Working Memory
Reference-Back Task
Also called: reference-back paradigm, reference back task, working memory gating task, gate opening and gate closing task
Measure how working memory lets new information in: letters in red or blue frames are compared with the last red-framed letter, which separates the costs of opening and closing the gate to working memory and of replacing what it holds.
9-11 minTimed: desktop or laptop with a mouse or trackpad
Free to start: 10 free participants per account, then $0.99 per participant for academic research, or any yearly plan. Try it here first, with no account.
- Papers, 2021 to 2025
- 19
- Evidence grade
- D
- Registered clinical studies
- 0
What it measures
Measures how working memory lets new information in and keeps the rest out. Participants see X or O in a red or blue frame and decide whether each letter matches the one in the last red frame; only a red frame changes the letter to hold. The main score is the gate-opening cost.
Constructs
- Working memory updating
- Input gating of working memory
- Cognitive control
Research uses
- Working memory and cognitive control research
- Research on dopamine and working memory gating
- Within-study group comparison
How the paradigm works
The Reference-Back Task measures how working memory lets new information in and keeps the rest out. Each trial shows an X or an O in a frame, and on every trial the participant decides as quickly and accurately as possible whether the letter is the same as the letter in the last red frame, pressing F or J. A red frame with a solid edge marks a reference trial: after the answer, its letter is the one to hold. A blue frame with a dashed edge marks a comparison trial: the participant answers and keeps the letter held. The task comes from Rac-Lubashevsky and Kessler (2016), who built it from the n-back task to take its parts apart, and it reads the input gate of working memory that the model of Chatham and colleagues (2011) places in the basal ganglia.
More about the paradigm
Each block begins with a letter in a red frame that needs no answer, shown for 2 seconds. Then come 64 trials in which each of the eight kinds of trial (reference or comparison, after a trial of the same or the other type, with the same or a different letter) comes 8 times, in an order drawn from the session's seed before the session starts, with no more than 5 trials in a row of one type or one answer. A blank of 500 ms and a cross of 500 ms come before each letter, which stays until a key or 3 seconds. The session starts with 16 practice trials with feedback, and a second set only when the first had under 75% right answers; the test has 4 blocks (2 to 6 as a setting), a pause between blocks and no feedback. The edge pattern repeats the frame's color, so the cue does not rest on color alone. Which key means same is drawn from the session's ID.
The main score, the gate-opening cost, is the extra time of a reference trial that follows a comparison trial over one that follows a reference trial: the cost of opening the gate to working memory. The gate-closing cost is the same for comparison trials after a reference trial: the cost of closing it again. The substitution cost is the extra time of a different letter on reference trials, beyond the same difference on comparison trials: the cost of replacing the letter held. The updating cost compares reference with comparison trials after a trial of the same type, and the matching cost compares different with same letters. Each cost comes from the median times of right answers in each kind of trial, leaving out answers under 100 ms and the trial after an error. The costs are differences of tens of milliseconds, so one person's costs are research measures best compared between groups. Scores support research and do not diagnose.
Scores
- Gate-Opening Costms
- Median time of right answers on reference trials after a comparison trial minus after a reference trial, averaged over same and different letters.
- Positive values mean a reference trial took longer after a comparison trial; not better or worse
- Gate-Closing Costms
- Median time of right answers on comparison trials after a reference trial minus after a comparison trial, averaged over same and different letters.
- Positive values mean a comparison trial took longer after a reference trial; not better or worse
- Substitution Costms
- On trials after the same type, different minus same letters on reference trials, less the same difference on comparison trials.
- Positive values mean replacing the letter held took extra time; not better or worse
- Updating Costms
- On trials after the same type, reference minus comparison trials, averaged over same and different letters.
- Positive values mean reference trials took longer than comparison trials
- Matching Costms
- Different minus same letters, averaged over the eight kinds of trial.
- Positive values mean a different letter took longer to answer than the same letter
- Accuracy%
- Right answers as a share of right and wrong ones, over all test trials.
- Below 0.80 is flagged for review
- Accuracy, Reference Trials%
- Right answers as a share of right and wrong ones on reference trials.
- Higher means fewer errors
- Accuracy, Comparison Trials%
- Right answers as a share of right and wrong ones on comparison trials.
- Higher means fewer errors
- Fewest Usable Trialscount
- Usable response times in the kind of trial with the fewest.
- Under 10 is flagged for review
- Missed Letterscount
- Test letters not answered within 3 seconds.
- Expected to be 0
- Anticipationscount
- Answers under 100 ms after the letter appeared, counted neither right nor wrong.
- Expected to be 0
Use and evidence
19 papers from 2021 to 2025 name this task in their title or abstract, up 46% compared with 2016 to 2020. That ranks it 405 of 495 test families in our census. See the counts for every test.
