, with the 2018 addendum: retrieval, extinction, and selection

Research date: 2026-09-22.

Citation and status. Daniela Schiller, Marie-H. Monfils, Candace M. Raio, David C. Johnson, Joseph E. LeDoux, and Elizabeth A. Phelps. Preventing the return of fear in humans using reconsolidation update mechanisms. Published in Nature 463, 49–53, issue 7 January 2010; online 9 December 2009. Original DOI. Read together with the same authors' Addendum: Preventing the return of fear in humans using reconsolidation update mechanisms, Nature 562, E21, published 26 July 2018, DOI, and its revised supporting material. The addendum changes the account of selection; it is not an independent experiment or replication.

Acquisition and reading. Read the complete six-page publisher-formatted original, including online Methods and references; all eleven pages of original supporting material; all fourteen pages of revised supporting material; and the complete substantive addendum paragraph available in publisher HTML. Visually inspected original Figures 1–3, supplementary Figures S1–S3 in both versions, and original/revised exclusion passages. The addendum PDF was not obtained: its PDF URL returned the retained HTML article. No raw participant data, OSF files, or analysis code were inspected. The separate Chalkia replication is assigned to another reader; this note's source audit was formed independently.

Local original PDF, original text, original supplement, original supplement text, addendum HTML, addendum text, revised supplement, revised supplement text, and provenance.

Source findings (approximately 150-word factual core). Experiment 1 analyzed 65 participants conditioned to colored squares. Next-day extinction followed an isolated reminder after ten minutes, after six hours, or without a reminder. Reported skin-conductance recovery occurred in the latter two groups; the ten-minute group's within-group change was nonsignificant. The omnibus interaction concerned early versus late re-extinction, a different contrast. A 10–14-month follow-up analyzed nineteen participants and pooled the two control groups; its recovery-index difference used a directional test. Experiment 2 analyzed eighteen participants, reminding one of two shock-associated cues before extinction. Separate tests reported recovery for the unreminded cue but not the reminded cue. The 2018 addendum substantially increased disclosed recruitment/exclusions and described retrospective qualitative selection, with exceptions to its reconstructed rules. The authors acknowledge that their behavioral experiment does not directly establish reconsolidation mediation. These laboratory autonomic-response results do not establish erasure of event knowledge or dependable modification of autobiographical memory. Original; addendum.

What was manipulated and measured. The original association was approximately one day old when treatment began. Experiment 1 used six reinforced plus ten unreinforced CS+ presentations and ten CS− presentations; reinforcement was approximately 38%. The reminder was one unreinforced CS+ presentation. Across reminder plus extinction, groups received eleven CS+ exposures, with their temporal grouping differing. The no-reminder group was split between immediate-after-break and six-hour extinction schedules. Day 3 assessed spontaneous recovery and then provided further extinction. Experiment 2 used two reinforced cues, each with five reinforced plus eight unreinforced acquisition presentations; one cue and the safety cue were reminded, followed ten minutes later by extinction of all cues. Its next-day test followed four unsignalled shocks and a ten-minute interval. These are specific exposure schedules, not a mapped universal six-hour updating window.

SCR amplitudes were measured 0.5–4.5 seconds after stimulus onset, thresholded at 0.02 microsiemens, square-root transformed, and divided by each participant's mean square-root-transformed shock response. Analyses used unreinforced cue trials. No reported event-recollection, contingency-expectancy questionnaire, subjective fear, or subsequent voluntary-action endpoint establishes that these other functions changed with SCR. Laboratory shock exposure here describes the published design, not a proposed session for this project.

Recruitment, selection, and follow-up — original versus clarified record.

The revised Experiment 1 arithmetic remains ambiguous: 126 minus 61 already equals 65, before its separately listed five technical cases. Preserve the printed numbers rather than silently changing the denominator to 131 or treating technical cases as an established subset. Follow-up analyzed 19/65, approximately 29% of the original analyzed sample; the ten-minute group retained 8/20 versus 11/45 across pooled controls. Returnees had received re-extinction on Day 3, so the one-year result follows additional intervention during the initial test. Four returnees were excluded for failed re-extinction or no measurable shock response at follow-up.

