PaperlayerView exact revision on arXiv

Source-linked AI summary

Gender, Sexual Orientation, and the One-Year Refugee Bar: A Computational Study of Bill C-12 and Canadian Refugee Adjudication

Sean Rehaag

arXiv:2610.30297v1cs.CYcs.AI

TL;DR

This paper examines whether late refugee claims in Canada are concentrated among gender- and SOGIE-based persecution claims, and whether refusals of those claims are more likely to be overturned. It analyzes published refugee tribunal and court decisions using AI-assisted categorization checked through human review, finding that these claims are overrepresented among late claims and more often succeed on review.

  • Problem

    The study addresses the lack of published Canadian evidence on how common late claims are, who makes them, and how they fare.

    Source: §5 · p. 36

  • Method

    The study computationally analyzes published refugee decision records using AI categorization, with a human reviewer checking 125 decisions.

    Source: §3.1 · p. 19, §3.3 · p. 25

  • Results

    Gender-based and SOGIE claims are overrepresented among late claims, and refusals of those late claims are more likely than other late refusals to be overturned on appeal or judicial review.

    Source: Summary · p. 31, §5 · p. 36

  • Takeaways & Limitations

    The findings support the conclusion that late gender- and SOGIE-based claims are not anomalous and that their refusals have more often been set aside on review.

    Source: §5 · p. 36

    The comparisons lack statistical controls, and the study cannot assess how late claims fare in the unpublished PRRA process.

    Source: §3.5 · p. 32

Abstract

from arXiv · show

In March 2026, Bill C-12 amended the Immigration and Refugee Protection Act to make refugee claims ineligible for referral to the Refugee Protection Division where they are made more than one year after the claimant's first entry to Canada. Affected claimants are diverted to a Pre-Removal Risk Assessment (PRRA), which is typically decided by an immigration officer on a paper record, with no guaranteed oral hearing, no appeal on the merits, and no automatic stay of removal pending judicial review. The government defended the bar partly on the premise that genuine refugees claim promptly. Critics countered that late claims are concentrated among vulnerable groups, particularly people fleeing persecution based on gender, sexual orientation, gender identity, or gender expression (SOGIE). This article tests those competing premises. Using artificial intelligence tools, checked through human review, it examines nearly 29,000 published decisions of the Refugee Protection Division, the Refugee Appeal Division, and the Federal Court from 2006 to 2025. The results of this study demonstrate that, first, while most claimants claim within a year, roughly one in nine in the published record does not, and gender and SOGIE claims are significantly overrepresented among these late claims. Second, among late claims, refusals of gender and SOGIE claims are more likely than refusals of other claims to be overturned on appeal or judicial review. Late claims on other grounds, by contrast, fare somewhat worse on review than timely claims. The one-year bar therefore removes the hearing, the appeal, and the stay from the late claimants for whom those safeguards have been doing the most work. The article recommends repealing the bar or, failing that, adopting exemptions, procedural protections, and greater transparency in PRRA decision making.

1. Introduction

The one-year bar shifts late refugee claims from the full determination system to a PRRA with fewer safeguards. The study tests whether late claims disproportionately involve gender or SOGIE persecution and whether refusals of those claims are more often overturned.

1. Introduction

  • Under Bill C-12, claims made more than one year after first entry lose access to an oral hearing, a merits appeal, and the full system’s other safeguards.They are instead assessed through a PRRA, typically by an immigration officer on a paper record.Source: §1 · p. 2
  • The debate pits the government’s premise that genuine refugees claim promptly against advocates’ claim that late claims concentrate among people fleeing gender- and sexual-orientation-based persecution.Source: §1 · p. 2
  • The study tests whether gender-based and SOGIE persecution claims are overrepresented among late claims and whether their refusals are more often overturned on review.It also measures how common late claims are in the published record and how they fare at each stage.Source: §1 · p. 3
  • The article uses computational methods to extract timing and claim-ground information from RPD, RAD, and Federal Court decisions, then statistically analyzes the resulting dataset.Source: §1 · p. 3

2. Context

The context for the reforms is rising refugee-claim volumes alongside processing capacity that did not keep pace. Government attention also focused on particular claimant groups, including people entering through irregular U.S. border crossings.

