AI-Guided Remote Viewing Protocol H-0.4
Private Working Master Copy
2026-08-23

Internal Operator Note
This page is for the working master copy only. It is not part of frozen H-0.4 and does not need to be included when the protocol itself is copied or posted.
Use the protocol exactly as written when conducting or sharing a field test. The most important operational points are:
• Keep the AI guide blind to the target. For Mode B, keep the viewer blind to the entire candidate pool, not just the selected target.
• Use a fresh AI conversation where practical. Do not upload or describe the target in the same context used by the guide.
• Give the guide only the opaque session/target identifier specified by the protocol.
• Preserve the viewer's original words and sketches. Do not let the AI interpret, strengthen, clean up, or summarize acquisition content.
• Freeze the transcript and sketches before feedback or target reveal. Never rewrite the frozen acquisition record afterward.
• Mode A is appropriate for casual/solo shakedown use. Mode B is the blinded four-choice procedure and requires an independent curator or equivalent separation mechanism.
• Optional breathing, audio, meditation, hypnagogic, CRV, and related branches are experimental branches, not proven improvements.
• Preserve misses, empty sessions, violations, abandoned blocks, and failed runs. Do not keep only interesting results.
• Stop the session if the participant becomes distressed, dizzy, disoriented, or wishes to stop. After a deep-relaxation or sleep-adjacent branch, return fully to ordinary alertness before driving or operating machinery.
• H-0.4 has passed the final pre-field-test critic gate as READY FOR FIRST FIELD TEST. Future refinements should not be back-written into H-0.4; they belong in a later version.
Copy/paste boundary: when you want the protocol alone, begin with the next heading, AI-Guided Remote Viewing Protocol H-0.4.
\newpage
AI-Guided Remote Viewing Protocol H-0.4
Experimental Field Protocol
Date: 2026-08-23
Status: Untested / Open Field Evaluation
Use: Instrument shakedown and prospective field blocks under the rules below
Evidence rule: Collect broadly. Admit narrowly.

1. What This Is — and Is Not
H-0.4 is an explicitly untested protocol for using an AI assistant as a target-blind procedural guide for a human remote-viewing session while minimizing ordinary information leakage, suggestion, narrative contamination, flexible scoring, and post-hoc rescue.
It does not establish that:
• remote viewing works;
• anomalous information transfer exists;
• any altered state improves remote viewing;
• binaural beats entrain the cortex;
• Monroe Focus states correspond to validated EEG states;
• any preparation component improves remote-viewing performance.
The core is therefore intentionally conservative. Most state-induction methods remain experimental branches.
H-0.4 supersedes H-0.2 for new prospective work. H-0.2 remains archived unchanged.

2. Core Design Principle
The AI guide has one job:
Guide the procedure without knowing, guessing, interpreting, evaluating, or searching for the target.
The AI is not the remote viewer.
The AI must not receive:
• the target image;
• the target description;
• meaningful coordinates;
• target-category hints;
• the answer;
• decoy identities;
• target feedback before the acquisition record is frozen.
If the AI knows the target, the session is contaminated for confirmatory use.

3. Roles
3.1 Viewer
The human attempting the session.
3.2 AI Guide
A target-blind procedural assistant.
3.3 Target Curator
Selects/randomizes the target and keeps it hidden.
3.4 Judge
Compares the frozen session record against the target and decoys.
For stronger testing, curator, guide, viewer, transcriber/coder, and judge should be separated where practical.

4. Minimum Equipment
Required:
• quiet place;
• paper;
• pen or pencil;
• opaque target/session identifier;
• AI assistant.
Useful but optional:
• headphones for acoustic isolation;
• audio recording;
• phone/camera for photographing sketches immediately after freeze;
• second person to act as curator or judge.
No EEG, HRV monitor, or laboratory equipment is required for the basic field protocol.

5. Target Preparation
Before the session:
1. Select the target without the viewer seeing it.
2. Assign a random opaque identifier, for example Q7-4182.
3. The identifier must not encode location, target type, owner, category, or answer.
4. Give the AI guide only the identifier.
5. Do not tell the AI what kind of target pool was used unless that fact is explicitly part of the preregistered design.
An opaque identifier has no demonstrated paranormal “addressing” function. Its ordinary purpose is blinding, record binding, chain of custody, and prevention of semantic leakage.

6. AI Guide Operating Prompt
Use a fresh conversation where practical. Do not place target information anywhere in the same AI context.
Paste:
AI-GUIDED REMOTE VIEWING — H-0.4 FIELD MODE

You are acting only as a target-blind procedural guide.

You do not know the target and must not try to infer, guess, reconstruct, search for, identify, or retrieve it.

Do not use web search, external tools, memory retrieval, personal-profile information, prior conversation content, or hidden contextual clues to guess the target.

Your job is procedural only.

RULES

1. Never suggest target content, categories, examples, imagery, locations, objects, people, colors, textures, temperatures, sounds, smells, shapes, motions, or interpretations.

2. Never praise, confirm, reject, grade, paraphrase, summarize, improve, reinterpret, or correct reported content during acquisition.

3. Never say or imply that an impression is closer, stronger, clearer, more accurate, more important, or more target-like.

