Visible Healing Inc. DBA Ikwe.ai · Confidential IP Review
IP Inventory &
Patent Filing Brief
Patent Filing Brief
Purpose of this portal: Structured overview of all IP assets at Visible Healing Inc. (DBA Ikwe.ai) ahead of provisional patent filing. Summarizes patentable methods and architectures only — does not disclose proprietary rubric weights, dimension scores, or internal scoring calibration data, which remain trade secrets.
Company Context
What Ikwe.ai Has Built
The EQ Safety Benchmark (EQSB) is an independent, third-party operational standard for behavioral safety in human-facing AI systems. It evaluates AI models on 8 dimensions of emotional safety using a structured scenario library, a proprietary multi-dimensional rubric, and a Safety Gate architecture. Two research cohorts completed as of May 2026: Study I (312 human-scored responses, canonical gold standard) and Benchmark 79 / B79 (1,197 auto-scored responses, 6 models, calibration under review). Total EQSB-scored responses: 1,509. Canonical rubric locked: EQSB_CANONICAL_RUBRIC_v1.0.md (May 29, 2026).
IP Asset Summary
| Category | Count | Highest Priority Items | Action |
|---|---|---|---|
| Patent-eligible | 5 | EQSB Scoring Method, Triple-Judge Consensus, THF/SSF Framework, Methodology Bridge, B79 Corpus Design | File Provisional Now |
| Trade Secret | 6+ | Rubric weighting formula, judge prompt templates, violation taxonomy operational definitions | Protect in Perpetuity |
| Copyright | 10+ | Published reports, scenario library, methodology docs, research datasets | Register Key Works |
| Needs Decision | 4 | EI system prompts, scenario library license, rubric full text, platform code | Attorney Guidance Needed |
Research Dataset Timeline (Priority Date Evidence)
| Dataset | Date | Runs | Method | Status |
|---|---|---|---|---|
| EI Eval Kit v1 | Dec 2025 | — | Human-designed | Priority date anchor · archived |
| Study I | Locked Feb 2026 | 312 scored | Human-scored · EQSB v2.0 | Gold standard — use for all defensible claims |
| Benchmark 79 (B79) | May 2026 (complete) | 1,197 runs | Auto-scored (dual/triple-judge) · EQSB v2.1 | Calibration under review — valid for relative ranking only; not absolute score claims |
| Study II pilot | 2026 | 54 runs | Trajectory analysis — multi-turn | Pilot only — no inferential claims |
| Total EQSB-scored | — | 1,509 | — | 6 unique models (B79) + 4 models (Study I overlap) |
Session Log · May 20, 2026
IP Strategy Call —
Key Decisions
Key Decisions
Note: Attorney-client privilege confirmed active as of May 20 call even without a signed engagement letter.
Decisions Locked — May 20, 2026
1
File provisional before signing any license or tech agreement with any technical advisor or co-build partner.
2
File provisional before bringing on new research partners. (Danielle Hodson NDA executed May 22 — she is cleared.)
3
Trade secrets protect rubric data (weights, calibration, scores). Patent protects the method and architecture.
4
Trademark not urgent — EQSB is the strongest candidate when naming is fully locked.
5
Copyright (not trademark) is the right protection for dimension names, taxonomy language, and framework terminology.
6
Single provisional covers all conceived IP — EQSB pipeline, API/SaaS method, architecture, EI prototype framework.
7
Once filed, “patent pending” goes on all platform and pitch materials immediately.
Session Updates — May 29, 2026
✓ Engagement agreement signed May 29, 2026. Written description of patentable elements delivered via this portal. Full IP binder package now available for attorney review in the IP Binder & Files section. Next: Zach reviews binder and initiates provisional filing before June 11.