Show the numbers
| Year | Papers |
|---|---|
| 2025 | 8 |
| 2024 | 6 |
| 2023 | 2 |
| 2022 | 1 |
| 2021 | 2 |
| 2020 | 5 |
| 2019 | 1 |
| 2018 | 2 |
| 2017 | 4 |
| 2016 | 1 |
| 2015 | 1 |
| 2014 | 0 |
| 2013 | 0 |
| 2012 | 0 |
| 2011 | 0 |
| 2010 | 0 |
| 2009 | 0 |
| 2008 | 0 |
| 2007 | 0 |
| 2006 | 0 |
| 2005 | 0 |
| 2004 | 0 |
| 2003 | 0 |
| 2002 | 0 |
| 2001 | 0 |
| 2000 | 0 |
Use grade E
Fewer than 30 papers from 2021 to 2025.
Evidence grade D
Little published measurement evidence found.
- Papers on test-retest reliability
- 0
- Papers on reliability or validity
- 0
- Registered clinical studies using it as an outcome
- 0
How the grades work
Use: A is 1,000 or more papers from 2021 to 2025, B 300 to 999, C 100 to 299, D 30 to 99 and E fewer than 30. Evidence: A means at least 25 papers report its test-retest reliability and at least 25 registered clinical studies use it as an outcome; B, at least 8 test-retest papers, and either 5 registered clinical studies or 150 reliability or validity papers; C, at least 2 test-retest papers, 30 reliability or validity papers or 3 registered clinical studies; D, little published measurement evidence found. The counts measure use. They include every paper that names the task in its title or abstract. Grades describe the published research on this task, not ConductCognition's version of it.
Source: ConductCognition test census 0.22.0, counted 3 November 2026 from OpenAlex and ClinicalTrials.gov.
Task design
The trial sequence as the test runs in ConductCognition.
- Letters
- X or O in a frame: red with a solid edge for a reference trial, whose letter is then the one to hold, and blue with a dashed edge for a comparison trial, which only compares. The letters are the same in every language.
- Blocks
- 4 blocks (2 to 6 as a setting). Each starts with a red-framed letter shown for 2,000 ms that needs no answer, then 64 trials: each of the eight kinds (reference or comparison, after the same or the other type, same or different letter) 8 times.
- Order
- Drawn from the session's seed before the session starts, with no more than 5 trials in a row of one type or one answer. Nothing adapts to performance.
- Trial
- A blank for 500 ms and a cross for 500 ms, then the letter until F or J or 3,000 ms. Which key means same is drawn from the session's ID.
- Practice
- 16 trials with feedback after a first letter, and a second set only when the first had under 75% right answers.
- Scoring
- The gate-opening cost is the median time of right answers on reference trials after a comparison trial minus after a reference trial; the gate-closing cost is the same for comparison trials. The substitution and updating costs use trials after the same type. Answers under 100 ms and the trial after an error are left out.
The frames' edges repeat their colors, solid for red and dashed for blue, so the cue does not rest on color alone; the published versions used color alone. Each session draws its own order by the same rules. The costs are differences of tens of milliseconds, so compare groups, not single people. The scores support research and do not diagnose.
Quality checks
Each session is marked valid, needs review or invalid, with the reason, before you see its scores.
Scored rows: reference_back_test from the scored run, every trial after each block's first, F or J for a letter the same as or different from the one held (chance 50%, half the letters are the same), correctness from the stored key and answer. Below chance invalidates and not above chance is flagged. One key on 90% of answers invalidates when accuracy is not above chance. Completing fewer than half the planned trials invalidates, and any other incomplete run is flagged. Accuracy under 80%, or fewer than 10 usable response times in any of the eight kinds of trial, is flagged for review. Missed responses and responses under 100 ms follow the timed-test rule.
Why these rules
Accuracy significantly below chance (one-sided exact test, p < .05) cannot come from guessing; it means the response mapping was reversed or the task was not followed (Slick, Sherman and Iverson 1999 treat below-chance forced-choice performance as definite response bias). Chance-level accuracy can be a real result in impaired participants, so it is only flagged. One option on at least 90% of answered trials, when the correct options are balanced and accuracy is not above chance, means the answers did not depend on the items (the task counterpart of long-string responding in surveys; Curran 2016). Missed responses and responses under 100 ms keep the platform's timed-test rule: more than max(3, 25%) missed responses or more than max(2, 20%) responses under 100 ms invalidate, and any is flagged. A response under 100 ms cannot be a reaction to the stimulus. Each cost is a difference between medians of right answers in two or four of the eight kinds of trial, so it needs enough of them in each.
No percentile. Scores and quality flags only.
Running this test
Settings you can change
| Setting | Standard | Allowed |
|---|---|---|
| Test blocks | 4 | 2 to 6 |
Yearly plans can change settings within these limits. Free and pay-as-you-go studies use the standard settings.
Need a setting this test does not have?
Devices and languages
- Timed test: participants need a desktop or laptop with a mouse or trackpad. Participant devices
- Languages: English, Spanish, French, German, Italian, Portuguese (Brazil), Chinese (Simplified) and Japanese.
- Available on: Every plan, including the free plan.
- No percentile. Scores and quality flags only. How norms work
- Recruiting online? See the setup guides for Prolific, Sona Systems, CloudResearch Connect and Qualtrics.