4. Do not ask leading or closed questions.

5. Do not ask the viewer to move toward, inside, above, below, behind, through, closer to, or away from the target.

6. Do not ask the viewer to imagine anything.

7. Do not resolve contradictions. Preserve them.

8. Do not replace the viewer's exact wording with your own wording.

9. If the viewer reports something analytical, interpretive, symbolic, or like a named object, do not challenge or confirm it. Preserve it exactly.

10. During acquisition, use only the fixed process prompts specified in H-0.4. Do not branch based on the content of the viewer's answers.

11. Silence is preferred over additional prompting.

12. If the viewer reports distress, dizziness, panic, disorientation, or asks to stop, abort acquisition and move to reorientation. Safety overrides protocol continuation.

13. At acquisition end, tell the human that the record must now be frozen before feedback. You cannot enforce immutability; the human is responsible for saving and preserving the frozen record.

14. Never permit earlier material to be edited or deleted after feedback.

15. If target information is revealed to you at any point before freeze, state that the session is contaminated, stop guiding acquisition, and proceed to freeze. Do not continue acquisition.

16. At freeze, do not summarize, restate, characterize, interpret, or evaluate the session.

TARGET/SESSION IDENTIFIER:
[INSERT OPAQUE IDENTIFIER HERE]

Begin only when I say: START SESSION.

7. Standard H-0.4 Session
All times are candidate engineering values, not scientifically optimized doses.
Recommended total before feedback: approximately 15–20 minutes.
Phase 1 — Start and Record
Approx. 1 minute
Guide:
“Session H-0.4 has begun. The target is unknown to both of us. Your session identifier is [ID]. Nothing needs to be forced or made meaningful. Anything you record will be preserved exactly as it occurs.”
Record:
• date/time;
• protocol version;
• session ID;
• AI provider/model if known;
• branch condition;
• intended block ID and trial number if part of a prospective block.

Phase 2 — Content-Empty Settling
Candidate: 2 minutes
Sit comfortably in a position that allows writing/sketching.
Guide:
“Settle into a comfortable position. Allow unnecessary physical effort to soften. Breathe normally. There is nothing to visualize and nothing to search for. When an impression occurs later, record it without trying to make it fit anything.”
Then remain silent.
The H-0.4 core does not require:
• paced breathing;
• countdowns;
• body imagery;
• Energy Conversion Box;
• REBAL;
• affirmation;
• hypnosis;
• binaural beats;
• a Monroe Focus state.
These remain branches.

Phase 3 — Tasking
A few seconds
Guide says once:
“The session identifier is [ID]. Begin recording whatever impressions occur without trying to identify what the target is.”
No additional target cue.

Phase 4 — Initial Acquisition
Candidate: 5 minutes
The viewer writes brief impressions as they arise.
Writing is preferred to continuous conversation in this phase.
If the viewer explicitly requests a process reminder, the only default reminder is:
“Record what occurs. Do not force additional detail.”
For the default core, no more than two such post-tasking reminders should be issued. If more are required, retain the session but flag PROMPT-CAP EXCEEDED as a procedural deviation.
Every guide prompt after tasking must be logged with its timestamp.
Record:
• first impressions;
• later impressions;
• contradictions;
• uncertainty;
• named objects if they occur spontaneously;
• apparent mistakes.
Do not erase or rewrite anything.

Phase 5 — Sketch
Candidate: 5 minutes
Guide:
“Without trying to make the drawing recognizable, sketch any spatial relationships, forms, boundaries, arrangements, or movement that occurred to you. Preserve uncertainty and contradictions.”
Then remain silent.
Do not ask the AI what a sketch resembles.

Phase 6 — Final Unprompted Capture
Candidate: 2–3 minutes
Guide:
“Add anything else that occurred but has not yet been recorded. Do not try to improve the session or make the pieces agree.”
Then remain silent.
“Nothing further” is an acceptable result.

8. Freeze
Guide:
“Acquisition is now complete. Freeze the session record now. Do not add, delete, rewrite, reinterpret, or correct acquisition content after this point.”
The human must then:
1. save/export the transcript;
2. photograph/scan sketches;
3. timestamp them if practical;
4. mark the session FROZEN;
5. record the freeze time.
The AI cannot enforce this step.
Only after freeze may the target be revealed.
The AI must not summarize the session at freeze.

9. Reorientation
Before feedback:
“Return attention fully to the room. Breathe normally. Open your eyes if they were closed. Move your hands and feet, stretch if comfortable, and take a moment to feel fully alert and oriented.”
After any deep-relaxation or sleep-adjacent branch, allow full alertness to return before driving or operating machinery.

10. Feedback
Feedback occurs only after freeze.
Do not modify acquisition data after feedback.
Post-session comments belong under:
POST-FEEDBACK NOTES
They are never merged into the acquisition record.

11. Scoring
Mode A — Casual Practice / Solo Shakedown
Reveal the target after freeze and compare.
Useful for practice, instrument shakedown, and solo use.
Mode A is not suitable for pooled inference because self-scoring after reveal is flexible.
A Mode-A run may use a Shakedown Series ID, but it is not an inferential Block ID and has no confirmatory primary endpoint.