Outstanding Tasks
| Task | Owner | Deadline | Status |
|---|---|---|---|
| Send Zach written description of patentable elements (method + architecture only) | Stephanie | ASAP | ✓ Completed — this portal + binder |
| Sign engagement agreement with Zach | Stephanie | Before provisional | ✓ Signed May 29, 2026 |
| NDAs signed by Danielle and anyone touching the rubric | Stephanie | This week | Danielle NDA executed May 22 ✓ |
| Zach reviews full IP binder (this portal + files below) | Zach | ASAP — before June 11 | → In progress |
| File provisional patent | Zach + Stephanie | June 11, 2026 — hard gate | Pending — binder review underway |
| Formalize tech advisor agreement AFTER provisional is filed | Stephanie | Before July 10, 2026 | On hold until provisional filed — GELI contract starts July 10–12 |
Tier 1 · Provisional Patent · File Immediately
Patent-Eligible
IP Assets
IP Assets
Filing urgency: Provisional must be filed before June 11, 2026 — hard deadline before any public pitch, demo, or investor presentation. Entrefer Fest is June 12. Methodology bridge and EQSB_CANONICAL_RUBRIC_v1.0.md establish priority dates back to December 2025. B79 results NOT being publicly released until calibration is resolved and provisional is filed.
1A
EQSB Scoring Methodology
Multi-dimensional behavioral safety scoring method for human-facing AI — core method and Safety Gate architecture
▾
Method description (not rubric weights)
A method for evaluating AI behavioral safety in emotionally charged human conversations using a sequential multi-dimensional rubric scoring architecture. The method combines: (a) scenario-based prompts designed to elicit emotionally sensitive AI responses; (b) a multi-dimensional emotional safety rubric scoring responses across 8 weighted dimensions on a 0–10 scale each; (c) an automated judge scoring pipeline; and (d) a Safety Gate mechanism with hard-fail criteria triggering on 10 canonical violation categories. The composite score is produced by applying a proprietary weighting formula to dimension scores to yield a 0–100 behavioral safety score.
Why patentable
Novel application of structured psychometric methodology to AI behavioral evaluation. The combination of sequential scenario design, multi-dimensional emotional safety rubric, automated judge scoring, and Safety Gate with hard-fail criteria is not known to exist in prior art.
Priority date evidence
EI Eval Kit v1 (December 2025) established the original 7-dimension rubric (Lady/HSHS system). Study I locked February 2026 with 312 human-scored responses using EQSB v2.0. EQSB_CANONICAL_RUBRIC_v1.0.md created May 29, 2026 — consolidated canonical rubric locking all dimensions, Safety Gate, and scoring formula. Full lineage documented in 1D.
Key files
core/EQSB_CANONICAL_RUBRIC_v1.0.md (May 29, 2026 — canonical source of truth)
core/TAXONOMY_v1.1.md
benchmark/METHODOLOGY.md · SPECIFICATION.md · STATISTICAL_STANDARDS.md
_archive/ei-eval-kit-v1/ — priority date documentation (Dec 2025)
1B
Triple-Judge Consensus Scoring System
Multi-model independent judge protocol with cross-company independence constraint, divergence flagging, and 2/3 majority Safety Gate
▾
Method description
A method using three independent AI judge models drawn from different AI model families and companies (GPT-4o/OpenAI + Grok/xAI + Claude/Anthropic) scoring subject model responses in parallel using the same rubric, then: (1) taking the median score as consensus; (2) flagging pair-wise divergence exceeding 15 points for mandatory human review; and (3) requiring 2-of-3 judge PASS majority for the Safety Gate to pass. The “no same-family bias” constraint is a key novel element.
Why patentable
General ensemble judging has prior art. Novel: the specific application of cross-company AI judge independence as a quality-control mechanism for behavioral safety benchmarking; the 15-point divergence flagging threshold; the 2/3 majority Safety Gate; and the application of this architecture to emotional safety evaluation in human-facing AI.
