Research queue

The lab's to-do list, in public

Project Aletheia is an independent lab that stress-tests contested science: take a claim, find the most boring explanation that could produce it, and run the test that decides between them. This page is everything the lab wants to test next — and, honestly, what each item still needs before it can run. Some need a dataset. Some need a better design. Some need a scientist with the right instrument. If you can supply the missing piece, any of these can move.

Experiments looking for a scientist

Fully designed experiments — question, protocol, budget, and the result that would kill them — waiting for a researcher with the right setup. The lab designs, you run, and the result publishes either way: a null here is a result, not a failure.

Package in preparationEXP-2026-001-RF-PULSE-AUDITIONPre-Foundry

Can simple radio-pulse patterns be heard and decoded?

Old military-era reports say people can hear certain radio-frequency pulses inside their head (a real, accepted effect called radio-frequency hearing) and one thin historical claim says simple pulse CODES could be interpreted. Nobody ever published the numbers. This experiment tests whether volunteers can decode simple pulse patterns above chance under modern safety limits.

Why it matters

It would replace a 60-year-old undocumented claim with the first real error-rate table for pulse-code radio hearing, and a clean null would finally bound what the historical record actually supports.

What it needs

  • RF exposure lab with safety (SAR) monitoring and human-subjects approval
  • pulse-modulated microwave source in the published RF-hearing range
  • sound-shielded test room
  • 10-20 adult volunteers

Budget

Gold

$100k+ standalone with dedicated exposure hardware and full board review

Shoestring

not runnable on a shoestring: RF human exposure requires a properly equipped lab

Standard

$15k-40k as a hosted study inside an existing RF/bioelectromagnetics lab (facility time, safety review, subject payments)

Time: 2-4 months including ethics review; sessions are under one hour per subject · Feasibility: HARD - needs a specialized RF exposure facility and full human-subjects review; realistic only as a collaboration with an existing bioelectromagnetics lab

Kill condition: Decoding accuracy indistinguishable from chance across the preregistered trial count at safe exposure levels

Claim ceiling: Bounded psychophysics only: whether simple pulse symbols are decodable above chance. Not speech, not communication devices, no clinical claims.

Source: Aletheia RF-ultrasound auditory investigation (2026-06); full protocol draft with safety preconditions exists and ships to the hosting lab

I could run this — talk to me
Seeking a scientistEXP-2026-002-VIBRATION-ONSETPre-Foundry

Do body-vibration episodes really come right before out-of-body experiences?

Thousands of people who report leaving their body describe a strong body-wide vibration just before it happens. All existing evidence is after-the-fact storytelling. This study has sleepers who often report such episodes log every vibration event prospectively (diary plus a simple wearable) so the timing claim gets tested forward instead of remembered backward.

Why it matters

The vibration-first sequence is one of the most consistent patterns in a century of these reports (Aletheia's own text analysis of 2,192 accounts found it splits cleanly across independent halves of the data). Nobody has ever tested it prospectively. Either outcome is a real result about how these experiences work.

What it needs

  • a sleep researcher or consciousness lab willing to host
  • 20-40 frequent experiencers (recruitable through existing experiencer communities)
  • consumer sleep wearables (accelerometer plus heart rate)
  • a preregistered diary protocol

Budget

Gold

$60k+ with in-lab polysomnography nights for a subset

Shoestring

$3k-6k fully remote: wearables shipped to participants, online diaries, subject payments

Standard

$10k-25k with a hosting lab, better wearables, and compliance monitoring

Time: 3-6 months of logging; setup one month; analysis is prewritten · Feasibility: MEDIUM - fully remote shoestring version is genuinely runnable; the hard part is disciplined recruitment, not equipment

Kill condition: Prospective logs show vibration episodes are NOT preferentially followed by the experiences within the preregistered window

Claim ceiling: Timing and sequence of self-reported episodes only. Says nothing about what the experiences ARE - only about whether the reported sequence survives prospective logging.

Source: Aletheia vibration-transition program (2026): corpus analyses, a full grant-grade study design, and the prewritten analysis plan exist

I could run this — talk to me
Validated package · resource blockedEXP-2026-003-LIGHT-TOUCH-HRVFoundry validated

Light Touch 30-minute HRV showdown

Does 30 minutes of Light Touch concha stimulation change paired log RMSSD more than a sensation-matched sham in healthy adults?

Why it matters

Give a qualified researcher a complete, pre-tested protocol for one narrow short-term HRV comparison.

