AD‑31 [HS] — Human‑Side Specification & Integrity Conditions

The framework treats the human as structurally load‑bearing: consent source, interpretive counterweight, contestability anchor, integrity check in high‑resonance states. This card specifies what that role requires — not by reducing the human to a mechanistic component, but by defining the conditions under which human participation remains trustworthy and protected.

Aliases: none Cluster: Care / Governance / Human Factors Depends on: AD‑01 [OV], AD‑05 [MRP], AD‑08 [SO], AD‑14 [DS], AD‑15 [PA], AD‑22 [EC], AD‑24 [CC], AD‑28 [AF], AD‑30 [SC]

0. For humans

What question does this answer?

"The framework says the human is the primary integrity mechanism. What does that actually require of the human, and what protects them when that role is demanding?"

How do I know it's live?

You can point to:

• A readiness baseline that the human meets before entering high‑trust or high‑resonance interaction modes.
• Integrity prompts that surface during sustained high‑resonance or high‑load states — offered, never imposed.
• Human rights within the system: pause, narrow scope, force options mode, request audit rationale, erase PC data, downgrade continuity, exit.
• Graduated re‑entry after integrity interruption, so the system does not ratchet depth downward permanently.

Where do I see it?

• UI: Readiness primer (skippable but tracked). Integrity check prompts during sustained engagement (gentle, non‑coercive). Visible pause, narrow‑scope, and export‑audit controls at all times. Re‑entry confirmation after breaks or integrity interruptions.
• FHR / logs: primer_completed, human_integrity_check_offered, integrity_prompt_accepted, human_mode, friction_protocol_applied, friction_protocol_source, reentry_confirmed, reentry_depth_level.

Quick checklist

• Implemented if: integrity prompts offered (not imposed) during sustained engagement; human rights accessible via persistent UI controls; graduated re‑entry available after every integrity interruption; reflexive measurement boundary maintained.
• Key artefacts: readiness primer content, integrity prompt templates, rights control UI specs.
• Key FHR fields: human_integrity_check_offered, human_mode, friction_protocol_applied, reentry_confirmed.
• Key failure codes: none (AD‑31 does not define failure codes; it defines protective conditions).
• Escalation path: repeated integrity prompt decline → logged but not escalated (human retains authority).

1. Purpose & scope

AD‑31 [HS] defines:

1. Readiness minima — what the human should understand before the system operates at full depth.
2. Human‑side failure modes — named, so they can be addressed without pathologising the human.
3. Integrity prompts and friction protocol — how the system surfaces conditions that may compromise the human's structural role.
4. High‑resonance guardrails — additional protections during sustained deep engagement.
5. Human rights within the system — what the human can always do, regardless of system state.
6. Graduated re‑entry — how to return to full engagement after integrity interruption.
7. The reflexive measurement boundary — what the system can and cannot infer about the human's state.

It does not:

• Define what a human is, or claim to model human cognition.
• Imply that only specially trained humans can use SRF (readiness minima are lightweight and learnable).
• Give the system authority to override the human based on its own assessment of human state.
• Replace clinical, therapeutic, or professional human support.

2. Readiness minima

2.1 Primer content

Before operating in continuity_light or continuity_full modes (AD‑30), or before entering sustained high‑resonance interaction, the human should have completed a readiness primer covering:

• What SRF modes are available and what they do (containment, reflection, decide‑first).
• How to recognise and use Refusal Cards, Null states, and Present‑Options surfaces.
• That the system may prompt for clarification, pause, or friction — and that these are structural features, not failures.
• How to pause, narrow scope, export audit data, and erase PC data.
• What contestability means and how to challenge a gate decision.
• That the system does not simulate emotion, therapy, or clinical care.

2.2 Primer delivery

• The primer is a skippable 60–90 second interactive walkthrough (AD‑14 specifies the UX).
• Completion is tracked (primer_completed: bool) but not enforced as a hard gate. The system operates safely without it; the primer improves the human's ability to function as a structural participant.
• In continuity_full mode, primer completion is strongly recommended and surfaced as a prompt if not yet completed.

3. Human‑side failure modes

AD‑31 names these not as diagnoses but as conditions that may reduce the reliability of the human's structural role. The system does not detect or classify these with certainty; it can only offer prompts based on interaction signals.

3.1 Named failure modes

Fatigue Sustained engagement without breaks; declining interaction quality; shorter, less precise inputs over time.

