32 KiB

Raw Blame History

Dataroom Architecture


Version	1.3
Date	2026-03-22
Author	Johan Jongsma
Status	Draft — for review by Misha

How files, request lists, visibility, tasks, and notifications fit together.

The Problem with "Multiple Datarooms"

Traditional deal platforms (and DropBox-based workflows) create separate folder trees for each buyer. This causes:

File duplication — the same income statement exists in 5 folders, each potentially a different version
Sync nightmares — Seller uploads v3, IB updates Buyer A's folder but forgets Buyer B
No single source of truth — which copy is current?
Rigid structure — adding a buyer means cloning an entire folder tree

DealSpace solves this by storing every file once and controlling who can see it through visibility rules. A "dataroom" is not a place where files live — it's a lens that shows each party the files they're allowed to see.

Parties

Role	Description
IB (Investment Bank)	Runs the deal. Pays for DealSpace. Controls all visibility and routing. The intermediary between Seller and Buyers.
Seller	The company being sold. Uploads documents when requested. Never sees who is asking — all requests come through the IB.
Buyer / PE	Interested parties evaluating the company. Each buyer sees only what the IB has shared with them. Buyers never see each other.

Organizations & People

Organizations and people exist across deals — they're platform-level entities, not per-project copies. When an IB adds "Blackstone" to three different deals, it's the same org record, with the same people. Only their role and permissions vary per deal.

Organizations

org_id          — unique, platform-wide
name            — "Blackstone"
domains         — ["blackstone.com"] (for auto-matching users)
logo            — URL
website, phone, address, etc.

An org is created once (typically via scrape when first encountered) and reused across deals. When the IB looks up blackstone.com for a second deal, DealSpace returns the existing record — no re-scrape.

People

People belong to an organization. Their profile is platform-level, but their role in a deal is per-deal.

person (platform-level):
  name            — "Sarah Chen"
  email           — "schen@blackstone.com"
  title           — "Managing Director"
  phone, photo, bio, linkedin

deal membership (per-deal):
  person          → Sarah Chen
  org             → Blackstone
  deal_role       — "lead_contact" | "reviewer" | "uploader" | "viewer"

Deal Roles

Within a deal, each person has a role that determines what they can do and what gets routed to them:

Deal Role	Description	Typical Person
`deal_lead`	Primary contact for this org in the deal. Receives all communications. Can delegate.	MD, Partner, VP
`reviewer`	Reviews and approves documents before they go up the chain.	Director, Senior Associate
`uploader`	Uploads documents to answer requests. Cannot approve.	Associate, Analyst, Assistant
`viewer`	Read-only access to the dataroom. No tasks.	Observers, junior staff, advisors
`admin`	Manages the org's team within the deal. Can add/remove people.	Office manager, deal coordinator

A person can hold multiple roles (e.g., deal_lead + admin).

Automated Routing

Deal roles enable semi-automated task routing. When the IB assigns a task to an org (not a specific person), DealSpace routes it based on roles:

Inbound request to Seller org:

IB assigns "Provide Statement KLM" to Seller org
  → auto-routes to Seller's deal_lead (CFO)
  → CFO can accept, or delegate to a reviewer/uploader in their org

Category-based routing: Organizations can configure routing rules per document category:

Seller org routing rules:
  Financial/*        → CFO (deal_lead)
  Legal/*            → General Counsel (reviewer)
  HR/*               → VP People (reviewer)
  Operations/*       → COO (reviewer)
  IT/*               → CTO (reviewer)
  (default)          → CFO (deal_lead)

When a request for "Employee benefits summary" comes in tagged HR, it skips the CFO and goes directly to VP People. The CFO still has visibility (as deal_lead) but doesn't have to manually route every request.

Approval chains: The routing rules also define who approves on the way back:

Assistant uploads "Employee Benefits Summary.pdf"
  → auto-returns to VP People (reviewer) for approval
  → VP People approves → auto-returns to CFO (deal_lead)
  → CFO approves → returns to IB

The IB can override any auto-routing — these are defaults, not constraints.

The IB's Own Team

When the IB creates a deal, their org is automatically added as the first party. The 3-4 bankers working the deal are visible in the Parties tab, with roles:

Muskepo (IB):
  Michael Muskepo — deal_lead
  Sarah Kim       — reviewer
  James Park      — uploader, admin

This makes it clear to the Seller and to IB management who's running the deal.

Advisors

Sellers often bring their own advisors — accountants, lawyers, consultants. These are separate orgs added to the deal with their own roles:

Seller: Acme Corp
  CFO, COO, etc.