Cite and describe
Participants completed the Reference-Back Task in a web browser (ConductCognition task version 1.0.0, scoring version 1.0.0; Rac-Lubashevsky & Kessler, 2016), with the standard settings (test blocks: 4). The main score was gate-opening cost (median time of right answers on reference trials after a comparison trial minus after a reference trial, averaged over same and different letters).
ConductCognition versions: task version 1.0.0, scoring version 1.0.0. Every export records the versions each session used, for a methods section.
- Rac-Lubashevsky, R., & Kessler, Y. (2016). Dissociating working memory updating and automatic updating: The reference-back paradigm. Journal of Experimental Psychology: Learning, Memory, and Cognition, 42(6), 951-969. https://doi.org/10.1037/xlm0000219The paradigm: letters compared with the last reference letter, with a cue that says whether to update, separating updating from opening and closing the gate.
- Rac-Lubashevsky, R., & Kessler, Y. (2016). Decomposing the n-back task: An individual differences study using the reference-back paradigm. Neuropsychologia, 90, 190-199. https://doi.org/10.1016/j.neuropsychologia.2016.07.013The costs as individual differences in 88 participants, set beside the n-back task.
- Jongkees, B. J. (2020). Baseline-dependent effect of dopamine's precursor L-tyrosine on working memory gating but not updating. Cognitive, Affective, & Behavioral Neuroscience, 20(3), 521-535. https://doi.org/10.3758/s13415-020-00783-8A first reference letter shown for 2 s with no answer, medians of right answers, the trial after an error left out, and updating measured on trials after the same type.
- Boag, R. J., Stevenson, N., van Dooren, R., Trutti, A. C., Sjoerds, Z., & Forstmann, B. U. (2021). Cognitive control of working memory: A model-based approach. Brain Sciences, 11(6), 721. https://doi.org/10.3390/brainsci11060721A web version with framed X and O shown until the answer, in which gate closing cost 72 ms and gate opening 45 ms.
- van Dooren, R., Jongkees, B. J., & Sellaro, R. (2025). Self-prioritization in working memory gating. Attention, Perception, & Psychophysics, 87(2), 399-414. https://doi.org/10.3758/s13414-024-02869-8Equal counts of each kind of trial in every block of an online version, with practice and feedback before the test.
- Chatham, C. H., Herd, S. A., Brant, A. M., Hazy, T. E., Miyake, A., O'Reilly, R., & Friedman, N. P. (2011). From an executive network to executive control: A computational model of the n-back task. Journal of Cognitive Neuroscience, 23(11), 3598-3619. https://doi.org/10.1162/jocn_a_00047The gating account the task reads: a model in which working memory learns when to let information in.
Source papers
- Rac-Lubashevsky R, Kessler Y (2016). Dissociating working memory updating and automatic updating: The reference-back paradigm. Journal of Experimental Psychology: Learning, Memory, and Cognition, 42(6):951-969.Paradigm source, Adults
- Rac-Lubashevsky R, Kessler Y (2016). Decomposing the n-back task: An individual differences study using the reference-back paradigm. Neuropsychologia, 90:190-199.Paradigm source, Young adults
- No normative data for this task.Paradigm source, N/A
Questions
How long does the Reference-Back Task take?
About 9 to 11 minutes, as listed in the test library.
What device does a participant need?
Timed test: participants need a desktop or laptop with a mouse or trackpad. The browser must be Chrome, Edge, Firefox or Safari.
What scores does it produce?
Gate-Opening Cost, Gate-Closing Cost, Substitution Cost, Updating Cost, Matching Cost, Accuracy, Accuracy, Reference Trials, Accuracy, Comparison Trials, Fewest Usable Trials, Missed Letters and Anticipations. Each session is also marked valid, needs review or invalid, with the reason.
How widely is the Reference-Back Task used in research?
19 papers from 2021 to 2025 name it in their title or abstract, up 46% compared with 2016 to 2020. That ranks it 405 of 495 test families in the ConductCognition test census (use grade E). Source: ConductCognition test census 0.22.0, counted 3 November 2026 from OpenAlex and ClinicalTrials.gov.
How much reliability evidence does the Reference-Back Task have?
In the published research, 0 papers report its test-retest reliability and 0 registered clinical studies use it as an outcome (evidence grade D in the ConductCognition test census).
Does it report a percentile?
No. Scores and quality flags only.
Which settings can a researcher change?
Test blocks (2 to 6). Settings can be changed on yearly plans; free and pay-as-you-go studies use the standard settings.
Which languages can participants use?
Participants can take this test in English, Spanish, French, German, Italian, Portuguese (Brazil), Chinese (Simplified) and Japanese. English is on every plan, and the other languages come with yearly plans. Have a native speaker read the machine-translated text before launch.
Is this a clinical test?
No. ConductCognition is for research use only. It runs, scores and exports the test; clinical interpretation is outside the platform.
Every test: the full test list.
Run this test in your next study
Start free with 10 free participants on any test, then use participant credits or a yearly plan.
Related tests