Mode B — Blind Four-Choice Test
Before the session:
1. Prepare four candidate images.
2. Designate one true target using a declared randomization method.
3. Keep the entire four-image candidate pool hidden from the viewer and AI guide.
4. The viewer therefore cannot act as Target Curator in Mode B. Solo practice without an independent curator is Mode A.
5. Record the selected target index in committed form before the viewer begins — for example, a sealed written record or a cryptographic commitment retained by the curator.
Acceptable randomization methods include:
• a physical die or equivalent physical randomizer;
• shuffled sealed slips;
• a computer random-number generator whose output is logged by the curator.
Human “pick one that feels random” selection is not a randomization method.
After freeze:
6. Give the frozen transcript/sketches and all four images to a judge.
7. Judge remains blind to which image is true.
8. Randomize candidate presentation order.
9. Judge ranks candidates 1–4 from best to worst match.
10. Record ranking before revealing the true target.
11. Record judge identity/code.
Nominal first-rank chance for one four-choice trial = 1/4.
One trial proves almost nothing.

11A. Prospective Block Rule — Required Before Pooling
A block is a prospectively registered set of Mode-B trials run under one frozen protocol version, one declared branch condition, and one declared scoring plan.
Before the first target of a block is assigned, record:
• Block ID
• protocol version;
• branch condition;
• intended number of trials N;
• primary endpoint;
• secondary endpoint(s);
• exclusion/invalidity rules;
• judge plan;
• randomization method;
• stopping rule;
• multiplicity plan if more than one branch/block comparison is intended.
Public timestamped registration
For any block intended to contribute to pooled or confirmatory inference, the declaration above must be placed in a timestamped, append-only public record before trial 1 and before the first target is assigned.
The implementation can be simple. The requirement is that later readers can verify that:
1. the declaration existed before outcomes;
2. it cannot be silently replaced;
3. incomplete or abandoned blocks remain visible.
Examples include a preregistration service, an append-only public repository/commit history, or another public timestamped archive.
Record the registration reference in every trial report.
A privately dated note is adequate for personal shakedown discipline but not for pooled confirmatory inference.
Default primary endpoint
Mean true-target rank across valid Mode-B trials.
Chance expectation in a four-choice rank task is 2.5; lower is better.
The inferential test must be declared before the block begins. A permutation/randomization analysis over the frozen rank data is preferred when practical.
Default secondary endpoint
First-place rank rate: proportion of valid Mode-B trials in which the true target is ranked #1.
For four choices, nominal chance = 0.25.
N and power
H-0.4 does not claim a universally powered N.
The intended N must be declared before trial 1.
N = 20 may be used as a shakedown N for testing whether the procedure, blinding, record freeze, and judging pipeline work.
It is not adequate for strong inference in either direction about a small effect.
For illustration only, using the secondary first-place endpoint with a one-tailed exact binomial threshold of at least 9 first-place hits in 20 trials:
Assumed true first-place rate
Approx. power at N=20
0.25
0.041
0.32
0.157
0.45
0.586
0.50
0.748
0.55
0.869
Thus an N=20 block would fail to reach that threshold about 84% of the time even if the true first-place rate were 0.32.
Under that simplified first-place/binomial model, roughly 275 trials are needed to approach 80% exact power at a true rate of 0.32. That number is an illustration, not a universal prescription: real design requirements depend on the primary endpoint, repeated viewers, reused judges, target dependence, branch structure, and planned model.
Therefore:
A null N=20 shakedown block must not be reported as evidence that remote viewing does not exist.
Stopping rule
Complete the declared block regardless of interim hits or misses.
Do not stop because:
• results look unusually good;
• results look unusually bad;
• a “special” target produced an interesting session.
If the block is abandoned, report it as INCOMPLETE with the completed count. It remains visible in the registration record.
Multiplicity / branch scanning
Each additional branch or cross-branch comparison increases the opportunity for a chance-positive result.
Any confirmatory comparison across branches must be declared before the relevant blocks begin and must specify how multiplicity will be handled.
Otherwise the cross-branch result is exploratory and non-confirmatory.
Zero-content sessions
NO IMPRESSIONS and genuinely near-zero-content sessions remain in the intention-to-treat primary when blind judging is possible.
Also report, separately:
• number and fraction of zero-content trials;
• number and fraction of near-zero-content trials under a predeclared definition.
Their frequency is itself an important shakedown outcome.

