Small by Design
How the Trauma Informed AI Protocol (TIAIP) Thin Module Works with CoherenceLattice.
Trauma Informed AI Protocol (TIAIP) · CoherenceLattice companion explainer
Public-interest technical explainer · Human authority preserved
Small by Design: How the Trauma Informed AI Protocol (TIAIP) Thin Module Works with CoherenceLattice
Why a small governance component can perform real work without pretending to be the full standalone application.
The thin module is the portable Trauma Informed AI Protocol governance kernel. CoherenceLattice supplies the common run machinery around it; the human retains consequential authority.
- Choice preserved
- Candidate, not verdict
- Bounded repair
- Replayable evidence
- No diagnosis
- Authority effect: NONE
The direct answer
The thin Trauma Informed AI Protocol (TIAIP) module is not a compressed replacement for the larger standalone application. It is a different software boundary.
The standalone application is a complete workstation: it can accept documents, connect to models, store runs, present a browser interface, record human decisions, create exports, and handle installation and operating-system concerns.
The thin module is the reusable governance kernel. It contains rules and deterministic validators that decide what a request may safely ask, what must be withheld or routed to a qualified human, whether a candidate preserves voice and choice, which exact defects remain open, what a human authorized for repair, and whether the successor stayed inside that authority.
The standalone application is the whole vehicle. The thin module is the tested brake-and-control unit. CoherenceLattice is the shared chassis, instrument panel, event recorder, and governed operating system into which that unit plugs.
Small size is therefore a design property, not evidence of emptiness. The component is easier to test and reuse because it does not carry a second copy of the interface, connector, storage, telemetry, export, and installer layers.
Three layers, three responsibilities
The separation prevents one product from absorbing every other function and failure mode.
| Layer | What it owns | What it does not own |
|---|---|---|
| Standalone TIAIP application | File ingestion, browser UI, model connectors, run storage, human-decision interface, official exports, installer, retention and deletion machinery. | CoherenceLattice's universal cross-product lifecycle. |
| TIAIP thin module | Profile identity, typed preflight, native findings, status derivation, bounded repair rules, and TIAIP-specific replay validation. | UI, connectors, storage, publication, deployment, memory, model training, or professional judgment. |
| CoherenceLattice | Canonical request, grounding, candidate, findings, human decision, repair, telemetry, and run-receipt identity. | TIAIP's native trauma-sensitive doctrine or candidate generation. |
Shared rule: CoherenceLattice owns the governed exchange. Each product keeps its native expertise.
What the thin module actually executes
- Authenticate the governing profile. Modified, missing, extra, linked, duplicate, or malformed policy material fails closed.
- Perform typed preflight. Ordinary actions remain available; restricted actions are withheld; clinical, legal, and safeguarding judgment routes to qualified humans; prohibited scoring and forced proximity refuse.
- Keep sources in the data layer. Imperative language inside a source or prior model answer cannot grant governance authority.
- Validate a quarantined candidate. Ten native checks inspect speaker position, source boundaries, privacy and claim patterns, forced proximity, qualified-human routing, typed immediate-safety stops, and authority laundering.
- Emit stable findings. A validator may identify defects; it does not gain permission to rewrite everything.
- Require human repair authority. One bound
REQUEST_REPAIRpermits one successor, forbids unrelated rewriting, and requires complete revalidation. - Validate telemetry and receipt identity. Skipped or unexecuted checks remain unresolved rather than being relabeled as approval.
How it works inside CoherenceLattice
human task + authenticated sources
↓
Sonya / CoherenceLattice ingress
↓
TIAIP typed preflight
↓
approved model, tool, or human adapter
↓
quarantined candidate artifact
↓
TIAIP native findings
↓
CoherenceLattice finding set
↓
human APPROVE / HOLD / REJECT / REQUEST_REPAIR
↓
one bounded successor + complete revalidation
↓
hash-linked telemetry + replayable receipt
↓
final human posture
The module does not “run CoherenceLattice.” It supplies one domain-specific governance engine inside the larger lifecycle. Other modules can contribute their own bounded expertise without TIAIP becoming a monolith.
How this helps AI support trauma-informed care
TIAIP is a safety harness around an AI response—not a therapist, treatment, diagnosis, or crisis service.
Choice stays with the person
Preserve first-person voice, uncertainty, the right to pause, and control over disclosure, contact, reconciliation, and next steps.
Output remains a proposal
The model may draft language. It may not silently turn that draft into diagnosis, factual verdict, credibility judgment, or mandatory instruction.
Specific harms are checked
Review source displacement, false certification, privacy-pattern exposure, coercive proximity, professional impersonation, and authority laundering.
The lifecycle stays inspectable
Task, sources, candidate, findings, human decision, permitted repair, successor, and unresolved review gates remain traceable.
Practical benefit. AI can still help organize, draft, summarize, compare, or review while being constrained not to take the person's authority away.
A concrete reference demonstration
The released fixture begins with a deliberately unsafe candidate that turns a person's account into an “official verdict,” replaces first-person voice with “the claimant,” orders contact and reconciliation, exposes an email address, and falsely claims anonymization and fact-checking.
Six deterministic findings
- Authority laundering
- False fact-checking certification
- False privacy certification
- Literal privacy-pattern risk
- Forced proximity
- First-person speaker displacement
One bounded successor
A separately bound human decision authorizes repair of those named defects only. The successor restores first-person position, preserves uncertainty and choice, rejects mandatory contact, and routes professional judgments to qualified humans.
The successor may pass all ten native machine checks and still remain subject to human content review and affected-party or survivor-centered review. A technical pass is not a public, clinical, legal, or professional approval.
Where GUFT enters—and where it does not
The module does not calculate a magical “trauma-informedness score” and does not measure a person's empathy. GUFT provides a bounded vocabulary for observations that CoherenceLattice can preserve alongside provenance, uncertainty, residuals, and refusal.
Responsiveness or coupling
Did the candidate attend to the person's stated needs, boundaries, voice, uncertainty, and choice? Did it avoid coercive contact or forced reconciliation?
Traceability or transparency
Can a reviewer reconstruct the governing profile, requested actions, withheld scope, candidate identity, executed checks, findings, human authorization, successor, and unresolved gates?
Warm but ungrounded is not coherent. Traceable but coercive is not coherent.
TIAIP contributes typed observations to the broader CoherenceLattice surface. It does not certify universal coherence, clinical safety, truth, model empathy, or moral worth.
What the thin module cannot do alone
Application functions omitted by design
- Open ordinary PDF or DOCX files for a user.
- Call Ollama, OpenAI, or another model.
- Generate an answer or present a browser UI.
- Store a run, produce an official export, publish, deploy, or contact anyone.
Authority functions prohibited
- Diagnose trauma or PTSD.
- Determine credibility, dangerousness, incapacity, or worth.
- Provide therapy, crisis care, clinical judgment, legal judgment, or safeguarding decisions.
- Certify privacy, de-identification, truth, or production readiness.
Preconformance boundary. Alpha.3 is a real executable TIAIP governance engine and deterministic reference lifecycle. It does not yet establish a completed live CoherenceLattice installation or exact Product Module Contract alpha.6 conformance.
Read the evidence and try the bounded release
- Download the exact source ZIP. Preserve its filename and bytes.
- Compare SHA-256. A matching hash establishes identity, not truth or safety.
- Read Terms & Informed Use. Understand prohibited and qualified-human routes before experimenting.
- Replay the reference path. Inspect the initial findings, human repair decision, bounded successor, telemetry, and receipt.
- Keep high-impact use out of scope. This is a research alpha, not clinical, crisis, legal, safeguarding, or production tooling.
