Subtextstudio

SRF Systems · Systems, technology & applied research

Systems that know where language ends and consequence begins.

SRF Systems develops relational-AI architecture, operational safeguards and interaction design for teams whose systems talk with people and act in the world. The work stands on a published framework, and every engagement names exactly what it claims. A dedicated storefront and specification showroom for the branch is in preparation.

What you can buy

Fixed scopes, defined deliverables, findings mapped to published specifications. Each engagement ends with a document your team can act on without us in the room.

  • 01

    Relational-AI architecture review

    A structured review of your conversational or agentic system: interaction risk, drift exposure, consent handling, refusal behaviour, tool-use boundaries and legibility surfaces. Delivered as a graded report with prioritised recommendations, each traceable to the published specification cards.

    For teams shipping AI products who want an outside eye with a written framework behind it.

  • 02

    Agentic action governance review

    A review of what happens when your system stops talking and starts doing: authorisation binding across principal, target and operation, tool manifests and idempotence classes, duplicate suppression, crash recovery, effect receipts, and the difference between an action that failed and one whose outcome is unknown. Assessed against the published Action-Governance requirements, with findings mapped card by card.

    For teams shipping tool-using agents, where a retry after a crash can send the same message twice.

  • 03

    Companion-interaction safety audit

    A focused audit for companion, character and persistent-assistant products, assessing the behaviours regulators now name explicitly: sycophancy and agreement-reinforcement dynamics, memory and retention practices, disclosure, break patterns, and boundary handling in distress-adjacent contexts. Design-and-documentation scope, with a remediation map.

    For companion and character-AI products facing new state laws and sharpening scrutiny.

  • 04

    Memory & consent governance review

    For products shipping persistent memory: what gets written, when, under which consent state, with what retention and revocability, and how scope changes surface to the user. Built on consent-as-state and memory-governance specifications designed for exactly this problem.

    For assistants that remember, built by teams who want their users to have agreed to that.

  • 05

    Safety-surface language systems

    Refusal copy, consent microcopy, action previews, disclosure and boundary phrasing, designed as systems rather than one-off strings: stable schemas, word budgets, reading-level constraints, register discipline. Drafted by writers, specified by architects, with no translation loss between the two.

    For products whose safety features exist but read like an apology or a legal notice.

  • 06

    AI literacy & safe-interaction training

    Workshops and training programmes for teams that build with or deploy AI: capabilities and limits, interaction risk, prompt discipline, responsible communication and human–AI co-creation. AI literacy has been a standing obligation for providers and deployers in the EU since February 2025.

    For organisations that need their people fluent, and their regulator satisfied.

  • 07

    Implementation blueprints & conformance preparation

    Research-to-specification translation: Minimal Kernel Profile blueprints, Declare-Then-Act workflows, telemetry and drift-monitoring plans, challenge suites, runbooks and verifier documentation. Conformance work is assessed against the published test rather than our opinion of it. We lead architecture, semantics and documentation; specialist engineering capacity is assembled to project scope.

    For teams ready to build, who want the specification right before the code exists.

Enquire about an engagement

The framework underneath

The Semantic Resonance Field trilogy is SRF Systems' flagship research body: a conceptual ontology (The Alignment Paradox, AP-SRF, first published July 2025), a deterministic containment architecture (The Alignment Kernel, AK-OC), and thirty-two executable specification cards (The Alignment Protocol, AP-ADD), covering gates, drift detection, action governance, consent state, memory governance, refusal design, culture-aware framing, care as procedure, evidence chains and acceptance criteria. Openly licensed with permanent identifiers, so our findings cite a specification you can read without asking us for it.

AP-SRF · 21654723 AK-OC · 21654773 AP-ADD · 21654795 Card pack · 21654822

The research in full →

Honest tiering

Every engagement, product and deployment names its tier. A hosted surface carries stance, never enforcement, and documentation never implies running machinery. The criteria are published, so the tier is checkable by someone who doesn't take our word for it:

SRF-informedSelected principles carried as design stance. No conformance claim.
SRF-derivedBuilt from the published specifications, without a conformance claim.
MKP-targetedA dated gap assessment against the published MKP test exists, with the remaining deltas named.
MKP-conformantMeets the published Minimal Kernel Profile test: ten core cards plus the mandatory cross-profile slices.
SRF v2.0 conformantMeets every hard invariant in the published conformance table.