2. Context

  • Annual claims rose from roughly 58,000 in 2019 to approximately 174,000 in 2024, after a brief pandemic-era dip.The figure reached approximately 94,000 in 2022 and 147,000 in 2023.Source: §2.1 · p. 4
  • By the end of 2025, pending claims had grown from roughly 71,000 in 2018 to approximately 300,000 as intake outstripped funded processing capacity.In 2024–25, the RPD’s funded finalization target of 85,000 was less than half its intake.Source: §2.1 · p. 4
  • Government attention to rising claims and backlogs also targeted claimants who transited through the United States, including those crossing irregularly.The Safe Third Country Agreement allows Canada to return most people arriving at a land border without adjudicating their claims, but originally did not cover irregular crossings.Source: §2.1 · p. 4

2.2 From Bill C-2 to Bill C-12

The one-year bar first appeared in the 2025 Strong Borders Act, Bill C-2, before immigration measures were separated and reintroduced as Bill C-12. Opposition to the original omnibus bill largely focused on its surveillance provisions rather than refugee measures.

2.2 From Bill C-2 to Bill C-12

  • The one-year bar first appeared in Bill C-2, introduced on June 3, 2025, as an omnibus bill covering border security, police powers, information sharing, money laundering, and immigration.Source: §2.2 · p. 8
  • Hundreds of civil society organizations opposed Bill C-2, with much of their criticism directed at surveillance and cross-border data-sharing provisions.Source: §2.2 · p. 8
  • In October 2025, the government separated the immigration measures from the omnibus bill and reintroduced them as Bill C-12.Source: §2.2 · p. 8

2.3 The one-year bar and the PRRA

The bar makes certain late claims ineligible for RPD referral, diverting them to a PRRA with more limited procedures and recourse. PRRA decisions are also unpublished, limiting public scrutiny and research.

2.3 The one-year bar and the PRRA

  • The bar applies to claimants who entered Canada after June 24, 2020, and applied more than one year after their first entry; leaving and returning does not reset the clock.Source: §2.3 · p. 9
  • The PRRA assesses the same substantive risks as refugee protection, but is typically decided by an immigration officer on a written record within tight timelines.An oral hearing is available only when prescribed factors are met; Bill C-12 did not extend a hearing requirement that applies to another category of PRRA-only claimants.Source: §2.3 · pp. 9–10
  • Unlike the RPD process, the PRRA does not provide a usual merits appeal, and judicial review is deferential with no automatic stay of removal.For many claimants, the PRRA will therefore be the final determination.Source: §2.3 · p. 10
  • PRRA decisions are unpublished, and obtaining granular, timely decision-making data through access-to-information requests is extremely challenging.The available high-level comparison reported PRRA acceptance at 33 percent versus a five-year IRB average of 63 percent, though the caseloads differ.Source: §2.3 · p. 11

2.4 The government’s case

The government argues the bar will ease backlogs, deter strategic claims, and preserve adequate assessment through PRRAs. Critics counter that delay can be legitimate and that an absolute bar replaces individualized assessment, disproportionately diverting vulnerable claimants.

2.4 The government’s case

  • The government defends the bar as a way to reduce asylum-system pressure, deter strategic claims, and direct clearly meritorious cases to paper-based PRRA decisions.Its arguments assume late claims are disproportionately strategic and that PRRAs can identify meritorious claims without an administrative appeal.Source: §2.4 · pp. 12–13

2.5 The critics’ case

  • The Convention defines refugee status through a well-founded fear of persecution, not the speed with which protection is sought.The Canadian Council for Refugees told a Senate committee that the UN Convention sets no timeline for asylum.Source: §2.5 · p. 13
  • Critics accept that delay can affect credibility and subjective fear, but object to converting that factor into an absolute bar.Case law establishes that delay can matter in adjudication; previously it was weighed with the claimant’s explanation.Source: §2.5 · pp. 13–14
  • Critics argue that the bar replaces case-by-case consideration of delay and its explanation with an absolute eligibility rule that admits no explanation.The CBA called the rule “arbitrary and without legal justification” and argued delay should not alone remove access to a Charter-mandated hearing.Source: §2.5 · p. 14
  • Critics say delayed disclosure can reflect trauma or lack of safety, and that a lower-procedure tier may disproportionately include gender-based and SOGIE claimants.The CBA, Amnesty International, and Canadian Civil Liberties Association warned that survivors may need years before feeling safe to disclose identities or experiences.Source: §2.5 · p. 14
  • A well-founded claim may also arise late because risk develops after arrival, including following political change, war, or activism in Canada.These are examples of sur place claims and other legitimate reasons identified by CARL.Source: §2.5 · p. 14

2.6 The American experience with the one-year bar

The US one-year deadline offers exceptions and still preserves adjudication, yet research found substantial denials and unequal burdens. The Canadian debate lacks direct empirical evidence, and the study identifies patterns that could show whose claims and safeguards are affected.