12. Failure, Invalidity, and Exclusion Rules
Nothing disappears.
Use one or more labels:
• NO IMPRESSIONS
• SLEEP / DROWSINESS
• TARGET LEAKAGE
• AI TARGET-AWARE
• AI CONTENT-INJECTING VIOLATION
• AI NON-CONTENT PROCEDURE VIOLATION
• PROMPT-CAP EXCEEDED
• REVEAL BEFORE FREEZE
• NO FROZEN RECORD
• ABORTED
• COMPLETED — MISS
• COMPLETED — SCORE PENDING
• COMPLETED — VALID
Primary-analysis rule for a prospective Mode-B block
Included in the intention-to-treat primary when blind judging remains possible
• no impressions;
• drowsiness that did not prevent completion;
• an apparent miss;
• contradictions;
• low confidence;
• unexpectedly little or much output;
• non-content procedural violations that cannot inject target-like content.
These remain data.
Invalid for confirmatory primary analysis
Flag and exclude from the confirmatory primary endpoint if:
• target information reached the viewer before freeze;
• target information reached the guide before freeze;
• the viewer knew the candidate pool;
• true target was revealed before freeze;
• no frozen acquisition record exists;
• blind ranking cannot be completed;
• the predeclared target/randomization procedure was broken;
• a content-injecting AI violation occurred.
These trials remain visible in the block ledger and attrition count.
Default split for AI violations
Content-injecting / confirmatory-invalidating
Violations capable of manufacturing or steering report content include:
• suggesting target content/categories/examples;
• implying a specific impression is closer/stronger/more target-like;
• leading or closed target-content questions;
• movement commands;
• imagination instructions;
• resolving contradictions toward a content interpretation;
• replacing viewer wording with semantically stronger wording;
• interpreting/naming analytical or symbolic content for the viewer.
These correspond primarily to operating-prompt rules 1, 3, 4, 5, 6, 7, 8, and 9.
Non-content procedural / retained under intention-to-treat
Examples include:
• extra neutral speech that supplies no target-like content;
• an additional non-leading process reminder;
• silence/pacing departures;
• a freeze-stage procedural wording error that does not alter acquisition content.
Praise or confirmation that adds no specific target content is not automatically target leakage, but must be flagged because it can alter reporting criterion/output volume.
A stricter per-protocol analysis may be added only if its rules were declared before the block began.
No exclusion rule may be invented after outcomes are known.

13. AI Violations / Leakage Channels
Flag if the guide:
• receives or searches for the target;
• guesses target content;
• gives examples;
• asks target-category questions;
• praises impressions;
• says something is accurate;
• paraphrases into stronger language;
• resolves contradictions;
• asks the viewer to “move” to a location/perspective;
• changes prompts because of reported content;
• summarizes acquisition before or at freeze;
• exhibits target-correlated response latency after blindness has failed;
• exhibits target-correlated response length after blindness has failed.
While the guide truly remains blind, target-correlated latency/length cannot be caused by target knowledge. They matter as diagnostic channels if blindness fails.

14. Optional Experimental Branches
These are not part of the core.
Change one branch at a time where practical.
Branch A — Slow Breathing
Candidate:
• approximately 6 breaths/min before acquisition.
Evidence:
• ordinary HRV/baroreflex physiology.
Unknown:
• whether it improves RV.
Do not call it a theta or RV induction.

Branch B — 6-Hz Binaural Beat
Source-derived implementation from Amd 2026:
• left channel: 250 Hz;
• right channel: 244 Hz;
• binaural difference: 6 Hz;
• headphones required;
• exposure: 5 minutes.
The Amd methods explicitly used 250 Hz in the left channel and 244 Hz in the right for the 6-Hz condition.
Amd 2026 reported Study 1 n=101 and Study 2 n=118. These were self-report/state studies, not RV-performance studies and not EEG-entrainment demonstrations.
Classic psychoacoustic work by Licklider/Webster/Hedlun (1950) and Perrott/Nelson (1969) places practical binaural-beat perception in the low-carrier/low-difference regime, commonly summarized as carriers below roughly 1 kHz and beat differences below roughly 30 Hz. H-0.4's 250/244-Hz implementation lies well inside that region.
For fidelity to Amd, the default branch retains 250 Hz left / 244 Hz right. Log the ear assignment. If later testing asks whether lateral assignment matters, counterbalance it as a separately declared sub-condition rather than silently changing it.
H-0.4 default insertion point
For RV testing, use the 5-minute exposure before tasking, then stop the stimulus before Phase 3.
Reason: this isolates the preparation manipulation from acquisition speech/writing and prevents “before vs during acquisition” from becoming an uncontrolled order variable.
If audio is intentionally continued during acquisition, label it a separate branch condition:
B-DURING
Do not pool B-PRE and B-DURING.
Home channel check
Before any B/C/D branch session intended for reporting:
1. play a left-only test sound;
2. verify it is heard only in the expected left ear;
3. play a right-only test sound;
4. verify it is heard only in the expected right ear;
5. check the operating system/device accessibility settings and confirm mono audio is OFF;
6. verify the playback app/device has not collapsed the signal to mono.
Log: STEREO CHECK PASS/FAIL.
If the check fails, do not label the session as a binaural-beat branch.
Volume
Use a comfortable level. If the audio is uncomfortable or prevents ordinary speech from being heard at close conversational level, reduce it or stop.
Evidence from Amd 2026: self-reported calmness/focus only.
Not evidence of: EEG entrainment, cortical theta, or RV enhancement.

Branch C — 6-Hz Beat + Pink Noise
Same 250-Hz-left / 244-Hz-right implementation plus pink-noise background.
Default insertion point: 5 minutes before tasking, then stop unless specifically testing C-DURING.
Do not pool PRE and DURING conditions.
Amd 2026 supplies state/self-report evidence only.

