AkashicINQUIRY010 · Research Proposal ·
Sense X — How Did I Know That?
AkashicABSTRACT
When an impression feels new or received, how accurately can we tell where it came from? Source-monitoring research shows that people can misattribute a familiar idea to their own generation. Study A proposes a randomised exposure experiment to estimate how often previously encountered details later appear in responses labelled new. Study B proposes a separate blinded test of whether locked choices match independently selected, already existing hidden targets at more than the 25% chance rate for four equally likely options. Both studies are proposals, with no participant results. The findings would apply to the tasks tested. An unusual result in Study B would call for an audit and independent replication before interpreting its cause.
Research Context
HOMESENSE842 introduces Sense X as a question about how an impression arrives. AkashicINQUIRY009 asks how readers check the source of an external claim. Here the source to check is an impression. Start with what can be recorded, what was available beforehand and which result would change the assessment. HOMESENSE842 · AkashicINQUIRY009
Proposed Studies
Study A · Source Monitoring
Prediction. Under matched prompts, a distinctive detail from briefly exposed material will appear in impressions labelled new more often than a comparable detail withheld during exposure.
- Prepare matched groups of unfamiliar, distinctive items and randomise brief exposure, explicit exposure and withholding within counterbalanced sets. Use the explicit-exposure condition as a manipulation check for remembering prior exposure. Fix the task, delay, permissible prompts and scoring rubric before the main run.
- Ask the participant for a fixed number of responses per cue. Capture the exact response, confidence and source judgement (remembered, inferred, apparently new or unsure) before any item list or feedback. Two independent scorers blinded to assignment code distinctive matches. Freeze synonym rules and how ties, ambiguity and absent responses are handled. Preserve the initial ratings and resolve disagreements using the prespecified rule.
- For the primary outcome, compare the proportion of eligible prespecified item opportunities yielding both a distinctive match and an apparently-new label in briefly exposed versus withheld items. Report the absolute difference and interval. Preserve the full assignment blocks and repeat the allowed randomisation for inference. Report match frequency, source labels conditional on a match, coder agreement and sensitivity to ambiguous matches separately.
The withheld set supplies a baseline for matches generated by the prompts and rubric. A credible difference supports an effect of experimentally assigned exposure in this task. A small or imprecise difference leaves the interpretation dependent on the manipulation check and interval. Johnson, Hashtroudi & Lindsay, 1993 · Brown & Murphy, 1989
Study B · Already Existing Concealed Target
Prediction. Locked single choices exceed 25% accuracy across trials with one independently and uniformly selected target among four fixed options.
- Generate each target on an isolated server before the choice. Publish a timestamped commitment combining the protocol version, trial ID, target index and independently generated high-entropy nonce in a canonical encoding. Put the commitment in an independently witnessed or append-only log. Before the response is locked, neither participants nor interacting staff have access to the target, nonce or target-dependent client data. Keep the interface, timing and option order independent of the selected target.
- Record the impression first, then lock one forced choice in a prespecified response window. Release the targets and nonces for verification once all choices in the block are locked or the response windows have expired. Audit logs for timestamp order, client payloads, replay and duplicate records. Run synthetic 25% null trials and injected leakage cases before the human pilot. Log every committed trial, duplicate, timeout, integrity failure and exclusion. Predefine what qualifies as a valid lock and test sensitivity to unresolved trials. If completion depends on target or access to the target, stop interpretation and audit the leakage.
- Freeze one primary outcome, one directional decision rule, the sample size and stopping rule before confirmatory data collection. Keep calibration and pilot trials out of the confirmatory analysis. If every analysed target is uniform and independent and unobservable at choice time, the null probability conditional on the response history is ¼. Simulate the actual logging and missing-data rule under that null. Use participant- or session-aware uncertainty for effect estimates and specify how any results generalise beyond the tested participants. Report effect size, uncertainty, misses and integrity failures alongside any p-value.
Planning only. Under ideal independent, complete Bernoulli trials, a one-sided exact binomial test with a 5% significance limit and 600 valid trials rejects at 169 or more hits. Its power is about 85% at a true 30% hit rate. This is one sensitivity example. Choose the smallest worthwhile effect and final design after the pilot, including any missing-trial allowance. A detectable uplift supports an anomaly in this protocol only if the integrity checks pass. Independent repetition is the next discriminating test. HOMESENSE813 · Inaccessible-record proposal · Nosek et al. · Preregistration
Start With the Journal
- Before Checking
- Record event time, logging time, exact impression or sketch, confidence, possible prior exposure and what could be inferred.
- After Checking
- Record the source and date checked, every reasonable comparison, the match rule and the outcome, including a miss or ambiguous result. Preserve the original entry.
The journal records an experience and its subsequent checks. In a prospective test, only entries locked before verification count towards the specified outcome.
Pilot and Evidence Checks
Study A. Give two independent reviewers the same response excerpts and candidate items in shuffled order. Hide exposure assignment, participant source labels and the other reviewer’s scores. Each reviewer records a candidate ID or “no match”, plus “distinctive”, “ambiguous” or “absent”. Use three calibration examples first, then score a separate pilot set that includes near-matches, items with overlapping features and blank responses. Preserve both score sheets, report agreement by category and settle ambiguities using a rule chosen before the main study. Apply the participant’s source label only after match coding is locked.
Study B. Before human testing, run a clean synthetic block and blocks containing a target exposed in a pre-lock client payload, replayed submissions, missing responses and reveal or commitment failures. Check the client, server and independent witness logs together. Review the flagged trial IDs and the resulting analysis denominator. Further live testing must check channels that simple event fixtures cannot exercise, including timing, caching, target-dependent option order and access controls.
What would change the assessment? In Study A, a credible excess of novel-labelled matches to briefly exposed items would support an exposure effect in this task. In Study B, chance-level accuracy with a sufficiently narrow interval would constrain effects of the size tested. A reproducible excess with intact controls would motivate independent replication and further checks of access routes. Comparable design cautions come from the Transparent Psi Project, a replication of a separate two-choice task with a future target.
Status. Editorial research proposal. Participant data have yet to be collected.