Original Experiment 1 criteria focused on second-half differential acquisition/extinction responses and a 0.1 threshold. The addendum instead acknowledges qualitative assessment during data collection and analysis. Its revised acquisition rule excludes only when all four summaries fall below an individually scaled threshold: first half, second half, final trial, and first-to-last increase. Its extinction rule combines second-half level, final-trial level, and insufficient first-to-last reduction, expressed using trials or halves. It also explicitly lists exceptions: two acquisition and one extinction inclusions despite failing criteria; three acquisition and six extinction exclusions despite not meeting every exclusion criterion. Experiment 2 is described as qualitatively selected without an equally explicit rule reconstruction. These are retrospective descriptions, not prospectively registered selection rules. Selection using post-treatment extinction can change the estimand and warrants sensitivity analysis; the documents alone do not quantify its effect on the reported difference.

Statistical audit — distinguish the actual contrasts.

• Next-day Experiment 1: the reported group-by-time interaction, F(2,62)=4.63, compares the first four re-extinction responses with the following four. The recovery index instead compares the final extinction trial with the first re-extinction trial. Reported within-group t values for that index are 2.69 without reminder, 2.66 after six hours, and 0.28 after ten minutes. A significant effect in one group and a nonsignificant effect in another is not itself a significant group difference. The report does not supply the corresponding direct between-group recovery-index estimate and interval. This does not negate its omnibus result; it restricts what that result tests.
• One-year follow-up: the report includes a direct index comparison, t=1.75, reported P<.05 in the context of one-tailed follow-up tests. The group-by-stage interaction is F(1,17)=2.78, reported P<.07 one-tailed; the group main effect is F(1,17)=5.89. It also gives a post-reinstatement group comparison, t=2.18, P<.03. Thus it would be inaccurate to say all long-term evidence consists only of separate nonsignificance tests. Its small selected sample, pooled controls, directional testing, and absent interval estimates limit the persistence claim. The standard F tail for 2.78 with 1,17 degrees of freedom is approximately .114; the manuscript's directional convention should be recorded explicitly.
• Experiment 2: the reported interaction, F(2,51)=5.14, concerns three stimuli and early/late re-extinction. Last-extinction to first-re-extinction tests separately give t=2.16 for the unreminded CS+, 0.22 for the reminded CS+, and 0.16 for CS−. No direct contrast between the two CS+ recovery indices is reported. Moreover, the denominator degrees of freedom require checking against the declared within-subject design; the paper does not provide enough implementation detail to resolve this from its text. No raw-data reanalysis is claimed here.

Mechanism and extraction — our interpretation. A reminder is an exposure within a learning history. Altered later responses can arise through changed associations, new inhibitory learning, cue grouping, context inference, or changes in retrieval/expression. The paper itself acknowledges both the absence of direct evidence for reconsolidation mediation and the unresolved distinction between timing and grouping of cue presentations. Reduced SCR cannot uniquely identify a rewritten trace, and an absence of statistically significant recovery is not proof that recovery is zero. Nor does this protocol establish that every retrieval destabilizes every memory.

For person modeling, acquisition therefore has two targets: what can be inferred about the person before an interview and how that interview may change later states. A faithfully stored transcript preserves an observation, not an assurance that the person's future response policy is unchanged. These results motivate testing this distinction; they do not supply a universal rule for predicting interview-induced change.

Falsifiable acquisition/transfer implication — proposal. Use controlled neutral episodes and independently recorded event details. Compare an additional retrieval request, matched re-exposure to existing information, and no additional elicitation; cross these with later corrective information. Record all cue, response, correction, and test times. Match information exposure and follow-up conditions where the contrast requires them, and treat response verification as part of the intervention history. Freeze scoring, exclusions, and the direct between-condition change contrast in advance; retain analyses of all assigned participants alongside clearly defined learning-criterion subsets. Different test schedules or separate samples can estimate effects introduced by the tests themselves.

Measure event/source accuracy, current reports and confidence, and consequential choices separately at new delays and contexts. Give recipient systems the same acquired observations, varying whether they also receive the elicitation/intervention history. Test forecasts of the source person's future responses and the recipient's own matched continuation actions. An improved forecast from history establishes its predictive value; attributing that gain to acquisition-induced transitions additionally needs the randomized acquisition-policy contrast and a baseline using the same evidence without a transition model. Score predicted response distributions separately from individual action agreement: irreducible variation in source choices can limit agreement even for an optimal predictor, while deterministic modal actions can outscore faithful independent sampling on that metric. Preserve event-optimal action scores as a separate target. This tests functional consequences of elicitation within the chosen task, without assuming the biological reconsolidation mechanism.