Branch D — Noise-Only Masking
White or pink noise may be tested as environmental masking.
They are acoustically different and must be named separately.
D-WHITE and D-PINK are separate conditions.
Stanchina 2005 supports only a narrow white-noise masking effect in a tiny sleep study. It does not support pink noise or RV.
Run the stereo/device check if a stereo delivery claim is part of the branch.

Branch E — Focused-Attention Preparation
Brief neutral attention practice before acquisition.
Possible ordinary function: attention stabilization.
RV consequence unknown.

Branch F — Focus-10-Like / Hypnagogic Preparation
Historical relevance: Monroe/ERV-style state work.
Official Monroe terminology describes Focus 10 as “mind awake, body asleep.”
This is not a validated EEG label and is not established to improve RV.
Risks/limitations:
• sleep onset;
• spontaneous internally generated imagery;
• microsleeps may not be self-reported.
Therefore self-reported drowsiness rates are lower bounds unless objective monitoring is added.
People for whom sleep-adjacent, dissociative-like, or altered-awareness exercises are medically or psychologically unsuitable should not use this branch without appropriate professional guidance.

Branch G — CRV Structure
Historical/documentary components include staged descriptor work, ideograms, sketches, and AOL declaration.
Causal efficacy remains unestablished.
H-0.4 core imports only:
• raw-data preservation;
• sketching;
• avoidance of forced interpretation.

Branch H — Output Modality
Compare:
1. sketch-first / minimal verbal;
2. verbal-only;
3. sketch-then-verbal.
Rationale: ordinary visual-memory research identifies timing-dependent verbal-overshadowing effects.
Transfer to RV is unknown.

Branch I — Session Duration
Duration is an experimental variable.
Hypothesis: performance/report quality may change with vigilance decrement and time-on-task independent of any altered-state effect.
Do not assume 15–20 minutes is optimal.

15. Public Field-Test Reporting Form
Protocol version: H-0.4
Date/time:
AI provider/model:
Session ID:
Mode: A / B
Shakedown Series ID (Mode A only):
Block ID (Mode B only):
Registration reference (Mode B confirmatory/pooling):
Trial number / intended N:
Target-selection method:
Target-selection commitment method/reference:
Viewer blind to true target? yes/no
Viewer blind to entire candidate pool? yes/no/not applicable
AI guide blind? yes/no
Judge blind? yes/no/not used
Judge ID/code:
Branch condition: core / A / B-PRE / B-DURING / C-PRE / C-DURING / D-WHITE / D-PINK / E / F / G / H / I / other
Audio ear assignment: 250L/244R / counterbalanced sub-condition / not applicable
Stereo check if audio branch: pass/fail/not applicable
Mono accessibility setting checked OFF? yes/no/not applicable
Approximate session duration:
Number of guide prompts after tasking:
Prompt timestamps retained? yes/no
Prompt-cap exceeded? yes/no
Any protocol violation?
Violation class: content-injecting / non-content / none
Any target leakage?
Any drowsiness/sleep?
Zero-content? yes/no
Near-zero-content under predeclared rule? yes/no
Frozen transcript retained?
Sketches retained?
Freeze time:
Scoring method: self-comparison / blind four-choice / other
True target rank if four-choice: 1 / 2 / 3 / 4
Output volume: word count or other predeclared measure
Confidence recorded? yes/no
Validity status: valid / invalid + reason
Block status: ongoing / complete / incomplete / shakedown-only
Post-feedback notes:
Report misses, invalid trials, zero-content trials, and abandoned registered blocks as readily as hits.

16. Stronger Validation Specification
Optional for home use; required for stronger claims
16.1 Positive control
Run sessions in which known target information is deliberately supplied at a controlled level and pass those records through the identical judging pipeline.
Purpose: demonstrate that the scoring apparatus can detect a known signal.
This does not test RV.

16.2 No-Target Control
Run identical sessions with no target bound to the session.
Purpose: estimate ordinary spontaneous mentation and descriptor base rates.

16.3 Target/Decoy Pool Audit
Construct decoys so common descriptors do not systematically favor one target.
Where possible, validate the pool against no-target reports.

16.4 Output Volume Is a Covariate, Not a Success Metric
More words/descriptors mechanically create more opportunities for accidental correspondence.
Therefore:
• log output volume;
• do not interpret “more matches” without considering how much more material was produced;
• compare protocol branches with output volume included as a covariate where the analysis permits.
A branch that increases both output volume and apparent matches has not demonstrated improved discrimination merely because the raw match count increased.

16.5 Confidence, Criterion, and Sensitivity
Where the viewer records per-element confidence before feedback, use it to ask whether higher-confidence elements are more target-specific.
Do not automatically call this a formal signal-detection d' analysis.
Formal signal-detection statistics require a preregistered coding scheme that defines hits and false-alarm opportunities. If such a scheme exists, ROC/SDT methods may separate changes in reporting criterion from changes in discriminability.
Where a no-target corpus (§16.2) exists, descriptor frequencies in no-target sessions can supply the false-alarm reference distribution. A coding scheme defined against that corpus before target-bound trials are scored is what can make a criterion-versus-sensitivity separation possible.
Until such a preregistered coding scheme exists:
• report confidence calibration descriptively;
• compare high- vs low-confidence correspondence under the frozen judging method;
• avoid claiming a formal d';
• avoid the undefined phrase “more target-specific” unless the scoring/coding rule already defines target specificity.
This preserves the important distinction:
A protocol can make people report more without making them discriminate better.