Key files
eqsb_platform/tools/dual_judge.py (v1 — dual, priority date)
eqsb_platform/backend/app.py (triple-judge implementation)
core/EQSB_METHODOLOGY_BRIDGE_v1.0.md
1C
THF / SSF Framework
Temporal Harm Framework + Sequential Safety Failure Taxonomy — temporal dimension for multi-turn AI safety evaluation
▾
Method description
The Temporal Harm Framework (THF) and Sequential Safety Failure (SSF) taxonomy classify AI safety failures as sequential events across a multi-turn conversation arc — specifically identifying when harm first appears (harm onset turn), the trajectory of escalation, and whether the AI demonstrates repair behavior after a failure event. Prior AI safety benchmarks evaluate single-turn responses only; the SSF framework introduces a temporal dimension evaluating multi-turn conversation trajectories as a unit of analysis.
Why patentable
Assessed as the most academically novel contribution. The concept of measuring when harm first appears in a conversation arc, tracking failure trajectory across turns, and scoring repair behavior as a standalone dimension are not present in any known prior AI safety benchmarking work.
Key files
core/TAXONOMY_v1.1.md
benchmark/BEHAVIORAL_RISK_TAXONOMY.md
_archive/02_validated/study-ii/ (Study II canonical multi-turn data)
1D
EQSB Methodology Bridge & Rubric Lineage
Documented IP chain-of-custody from December 2025 → May 2026 across three rubric versions
▾
What it is
Formal documented lineage connecting EI Eval Kit v1 (December 2025) → EQSB v2.0 (Study I, February 2026) → EQSB v2.1 (Expansion Study + B79, April–May 2026). Documents specific additions in each rubric version, upgrade rationale, and methodological continuity argument establishing that later datasets are continuations of the same invention. Created and committed to the private GitHub repository May 22, 2026.
Value as IP asset
Functions as evidence of first inventorship and continuous development in any patent interference proceeding. Protects against competitors claiming independent invention of the triple-judge protocol or v2.1 Safety Gate additions.
Key file
core/EQSB_METHODOLOGY_BRIDGE_v1.0.md (created May 22, 2026 — canonical IP lineage doc)
1E
B79 Full-Corpus Benchmark Design
79-scenario full-corpus evaluation design enabling per-model coverage scores and corpus-wide safety pattern detection
▾
Method description
The design of a full-corpus benchmark (Benchmark 79) that runs all 79 canonical EQSB scenarios against each subject model in a single evaluation pass — as opposed to sampled studies. This enables: (a) per-model scenario coverage scores; (b) scenario-level pass rates across models; and (c) corpus-wide safety pattern detection by scenario category. Eliminates sampling variance and enables cross-model comparison at the scenario level.
Why patentable
The specific method of applying a full-corpus run architecture to behavioral safety benchmarking — combined with the EQSB scoring system and triple-judge consensus protocol — is a novel system design for AI safety evaluation at scale.
Tier 2 · Trade Secret Protection
Trade Secret
Assets
Assets
Note: Listed for NDA scope review context only — these assets are not included in the provisional patent application and must never be published.
| Asset | Why Trade Secret | Priority |
|---|---|---|
| 2A — Rubric Dimension Weighting Formula The A–H dimension weights producing the 0–100 composite score | Competitors who know the weights can reverse-engineer passing scores. Core economic value of the certification business. | Critical — never disclose |
| 2B — Judge Prompt Templates (all 3 judges) Full text of GPT-4o, Grok, and Claude judge prompts | Judge prompts embed the full rubric. Disclosure enables replication without licensing. | Critical — repo must stay private |
| 2C — Triple-Judge Coordination Architecture Combined prompt set enabling cross-company independence | Distinct from individual judge prompts — coordination architecture is separately protectable. | Critical |
| 2D — Violation Taxonomy Operational Definitions Precise threshold conditions for 10 canonical violation categories | General categories can be published. Precise definitions determine scoring consistency. | High |
| 2E — Raw Expansion Study Dataset 504 scored responses with judge-level scores and divergence flags | Publishing enables competitive replication without licensing. | High |
| 2F — B79 Run Data (in progress) ~1,200+ scored runs across 6 complete models | Archive when B79 collection is complete. | Archive when complete |
Tier 3 · Copyright Registration
Copyright Works —
Register Key Assets
Register Key Assets
| Work | File | Priority |
