AI Transparency & Data Disclosure

The Stroop LLC dba Strooply  ·  Effective Date: May 20, 2026  ·  Page Version 2.0

This page is maintained in accordance with California Civil Code § 3111 (AB 2013, effective January 1, 2026) and Washington state AI transparency expectations. It should be read in conjunction with our Terms of Service and Privacy Policy.

Jump to section

1. Our Role: Deployer, Not Developer

The Stroop LLC ("Strooply") is a deployer of third-party generative AI systems, not a developer of foundational AI models. We do not train, pre-train, or fine-tune large language models or multimodal models from scratch. We integrate commercially available models provided by Google LLC into our Service via API access.

Under AB 2013, the obligation to document training datasets rests primarily with the model developer. As a deployer, we satisfy our transparency obligations by: (1) identifying the specific models integrated into our Service; (2) linking to the developer's published training data documentation; and (3) disclosing how we use these models and what safeguards we apply.

2. Generative AI Systems Integrated Into Strooply

Strooply utilizes a tiered architecture of generative AI models to provide its "Agentic Vision" research, automated listing generation, and financial analytics. The following Google Gemini models are currently integrated into the Service:

Model Status Primary Use
Gemini 2.5 Flash Generally Available Listing generation, market analytics, OCR receipt extraction
Gemini 2.5 Flash-Lite Generally Available Lightweight classification, category suggestion
Gemini 3.1 Flash Preview Enhanced listing generation, complex item research
Gemini 3.1 Flash-Lite Preview Fast-path object identification, Item Specifics extraction

Preview Models: Models designated as "Preview" are pre-general-availability releases. Their technical specifications, capabilities, and terms may evolve. We integrate Preview models to improve Service quality but maintain Generally Available models as fallbacks to ensure continuity.

Model Update Policy: When we add, replace, or upgrade a model in production, we will update this page within 14 days. If a model change materially alters how your data is processed (e.g., a change in the developer's data retention or training data usage terms), we will notify you via the app interface or email before the change takes effect, consistent with our Privacy Policy's sub-processor update commitment.

3. Training Data Documentation (AB 2013 Compliance)

In accordance with California Civil Code § 3111, documentation regarding the training datasets for the foundational models integrated into Strooply is publicly accessible via the model developer's (Google LLC) official reports:

Link Verification Notice: We verify these links quarterly. If a link is broken or the developer has reorganized its documentation, please contact us at [email protected] and we will provide an updated reference within 5 business days.

4. High-Level Summary of Training Datasets

Per the disclosures provided by the model developer (Google LLC), the datasets used to train the foundational models integrated into Strooply include the following characteristics:

Attribute Description
Sources Public web data, licensed content, code repositories, and multimodal media
Purpose To enable reasoning, creative generation, visual object identification, and language understanding
Scale Estimated in the trillions of multimodal tokens
Data Types Interleaved text, high-resolution imagery, audio, video, and source code
IP Status Datasets include a mix of public domain material and intellectual property (copyrighted and trademarked works)
Licensing Google licenses and purchases specific high-quality datasets for training purposes
Personal Information Datasets may include publicly accessible personal information as defined by the CCPA (Cal. Civ. Code § 1798.140(v))
Aggregate Data Datasets include aggregate consumer information
Data Cleaning Data underwent extensive deduplication, safety filtering, and quality modification
Time Period Data collection is ongoing; current models use data through early 2026
Synthetic Data Models may utilize synthetic data for reasoning and logic refinement

Important: This summary is derived from Google's published documentation. The Stroop LLC does not independently audit or verify Google's training data disclosures. If Google updates its documentation, this summary will be updated accordingly.

5. Strooply's Proprietary Logic: No Additional Training Data

Strooply's "Agentic Vision" workflows and proprietary prompt logic do not constitute a separately trained generative AI system. Specifically:

This prompt logic is a Trade Secret of The Stroop LLC (protected under RCW 19.108) and is not disclosed publicly. However, because it does not involve additional model training, it does not trigger independent training data documentation obligations under AB 2013.

6. How AI Is Used in the Service & Human Oversight

Scope of AI-Assisted Functions:

Feature AI Function Human Oversight Requirement
Listing Generation Generates titles, descriptions, Item Specifics, and category suggestions from photos Mandatory review before publish: you must verify and edit all fields
Market Analytics Estimates average sold prices, demand velocity, and sell-through rates Advisory only: presented as estimates, not guarantees
Receipt Scanner (OCR) Extracts cost-basis, dates, and line items from receipt images Mandatory verification: you must confirm all extracted values
Sourcing Research Provides item identification, brand recognition, and value indicators Advisory only: you must independently verify before purchasing decisions
Profit Calculator Computes estimated ROI, net profit, and fee projections Advisory only: estimates only, not audited figures

The "Pilot Rule" (Human-in-the-Loop): In accordance with Washington state expectations regarding automated decision-making and our Terms of Service (Section 4), Strooply maintains a mandatory human-in-the-loop architecture. No AI-generated content is published to eBay, and no financial figure is finalized, without your explicit review and approval. You are the "Pilot in Command": the AI is your co-pilot, not an autonomous agent.