2.6 The American experience with the one-year bar

  • Unlike the Canadian bar, the US rule has exceptions and late applicants still receive an immigration-judge hearing on withholding of removal and Convention Against Torture protection.The US clock runs from most recent arrival; Canada’s runs from first entry and contains no exceptions.Source: §2.6 · pp. 17–18
  • The US one-year deadline led to rejection of 54,141 late applicants, about one in six adjudicated applicants, and researchers estimated roughly 16,000 additional asylum grants might otherwise have occurred.Those estimated grants would have covered more than 21,000 refugees including family members.Source: §2.6 · p. 17
  • Women in the US dataset filed late at a rate of 32.7 percent versus 29 percent for men, and filed four or more years after entry at a rate over 50 percent higher.Other researchers also found disproportionate effects on women facing gender-based violence and people facing persecution based on sexual orientation or gender identity.Source: §2.6 · pp. 17–18

2.7 The missing Canadian evidence

  • No published Canadian study had measured how common late claims are, who makes them, or how they fare once decided.Existing guidelines, qualitative scholarship, and US evidence suggested overrepresentation of gender-based and SOGIE claims, but were indirect.Source: §2.7 · p. 18
  • The study predicts that critics’ account would be reflected in overrepresentation of gender and SOGIE claims among late claims and more frequent reversals of their refusals.These patterns matter because the bar removes the oral hearing, appeal, and automatic stay from claimants diverted to PRRA.Source: §2.7 · p. 18

3. The Study

The study examines late claims, their grounds, and outcomes across Canadian refugee adjudication. It categorizes published decisions using computational methods and human review, while cautioning that selective publication limits what absolute rates can show.

3. The Study

  • The study asks whether late claims more often involve gender-based or SOGIE persecution and how those claims fare, while also describing late-claim frequency and outcomes at each adjudication level.The levels are first-instance RPD decisions, RAD appeals, and Federal Court judicial review.Source: §3 · p. 19

3.1 The data

  • The study categorized 28,852 published decisions from 2006 to 2025 drawn from RPD, RAD, and Federal Court records.The corpus was assembled from four sources, including positive RPD reasons obtained through access-to-information requests; the study also categorized 32,462 decisions overall.Source: §3.1 · pp. 19–20
  • Because each source publishes selectively, absolute rates in the corpus say little about the refugee system as a whole, while within-source and within-level group comparisons are firmer.Published RPD reasons lean negative, the RLLR source is exclusively positive, and RAD and Federal Court decisions mostly concern claimants contesting earlier losses.Source: §3.1 · p. 20

3.2 What was categorized

Decisions were categorized by timing, outcome, and alleged persecution grounds, with an additional Federal Court screening step. Timing follows the statutory first-entry test and includes an unclear category when the published text does not establish when the claim was made.

3.2 What was categorized

  • Federal Court cases were screened to retain decisions reviewing RPD or RAD refugee determinations and exclude unrelated immigration matters, PRRAs, and other listed proceedings.Decisions outside 2006–2025 or without English text were excluded.Source: §3.2 · pp. 20–21
  • A claim was categorized late only when the text or a reasonable inference showed it was made more than a year after the claimant’s first-ever Canadian entry.Leaving and returning does not reset the clock; unclear was used when the published record did not establish timing.Source: §3.2 · p. 21
  • Outcome was measured from the claimant’s perspective as protection granted, an appeal allowed, or judicial review granted and remitted, with non-dispositive procedural rulings unclear.At RAD, a Minister’s appeal dismissal counted as success; at Federal Court, dismissal of a Minister’s application counted as success.Source: §3.2 · p. 21
  • Grounds categories captured alleged gender-based or SOGIE persecution, both, neither, or unclear, rather than what the tribunal found credible.Gender-based grounds included domestic violence, forced marriage, sexual violence, and reproductive coercion; SOGIE covered sexual orientation, gender identity, and gender expression.Source: §3.2 · p. 21

3.3 Method: machine categorization and verification

The study used a three-layer AI categorization workflow, checking model outputs against a stronger-model benchmark and human review. Benchmark agreement was high on grounds but lower on timing, and human review found most categorizations defensible.