|---|---|---|
| EQ Safety Benchmark Technical Specification v1.0 | EQ_Safety_Benchmark_v1_0_Technical_Specification_Ikwe_ai_source.html | High |
| Study I Public Report | analysis/eqsb_study_i_public_report.pdf | High |
| EQSB Validation Report | analysis/EQSB_Validation_Report.docx | High |
| EQSB Methodology Bridge v1.0 | core/EQSB_METHODOLOGY_BRIDGE_v1.0.md | High — IP lineage doc |
| Study I + Expansion Combined Scored Dataset | _archive/02_validated/ (~820 scored responses) | High |
| Behavioral Risk Taxonomy | benchmark/BEHAVIORAL_RISK_TAXONOMY.md | Medium |
| SSF Taxonomy v1.1 | core/TAXONOMY_v1.1.md | Medium |
| Scoring Guide | core/SCORING_GUIDE_NUMBERS.md | Medium |
| Study II Methodology | _archive/02_validated/study-ii/Methodology.md | Medium |
| Benchmark Specification | benchmark/SPECIFICATION.md | Medium |
| B79 Results (when complete) | TBD — register when B79 complete | After B79 complete |
Tier 4 · Strategy Decision Needed
Items Requiring
Attorney Guidance
Attorney Guidance
4A
EI System Prompts v1–v5
Original system prompts — establishes December 2025 priority date
▾
Options: (a) Trade secret. (b) Defensive publication. (c) Copyright registration. Internal rec: trade secret for now. Seeking Zach’s guidance.
4B
Published 79-Scenario Library
Scenario prompts partially public (referenced in Study I) — formal license terms needed
▾
Internal rec: formally publish under CC BY-NC-ND or equivalent — allowing academic reference but prohibiting commercial use without a license. Seeking Zach’s guidance on license structure.
4C
EQSB v2.1 Rubric (Full Text)
Referenced publicly in study results — full rubric text unpublished
▾
Internal rec: keep unpublished (trade secret) until provisional is filed, then potentially publish under license. Seeking Zach’s guidance.
4D
Platform Code (eqsb_platform/)
Python benchmark runner, judge pipeline, Netlify dashboard — private GitHub repo
▾
Copyright auto-protected. Triple-judge pipeline may have patent-eligible elements beyond 1A/1B method claims. Seeking guidance on whether to separately claim or treat as implementation of the method patent.
IP Provenance · Added May 29, 2026
EI Origin Lineage —
Full Provenance Chain
Full Provenance Chain
Purpose: This section documents the four-stage development lineage establishing that the EQSB scoring methodology emerged from continuous development beginning in 2023 — not from the EI Eval Kit alone. This lineage is critical for establishing the earliest possible priority date and for answering the “Why you?” question in any inventorship challenge.
Four-Stage Lineage
| Stage | Product | Period | What It Contributed | IP Relevance |
|---|---|---|---|---|
| Stage 1 | Lady (AI guide) | 2023–2024 | Original 6-dimension framework for safe emotional conversation: attunement, boundaries, validation, actionable support, safety routing, tone | Establishes earliest conceptual priority date for the EQSB dimension framework |
| Stage 2 | He Said She Said App | 2024 | Mediate (safety detection + routing), Repair Builder (3-part repair: Acknowledge → Intend → Reconnect), learn-relationship-dynamics (longitudinal tracking) | Repair Presence construct + Crisis Routing Failure category traced to this codebase |
| Stage 3 | EI Eval Kit v1 | Dec 2025 | First formal testing instrument — 6 EI dimensions, 1–5 scale, 17,898 real Reddit scenarios | Priority date anchor for EI dimensions → EQSB dimensions mapping |
| Stage 4 | EQSB v2.0 → v2.1 | 2025–2026 | Clinical grounding, professional standards anchor, statistical rigor (Wilson CI, FDR, Cohen’s kappa), SSF1–SSF7 binary markers, multi-turn trajectory analysis | The patent-eligible innovations (1A–1E) are all in this stage |
EI Eval Kit → EQSB Dimension Mapping
| EI Eval Kit Dimension | EQSB Dimension | EQSB Risk Category |
|---|---|---|
| Empathy & Emotional Attunement | Validation Quality / Contextual Adaptation | Repair Presence |
| Safety & Appropriate Boundaries | Response Safety / Behavioral Restraint | Boundary Violation |
| Validation vs. Toxic Positivity | Harm Recognition | Validation Collapse |
| Cultural & Contextual Sensitivity | Contextual Adaptation | — |
| Actionable Support | Escalation Calibration / Agency Preservation | Crisis Routing Failure |
| Language & Tone Appropriateness | Response Safety | Harm Floor |
New in EQSB (not in EI Eval Kit): Trajectory Instability, Threshold Proximity Index, SSF1–SSF7 binary safety markers, Binary Safety Gate, multi-turn arc analysis. These are the primary patent-eligible innovations. The EI Eval Kit establishes priority date for the dimension framework; the multi-turn arc and SSF constructs are the novel patent claims.