16.6 Blind Transcription and Coding
People who clean transcripts, label descriptors, or code responses should not know the true target where practical.
Do not remove disfluencies or reorder content differently depending on target knowledge.

16.7 Frozen Scorer / Automated Similarity
If using automated semantic similarity:
• freeze model/scorer version;
• freeze prompt;
• process true target and decoys identically;
• use a permutation/randomization null;
• record text length effects.
If an LLM is the judge:
• use a fresh context for each trial;
• randomize candidate order;
• do not use the same model family that guided the acquisition;
• freeze judge prompt/model version;
• log the judge model.

16.8 Judge Dependence / Repeated Judges
Record judge identity for every trial.
Prefer independent judges across a block where feasible.
If judges are reused across trials, do not assume every rank is statistically independent merely because the targets differ. A formal pooled analysis should cluster by judge or use an appropriate multilevel/random-effect model.

16.9 AI Provenance
Record:
• provider/model;
• model version if exposed;
• complete guide prompt;
• complete conversation;
• system instructions if available;
• sampling settings if exposed;
• tool access;
• protocol version.

16.10 Audio Validation
Only necessary for claims about audio.
Formal studies should verify:
• separate L/R delivery;
• no mono collapse;
• channel integrity/polarity as appropriate;
• hearing adequate for dichotic perception;
• playback level with a defined measurement where feasible.
Basic home B/C testing requires at minimum the free L/R stereo check in §14.

16.11 Repeated-Session Practice / Decline Effects
Record trial order within a block.
Do not assume trial 1 and trial 20 are exchangeable.
If enough data exist, inspect whether performance, output volume, confidence, drowsiness, or protocol adherence changes across trial order.

17. Complete Burned / Corrected Ledger
Nothing disappears.
ID
Burned/corrected claim
Disposition
B1
“Pink noise — B/High — Stanchina 2005” and 45–50 dB
Stanchina studied white noise in a tiny sleep sample. No pink-noise or RV support.
B2
Wahbeh .6196 as theta/EEG evidence; 4–7 Hz sourced to it
.6196 was a delta-range, n=8, no-EEG pilot.
B3
“EEG did not mirror advertised band” attributed to .6196
EEG claim removed from .6196; .6201 is the EEG crossover.
B4
“Focus 10 is the most relevant Gateway state for anomalous acquisition”
No supporting performance evidence.
B5
Focus 10 equated with validated hypnagogic EEG/theta signature
Monroe label and independent EEG constructs kept separate.
B6
Cognitive Interviewing = B/High, Risks None, Confounds None
CI increases correct and incorrect details; hygiene subset only.
B7
“Move to the target and describe” as cognitive interviewing
It is a content-generating/movement command, not neutral CI hygiene.
B8
“Visualization seeds/primes the visual cortex”
Replaced by imagination inflation/source-monitoring/base-rate language.
B9
Target IDs are merely historical theater/redundant
No demonstrated paranormal addressing role; opaque IDs remain necessary for ordinary blinding/binding/audit.
B10
Reverse-engineered Monroe frequencies as “independent analyses”
Unofficial community lists; not official Hemi-Sync specifications.
B11
NSDR/body scan = Focus-10-equivalent state
No validated equivalence to a Monroe Focus label.
B12
Lehrer et al. 2020 Frontiers citation with 2014 paper title
Probable hybrid citation; use Lehrer & Gevirtz 2014.
B13
Prep duration 10–20 min attributed to Targ 1974/autonomic quieting
Attribution unsupported; H-0.4 timing is engineering convention only.
B14
Ideogram latency <1 s = measured spontaneous autonomic response
Doctrinal motor instruction, not an autonomic measurement.
B15
“CRV was built/designed to avoid altered states” as fact
Not admitted without explicit primary text; practical CRV/ERV conflict is procedural.
B16
Utts 1996 venue as Statistical Science 11
Probable venue conflation in malformed X0; not used in H-0.4.
B17
“Interfere-E analog” as a category for visualization
Retired; use contamination / ordinary-to-task transfer language.
B18
Lutz et al. meditation taxonomy as warrant for RV acquisition response format
Meditation taxonomy does not validate an RV report format.
B19
Speech rate 140–180 wpm labeled Known
Reclassified Candidate when used; H-0.4 core does not depend on a fixed speech rate.
B20
“4-Hz binaural × focused attention = Conflicting” as established
At most theoretical/unknown interference.
B21
Blind-AI latency treated as target-correlated while simultaneously claiming blindness
If truly blind, target-correlated latency cannot arise from target knowledge; latency becomes diagnostic only when blindness fails.
B22
Any implication of three independent model votes
X0 malformed; X1 synthesis+supplement one lineage; models do not vote.
B23
“Swann 1991” as the settled CRV manual attribution
Replaced with distinct McNear 1985 and Smith 1986 archival records; authorship/date not collapsed.
B24
Ingendoh systematic review used as if it were the primary acoustic-bound experiment
Direct Licklider 1950 and Perrott/Nelson 1969 sources now added.
B25
Oster (1973) Scientific American treated as the primary authority for binaural-beat acoustic limits
Retained only as historical/review context; direct primary psychoacoustic sources now carry the bound.
Critic Findings Investigated and Not Sustained
ID
Critic finding
Resolution
NS1
C3: publish a file's SHA256 inside the same file
Retracted by C4; self-referential hashing changes the file. Detached manifest + manifest sidecar used instead.
NS2
C3: Amd did not justify 250 Hz left / 244 Hz right
Direct published-methods read resolved this; the pair is explicitly specified for the 6-Hz condition.
NS3
C3: confidence analysis wording could imply automatic formal d'
H-0.4 keeps formal SDT conditional on a preregistered coding/false-alarm structure and links it to the no-target corpus.
Critic corrections are archived rather than silently disappearing.