3.3 Method: machine categorization and verification

  • The three-layer workflow used a 600-decision benchmark, smaller-model corpus categorization, and human review.The benchmark tested whether a smaller model could reproduce a stronger model's answers; human review checked a sample of categorizations.Source: §3.3 · p. 22
  • 95 percent of the smaller model's late classifications were confirmed by the benchmark, indicating late claims were undercounted rather than inflated.Most timing disagreements (41 of 52) were borderline timely-versus-unclear calls.Source: §3.3 · p. 23
  • 97 percent agreement on whether any gender or SOGIE allegation was present, while timing was hardest at 91 percent agreement.The smaller model agreed with the benchmark on the gender-or-SOGIE question in 97 percent of cases; Table 2 summarizes model agreement on 600 decisions.Source: Table 2 · p. 23, §3.3 · p. 23
  • 292 of 293 categorizations in the random human review were judged defensible under a reasonableness, not correctness, standard.The author read 100 decisions in full; every screening, timing, and outcome answer reviewed passed, while one basis answer failed.Source: §3.3 · p. 24
  • 32 of 33 late-claim categorizations across random and targeted samples were judged reasonable.The targeted review examined 25 additional decisions categorized as late; 24 were upheld and one was judged unclear rather than late.Source: §3.3 · pp. 24–25
  • Manual checking suggested the original text filter missed roughly 5 to 7 percent of Federal Court refugee judicial reviews; widening it captured 397 refugee matters among 1,995 additional decisions.The newly captured decisions were categorized with the same model and instructions, and screening filtered out non-refugee matters.Source: §3.3 · p. 25
  • The checks do not establish human-gold-standard accuracy: one reviewer knew expected findings, and model-led review flagged a benchmark answer inconsistent with instructions.The benchmark was produced by an AI model, and the human review used a reasonableness rather than correctness standard.Source: §3.3 · pp. 23, 25

3.4 Findings

The published record shows that late claims are a substantial minority, disproportionately involve gender/SOGIE allegations, and have distinct review outcomes. Late gender/SOGIE claims succeed more often on review than other late claims, while late claims on other grounds fare worse than timely ones.

How many claims are late

  • 11.7% of decisions with clear timing were late, while 88.3% were timely.The authors say the true late-claim share is likely higher because earlier visits that could trigger the bar may not be recorded precisely in the published decisions.Source: How many claims are late · p. 26

Who claims late

  • 41.5% of late claims involved gender/SOGIE allegations, versus 31.1% of timely claims (odds ratio 1.58).The association recurred in all four sources; SOGIE claims were overrepresented in each, while gender-based results varied by source.Source: Who claims late · pp. 26–27, Table 4 · p. 27
  • 28% versus 19% at the RAD and 38% versus 27% at the Federal Court: late gender/SOGIE claims more often succeeded than other late claims.Late claims on other grounds fared worse than timely claims; these outcomes describe published appeals and judicial reviews brought and decided, and do not establish why the gap appears.Source: Table 5 · p. 29, Do late claims succeed, and are errors… · pp. 30–31
  • The Board granted protection in 98 late claims, including 57 involving gender/SOGIE allegations, but first-instance sources do not support comparative success-rate estimates.Published RPD decisions skew negative, while RLLR decisions are positive by construction; the late-claim grants in the RPD sample are too few for comparison.Source: Do late claims succeed, and are errors… · pp. 29–30

Summary

  • Both research questions are answered yes: gender/SOGIE claims are overrepresented among late claims and succeed more often on review than other late claims.Late claims on other grounds fare somewhat worse on review than timely claims; the published record is a fraction of all claims and first-instance decisions skew overwhelmingly negative.Source: Summary · p. 31

3.5 Limitations

The study’s findings are bounded by publication and measurement limits, historical scope, coarse claim categories, and simple comparisons that cannot evaluate PRRA outcomes.

