KurisutinaAnastasia Chalkia, Natalie Schroyens, Lu Leng, Niels Vanhasbroeck, Ann-Kathrin Zenses, Lukas Van Oudenhove, and Tom Beckers. No persistent attenuation of fear memories in humans: A registered replication of the reactivation-extinction effect. Cortex 129, 496–509. Peer-reviewed primary human experiment, published online 18 June 2020. DOI: 10.1016/j.cortex.2020.04.017. Author-hosted final PDF; PMC accepted manuscript.
Evidence from this source
Of 246 recruits, 124 were retained, 62 per randomized group. Exclusions comprised 82 acquisition failures, 15 extinction failures, two medical exclusions, and 23 other losses. Day 1 paired colored squares with shocks; Day 2 compared extinction preceded by one unreinforced reminder ten minutes earlier against extinction without that reminder, equating total cue presentations. The primary endpoint was the change in differential skin conductance from final extinction to initial Day-3 re-extinction. Both groups recovered. Recovery means were .21 (SD .34) with reactivation and .16 (.28) without: t(122)=.93, p=.35, d=.17, numerically opposing benefit. Same-day reinstatement also showed no advantage; alternative signal processing agreed. This tests conditioned response recovery, not a direct readout of a stored trace. There was no six-hour reminder arm or one-year follow-up; reinstatement was supplementary rather than an exact replication of the original follow-up. Methods, Results, Figures 1–3.
Registration and comparison audit
The retained Stage 1 protocol states initial journal acceptance on 24 February 2017 and an editor-agreed exclusion amendment on 21 September 2017. The OSF registration was filed on 22 April 2018. Its stored hash exactly matches the downloaded protocol. Thus, the public timestamp alone is not evidence that registration preceded all collection; the paper explicitly describes collection before the amendment. The original, pre-amendment manuscript was not retrieved.
The amended protocol specifies six alternative acquisition inclusion routes, five extinction routes, replacement until 62 per arm, two processing pipelines, and the direct between-group recovery-index test. It powers that test at 90% for g=.53 with one-sided alpha .05; it attributes g=.732 to the original study. These are planning assumptions, not a replication confidence interval. Protocol, pp. 8–16. The 2017 rules should not be called identical to the later Schiller addendum. The reported final p=.35 is two-sided; choosing the planned benefit direction would not rescue a result pointing oppositely.
The comparison that matters is the difference between recovery indices. Significant recovery in one arm and nonsignificant recovery in another would not establish that difference. Conversely, observed recovery in both arms would still permit a modest differential benefit. The explicit between-group test avoids those errors.
Original statistical interpretation and limits
Precision calculation, not a reported paper result. Using the rounded t=.93 and equal group sizes gives standardized reactivation-minus-control recovery d=t√(2/62)=.167. The conventional large-sample standard error √(2/62+d²/(2×122)) is .180. A normal-approximation 95% interval is therefore approximately [−.186,.520], or [−.520,.186] when positive means benefit. This calculation assumes the usual independent-group standardized-mean model, uses rounded published inputs, and is neither an exact interval nor a raw-data reanalysis. It places the targeted positive effects around .53/.73 outside the approximate interval for this selected sample and procedure. It does not establish exact equality, exclude small benefits, or adjudicate every proposed boundary condition.
Selection requires care. Acquisition eligibility concerns behavior before the Day-2 manipulation; extinction eligibility conditions on a response after randomization. The latter can disturb randomized comparability if treatment affects eligibility. Prespecifying a selection rule prevents discretionary application but does not remove this causal issue. Replacing exclusions until balanced retained groups is not an intention-to-treat analysis. Without inspecting group-specific losses and all randomized outcomes, a population-average causal estimate cannot be certified here. Selection limits both positive and negative generalizations; it does not itself explain away the retained-sample result.
Algebraic caveat to a discussion argument. The paper suggests that the US-normalized inclusion threshold preferentially admits high-US responders. Under its stated processing, however, both differential CS response and threshold share the same positive participant denominator U. If A is the relevant difference of square-rooted CS responses, A/U > .1/U is equivalent to A > .1; U also cancels for linear averages and changes. The denominator alone therefore cannot supply that selection mechanism under those assumptions. This is an audit of the written equations, not a verified implementation finding. Behavioral selection and possible associations with US responsiveness remain separate issues.
Response attenuation does not uniquely identify deletion, overwrite, inhibitory learning, or context-dependent retrieval. Likewise, failure of this behavioral intervention does not refute all biological reconsolidation mechanisms. This experiment supplies no independent manipulation check that identifies destabilization as the operative latent state, and skin conductance is not equivalent to conscious fear or autobiographical content. Unknown moderators are possible explanations to test prospectively, not established explanations for this result.
Original functional-transfer implication
Do not treat an elicited memory as automatically overwritten whenever later information changes the current answer. Compare an overwrite model with models retaining earlier associations, later evidence, and retrieval context. In a benign learned-cue task, test delayed responses, changed contexts, and renewed evidence using equal observations and acquisition costs. Prespecify the between-policy contrast and retain all assigned cases; report any eligibility-restricted estimand separately.
For extraction, distinguish effects of the elicitation/update procedure on the source from effects of releasing its acquired information to a recipient. Pair recipient release comparisons on the same acquisition history, then separately evaluate their committed actions in new contexts. A recipient's report that it has “updated” is insufficient. Successful functional prediction would still not demonstrate biological trace replacement or transferred subjective experience. No sessions or interventions were run.
Reading and acquisition record
Read the complete main introduction, methods, results, discussion, and all eight footnotes in the retained PMC text, cross-checked against the 14-page publisher-formatted PDF. Visually checked Figures 1–3; no graph digitization. Read the complete Stage 1 substantive text, pp. 1–19 of its 26-page PDF; pp. 20–26 are bibliography, retained but not independently followed. Read the OSF README, file/version metadata, and registration record. The registered protocol's SHA-256 agrees with the OSF registration snapshot. No raw participant records, lab log, task code, or analysis datasets were retrieved or reanalyzed.
Supplementary Figures S1–S2, including the detailed collection/amendment timeline and alternative-processing plot, were identified but not retrieved or read. PMC returned a download challenge rather than the DOCX; Europe PMC and publisher/mirror attempts failed. No challenge or error file was retained as a source. Consequently, the full supplement timeline and exhaustive protocol-compliance audit remain incomplete. The Stage 1 age range is 18–48; the final recruitment range is 17–45. This observed difference is recorded without inferring its effect. The methods comparison does not certify implementation fidelity or establish when particular analyses were first inspected.
Local sources: final PDF, PDF text, PMC HTML, Stage 1 PDF, Stage 1 text, registration metadata, provenance and hashes. Accessed 22 September 2026. The final PDF states ©2020 Elsevier, all rights reserved; public availability is not an open reuse license.