Goldfish remember what they learn for days or longer.

1. Pass Title matches the evidence

   Title 'Goldfish remember what they learn for days or longer.' matches finding False and the multi-day and month-scale retention evidence.

2. Pass No obvious contradictions

   Finding False, answer and conclusion agree; the article consistently avoids asserting a single maximum duration.

3. Pass Plausible justification

   Conclusion gives a sensible reason: conditioned responses and learning are retained across days and beyond, which contradicts a universal three-second limit.

4. Pass Sources are relevant

   PNAS memory-fixation study, Overmier conditioning study, Brown fish cognition review and Givon navigation study are on topic, judged by name, title and description only.

5. Pass Common questions answered

   Three common questions (bowl swimming, reflex versus memory, 11-month report) each get a responsive answer.

Coverage: Read the full article payload; sources judged by name/title/description/role only, per First Look's rules; no source snapshots read; no web access.

1. Concern (minor) Sourcing is strong and transparent

   Cite Overmier and Curnow (1969) by journal and DOI instead of a ResearchGate page that blocks access, since the summary's month-scale claim rests on it.

   Sourcing is mostly primary and peer-reviewed: PNAS 1965 (PMC full text), Behavioural Brain Research 2022 (DOI 10.1016/j.bbr.2021.113711), Animal Cognition 2015 review (DOI 10.1007/s10071-014-0761-0). However, the month-scale retention claim, which appears in the summary ('month-scale intervals') and a key point, rests solely on source:goldfish-source-overmier, cited as a ResearchGate profile URL with no journal name or DOI. The snapshot returned HTTP 403 with no archive, so readers and auditors cannot verify it; the publisher's record of the paper should be readily available. The article candidly flags this in its editorialNote, which keeps this Minor: the central finding does not depend on it. Confidence: High.

   About source:goldfish-source-overmier, evidence:goldfish-classical-retention The finding

2. Pass Central claims are accurate

   Summary of article: goldfish memory is not capped at three seconds; controlled experiments show retention over days or longer, so the claim is False (High). Claim ratings: 1. Central finding that a universal 3-second limit is false - True. Givon et al. abstract: goldfish were 'trained' to drive a vehicle and 'reach the target regardless of the starting point'; snippets compare 'first and last sessions' and 'last days of training', confirming multi-day retention. Brown's abstract: 'fish perception and cognitive abilities often match or exceed other vertebrates.' 2. e1 Agranoff 1965 avoidance learning retained across days - Unverifiable from the snapshot (only bibliographic page: 'Memory fixation in the goldfish', PNAS 1965;54(3):788-793), but consistent with the well-known design of this classic study; title, authors, journal, year all match. 3. e2 Overmier and Curnow month-scale retention - Unverifiable (source unreadable, 403). The article itself flags this in editorialNote. 4. e3 Brown review places fish learning in context - Mostly True; abstract supports broad cognitive sophistication, though it does not mention the three-second claim explicitly. 5. e4 Givon six goldfish, Ben-Gurion, multi-day navigation - True ('A total of six fish', 'Ben-Gurion University of the Negev'). The central finding is supported by readable sources and by an uncontested scientific consensus; unverifiable items do not change it. Confidence: Medium-High (two evidence items could not be read in full).

3. Pass Claims are supported and labeled

   Each evidence item cites a source; the conclusion ('all require retained information beyond seconds') follows from the evidence. Inferences are labeled as such, e.g. e3 'provides an independent scholarly context', and the article explicitly declines to assert the 11-month figure ('not asserted here'). It separates what is established from what is not ('does not justify replacing the myth with a single universal lifespan'). Confidence: High.

4. Pass Context is complete

   The article states memory duration 'depends on the task, training, memory system, and test conditions', addresses the anecdotal bowl-swimming argument, the 'conditioning is only a reflex' objection, and the 11-month report. No contrary evidence supporting a 3-second cap is known to exist, so nothing omitted would change the impression. The article could mention the origin of the myth, but that is not needed for the reader's understanding. Confidence: High.