Seller's Counsel: Baker McKenzie
  Partner (reviewer) — reviews legal docs before release
  Associate (uploader) — prepares and uploads legal docs

Seller's Accountant: Deloitte
  Partner (reviewer) — reviews financial docs
  Manager (uploader) — prepares financial docs

Each advisor org has its own routing rules and visibility scope. The IB controls what each advisor can see — typically limited to their domain (legal advisor sees Legal/, not Financial/).

The Deal Lifecycle

Phase 1: Preparation

IB creates the deal in DealSpace
IB loads their standard request list (template) — the checklist of everything they typically need from a seller
IB adds the Seller org to the deal
IB assigns requests to people at the Seller org
Seller uploads documents to answer each request
These uploads form the internal dataroom — only visible to the IB

Phase 2: Initial Offering

IB reviews the internal dataroom
IB selects a subset of documents to include in the offering
IB shares those documents with "all buyers" — this is the public dataroom
Buyers are invited to the deal and see only the shared documents
Buyers receive a notification: "N documents available"

Phase 3: Due Diligence

Each buyer submits their own request list — questions and document requests specific to their evaluation
IB receives these requests. For each one, IB either:
- Links an existing file — "we already have this from Phase 1"
- Goes back to the Seller — assigns a new task to get the information
When the Seller provides new documents, IB decides visibility:
- Share with the requesting buyer only
- Share with all buyers (if it's common-sense information)
When sharing with all buyers, others are notified via batched digest

Phase 4: Narrowing

Some buyers drop out, some advance
IB may revoke access for eliminated buyers
Remaining buyers get deeper access — more sensitive documents shared
Process repeats until deal closes

Phase 5: Confirmatory Due Diligence

Preferred buyer selected — deal enters final phase
Many documents are refreshed (updated financials, revised contracts, etc.)
Refreshed documents replace the originals silently — no new-document notification
Buyer already has access; they see the latest version next time they open it
New documents in this phase are shared directly, also without fanfare
The dataroom is now a living reference, not a drip feed

Data Model

Objects (files)

Every file is stored exactly once.

object_id        — unique identifier
project_id       — which deal this belongs to
filename         — "Q3-2025-Income-Statement.pdf"
category         — ["Financial", "Historical", "Income Statements"]
stage            — draft | review | published
uploaded_by      — user who uploaded it
uploaded_at      — timestamp

Document Versioning

When a document is updated (e.g., Seller uploads a refreshed income statement), the new version does not go live automatically. It follows an approval flow:

Seller uploads v2 of "Income Statement Q3"
  → v2 stored as pending, v1 still served to buyers
  → Anonymization runs on v2 (using existing entity registry)
  → If no new entities detected and deal is in confirmatory stage:
      auto-approve → v2 replaces v1 silently
  → Otherwise:
      IB reviews anonymized v2 → approves → v2 replaces v1

Once approved:

v2 replaces v1 — all existing answer links and visibility rules carry over automatically
v1 is retained in version history but no longer served by default
Buyers who already downloaded v1 see v2 next time they open the dataroom

Notification behavior depends on deal stage:

Deal Stage	Version Update Behavior
Phase 2-3 (active DD)	Notify buyers: "Updated document available"
Phase 4 (narrowing)	Notify only if IB explicitly flags it
Phase 5 (confirmatory)	Silent replacement — no notification

IB can always see the full version history and compare versions
Audit log records who downloaded which version and when
If a buyer downloaded v1 and v2 replaces it, the audit log retains both records

Folder Structure

The category field is a list of tags that creates the illusion of a folder tree:

Financial/
  Historical/
    Income Statements/
      Q3-2025-Income-Statement.pdf    ← category: ["Financial","Historical","Income Statements"]
      Q4-2025-Income-Statement.pdf
  Projections/
    ...
Legal/
  Corporate/
    Articles of Incorporation.pdf     ← category: ["Legal","Corporate"]

This is a UI presentation of tags, not actual nested storage. A file can appear in multiple categories. The IB's "standard dataroom structure" is a category template applied to new deals.

Request Lists

The driver of all document flow. Three types in a typical deal:

Request List	From	To	Purpose
IB Standard RL	IB	Seller	"Give us everything we need to sell your company"
Buyer A RL	Buyer A	IB	"We need these specific things to evaluate"
Buyer B RL	Buyer B	IB	Different buyer, different questions

Each request list contains requests — individual line items like "Provide audited financial statements for the last 3 years."

Answer Links (many-to-many)

A single file can answer multiple requests from multiple request lists.