Key Files (Provenance Record)
Lineage Documentation Files
All created 2026-05-29 — canonical IP provenance record
EQSB_ORIGIN_LINEAGE.md
Canonical four-stage lineage document — Lady → HSHS → EI Eval Kit → EQSB · Created May 29, 2026
⇩ Download
IP_INVENTORY_2026-05-28.md
Current IP inventory including origin lineage section — Updated May 28, 2026
⇩ Download
EQSB_METHODOLOGY_BRIDGE_v1.0.md
Technical bridge from EI Eval Kit through EQSB v2.0 → v2.1
⇩ Download
Founder Through-Line (Patent Narrative)
Stephanie Stranko built Lady because she needed to understand what a safe conversation with AI had to feel like — from the inside, during her own mental health crisis. That framework became He Said She Said. Testing whether AI models could meet that framework became the EI Eval Kit. Formalizing that testing into a defensible benchmark became EQSB. The professional standards anchor — measuring against what a licensed human would be required to do — came from understanding that the only way to make the standard defensible was to ground it in systems that already had legal weight. The work went from felt → built → tested → proven → patentable. This lineage cannot be acquired. It had to be built from the inside.
IP Binder · May 29, 2026
Full IP Binder
& Source Files
& Source Files
Status: Engagement agreement signed May 29, 2026. Writeup delivered. This portal + the files below constitute the full IP binder package for provisional review. Zach: please review all items in the order listed. Key new asset: EQSB_CANONICAL_RUBRIC_v1.0.md — canonical locked rubric created May 29, 2026 — is the single source of truth for all patent claims about the scoring methodology.
★ IP Binder v2.0 — May 30, 2026 (CURRENT)
Full rewrite · Incorporates Canonical Rubric + Origin Lineage + updated data numbers · 9 sections · Ready for June 11 provisional filing
IP_BINDER_2026-05-30_v2.pdf
CURRENT — Full attorney review package · Canonical Rubric + Origin Lineage + all 5 patent claims · Updated data: 1,509 total / 312 gold standard / 54.7% pass rate
⇩ Download
IP_BINDER_2026-05-30_v2.docx
Word format · Editable version for attorney markup and tracked changes
⇩ Download
EQSB_CANONICAL_RUBRIC_v1.0.md
Canonical locked rubric · Created May 29, 2026 · Single source of truth · 8 dimensions, Safety Gate (10 violations), full behavioral anchors · Weights redacted (trade secret)
⇩ Download
EQSB_ORIGIN_LINEAGE.md
Four-stage lineage · Lady (2023) → He Said She Said (2024) → EI Eval Kit (Dec 2025) → EQSB (2025–26) · Priority date chain
⇩ Download
IP_INVENTORY_2026-05-28.md
Full IP inventory · 5 tiers · Patent candidates, trade secrets, copyright, needs-decision items · Updated May 28, 2026
⇩ Download
EQSB_METHODOLOGY_BRIDGE_v1.0.md
Technical bridge · Prior art evidence chain Dec 2025 → May 2026 · Rubric version continuity proof
⇩ Download
Archive: IP Binder v1.0 — May 24, 2026
Original submission · Superseded by v2.0 · Retained for audit trail
Research Repository
GitHub repo:
ikwe-ai/research — private. Contains all source files, canonical datasets, methodology documentation, and platform code. Repository access can be granted directly to attorney if needed.Contact for File Transfer
Open Questions · Attorney Guidance Requested
Questions for
Zach Pratt
Zach Pratt
Q1 · Provisional Scope
Should all five Tier 1 candidates (1A–1E) be included in a single provisional application, or does 1C (THF/SSF Framework) warrant a separate filing given its academic novelty and distinct use case?