AI Does Not Make Consequential Decisions: Strooply's AI systems do not autonomously publish listings without your approval; make purchasing decisions on your behalf; file taxes or submit financial documents; determine your eligibility for any service, credit, or benefit; or make decisions that produce legal effects concerning you.

7. Output Safety & Quality Measures

To mitigate risks associated with generative AI (including hallucinations, inaccuracies, and inappropriate content), we apply the following safeguards:

(a) Input Filtering: User-submitted images are screened for content that violates our Terms of Service before AI processing.

(b) Structured Output Constraints: AI responses are constrained to structured schemas (JSON with defined fields) to reduce free-form hallucination and ensure outputs conform to eBay's listing format requirements.

(c) Confidence Indicators: Where feasible, the Service flags low-confidence outputs (e.g., uncertain category matches or price estimates with high variance) to prompt additional user review.

(d) Model Fallback Logic: If a primary model returns an error, malformed output, or fails safety checks, the system retries with an alternative model or returns a graceful failure rather than surfacing potentially harmful content.

(e) Prohibited Output Categories: Our system prompts explicitly instruct models not to generate: medical claims, legal advice, guaranteed financial returns, discriminatory content, or content that infringes known intellectual property.

(f) Continuous Monitoring: We monitor AI output quality through user feedback signals (edits made to AI suggestions, error reports) and periodically review output samples to identify systematic issues.

(g) NIST AI RMF Alignment: We strive to align our AI risk management practices with the NIST AI Risk Management Framework (AI RMF 1.0), particularly the Govern, Map, Measure, and Manage functions as they apply to a deployer of third-party AI systems.

8. Data Siloing & User Privacy Guarantee

Your data is not used to train public AI models.

Strooply accesses Google Gemini models primarily through Google Cloud Vertex AI using enterprise-grade configurations. This ensures that your inputs, including photos of your inventory, purchase costs, sourcing locations, and listing details, are:

Google's API Data Retention: Under Vertex AI enterprise terms, Google may retain API inputs and outputs for up to 30 days solely for abuse monitoring and debugging purposes, after which they are deleted. Google does not use this data for model training. For current details, see Google Cloud Vertex AI Data Governance.

AI Studio Fallback Path: In the event of a Vertex AI service interruption, capacity constraint, or rate limiting, requests may fall back to Google AI Studio under Google's standard API terms. Under these terms, Google may use API inputs for model improvement (subject to Google's then-current API terms). We minimize use of this fallback path and are committed to Vertex AI as our primary processing environment.

Your Opt-Out Right: You may disable the AI Studio fallback path entirely in Account Settings > Privacy. If you disable the fallback:

For full details on data processing paths and your choices, see our Privacy Policy, Section 3.

9. What This Page Does Not Cover

For clarity, the following topics are addressed in our other legal documents:

Topic Document
How your personal information is collected, stored, and deleted Privacy Policy
Your rights under CCPA/CPRA, WPA, and other state laws Privacy Policy, Section 8
Limitation of liability for AI-generated content Terms of Service, Section 4 and Section 16
The "Pilot Rule" and your mandatory review obligation Terms of Service, Section 4
Intellectual property and trade secret protections Terms of Service, Section 12
Sub-processor list and data sharing Privacy Policy, Section 5

10. Change Log

Date Version Change Description
May 20, 2026 2.0 Full expansion: added Deployer/Developer distinction; model table with status; AB 2013 links; training data attribute table; proprietary logic disclosure; AI use-case table with human oversight requirements; output safety measures (NIST RMF); expanded fallback opt-out instructions.
March 30, 2026 1.1 Initial publication

This page is reviewed and updated at minimum quarterly, or within 14 days of a model addition or removal.

11. Accessibility

If you need this disclosure in an alternative format (large print, plain language summary, or screen-reader-optimized text), please contact us at [email protected] and we will provide a reasonable accommodation within 10 business days.

12. Contact for Compliance Inquiries

For questions regarding our AI transparency practices, training data documentation, or data governance:

The Stroop LLC
Attn: Zachary Willette, Data Protection & AI Compliance
PO Box 274, Selah, WA 98942
Email: [email protected]
Phone: (509) 834-8027

For questions about your privacy rights or data processing, see our Privacy Policy or email [email protected] with the subject line appropriate to your jurisdiction (e.g., "California Privacy Request" or "WPA Appeal").


Version v2.0-20260520