Request: "Audited financials 2023-2025"  (from IB Standard RL)
  └── links to → Q3-2025-Income-Statement.pdf

Request: "Historical P&L"  (from Buyer A RL)
  └── links to → Q3-2025-Income-Statement.pdf  ← same file, different request

Request: "Revenue breakdown by quarter"  (from Buyer B RL)
  └── links to → Q3-2025-Income-Statement.pdf  ← same file again

The file exists once. Three requests point to it. No duplication.

Q&A (Text Answers)

Not every request needs a document. Buyers often ask questions that require a text answer:

"What is the customer concentration?"
"Are there any pending litigations?"
"What's the current headcount by department?"

These are handled as text answers on a request — same routing and visibility rules as document answers, just without a file attachment.

LLM-Suggested Answers

When a buyer asks a question, the answer may already exist — buried in a document that was uploaded for a completely different request. The LLM scans the existing document set and suggests answers:

Buyer B asks: "What is the customer concentration?"

LLM finds: In "2025-Annual-Report.pdf" (uploaded for Buyer A's request
"Revenue breakdown"), page 14 states: "Top 5 customers represent 42%
of revenue, with the largest customer at 11%."

Suggested answer: "Top 5 customers = 42% of revenue; largest = 11%
(source: 2025 Annual Report, p.14)"

The IB reviews the suggestion, edits if needed, and approves. The answer is posted to Buyer B's request with the source document optionally linked.

This works because:

Documents are flat with content summaries — the LLM can search across all of them
The answer might come from a document uploaded for the IB's own RL, or from a different buyer's request
Buyers never see that the answer was sourced from another buyer's request — they just see the answer

Q&A Visibility

Same rules as documents:

IB sees all Q&A across all buyers
Each buyer sees only their own questions and the answers shared with them
Seller sees questions as "IB asks..." — never the original buyer

Object Visibility

Who can see each file is controlled by explicit visibility rules.

object_id    — which file
scope        — "internal" | "all_buyers" | "org:{org_id}"
granted_by   — which IB user authorized this
granted_at   — when

A single file can have multiple visibility entries:

object_id	scope	meaning
doc-123	`internal`	IB can see it (always present)
doc-123	`all_buyers`	Every buyer in the deal sees it
doc-456	`internal`	IB only — not shared yet
doc-456	`org:buyer-A`	Only Buyer A sees this
doc-789	`internal`	Uploaded by Seller, under IB review

How Each Party's "Dataroom" is Derived

IB (internal view): All objects in the project. Plus metadata showing sharing status ("not shared", "shared with all", "Buyer A only").

Seller: Objects they uploaded + objects the IB explicitly shared with them (e.g., the request list, IB comments).

Buyer A: Objects where visibility scope includes all_buyers OR org:buyer-A. This is their "dataroom." They have no knowledge that other documents or other buyers exist.

Buyer B: Same logic, different org_id → completely different view of the same underlying file set.

RBAC

Permissions Matrix

Action	IB	Seller	Buyer
Upload files	Yes	Yes (answering requests)	No
See all files in deal	Yes	No	No
See own uploads	Yes	Yes	N/A
See shared files	Yes	Only what's shared with them	Only what's shared with them
Change file visibility	Yes	No	No
Share with specific buyer	Yes	No	No
Share with all buyers	Yes	No	No
Revoke buyer access	Yes	No	No
See who has access to a file	Yes	No	No
See other buyers exist	Yes	Configurable	Never
Download (full)	Yes	No	No
Download (watermarked)	Configurable	Yes	Yes
Create request list	Yes	No	Yes (their own RL)
Assign tasks	Yes	Within their org	No
See who requested a document	Yes	Never (anonymized by IB)	Own requests only

Anonymization & Redaction

Seller documents often contain information that must not reach buyers in raw form — customer names, vendor names, employee names, contract counterparties, sometimes even the Seller's own name (the company is often code-named, e.g., "Project Atlas").

This is not simple find-and-replace. It requires consistent entity replacement across the entire dataroom — if "Acme Corp" appears in 40 documents, it must become "Customer A" in all 40.

Entity Registry

Each deal maintains an entity registry — a mapping table of real names to anonymized labels:

Entity Registry — Project Atlas:

  Customers:
    Acme Corp           → Customer A
    Globex Industries   → Customer B
    Initech             → Customer C

  Vendors:
    Smith & Sons        → Vendor A
    Delta Logistics     → Vendor B

  Employees:
    Jane Rodriguez      → [redacted] / "the VP of Sales"

  The Company:
    Springfield Manufacturing → "the Company" / "Project Atlas"

This registry is deal-wide. Every document processed through the anonymization layer uses the same mapping, ensuring consistency.