3.5 Limitations

  • The corpus contains published decisions rather than a random sample, and selection into publication differs by source.Within-source comparisons avoid differences between sources but do not address selection within them; the source composition may affect observed associations.Source: §3.5 · pp. 31–32
  • Roughly a quarter of decisions have unclear timing, and the measure approximates whether claims were made within a year.Unclear timing largely reflects redacted dates or omitted entry dates; the undercount may differ across claim types because delay may be discussed more often in gender and SOGIE decisions.Source: §3.5 · p. 32
  • The 2006–2025 corpus predates the bar’s application to people entering after June 24, 2020, and claiming on or after June 3, 2025.The findings describe the population adjudicated historically, not precisely the population the bar will divert.Source: §3.5 · p. 32
  • The categories pool distinct claim types, and the study records allegations and timing rather than why a claim was late.Intersecting claims are 1.1 percent of the dataset and cannot be examined separately; the study establishes a pattern, not its proposed mechanisms.Source: §3.5 · p. 32
  • The comparisons are uncontrolled, and unpublished PRRA decisions prevent comparing how the same claims fare on paper and at a hearing.The analysis does not control for country of origin, year, representation, or decision-maker; more elaborate analysis might sharpen or qualify its estimates.Source: §3.5 · p. 32

4. Implications and Recommendations

The published record shows that late claims include well-founded gender- and SOGIE-based claims and that refusals in these categories are often corrected on review. The article recommends repealing the bar or, if it remains, adding exemptions, procedural safeguards, and transparency in PRRA decisions.

4. Implications and Recommendations

  • Among late claims, gender- and SOGIE-based claims are more likely than other late claims to succeed on appeal or judicial review.The study’s comparisons indicate that late claims on other grounds fare somewhat worse on review than timely claims.Source: §4 · p. 33, §4.1 · p. 34

4.1 The government’s case, revisited

  • About one in nine claims with clear timing was late, and 98 late claims in the dataset were found well-founded by the Board.Roughly 88 percent of claims with clear timing were timely, but late claims were tilted toward gender- and sexual-orientation-based claims.Source: §4.1 · p. 33
  • The published record includes 282 late-claim appeals decided for claimants by the RAD and 187 late-claim judicial reviews granted by the Federal Court.Most set aside refusals, though the data do not separate the remainder, which involved dismissal of a Minister’s challenge to a protection grant.Source: §4.1 · p. 33
  • The study cannot identify claimants’ motives, but late gender and SOGIE claims succeeded about as often as timely claims and more often than other late claims.Late claims on other grounds fared somewhat worse on review than timely claims.Source: §4.1 · p. 34
  • The authors argue that diverting these claims removes safeguards that the record shows are correcting errors, including a hearing by default, an appeal, and a stay.The study finds late claims are disproportionately based on gender-based and SOGIE-based persecution, for which refusals are reversed at higher rates than for other late claims.Source: §4.1 · p. 34

4.2 Recommendations

  • The authors prefer repealing the bar so claimants retain access to the regular refugee determination system.They propose that the government may raise program-integrity concerns in individual cases and seek a manifestly-unfounded designation, which retains an oral hearing and a chance to explain delay.Source: §4.2 · p. 35
  • If the bar remains, the authors recommend exempting gender and SOGIE claims and requiring hearings, administrative appeals, and stays for those still subject to it.These protections would apply to claimants subject to the bar, with exemptions for claims based on gender, sexual orientation, gender identity, or gender expression.Source: §4.2 · p. 35
  • If PRRA becomes many claimants’ first and only chance to demonstrate risk, the authors call for more published decisions, accessible data, and research partnerships.They recommend transparency to help assess whether PRRA decision-making replicates problems found in regular refugee adjudication.Source: §4.2 · p. 35

5. Conclusion

The conclusion argues that Bill C-12’s one-year bar will affect claimants whose gender- and SOGIE-based claims were often late and whose refusals were more likely to be set aside. It calls for the bar to be corrected by government or the courts.

5. Conclusion

  • Claims like the Nigerian student’s were late more often than other claims, were found well-founded in dozens of Board decisions, and had refusals set aside more often than other late claims.These patterns come from twenty years of published decisions.Source: §5 · p. 36
  • The conclusion says the bar removes safeguards from late claimants whose refusals have most often been set aside on appeal and judicial review.The author connects this concern to repeated adjudication errors and guidelines cautioning against negative delay inferences in gender- and sexual-orientation-based claims.Source: §5 · p. 36
  • The author concludes that the one-year bar was a mistake and should be corrected by the government or, if necessary, the courts..Source: §5 · p. 36
Loading 2610.30297v1…