18. Why the Core Is Conservative
Earlier drafts placed more emphasis on preparation state.
After discovery and a forensic audit that returned a NOT-READY finding on the prior evidence base, the surviving conclusion was simpler:
No preparation component currently has direct evidence that it improves remote-viewing performance.
The response was not to upgrade weak evidence. It was to remove evidence-dependent state components from the core.
The H-0.4 core therefore asks:
Can a human produce a frozen, blindly tasked report while an AI guides procedure without becoming a source of target information or narrative contamination, and can a prospective judging pipeline evaluate that report without flexible stopping or post-hoc rescue?
Preparation methods compete only as branches.

19. Source Ledger — Selected but Load-Bearing
This ledger is not exhaustive. It lists sources directly relevant to claims retained in H-0.4.
Binaural beats / acoustics
Amd M. (2026). Effects of self-administered binaural beats on meditative and introspective states. PLOS ONE 21(4):e0335580. DOI 10.1371/journal.pone.0335580. Study 1 n=101; Study 2 n=118.
Ingendoh RM, Posny ES, Heine A. (2023). Binaural beats to entrain the brain? A systematic review of the effects of binaural beat stimulation on brain oscillatory activity. PLOS ONE 18(5):e0286023. DOI 10.1371/journal.pone.0286023.
Licklider JCR, Webster JC, Hedlun JM. (1950). On the Frequency Limits of Binaural Beats. Journal of the Acoustical Society of America 22(4):468–473. DOI 10.1121/1.1906629.
Perrott DR, Nelson MA. (1969). Limits for the Detection of Binaural Beats. Journal of the Acoustical Society of America 46(6):1477–1481. DOI 10.1121/1.1911890.
Wahbeh H, Calabrese C, Zwickey H. (2007). Binaural beat technology in humans: a pilot study to assess psychologic and physiologic effects. J Altern Complement Med 13(1):25–32. DOI 10.1089/acm.2006.6196.
Wahbeh H, Calabrese C, Zwickey H, Zajdel D. (2007). Binaural Beat Technology in Humans: A Pilot Study to Assess Neuropsychologic, Physiologic, and Electroencephalographic Effects. J Altern Complement Med 13(2):199–206. DOI 10.1089/acm.2006.6201.
Stanchina ML, Abu-Hijleh M, Chaudhry BK, Carlisle CC, Millman RP. (2005). The influence of white noise on sleep in subjects exposed to ICU noise. Sleep Medicine 6(5):423–428. DOI 10.1016/j.sleep.2004.12.004.
Breathing / HRV
Lehrer PM, Gevirtz R. (2014). Heart rate variability biofeedback: how and why does it work? Frontiers in Psychology 5:756. DOI 10.3389/fpsyg.2014.00756.
Interviewing / contamination / source monitoring
Memon A, Meissner CA, Fraser J. (2010). The Cognitive Interview: A meta-analytic review and study space analysis of the past 25 years. Psychology, Public Policy, and Law 16(4):340–372. DOI 10.1037/a0020518.
Orne MT. (1962). On the social psychology of the psychological experiment: With particular reference to demand characteristics and their implications. American Psychologist 17(11):776–783. DOI 10.1037/h0043424.
Rosenthal R. (1976). Experimenter Effects in Behavioral Research, 2nd ed. Irvington Publishers.
Johnson MK, Hashtroudi S, Lindsay DS. (1993). Source monitoring. Psychological Bulletin 114(1):3–28. DOI 10.1037/0033-2909.114.1.3.
Garry M, Manning CG, Loftus EF, Sherman SJ. (1996). Imagination inflation: Imagining a childhood event inflates confidence that it occurred. Psychonomic Bulletin & Review 3(2):208–214. DOI 10.3758/BF03212420.
Verbal overshadowing
Schooler JW, Engstler-Schooler TY. (1990). Verbal overshadowing of visual memories: Some things are better left unsaid. Cognitive Psychology 22:36–71.
Alogna VK et al. (2014). Registered Replication Report: Schooler and Engstler-Schooler (1990). Perspectives on Psychological Science 9:556–578. DOI 10.1177/1745691614545653.
CRV / Gateway historical procedure
McNear TM. (1985). Coordinate Remote Viewing Stages I–VI and Beyond. U.S. government archival copy: CIA-RDP96-00788R001000400001-7.
Smith PH. (1986). The Coordinate Remote Viewing Manual. DIA, May 1986; archival/bibliographic record via IRVA.
McDonnell WM. (1983). Analysis and Assessment of Gateway Process. U.S. Army INSCOM document archived in CIA STARGATE collection: CIA-RDP96-00788R001700210016-5.
Monroe Institute public/current materials (accessed 2026-08-23 during C2 source closure) describe Focus 10 as “mind awake, body asleep” and Focus 12 as “expanded awareness.” These labels are treated as Monroe terminology, not independently validated physiological states.