5. Pass Framing is fair and neutral

   The claim is stated as it circulates ('A goldfish can remember information for only three seconds.'), objections are answered respectfully with evidence, and the scope is explicitly limited ('does not assign a single maximum duration'). No loaded language or advocacy. Confidence: High.

6. Pass Uncertainty is stated

   Each evidence item lists a specific limitation (aversive lab task, small sample of six, review not a new experiment, access restriction). The conclusion says duration 'depends on what is learned and how it is tested', and the editorialNote notes no external specialist review and that Overmier's claims 'should be checked against the full paper'. High confidence is appropriate given converging methods across decades and an uncontested consensus. Confidence: High.

Coverage: Read all four cited sources via get_source_snapshot. Agranoff et al. (PMC219745): live PMC page read, but it exposes only bibliographic data, page scans and references, not the methods/results text, so the 'retained across days' detail was not verifiable from the snapshot. Overmier and Curnow (ResearchGate): unreadable (HTTP 403, no Wayback capture), so the month-scale retention claim could not be checked against the source. Brown 2015 (Springer): abstract and reference list read; full text paywalled. Givon et al. 2022 (ScienceDirect, Wayback capture): abstract and section snippets read. Own knowledge used only to judge plausibility and raise questions.

1. Concern (minor) Confidence matches the evidence

   Qualify the title, e.g. 'Goldfish can remember what they learn for days or longer', to match the conclusion that duration depends on the task.

   High confidence is justified: three primary experiments using different methods (pharmacological memory disruption, classical conditioning, operant spatial learning) across 1965-2022 and three institutions, each by itself enough to falsify a universal 3-second cap, and limitations are stated per item. I would also give High. However, the title 'Goldfish remember what they learn for days or longer.' reads as a general rule about everything goldfish learn, while the conclusion says the evidence 'does not justify replacing the myth with a single universal lifespan' and 'duration depends on what is learned and how it is tested', and key point 5 says it 'does not establish one universal maximum duration'. The 'or longer' rests mainly on e2, whose access the article itself flags. A reader's belief (goldfish memory far exceeds 3 seconds) would not change, so this is Minor.

   About conclusion The finding

2. Pass Rates the real claim

   Claim: 'A goldfish can remember information for only three seconds.' This is how the myth circulates; a reader repeating it means goldfish forget everything almost immediately. The article rates exactly that ('the claim assessed is the alleged universal three-second maximum') and the conclusion stays in scope, explicitly refusing to substitute a new universal duration. No strawman or overreach.

3. Pass Rating follows from the evidence

   Taking the evidence as stated: avoidance learning retained across days (e1), conditioned responses to a month-scale test (e2), a review (e3), and multi-day vehicle navigation learning (e4). Any one controlled retention beyond three seconds falsifies a universal three-second cap, so I would choose False before reading the finding. The article's finding is False and the answer ('keep learned information for days, far beyond three seconds') agrees.

4. Pass Strongest objections answered

   Strongest objections I would raise: (1) everyday pet behavior looks like forgetting - answered in objection 1 ('Repetitive swimming is not a timed memory test'); (2) conditioning is reflexive, not 'real' memory - answered in objection 2, conceding it tests 'a particular kind of memory' while still contradicting a universal cap; (3) trained lab fish differ from ordinary goldfish - addressed in limitations ('does not define ... everyday behavior in untrained fish'). Answers rest on evidence and scope, not dismissal. The 11-month report is handled conservatively.

5. Pass Evidence is independent

   e1 (University of Michigan, 1965, PNAS, pharmacological disruption), e2 (University of Minnesota, 1969, classical conditioning), and e4 (Ben-Gurion University, 2022, spatial learning) cite different sources, institutions, decades and methods. e3 is a review that may summarize literature overlapping with the primary studies, so it adds context rather than fully independent data, but three independent primary lines remain, each sufficient on its own for the conclusion.

Coverage: Evaluated the article payload only (claim, finding, answer, confidence, title, summary, keyPoints, context, conclusion, four evidence items with limitations, three objections, five method steps). Per the Skeptic's division of labor, source snapshots were not used and source accuracy was not assessed.

