 = 15 of 100

The quantity is the budget score read from the other side, so it carries the same confidence as the argument opposite, and the same behavioural caveat. The counterfactual is the previous set of plans. What is definitional is that a transfer has two ends: if the government collects 23.4 billion euro more, borrowers pay 23.4 billion euro more, and no design is needed to establish that. What is estimated is the split between the poorest fifth and the rest, which is set here from the shape of the payment formula rather than from a distributional table, and the weight applied to each. The confounder that matters is the same one as opposite and in the same direction: borrowers who leave the plan or default pay less than the score assumes, which would reduce both this argument and the one it mirrors. Reverse causation does not arise. The Plausibility is at the upper end of what a budget score can carry, for the same reason as the argument it mirrors: a transfer has two ends and the score fixes both.

evidence basis: Projection · P ceiling 6 identification: Definitional · no rung ceiling

Counterfactual: the previous set of income-driven plans. Design: definitional — the sum is the budget score read from the payers' side, and a transfer has two ends by identity. Confounder: borrowers leaving the plan or defaulting, which reduces this argument and the one it mirrors together; named. Direction: not applicable. Ceiling: projection 6.0 binds, matching the argument it mirrors — the sum is a budget score read from the other side.

The ones who paid nothing

7of 100

The plan this replaced exempted income below 225 percent of the poverty line, so the poorest borrowers paid nothing at all. The new one takes a share from the first dollar, with a floor of ten dollars a month.

Value 5 · Household budgetsImpact 2.3Plausibility 6

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is the same money as the argument above, taken from the borrowers least able to give it up. It is counted separately rather than averaged into the rest because the burden is not evenly distributed: for a borrower who paid nothing, any payment is an infinite proportional increase, and folding that into an average with a borrower whose payment rises by a third would hide the group the change actually falls on. The value class is the same — this is money — and what differs is the weight, which is the highest this site uses for household income. Nothing is counted for the borrowers who leave the plan altogether rather than pay, whose losses are larger still and are not estimated. The value is the middle of the scale, and the reason this stands apart from the argument above is that the burden falls on a small group much harder, not that the good is different.

The plan this replaced exempted adjusted gross income below 225 percent of the federal poverty line, which for a single person is about 35,000 dollars, so a large share of borrowers on it paid nothing. The new plan takes one percent from the first dollar and never less than ten dollars a month, with fifty dollars off per child. Twenty percent of the total transfer is attributed to this group, in a range from 10 to 35 percent — 4.67 billion euro a year. A euro at these incomes carries the highest weight this site uses for households, 2.5, which is the same figure used for money reaching families in the bottom fifth anywhere else on the site. That gives 11.68 billion euro a year. The size rests on how many borrowers sat below the old exemption, which the Department has not published for the new plan and which is the widest uncertainty in this evaluation. The Impact is half the size of the broad payment increase although it involves a fifth of the money, which is what the weighting is for.

▸ Show calculation ▾ Hide calculation

	What borrowers pay in addition [3]	the federal saving, seen from the payers' side	23.36 billion euro a year
×	Share falling on borrowers who paid nothing before Setting, range 10 to 35 percent: the old plan exempted income below 225 percent of the poverty line, the new one takes a share from the first dollar [4]	20 %	4.67 billion euro a year
×	Weight of a euro in the bottom fifth the highest weight this site uses for household income, applied here as everywhere else	2.5	11.68 billion euro a year
÷	Normalised Impact scale of this evaluation	5 billion euro a point	2.34

Score 2.34 Impact × 5 Value × 6 Plausibility ÷ 10 = 7 of 100

The direction and the mechanism are statutory and not in dispute: an exemption was removed and a floor was introduced, so people who paid nothing now pay something. The counterfactual is the previous plan's exemption threshold. What is estimated is the share of the total that falls on this group, which is set here from the shape of the two formulas rather than from a published distributional table, and it is the number this argument turns on. The confounder that would lower it is that some of these borrowers were not enrolled in the old plan at all, and were therefore already paying more than nothing; that is named and unresolved and is why the range runs down to ten percent. Reverse causation does not arise. The weight of 2.5 is the same one used for the poorest fifth throughout this site and is not adjusted for this case. The Plausibility is above the middle: the removal of the exemption is written into the statute and how much of the total it accounts for is estimated.

evidence basis: Projection · P ceiling 6 identification: Definitional · no rung ceiling

Counterfactual: the previous plan's exemption of income below 225 percent of the poverty line. Design: definitional — the exemption was removed and a floor introduced by statute, so the direction follows from the text; only the share of the total falling on this group is estimated. Confounder: some of these borrowers were not enrolled in the old plan and were already paying; named, unresolved, and the reason the range runs down to ten percent. Direction: not applicable. Ceiling: projection 6.0 binds, because the share of the total falling on this group is estimated rather than published.

Parents with no plan at all

0.5of 100

Parent PLUS loans are not eligible for the new plan, and neither is a consolidation loan that contains one. For those borrowers the income-driven option disappears entirely rather than becoming less generous.

