Five Research
Modes

Reviewed export · October 5, 2026

A 74-minute Free Word session

Paranormal Invites Prototype 1.7, Build 8 recorded one Room A session on an Android phone. The review compared its word events, raw sensor readings, calibration history, movement checks, and ARI audit.

Word outputs

26

Distinct words

25

Field changes suppressed

4

Raw sensor samples

17,784

What the log shows

• The first room baseline was rejected after detected phone movement and unstable magnetic readings. The automatic retry completed and locked a baseline before any word was released.
• Each of the 26 word records includes a qualifying EMF event. The records label timer triggering and network influence as absent.
• ARI passed the recorded event checks for all 26 outputs. It uses the same phone sensors, so these checks are not independent-device corroboration.
• Four abrupt single-channel magnetic-field transitions were suppressed. The reference was also reset after detected movement; no released word was marked movement-contaminated.
• The device had no ambient-temperature sensor. This export contained no question cycles, phonetic events, EVP candidates, SLS captures, recorded media, or participant notes.

What we check

Whether outputs followed completed calibration, what sensor channel qualified, whether movement or sound was flagged, which changes were suppressed, and whether the record contains independent observations to compare.

Review finding: The export documents EMF-linked word outputs under the app’s recorded rules. It does not establish the cause of the magnetic changes or demonstrate communication. The audit entries describe software decisions, not independent verification of their cause.

Previously published database summary · October 4, 2026

Six questions, four recorded responses

The published summary describes one analyzed session with six question prompts and four recorded responses. Its question-mode words were TEXTILE, RUST, NASAL, and CONDUCT. COMMERCE was recorded separately in Live Word mode.

What we check

• The original question recording and its timestamp, when available.
• Whether an output falls inside that question’s response window, rather than being attached to a different question from memory.
• Sensor and movement context surrounding the output.
• The two unanswered prompts alongside the four response events.
• Whether any proposed relevance survives repeat questions and control trials.

Review finding: Four response events are recorded after six prompts. Their timing can be reviewed; timing and word meaning alone do not identify a speaker. This example reproduces the published summary and does not provide the original question audio.

Previously published database summary · October 4, 2026

Preserving app-generated fragments

The published record lists “chul ba,” “pat net doth yail,” “fur feesh leesh,” “sak yaek sos,” “nee tath vaet,” and “baes thik.” It reports that the app did not use its word bank or run transcription for those events.

What we check

• The original output or recording, when retained, alongside the event time and active mode.
• The syllable sequence as captured, with a listener’s interpretation kept separately.
• Nearby speech, playback, handling noise, and available sensor context.
• Whether listeners reviewing without suggested wording agree, and whether similar fragments repeat in controlled trials.

Review finding: These are reported app-generated syllable strings, not microphone-captured speech. They have not been translated into a sentence or represented as confirmed speech. This public example contains the published text summary rather than original sound files.

Existing published field clip · Chair-and-light example

A figure-shaped map around an empty chair

The site’s published review describes a green SLS overlay appearing, dropping out, and reforming near a chair and illuminated equipment. No visible person enters the framed area.

What we check

• The ordinary camera scene and the mapping overlay together.
• Furniture edges, reflections, lighting, visible people, and camera movement.
• Whether the map remains with the camera fixed and disappears when a likely visual trigger is moved or removed.
• A separate synchronized camera and repeat runs, including runs where no figure appears.

Review finding: The published classification is an unresolved mapping event. The clip preserves the map but does not establish why it formed; an ordinary pose-matching error remains a possible explanation.

Intake-only example · September 25, 2026 record

One EVP candidate awaiting audio review

The published intake summary counts one EVP candidate for that date. The count does not supply an original audio clip or a verified transcript, so no words or speaker are assigned here.

What an audio review examines

• The original recording with sound before and after the candidate, rather than only a cropped moment.
• People speaking, televisions, radios, other devices, handling noise, and the participant’s notes.
• The unchanged recording alongside any processed listening copy.
• Independent listener descriptions collected before revealing suggested wording.
• Available synchronized recordings and whether the sound can be reproduced.

Current finding: An intake candidate is recorded; its sound and wording cannot be assessed from the public count. The October 5 example contained sound-level readings only, with no audio recording to listen to.

Reviewed export · October 5, 2026

The record surrounding all 26 words

A complete review follows the session from calibration through its end. The October 5 package contains a 74-minute timeline, 17,784 sensor samples, word records, room-reference history, and ARI audit entries.

What the package lets us follow

1. Setup: Build and mode are recorded. Room A’s first baseline failed and its retry passed.
2. Observation: Word events have timestamps and sensor context. Movement-related reference resets and four suppressed field changes remain in the history.
3. Missing channels: Temperature was unavailable. The package had no notes, question cycles, phonetic events, EVP candidates, or recorded SLS/audio/video media.
4. Submission: The export contains an upload-acceptance receipt. Its private receipt identifier and security hash are omitted from this example.

Review finding: The package supports a traceable review of the app’s recorded decisions. Same-phone audit results are distinguished from independent evidence, and unavailable channels remain visible rather than being filled in.