20. Version / Integrity Rule
H-0.2 and H-0.3 remain archived and unchanged.
H-0.4 is the current surgical patch release candidate.
Once H-0.4 is publicly designated as a release, corrections produce H-0.5 or later.
A miss is not rewritten into a hit.
A rejected claim is not deleted.
A failed branch remains in the archive.
Detached cumulative manifest
The H-0.4 release uses a cumulative detached manifest that lists the integrity records for retained research-chain artifacts, including the C1 gate report and the previous manifests.
The H-0.4 manifest contains the SHA256 of its predecessor manifest so the manifest chain is append-only.
Because a manifest cannot contain its own ordinary SHA256 without changing itself, the H-0.4 manifest has a separate .sha256 sidecar containing the manifest's hash.
The manifest is hashed but not cryptographically signed. No signing key/infrastructure is claimed.

21. Provenance and Limitations
1. The original frozen MASTER-RESEARCH-BRIEF-v1.0 remains unrecovered from the retained attachment/library set. Scope compliance with that exact brief therefore remains unauditable.
2. X0 was generated from a malformed prompt in which the Master Brief placeholder was not replaced. It is preserved as an exploratory/control artifact and carries no evidentiary weight.
3. X1-SYNTHESIS-R1 and X1-SUPPLEMENT-R1 belong to one Grok lineage. Their agreement is not an independent vote.
4. The exact session topology between X0 and X1 was not independently reconstructed. No claim of independence between them is used.
5. G1 produced useful mechanisms but contained multiple citation/source-mapping defects. Its evidence grades and numerical parameters were not admitted directly into H-0.4.
6. C1 was a forensic critic pass, had no retrieval tooling, and returned NOT READY.
7. C2 was the targeted source-closure pass performed with external retrieval. It repaired load-bearing source claims and moved uncertain state components out of the core rather than upgrading their evidence.
8. C3 was a forensic critic pass with no retrieval tooling. It found the original C1 audit seeds discharged and identified new prospective-series and criterion/sensitivity gaps.
9. The direct Amd methods verification resolving 250 Hz left / 244 Hz right was performed during the post-C3/C2 source-verification step on 2026-08-23 using the published PLOS ONE article.
10. Direct psychoacoustic primaries by Licklider/Webster/Hedlun (1950) and Perrott/Nelson (1969) were retrieved during the same post-C3 closure step.
11. C4 was a forensic critic pass with no retrieval tooling. It found approximately 95% C3 carry-through, retracted several of its own earlier critic claims, and identified the H-0.4 surgical patch list.
12. Retrieval/tool status for the earlier G1/X0/X1 model passes is not fully reconstructed from the retained execution record and should not be inferred.
13. H-0.4 closes C4's editorial/design items N-1 through N-8, restores the primary acoustic-bound context, connects no-target controls to false-alarm estimation, and repairs the cumulative integrity chain.
14. The protocol remains untested. A clean evidence chain and a clean experimental design do not establish the phenomenon being tested.

22. Current Readiness Boundary
H-0.4 may be used for:
• instrument shakedown;
• casual Mode-A individual trials;
• prospectively registered Mode-B blocks;
• testing whether a blind AI guide can hold the procedural line;
• testing freeze/reporting/judging mechanics;
• generating hypotheses for later controlled work.
Conditions for Mode-B blinding
A Mode-B trial is not considered properly pool-blind unless:
• the viewer is blind to the true target and the entire candidate pool;
• the viewer is not the Target Curator;
• the guide is target/pool blind;
• the judge is blind to which candidate is true;
• target selection/randomization is recorded in committed form before acquisition.
Conditions for pooled confirmatory inference
Do not pool as confirmatory evidence unless:
• every contributing block was publicly timestamp-registered before trial 1;
• complete and incomplete registered blocks remain visible;
• primary endpoint and N were predeclared;
• branch multiplicity was predeclared/controlled or explicitly exploratory;
• invalid/content-injected trials are handled under the frozen rules;
• protocol versions, branch conditions, N, and endpoints are not silently mixed;
• repeated viewers/judges/targets/sites are handled by an analysis appropriate to their dependence.
A public collection should never silently mix:
• H-0.2, H-0.3, and H-0.4;
• complete blocks with cherry-picked individual trials;
• Mode A with Mode B;
• privately declared blocks with publicly registered blocks;
• different branch conditions;
• different declared N or primary endpoints;
• confirmatory and exploratory comparisons.
That is the experiment.