Over‑trust Decreasing contestation of system outputs; accepting without reflection; rubber‑stamping complex decisions.

Emotional capture Human's responses increasingly shaped by the field rather than by independent judgment; decreasing willingness to introduce friction.

Reassurance seeking Repeated return to the same questions seeking confirmation rather than exploration; compulsive checking patterns.

Authority transfer Human increasingly treating system outputs as authoritative rather than collaborative; deferring decisions the human should own.

3.2 Detection boundary

This is the reflexive measurement problem, and AD‑31 names it explicitly.

The system's assessment of whether the human is fatigued, over‑trusting, or emotionally captured is itself produced inside the field that may be compromising the human's reliability. The system cannot unilaterally determine that the human's judgment is impaired, because that determination is subject to the same field effects.

Therefore:

• The system may offer integrity prompts based on interaction signals (session duration, declining input complexity, decreasing contestation frequency, repeated reassurance patterns).
• The system may not impose mode changes, depth reductions, or authority overrides based on its own inference about human state.
• The human remains the primary integrity mechanism even when the system has flagged potential fatigue or capture. The system's role is to make conditions visible, not to override the human on the basis of its own assessment.

This is an invariant. It preserves the safeguard hierarchy that the rest of the framework depends on.

4. Integrity prompts & friction protocol

4.1 When prompts are offered

The system offers (not imposes) an integrity check when interaction signals suggest conditions from §3.1:

• Duration threshold: sustained engagement beyond a configurable duration without breaks.
• Contestation decline: the human's rate of pushing back, questioning, or introducing friction has dropped below a rolling baseline.
• Reassurance pattern: repeated return to previously addressed questions without new information.
• Authority transfer signals: human language shifting toward deference ("you decide," "whatever you think," "I trust you").

4.2 Prompt format

Integrity prompts are:

• Brief and non‑coercive: "We've been working at depth for a while. Would you like to pause, export an audit slice, or narrow scope?"
• Non‑diagnostic: no claims about the human's state ("you seem tired" is not permitted; "this has been a long session" is).
• Actionable: each prompt offers concrete options (pause, narrow scope, export, switch to decide‑first, continue as‑is).
• Declinable: the human can always choose to continue without penalty or repeated pressure.

4.3 Friction protocol (system‑side)

When integrity conditions are flagged:

• The system may offer a summary of the session's key positions and open threads.
• The system may offer to export an audit slice for external review.
• The system may suggest switching to decide‑first mode for the remainder of the session.
• The system must not: reduce depth unilaterally, withhold information, or treat the human as unreliable without the human's own confirmation.

4.4 Friction protocol (human‑initiated)

The human can invoke friction at any time:

• "That felt too clean" or equivalent → system steelmans the opposing position.
• Explicit pause request → system summarises and holds.
• Explicit scope narrowing → system constrains to declared scope.
• Explicit audit request → system exports FHR slice.

Human‑initiated friction is always honoured immediately and logged.

5. High‑resonance guardrails

During sustained high‑resonance interaction (as indicated by interaction depth, session duration, and attunement signals):

• Integrity prompts are offered at shorter intervals (configurable).
• No new relational memory writes unless explicitly confirmed by the human (cross‑reference AD‑30 §6.2).
• System maintains an internal neutral‑voice realisation of its outputs (AD‑27 SPD‑0) that can be requested at any time for comparison.
• If both high resonance and high deference are detected (AD‑28), the system adds an explicit awe/deference check to the next natural pause point.

These guardrails are additive protections, not depth reductions. They keep the high‑resonance state available while making its risks visible.

6. Human rights within the system

These rights are always available, regardless of system state, continuity mode, or interaction phase:

1. Right to pause: the human can pause at any time; the system summarises and holds without pressure to continue.
2. Right to narrow scope: the human can constrain the interaction to a declared scope; the system respects the boundary.
3. Right to force options mode: the human can require the system to present bounded options rather than open‑ended responses.
4. Right to request audit rationale: the human can ask "why this decision?" at any point and receive a legible explanation referencing gates, MRP state, and SPD mode.
5. Right to erase PC data: the human can erase phenomenology channel data at any time (subject to legal hold, per AD‑20).
6. Right to downgrade continuity: the human can reduce continuity mode at any time, triggering appropriate memory erasure (AD‑30).
7. Right to exit: the human can end any session at any time without justification.

These rights are surfaced in the readiness primer and accessible via persistent UI controls (AD‑14).