What it needs

  • Light Touch active and sensation-matched sham configurations
  • Beat-level RR recording with Polar H10 or laboratory ECG
  • Recruit 48 healthy adults to retain 40 complete pairs

Budget

Gold

$16092.84 cash; 210 person-hours; 90 calendar days

Shoestring

$8773.96 cash; 110 person-hours; 90 calendar days

Standard

$10025.84 cash; 150 person-hours; 75 calendar days

Time: Standard tier: 150 person-hours across 75 calendar days. · Feasibility: Grade D (44.5/100) from the frozen cost, calendar, instrument, and recruitment formula.

Kill condition: Kill the positive-effect route if 40 valid pairs do not produce a positive mean with two-sided p below 0.05.

Claim ceiling: At most, this estimates one short-term paired RMSSD contrast in healthy adults under the exact tested dose and sham.

Source: /Users/bo/data/adp/investigations/experiment_foundry_plan_2026-08-24/known_answer_light_touch/FINDING_EXPANSION.json

I could run this — talk to me

The question queue

Every open question the lab has vetted, with its status. Each card expands to show the honest state of its test: what data exists, what is blocking it, the boring explanation that has to be ruled out first, and the most the test could ever prove even if everything goes right.

New here? What the labels mean
Ready now —
the data, the checks, and the kill rule all exist; this could run today.
Needs data —
the question is solid but a dataset is missing. If you have it, that is the whole blocker.
Needs design —
we do not yet have a test that could not fool itself.
Needs a collaborator —
requires an instrument, lab, or subject pool the lab does not have.
“The most this could ever prove” —
every test states its ceiling up front, so a narrow result can never quietly inflate into a big claim.
Needs dataQ-BIOLOGY-RECEPTOR

Do the 27 genes really form a biological link, or did name changes and mixed gene families create it? A false link could send later research in the wrong direction.

Precise form: Do receptor-family bridges survive exact 27-gene identity and provenance checks?

1 linked test · last triaged 2026-07-23

The honest state of this test
What data we already have
Existing 27-gene ledgers; expression and true receptor-family universes are absent
What's stopping us
expression matrix and receptor-family annotations
The boring explanation we must rule out first
gene-symbol aliases and mixed receptor families
How we'd check it honestly
Deterministic provenance join plus expression/family-matched nulls with planted controls
What would kill it
Kill any bridge whose members lack exact source-backed receptor-family identity.
The most this could ever prove
Dataset-specific provenance result; no biological mechanism.

TEST-BIOLOGY-RECEPTOR-V1

Needs dataQ-BODY-BOUNDARY

Do fainting, coma, anesthesia, sleep paralysis, near-death experiences, and extreme acceleration produce different patterns of feeling outside the body? Fair samples could show whether experiences that sound alike have different triggers.

Precise form: Do G-LOC, syncope, coma, anesthesia, sleep-paralysis, and NDE data preserve trigger-specific body-boundary patterns?

1 linked test · last triaged 2026-07-23

The honest state of this test
What data we already have
Some coded narratives; raw denominators incomplete
What's stopping us
raw denominators and independent labels
The boring explanation we must rule out first
keyword extraction, selected case series, unequal denominators
How we'd check it honestly
Blinded precision audit followed by diagnosis-matched proportion contrasts
What would kill it
Kill if validated labels or non-selected denominators remove the trigger contrast.
The most this could ever prove
Trigger/corpus contrast only; no shared mechanism.

TEST-BODY-BOUNDARY-V1

Needs dataQ-CORT-INTERMISSION

Do reports of an interval between lives form a repeatable pattern when every detail is traced to the original case record? This would show whether the proposed subtype is more than a set of memorable examples.

Precise form: Do intermission features survive a source-traceable case-level matrix?

1 linked test · last triaged 2026-07-23

The honest state of this test
What data we already have
Public CORT sources exist; case matrix incomplete
What's stopping us
case-level extraction
The boring explanation we must rule out first
secondary summaries and selected memorable cases
How we'd check it honestly
Double-entry source verification with blind feature coding
What would kill it
Kill if source-verifiable case rates do not reproduce the proposed subtype.
The most this could ever prove
Documented-case pattern only.

TEST-CORT-INTERMISSION-V1

Needs dataQ-DREAM-PRIVACY-LINKAGE

Can dream reports be linked to the same writer in a new group when the possible author is not limited to a small known set? This matters because reliable linkage could create a privacy risk.

Precise form: Does dream-report linkage generalize to independent open-set contributors?

1 linked test · last triaged 2026-07-23

The honest state of this test
What data we already have
DreamBank result only; external cohort absent
What's stopping us
independent consented cohort
The boring explanation we must rule out first
recurring content, recorder/transcriber, closed-set advantage
How we'd check it honestly
Open-set external cohort with contributor and recorder separation
What would kill it
Kill generalization if open-set linkage falls to the frozen null.
The most this could ever prove
Current finding remains one-archive closed-set linkage.