1. Concern (minor) Verdict fields agree and stand alone

   Replace 'is contradicted with high confidence' in the conclusion with wording in the current label's terms, e.g. 'is false, with high confidence'.

   Verdict reads: FactHub rates the claim 'A goldfish can remember information for only three seconds.' False (High confidence): 'Experiments show goldfish keep learned information for days, far beyond three seconds.' That stands alone and the claim is worded as it circulates. But the changelog records the label changed from 'Contradicted' to False with 'the evidence and conclusion ... unchanged', and the conclusion still opens 'The three-second-memory claim is contradicted with high confidence.' The rule requires all verdict fields to express the conclusion 'in the current label's terms'; stale wording from an earlier label is Minor.

   About conclusion The finding

2. Concern (minor) Each section does its job

   Merge the summary into one sentence naming the strongest evidence lines; move the 'without implying one fixed duration' caveat to context or conclusion.

   Rule: 'summary is one sentence naming the strongest independent lines of evidence.' The summary is two sentences, the second ('These experiments directly contradict ... without implying one fixed duration for every kind of memory') carrying a caveat that the conclusion and keyPoints already state. Other fields comply: context is one short framing paragraph; conclusion weighs the lines of evidence and states what remains unproven; editorialNote states scope, the Overmier access caveat, and 'No external specialist review has been recorded.' keyPoints mostly state what evidence establishes.

   About conclusion The finding

3. Concern (minor) Readers can verify it

   Give the Overmier and Curnow source its journal, year and DOI or publisher link instead of only a ResearchGate profile URL.

   Rule: sources need full titles, DOIs or stable identifiers, and key claims need primary anchors. Agranoff (PMC permalink), Brown (DOI 10.1007/s10071-014-0761-0) and Givon (DOI 10.1016/j.bbr.2021.113711) comply, with role labels and descriptions. The Overmier and Curnow source is named only 'Overmier and Curnow' with no journal, year or DOI, anchored to a ResearchGate profile page the article says 'may restrict access'; it supports the month-scale claim in the summary and keyPoints. The title keeps it identifiable, so Minor. Method steps cover claim definition (1), sources reviewed (2), independent evidence (3) and recorded limits (4).

   About source:goldfish-source-overmier, evidence:goldfish-classical-retention The finding

4. Concern (minor) Consistent house style

   Explain 'puromycin' (a protein-synthesis inhibitor) and 'pseudoconditioning' and 'sensitization' at first use in the evidence summaries.

   Rule: technical terms a general reader would not know are explained at first use. e1's summary says 'tested how the timing of puromycin affected later performance' without saying what puromycin is or does, and e2's summary uses 'pseudoconditioning, and sensitization' unexplained. Otherwise consistent: American spelling, names consistent, no drafting text, figures consistent.

   About evidence:goldfish-memory-fixation, evidence:goldfish-classical-retention The finding

5. Pass Title states the finding

   Title 'Goldfish remember what they learn for days or longer.' states the finding in positive form, names the subject, resolves the claim, and does not repeat the myth (rule: lead with what is true). seoTitle 'Do Goldfish Have a Three-Second Memory? No' uses search wording and states the finding ('No'), which the seoTitle rule permits.

6. Pass Limitations are specific

   Each limitation is specific: e1 'aversive laboratory task and a pharmacological intervention'; e2 access plus 'specialized preparation and conditioned response should not be generalized'; e3 'not a new retention experiment focused only on goldfish'; e4 'sample was small and the task was highly engineered'. None is boilerplate or empty.

7. Pass Tone is measured and respectful

   Language is measured ('contradicted with high confidence', 'does not justify replacing the myth with a single universal lifespan'). Objection responses are short, evidence-based and respectful, e.g. 'Repetitive swimming is not a timed memory test.' No ridicule or unjustified absolutes.

Coverage: Used only the article payload: title, seoTitle, description, claim, finding, answer, confidence, summary, keyPoints, context, conclusion, evidence items and limitations, objections, method steps, editorialNote, sources, image caption/alt text. No web or source snapshots used. Out-of-scope items (title length, question titles, schema rules, generated independence note) not assessed.