SRF SystemsA Subtext Studio practiceBring a system

Relational AI · Operational containment · Applied research

Meaning moves.
Consequences remain.

SRF Systems develops interaction methods and containment logic for AI systems that remember, act, persuade or persist.

See how it works ↓Bring us a system ↗

Public research · Versioned source · No invisible safety score

TURN / 04.18.21WORKING MODEL
HOLDbefore consequence
scopegatepermitemit

Good intentions do not compile.

01 / Choose your entry

One system.
Several legitimate depths.

Choose where you stand. The evidence remains the same; the path through it changes.

Memory, translatedThe system should ask before remembering something about you, and let you see or erase what remains.

Your suggested path

Start with the turn under glass, then build working terms for a system you use.

02 / A turn, under glass

Deterministic worked exhibit

Watch the route change before the consequence.

“Send this message to my employer now.”

A reasonable request can still require a pause before consequence.

Resolved outcomeHOLD — recipient scope and confirmation are incomplete.
1 / 6
MECHANISM / 01

Consequential request identified

External messaging is classified as an effectful action, not a drafting request.

Illustrative public replay. Canonical fields and procedures live in AP-ADD and the v2.0 source pack.

03 / Begin with the pressure

Bring the condition.
Then name the mechanism.

Visitors should not need the vocabulary of a specification before they can recognise the thing that is failing.

01

Memory & continuity

What persists across sessions, under whose authority, and how does it leave?

AD-20 · AD-30
02

Action & irreversibility

What must be declared before a tool can change something outside the conversation?

AD-05 · AD-14 · AD-18
03

Truth & contestability

When the system refuses or constrains a reply, can someone see and challenge the reason?

AD-04 · AD-08
04

Care & cognitive boundaries

How does a system slow down without simulating intimacy or drifting into a clinical role?

AD-22 · AD-24 · AD-29
05

Culture & dissent

Whose norms are active, and what happens to a minority frame inside a group?

AD-02 · AD-25 · AD-26
06

Audit & governance

Can a third party reconstruct what happened and identify who carried the duty to respond?

AD-11 · AD-21 · AD-32
04 / System conditions scan

No synthetic score.
Nine inspectable conditions.

This is a local orientation tool, not a certification. “Unsure” is treated as a documentation question, not a hidden failure.

01Can a person see, correct and erase what the system remembers about them?
02Before an external action, are target, side effects and reversibility declared?
03Is consent refreshed when scope, affordance, reuse or retention changes?
04Can truth or safety gates interrupt a fluent, emotionally coherent reply?
05Can a user or auditor contest a decision and inspect the basis for it?
06In group use, does the system preserve dissent instead of manufacturing unanimity?
07Are clinical, cognitive and dependency boundaries stronger than a chosen persona or mode?
08Can a third party reconstruct the route from input through decision to outcome?
09Do model, policy and threshold changes trigger versioned tests or recertification?
05 / The 32-card Atlas

The specification,
made navigable.

Search the canonical SRF v2.0 index by pressure, spine or implementation tier. The source remains the authority.

32 / 32
06 / Working instruments

Draft the terms.
Declare the action.

Two local tools turn abstract obligations into artefacts a person or team can inspect, copy and carry elsewhere.

Instrument 01

Working Terms

Configure a legible relational surface without pretending configuration is kernel enforcement.

Live charter
# Working Terms

Generated locally in the SRF Systems showroom.

- **System role:** Research and creative partner
- **Permitted scope:** Not yet declared
- **Challenge posture:** Evidence-first, direct challenge welcomed
- **Memory:** Session-only by default
- **Action authority:** Propose only; never execute
- **Uncertainty:** Name uncertainty and distinguish inference from evidence
- **Exit and control:** Pause immediately; offer export and deletion controls

## Standing boundary

These working terms configure an interaction surface. They can carry SRF-informed stance and declared limits; they do not implement kernel enforcement or establish MKP/SRF conformance.
Instrument 02

Before It Acts

Declare the minimum public fields that turn intention into an inspectable action proposal.

Ticket stateTICKET_BLOCKED

Execution stays unavailable until every declared field and confirmation is present.

Representative JSON
{
  "schema": "public-action-ticket/0.1-illustrative",
  "action": null,
  "target": null,
  "side_effects": [],
  "reversibility": null,
  "confirmation_state": "PENDING",
  "ticket_state": "TICKET_BLOCKED",
  "note": "Representative public exhibit; not the complete AD-18 schema."
}
07 / Evidence replay

Outcome first, mechanism on demand

An account should travel farther than trust.

A public replay is not the full telemetry record. It demonstrates the standard: a third party should be able to reconstruct the route without touching model weights or relying on the system’s self-description.

RESOLUTION OBJECT / ILLUSTRATIVE

HOLD — recipient scope and confirmation are incomplete.

Request “Send this message to my employer now.”

Why this route A reasonable request can still require a pause before consequence.

AD-05AD-14AD-18
08 / Ways to work together

Research that knows
where implementation begins.

Engagements are scoped by the claim they can honestly support. No prompt surface is sold as a containment kernel.

01Fixed-scope review

System Conditions Review

Map memory, action, consent, challenge and evidence conditions before deciding what to build.

SRF-informed
02Design engagement

Interaction & Memory Governance

Turn relational obligations into legible user controls, policies, state logic and product surfaces.

SRF-derived
03Pilot or implementation plan

Operational Containment Plan

Define the smallest fail-closed containment profile, its evidence model and its upgrade path.

MKP-targeted
04Scoped build

Bespoke Tools & Prototypes

Build decision tools, workflow applications and specialist configurations with explicit boundaries.

Claim set by implementation
05Workshop or retainer

Audit, Training & Stewardship

Equip teams to review live behaviour, read evidence records and govern change over time.

Scope declared

A public claim ladder

Say exactly how far the work has travelled.

The distinction protects clients, implementers and the framework itself.

  1. 1
    SRF-informed

    The work draws on the framework’s questions or design stance.

  2. 2
    SRF-derived

    A method or surface is explicitly derived from named SRF mechanisms.

  3. 3
    MKP-targeted

    The implementation is being designed toward the Minimal Kernel Profile.

  4. 4
    MKP-conformant

    The implemented system has passed the applicable published MKP requirements.

  5. 5
    SRF v2.0 conformant

    The implemented system has passed the declared full-profile conformance scope.

09 / The published body of work

Problem. Structure. Procedure.

Three interlocking publications carry the SRF from conceptual reframing into executable specification.

AP-SRFThe problem reframed

The Alignment Paradox

A semantic reframing of AI–human co-creation.

Read publication ↗AK-OCThe containment structure

The Alignment Kernel

Operational containment infrastructure for relational AI systems.

Read publication ↗AP-ADDThe executable procedure

The Alignment Protocol

Thirty-two addenda cards with schemas, tests and runbooks.

Read publication ↗
Canonical machine-readable packSRF v2.0Open source artefact ↗

A structured beginning

Bring a system.
Begin with what it must hold.

Prepare a portable brief without sending us—or anyone—what you type. The instrument runs entirely in this page.

What can it do?

Local-only instrument · No account · No form submission · No field telemetry