TEST-DREAM-PRIVACY-LINKAGE-V1

Needs dataQ-HESSDALEN-ECOLOGY

Do reported lights in Hessdalen still show the same links to weather and solar activity when direct station records and new dates are used? This would test the first result with better local data.

Precise form: Does the matched weather/Kp result replicate on direct station data and held-out dates?

1 linked test · last triaged 2026-07-23

The honest state of this test
What data we already have
Current dated corpus; direct station coverage incomplete
What's stopping us
station logs and independent dates
The boring explanation we must rule out first
reanalysis-grid error and observer opportunity
How we'd check it honestly
Held-out case-crossover using station logs and observer-density controls
What would kill it
Kill extension if Kp elevation appears or weather differences vanish.
The most this could ever prove
Current finding stays reanalysis-based observation ecology.

TEST-HESSDALEN-ECOLOGY-V1

Needs dataQ-LZC-APERIODIC

After ordinary changes in the brain-wave spectrum are removed, does the signal-complexity measure still show no drug effect in new recordings? This would test whether the earlier null survives a cleaner measurement.

Precise form: Does the LZc effect remain absent after preregistered aperiodic control in new raw EEG?

1 linked test · last triaged 2026-07-23

The honest state of this test
What data we already have
Two retrospective datasets; prospective replication absent
What's stopping us
independent raw EEG
The boring explanation we must rule out first
spectral slope, power, state mixing, and small subjects
How we'd check it honestly
Person-level preregistered mixed model with spectral-surrogate comparison
What would kill it
Falsify the current claim if a powered drug effect survives slope control.
The most this could ever prove
Current claim remains two-dataset measurement result.

TEST-LZC-APERIODIC-V1

Needs dataQ-NDE-LIFE-REVIEW

Can the reported link between life reviews and encounters with deceased relatives be reproduced from checked, row-level near-death reports? The original percentages cannot yet be traced to a validated table.

Precise form: Is the reported life-review/deceased-relative subtype reproducible from validated NDE rows?

1 linked test · last triaged 2026-07-23

The honest state of this test
What data we already have
Claim summary exists; validated row packet not found
What's stopping us
recover coded rows
The boring explanation we must rule out first
generated labels and denominator loss
How we'd check it honestly
Blind label validation plus exact denominator recomputation
What would kill it
Kill if 58.14% and 16.279% cannot be reproduced on validated labels.
The most this could ever prove
Within-archive co-occurrence only.

TEST-NDE-LIFE-REVIEW-V1

Needs dataQ-EVOBC-VARIANT-SCORING

How much do the published image-model scores change when each symbol variant is judged against a source-backed list of acceptable forms? Exact-copy scoring may mark valid variants wrong.

Precise form: How do the paper's trained-model scores change under a source-backed variant-aware evaluation?

1 linked test · last triaged 2026-07-23

The honest state of this test
What data we already have
Pinned EVOBC releases and frozen variant-sensitivity receipts; original split and predictions unavailable
What's stopping us
original evaluation split and per-image predictions
The boring explanation we must rule out first
the exact-copy lookup channel may not represent the trained models' errors
How we'd check it honestly
Hash-bind the original split and per-image predictions and rescore without changing the accepted-label table
What would kill it
Kill any trained-model correction claim if direct rescoring is negligible or reverses direction.
The most this could ever prove
Direct trained-model score change under a frozen source-backed relation table; no decipherment claim.

TEST-EVOBC-VARIANT-SCORING-V1

Needs dataQ-QUANTUM-SWITCH-REPRO

Does the reported quantum-device inequality stay above its target when measurement settings are randomized and results are checked in separate time blocks? This tests whether slow drift and fixed measurement order explain the result.

Precise form: Does the quantum-switch inequality remain above 1.75 when settings are randomized and analyzed in repeated time blocks?

1 linked test · last triaged 2026-07-23

The honest state of this test
What data we already have
Published aggregate YAML counts reproduce exactly; no repeated time blocks are public
What's stopping us
timestamped or repeated-block event data with setting order
The boring explanation we must rule out first
predefined setting order plus source or interferometer drift can bias aggregate conditional probabilities
How we'd check it honestly
Freeze randomized setting order and repeated blocks, then test the inequality within blocks and under order-preserving drift controls
What would kill it
Downgrade robustness if the block-aware lower bound reaches 1.75 or the violation tracks measurement order.
The most this could ever prove
Robustness of this apparatus and protocol to measured time-linked drift; no universal quantum-switch claim.

TEST-QUANTUM-SWITCH-REPRO-V1

Looking for what already ran? Results live on Findings. Investigations that ended without an answer — and what would revive them — live on Cold Cases.