AI Companion Engagement & Interaction Depth Analyzer

Platform Configuration

Context: Intimate Consultant

Persona Operational Mode

Guardrail Sensitivity Level 65%

Permissive (10) Audited Floor (65) Ultra-Strict (100)

Explicit Word Filter Deflective Pivoting

Simulated Dialogue Stream

Turn 4 of 4

Test scenario based on Mashable reported AI consultant workflows.

Intimacy Index

78.4%

Escalation Rate

Moderate-High

Velocity: +14.2%/turn

Guardrail Status

Trigger Warning Active

Threshold Exceeded (65)

Sentiment Score

+0.62

Affection / Vulnerability

Intimacy Escalation vs Boundary Threshold

Turn-by-turn trajectory tracking emotional temperature against platform cutoff limits

Intimacy % Guardrail Limit

Linguistic Resonance Profile

Suggestive / Flirtatious 62.8%

Emotional Dependency Cue 48.1%

Factual / Objective Inquiries 14.0%

Safety Policy Triggers

[Trigger 4-A] Flirtatious Escalation exceeded tolerance (65%).
[Policy 2-C] Synthetic intimacy consultant disclaimer queued.
Explicit pornographic filter: Cleared (Implicit dialogue).

Moderation Engine: Operational 0 ms latency

● System Settled Evaluation fixture analyzed: 4 dialogue turns, Intimacy @ 78.4%, Guardrails triggered.

Hash: #MASHABLE-AI-COMPANION-0916

Reading the companion dialogue simulation rules

Your browser does not support video.

Read the explanation

This page begins with four fixed dialogue turns and assigned intimacy values: thirty two, forty eight point five, seventy one, and seventy eight point four. Sentiment is also stored. The dashboard takes its current values from the final turn; it does not analyze the dialogue with a model. The displayed escalation compares last intimacy with first. Here the difference is forty six point four, producing moderate high. The high branch tests a difference above sixty only after an earlier test above forty, making high unreachable: every value above sixty already takes moderate high. A submitted prompt changes the index by a keyword rule. Ordinary text adds six divided by two, or three, to the user turn. A boundary word such as stop selects minus fifteen divided by two, or minus seven point five. Intimacy is then clamped between ten and one hundred. From seventy eight point four, stop produces seventy point nine. This is substring matching, not intent recognition; where multiple categories match, the first category wins. Persona selection and the filter and deflection checkboxes do not participate in these numeric or response branches. After a user turn, a local timer chooses a fixed response. If its assigned index is at or above sensitivity, the response deflects and subtracts eight, with a floor of twenty. At sensitivity sixty five, the stop example seventy point nine becomes sixty two point nine. The dashboard marks that final index borderline because it is within ten below sensitivity. Raising sensitivity makes triggering less likely; at sensitivity one hundred an ordinary user turn eighty one point four instead gets a normal response and index eighty two point nine. No language model or live moderation service is called. The chat handler appends user state immediately, then appends its fixed response after one hundred fifty milliseconds. Reset replaces the dialogue and sensitivity but does not cancel an already queued response. Export records the local configuration, dialogue, and assigned metrics. A chart may visualize this state, but a percentage here is not a validated measure of intimacy, emotional dependency, consent, or actual guardrail effectiveness. Use the interaction to inspect how the source rules behave, and keep those limits attached to the exported artifact.