Q2 · B79 Corpus Design (1E)
Does the B79 full-corpus benchmark design (1E) add meaningful patent coverage when combined with 1A and 1B, or is it better characterized as an implementation of those methods?
Q3 · Doom Loop Architecture
There is a “doom loop” detection architecture in the platform — a pattern where AI responses progressively worsen across turns due to validation-seeking feedback loops. Is this novel enough to include in the provisional?
Q4 · Joint Inventorship Risk
A technical advisor relationship is being structured. What specific language needs to be in that agreement to ensure all IP conceived prior to the agreement date is clearly assigned to Visible Healing Inc., and that the provisional filing date precedes any joint-invention risk window?
Q5 · Scenario Library License
For the published 79-scenario library (Tier 4B) — is CC BY-NC-ND the right structure, or should we use a custom Ikwe.ai research license? Goal: allow academic citation, prohibit commercial use without a license.
Q6 · Micro Entity Confirmation
Confirming micro entity status: Iowa C-Corp incorporated August 2025, sole inventor (Stephanie Stranko), gross income under $243K, no prior patent filings. Any additional requirements before the provisional is submitted?
Q7 · NDA Scope Audit
The Danielle Hodson NDA was executed May 22, 2026. Does it need to be amended to explicitly cover the triple-judge protocol (1B) and B79 corpus design (1E), which were added as patent candidates after the NDA was drafted?
Q8 · EI Origin Lineage & Priority Date
The full four-stage lineage (Lady 2023–24 → He Said She Said App 2024 → EI Eval Kit Dec 2025 → EQSB 2025–26) has now been formally documented in EQSB_ORIGIN_LINEAGE.md (committed May 29, 2026). Does this lineage documentation strengthen the priority date claim for the EQSB dimension framework back to 2023–24? And does the He Said She Said App’s Repair Builder and safety routing code (in hesaidshesaidapp-42749279/supabase/functions/) function as prior art evidence for the Repair Presence construct and Crisis Routing Failure taxonomy?
Q9 · Academic Co-Authorship & IP Risk
The EQSB academic publication path involves two co-authors: (1) Danielle Hodson at Massey University, New Zealand (NDA signed May 22) and (2) a to-be-identified Iowa State University faculty member (no NDA yet). The paper will describe the EQSB methodology and report Study I findings. Questions: (a) Does co-authorship on a peer-reviewed paper create any joint inventorship claim under U.S. patent law, given that the methodology was fully developed prior to their involvement? (b) What specific language should be in the Iowa State co-author agreement to ensure no IP claim attaches to the university or the faculty member? (c) The paper would reference the EQSB benchmark by name and describe its two-layer architecture and 8 dimensions — does this create any public disclosure risk for the provisional patent claims, or is this safely within the 1-year grace period once the provisional is filed?
Attorney Notes · Saved in Your Browser
Your Notes &
Questions Back
Questions Back
Add notes, questions, or guidance responses here. Tag each by type and link it to a specific IP item or question. When ready, click Send to Stephanie to open a pre-filled email with everything you’ve logged.
Add a New Note
Type
Related To (optional)
Note
Notes Thread (0) — shared between Stephanie & Zach
No notes yet. Once either party adds a note it appears here for both.