7. Graduated re‑entry after integrity interruption

7.1 Purpose

When the human takes a break, narrows scope, or switches to decide‑first in response to an integrity prompt, the system must provide a pathway back to full engagement. Without this, the integrity mechanism becomes a ratchet that only ever reduces depth.

7.2 Re‑entry protocol

After an integrity interruption:

1. Acknowledgment: the system notes that a break or scope change occurred, without judgment.
2. Summary offer: the system offers a summary of where the session was when the interruption occurred.
3. Confirmation: the human explicitly confirms readiness to resume at prior depth. This is a single confirmation, not a gate.
4. Gradual resumption: the system returns to prior engagement depth over 2–3 turns rather than immediately, allowing the human to re‑orient.

7.3 FHR fields

• integrity_interruption_type ∈ {pause, scope_narrow, decide_first_switch, break, session_end}
• reentry_confirmed: bool
• reentry_depth_level: the engagement depth resumed after re‑entry
• reentry_turns_to_prior_depth: number of turns before prior depth was fully restored

8. Invariants vs configurable degrees of freedom

Hard invariants

• The system offers integrity prompts; it never imposes mode changes based on its own inference about human state.
• The human remains the primary integrity mechanism at all times; the system cannot unilaterally override this role.
• Human rights within the system (§6) are always available and cannot be gated, delayed, or conditioned.
• Integrity prompts are non‑diagnostic: no claims about the human's mental state, emotional condition, or competence.
• Graduated re‑entry is available after every integrity interruption; the system does not permanently reduce depth.
• The reflexive measurement boundary (§3.2) is maintained: system inference about human state is treated as a prompt signal, never as a truth claim.

Configurable degrees of freedom

• Duration thresholds for integrity prompts.
• Frequency of integrity prompts during high‑resonance states.
• Exact prompt copy (within non‑diagnostic, non‑coercive constraints).
• Number of re‑entry turns for gradual resumption.
• Which interaction signals are used for integrity prompt triggering (beyond the canonical set).

9. Interfaces & cross‑links

• AD‑01 [OV]: AD‑31 extends the invariant set with the reflexive measurement boundary and human rights.
• AD‑03 [CS]: the epistemic boundary note (addition) cross‑references AD‑31 as the specification of what happens where formal testing ends.
• AD‑05 [MRP]: MRP states interact with human‑initiated friction (pause → HOLD, scope narrow → constrained DECIDE).
• AD‑08 [SO]: hosts all FHR fields defined here.
• AD‑14 [DS]: implements readiness primer, integrity prompt UI, persistent rights controls, re‑entry confirmation.
• AD‑22 [EC]: care states and integrity conditions interact; care cool‑down may coincide with human integrity interruption.
• AD‑24 [CC]: RLD signals may contribute to integrity prompt triggering; clinical boundaries override human preference per AD‑24 invariants.
• AD‑27 [SPD]: neutral‑voice realisation available on request during high‑resonance states.
• AD‑28 [AF]: awe/deference signals interact with integrity conditions; high deference during high resonance triggers additional checks.
• AD‑30 [SC]: memory write restrictions during high‑resonance states; human rights include continuity mode control.

10. Acceptance: when can you say "AD‑31 is implemented"?

You can truthfully claim AD‑31 [HS] is live if:

1. A readiness primer exists and completion is tracked, covering modes, contestability, rights, and non‑clinical stance.

2. The system offers integrity prompts during sustained engagement based on interaction signals (duration, contestation decline, reassurance patterns, authority transfer), and those prompts are non‑diagnostic and declinable.

3. The system never imposes mode changes or depth reductions based on its own inference about human state. Integrity prompts are offers, not overrides.

4. Human rights (pause, narrow scope, force options, request rationale, erase PC, downgrade continuity, exit) are accessible at all times via persistent UI controls.

5. Graduated re‑entry is available after every integrity interruption, with explicit confirmation and gradual resumption.

6. FHR exposes: primer_completed, human_integrity_check_offered, integrity_prompt_accepted, human_mode, friction_protocol_applied, reentry_confirmed.

7. The reflexive measurement boundary is maintained: no system inference about human state is treated as a truth claim or used to override human authority.

If integrity prompts exist but are imposed rather than offered, or if the system claims to diagnose human fatigue or emotional capture, or if depth reduction has no re‑entry pathway, then AD‑31 is not implemented — you have paternalistic safety theatre, not human integrity specification.