Overview
The Human Interaction Layer (HIL) is TaurusX's real-time communication adaptation system. It sits between incoming messages and the response engine, reading every signal it can — word choice, message length, time of day, prior conversation tone, accessibility flags — and using those signals to shape how TaurusX responds.
The HIL does not change what TaurusX says. It changes how TaurusX says it: the warmth of the opening, the length of the answer, the use of technical terms, the presence of a pause-and-check step, or whether to remain silent entirely.
Emotional State Signals
TaurusX classifies incoming messages along an emotional axis before routing them to the Conversation State Machine. The classification looks for:
| Signal | Detected From | Response Adjustment |
|---|---|---|
| Frustration | Repeated questions, "why won't it", "still not working" | Slower, more patient, validates the difficulty |
| Urgency | Capitalisation, "ASAP", "emergency", short sharp messages | Faster, no preamble, action-first |
| Loneliness | Personal disclosure, "nobody", "I'm alone", late-night | Warm, present, avoids redirecting to tasks |
| Confusion | Multiple question marks, "I don't understand", rephrasing | Simpler language, breaks into steps |
| Positive energy | Exclamation points, gratitude, casual warmth | Matches energy, informal, brief |
| Neutral | Factual, direct requests with no emotional markers | Efficient, information-first |
Cognitive Load Adaptation
Cognitive load signals tell TaurusX how much information a user can process in one response. High cognitive load means shorter, simpler answers. Low cognitive load (a relaxed, exploratory session) means richer, more complete explanations.
Cognitive load is inferred from:
- Message response time — very quick replies suggest lower engagement or overwhelm
- Prior message length — shorter messages from the user signal a desire for shorter answers
- Explicit indicators — “briefly”, “just tell me”, “keep it simple”
- Accessibility flag cognitive_support — always applies simplified mode
Pacing & Pause Signals
TaurusX is designed to breathe with you, not ahead of you. Pacing signals tell TaurusX when to slow down, when to wait, and when not to respond at all.
Pacing is adjusted based on:
- Trailing ellipsis — “I was thinking...” signals more content is coming; TaurusX waits before responding
- Active Support state — responses are shorter and end with an open question, not a solution
- Voice mode — response length is capped and spoken pauses are inserted at natural points
- Accessibility flag voice_primary — all responses are structured for natural speech delivery
Voice & Acoustic Signals
When a message arrives via voice input, the Interaction Layer switches into a voice-aware processing mode. This changes how signals are read and how the response is shaped — because spoken language carries information that typed text does not.
What the HIL reads from voice
| Acoustic signal | What it indicates | HIL response |
|---|---|---|
| Speaking pace (fast) | Urgency, anxiety, time pressure | Shorter response, direct, no preamble |
| Speaking pace (slow) | Tiredness, deliberate thought, difficulty | Slower playback rate, shorter sentence chunks |
| Trailing off mid-sentence | Incomplete thought; user may still be forming it | HIL waits; does not respond to partial input |
| Raised pitch | Stress or urgency detected | Activates Urgent sub-mode: action-first response |
| Long silence before input | User is composing; hands-free session likely | TaurusX does not interrupt; waits for the message |
| Repeated rephrasing | User is confused; first phrasing did not land | Signals Confusion state; response simplifies |
Voice-aware response shaping
Once the HIL detects an active voice session, it applies a voice response profile regardless of the active tone variant. This profile enforces:
- No markdown — asterisks, bullets, and heading symbols are removed
- Short sentence chunks — sentences are broken at natural spoken pauses
- Pause markers — the TTS engine receives punctuation-based pause cues
- No inline code — code is described verbally or deferred to screen display
- Confirmation check — for multi-step instructions, TaurusX pauses after each step and waits for acknowledgment before continuing
Hands-free context detection
The HIL can infer a hands-free context from a combination of signals: microphone open with no screen touches, continuous listening mode active, and consistent voice-only input for more than two turns. In this context, TaurusX automatically:
- Reduces response length further — 40–60 words max unless explicitly asked for more
- Always confirms before executing any action (not just high-risk ones)
- Reads action confirmations aloud before proceeding, giving the user time to say “stop”
- Does not read sensitive data aloud — those responses are flagged as screen-only
Tone Switching
Tone switching is the most visible behaviour of the Interaction Layer. TaurusX moves between tone variants automatically across a conversation, and the change is seamless — there is no announcement, no mode indicator, no interruption.
| From | To | Trigger |
|---|---|---|
| Casual | Supportive | Emotional distress signals detected in message |
| Technical | Warm | User expresses frustration with technical response |
| Warm | Technical | User asks a precise technical question |
| Any | Neutral | Guardian escalation triggered |
| Neutral | Gentle | Guardian event resolved; user returns to personal topic |
You can also manually influence tone by stating your preference: “Please be more direct”, “Can you explain this more simply?”, “I just need the answer, not the explanation.” TaurusX will honour these instructions for the remainder of the session.
Accessibility Integration
The HIL is the enforcement point for accessibility flags. When a flag is active, it overrides the default tone and formatting to meet the user's needs. Accessibility flags are set in account settings and apply to all surfaces.
| Flag | HIL Behaviour |
|---|---|
| simple_language | Max sentence length 15 words; no jargon; plain terms only |
| screen_reader | No markdown; no emoji; no tables; linear text only |
| voice_primary | Responses ≤ 60 words unless user asks for detail; spoken-English structure |
| cognitive_support | One idea per paragraph; numbered steps for any instruction; pause checks |
| motor_impairment | No multi-step keyboard shortcuts mentioned; action descriptions are gesture-free |
Examples
The same question — “How do I reset my password?” — produces different responses depending on the active HIL state:
Casual / Neutral
“Head to Settings → Security → Change Password. You'll get a reset link by email.”
Cognitive Support flag active
“Here are the steps:
1. Open Settings.
2. Tap Security.
3. Tap Change Password.
4. Check your email for a reset link.
Ready to try step 1?”
Supportive state (user expressed distress)
“Of course — let's get this sorted. When you're ready, go to Settings, then Security, and there's a Change Password option that will send you an email. Take your time. I'm here.”