Banyan Grove is a structured evidence review workbench for document-heavy professional workflows. It assists reviewers in organising submitted documents, assessing criteria coverage, detecting gaps and inconsistencies, and producing reviewer-authored reports. It does not make determinations autonomously and does not treat AI-generated output as authoritative fact.

---

1. What This Tool Does

• Provides a structured workspace for organising and evaluating a document packet against a configurable set of review criteria.
• Optionally invokes OpenAI's GPT models to extract evidence items from uploaded documents — see §3 for details on external data transmission.
• All AI-extracted evidence is immediately marked provisional. It does not enter the reviewer record until a human reviewer explicitly accepts, edits, or dismisses each item.
• Runs rule-based disposition and coherence analysis on reviewer-entered data to surface missing information and logical inconsistencies. This analysis is fully deterministic and contains no AI or probabilistic scoring.
• Generates two outputs: a full Technical Report (reviewer-facing) and an Executive Summary (decision-maker-facing). Reports reflect only reviewer-accepted or reviewer-authored content — AI-provisional items are excluded unless accepted.
• Extracts plain text from uploaded PDF and TXT files to reduce manual transcription. Keyword detection highlights relevant terms found in documents.
• Maintains a persistent audit trail of all reviewer actions, AI runs, and field changes throughout the review lifecycle.

---

2. What This Tool Does Not Do

• No autonomous determination. The disposition output is advisory only. All final decisions remain with the authorised reviewer and their organisation.
• No authoritative AI output. AI-extracted items are labelled as provisional. The tool is designed so that reviewers must actively accept, edit, or reject each AI finding. AI output is never promoted to the record without human action.
• No legal opinion. Nothing produced by this tool constitutes legal, financial, or compliance advice.
• No OCR. Scanned or image-based PDFs will not yield extractable text. Text-layer PDFs and TXT files are supported.
• No guarantee of AI completeness or accuracy. Language model outputs may be incomplete, imprecise, or incorrect. All extracted items require reviewer verification before reliance.

---

3. AI Extraction & External Data Transmission

• Banyan Grove uses the OpenAI API (specifically the Responses API with Structured Outputs) to extract evidence items from uploaded documents when the reviewer clicks "Analyse Documents".
• The text content of all documents in the current case is sent to OpenAI's servers for processing. This transmission occurs only when the reviewer explicitly initiates an AI analysis run.
• Banyan Grove sends extraction requests with store=False, which instructs OpenAI not to retain the request or response content as application state. However, OpenAI's standard data processing terms apply and users should consult their organisation's data governance policy before uploading sensitive documents.
• If your organisation prohibits transmission of case-related documents to third-party AI services, do not use the AI extraction feature. The tool remains fully functional without it — reviewers can enter evidence items manually.
• Disposition analysis, coherence checking, and report generation do not involve any external API calls. These are performed locally and deterministically.

---

4. Data Storage

• Banyan Grove stores case data persistently in a local database. Cases, documents, evidence items, criterion assessments, gap/conflict findings, AI run records, and reviewer audit events are all written to the database and are preserved across sessions.
• Data is stored on the application server and is not automatically deleted when a session ends or a browser window closes. Cases remain until explicitly deleted by a reviewer.
• Uploaded document files are not stored to disk. Document text is extracted in memory at upload time and stored as text in the database. The original binary file is discarded after extraction.
• The database is local to the deployment and is not shared with or accessible to OpenAI or any other third party except as described in §3 above.
• Do not upload documents containing highly sensitive personally identifiable information unless your organisation's data handling policies explicitly permit use of this tool for that purpose.

---

5. Reviewer Responsibility

• The reviewer is responsible for the accuracy and completeness of all information entered and accepted into the tool.
• AI-extracted items are provisional suggestions. Accepting an item is an affirmative reviewer decision and constitutes the reviewer's endorsement of its accuracy.
• Disposition outputs reflect the data the reviewer has entered and accepted, evaluated against the rule set — they are only as reliable as the inputs.
• All disposition recommendations should be reviewed by a qualified professional before any action is taken.
• The "Gaps & Concerns" and "Reviewer Notes" fields should be used to document reasoning and any deviations from the system-generated disposition.

---

6. Disposition Rule Transparency

All disposition and coherence rules are deterministic and fully disclosed. When a disposition is generated, the Transparent Reasoning section shows exactly which rules were applied and why. There are no hidden weights, black-box models, or opaque scoring algorithms in the disposition layer. Reviewers can inspect and override any output.

---