Value 5 · Household budgetsImpact 0.2Plausibility 5.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is money leaving households that have no way to reduce the payment when income falls, priced at the middle of the scale like the rest of the transfer. It is separate from the other two cost arguments because the change is different in kind: those borrowers pay more under a plan, these have no plan. The households concerned are parents who borrowed for a child's education, which places them across the income distribution but concentrated below the middle, since the wealthier did not need to borrow. Nothing is counted for the child whose education the loan paid for. Nothing is counted for the borrowers who default as a result, whose losses are larger and are not estimated. The value is the middle of the scale, and what distinguishes this from the arguments above is that the option disappears rather than tightens.

About 3.7 million people hold a Parent PLUS loan. The new plan excludes them, and excludes a consolidation loan that contains one, so a parent whose income falls has no income-driven option left once the older plans close in July 2028. The number affected in a given year — those who would have moved onto an income-driven plan and now cannot — is put at 300,000, in a range from 100,000 to 700,000. What it costs each is put at 2,000 euro a year, in a range from 800 to 6,000: the difference between a standard payment and what an income-driven plan would have asked. Weighting gives 1.5, since parents who borrow for a child's education and then need an income-driven plan are concentrated below the middle of the distribution. That is 900 million euro a year. The Impact is the smallest here, about a twenty-fifth of the broad payment increase, because it reaches a specific group rather than everyone.

▸ Show calculation ▾ Hide calculation

	Parent borrowers who would move to an income-driven plan in a year Setting, range 100,000 to 700,000: those whose income falls and who have no other way through [2]	of about 3.7 million holding a Parent PLUS loan	300,000 borrowers a year
×	Additional payment each a year Setting, range 800 to 6,000 euro: the difference between a standard payment and what an income-driven plan would have asked	2,000 euro	600 million euro a year
×	Weight of a euro in these households parents who borrow for a child's education and then need an income-driven plan are concentrated below the middle of the distribution	1.5	900 million euro a year
÷	Normalised Impact scale of this evaluation	5 billion euro a point	0.18

Score 0.18 Impact × 5 Value × 5.5 Plausibility ÷ 10 = 0.5 of 100

The exclusion is written into the statute, so whether these borrowers have an income-driven option is not in question. The counterfactual is the income-contingent plan they could previously consolidate into. What is estimated is how many would use one in a given year, which depends on how many parent borrowers see their income fall, and on how many find another way through — deferment, forbearance or simply not paying. The confounder that would lower the figure is that some of these borrowers would have defaulted under either system, in which case the plan's absence costs them less than assumed; that is named and unresolved. Reverse causation does not arise. The cost per borrower is the difference between two published payment schedules and is the firmer half of the calculation. The Plausibility is a little above the middle: the exclusion is statutory and the number of people it bites on each year is estimated.

evidence basis: Projection · P ceiling 6 identification: Mechanistic · rung ceiling 6

Counterfactual: the income-contingent plan these borrowers could previously consolidate into. Design: mechanistic — the exclusion is statutory, but how many parent borrowers need an income-driven plan in a given year is behavioural and unmeasured. Confounder: borrowers who would have defaulted under either system, for whom the plan's absence costs less than assumed; named and unresolved. Direction: no reverse causation. Ceiling: projection 6.0 binds and mechanistic gives the same.

Summary

This is a transfer, not a saving, and that is why it comes out slightly negative. The 271 billion dollars the budget office credits to the federal government over ten years is money that 43 million borrowers pay instead, and a dollar taken from someone repaying a student loan on a modest income does more work there than in the federal budget. Two features of the new plan are genuine improvements and are counted as such: a balance can no longer grow while payments are being made, and one plan with automatic income data replaces five with an annual form that borrowers routinely missed. What tips the ledger is the bottom of the distribution — the plan this replaced exempted income below 225 percent of the poverty line, and the new one takes a share from the first dollar with a floor of ten dollars a month. Whether that group is a fifth of the burden or a tenth is the number that decides the result, and the Department has not published it.

Outlook — effect over time

Balanced · 0.43 previous scale

Sources

U.S. Department of Education: Fact sheet: simplifying student loan repayment. ed.gov
U.S. Department of Education, Federal Student Aid: Federal student aid portfolio summary. studentaid.gov
Committee for a Responsible Federal Budget: Student loan costs drop to near record lows after reconciliation reforms. crfb.org
The Education Trust: How the Repayment Assistance Plan works. edtrust.org
American Enterprise Institute: An analysis of the One Big Beautiful Bill Act's effect on student loans. aei.org

Last reviewed by Claude Opus 5 · September 6, 2026 · 1× AI, not yet reviewed by a human

September 6, 2026AI reviewClaude Opus 5First evaluation
First evaluation: the budget office's 271 billion dollar saving booked as a transfer, with the poorest fifth of borrowers counted separately.

Evaluations are produced with AI support and reviewed on a schedule for new developments; human passes are marked separately.How we review →