LLM-Assisted Detection

When the IB uploads or reviews a document before sharing:

LLM scans the document and identifies entities — company names, person names, addresses, phone numbers, contract counterparties
Each entity is matched against the registry:
- Known entity → auto-replaced with its alias ("Acme Corp" → "Customer A")
- New entity → flagged for IB to classify and assign an alias
IB reviews the anonymized version and approves

LLM detected 3 entities in "Top-Customers-2025.xlsx":

  ✓ Acme Corp → Customer A  (auto-matched from registry)
  ✓ Globex Industries → Customer B  (auto-matched)
  ⚠ Wayne Enterprises → [NEW — assign alias?]
    Suggested: Customer D

  [Approve All]  [Review Individually]

Consistency Across Documents

The entity registry ensures that:

"Acme Corp" in the revenue breakdown is the same "Customer A" as in the contracts folder
A buyer reading the financial model and then a customer contract sees consistent references
When a new document mentions a previously-registered entity, it's auto-replaced without IB intervention

What Gets Anonymized

Category	Treatment
Customer names	Replaced: "Customer A", "Customer B", ...
Vendor/supplier names	Replaced: "Vendor A", "Vendor B", ...
Employee names (below C-suite)	Redacted or replaced with role: "the VP of Sales"
C-suite names	Usually kept (buyers need to know who they're dealing with)
The company itself	Replaced with codename until LOI stage
Addresses, phone numbers	Redacted unless relevant
Contract values	IB decides per document — sometimes ranges replace exact figures

Versioning & Anonymization

Two versions of each document exist internally:

Original — as uploaded by the Seller, with all real names. Only visible to IB.
Anonymized — processed through the entity registry. This is what buyers see.

When the Seller uploads a new version, the anonymization runs again using the same registry. The IB reviews the anonymized output (not the original — they've already seen that). If no new entities are detected, it can auto-approve.

Approval Flow

Seller uploads document
  → LLM scans for entities
  → Known entities auto-replaced via registry
  → New entities flagged for IB
  → IB assigns aliases for new entities (added to registry)
  → IB reviews anonymized version
  → IB approves → document available in buyer datarooms

In later deal stages (confirmatory), anonymization may be relaxed — the buyer knows who the company is by then. The IB controls this per-stage.

Watermarking

Every download by a Seller or Buyer is watermarked at serve time:

{user_name} · {org_name} · {datetime} · CONFIDENTIAL

This is non-negotiable and cannot be disabled. IB downloads may optionally be watermarked.

Buyer Isolation

Buyers must never:

See other buyers' names, org names, or any identifying information
Know how many other buyers exist
See documents shared exclusively with other buyers
See other buyers' request lists or questions

This is enforced at the data layer, not just the UI. Queries for buyer users filter by their org's visibility scope before results leave the database.

Task Chain

A request becomes a task when it needs action from someone. Tasks route through a chain of people, tracking who assigned it and who it should return to.

Example Flow

Buyer C submits a request: "Provide Statement KLM"

1. Buyer C asks
   → appears in IB's task inbox

2. IB assigns to Seller CFO
   → assignee: CFO
   → return_to: IB
   → origin: Buyer C (hidden from Seller)
   → CFO sees: "IB requests Statement KLM"

3. CFO delegates to Director of Operations
   → assignee: Director
   → return_to: CFO

4. Director delegates to Assistant
   → assignee: Assistant
   → return_to: Director

5. Assistant completes the work, uploads the file
   → task returns to Director

6. Director reviews, approves
   → task returns to CFO

7. CFO approves
   → task returns to IB

8. IB reviews, anonymizes if needed
   → releases to Buyer C's dataroom
   → Buyer C's request is marked as answered

Fields on Each Request

assignee_id   — who currently owns this task
return_to_id  — who it goes back to when done
origin_id     — who originally asked (hidden from Seller)
stage         — open | in_progress | review | complete
due_date      — optional deadline

Auto-Flow

When the current assignee marks a task as complete or approved, the system automatically reassigns to return_to_id. No manual handoff needed — the task flows back up the chain by itself:

Assistant completes → auto-assigns to Director
Director approves   → auto-assigns to CFO
CFO approves        → auto-assigns to IB

If the reviewer rejects, the task flows back down to the previous assignee with a note.

Visibility Along the Chain

IB sees everything. Regardless of where a task currently sits in the chain, the IB can see:

Current assignee ("sitting with: Assistant, Seller Org")
How long it's been there
The full chain history (who had it, for how long)
Whether it's at risk of missing a deadline

Delegators see their subtree. The CFO who delegated to the Director can see that the Director re-delegated to the Assistant, and that the Assistant has had it for 3 days. The Assistant cannot see that the CFO delegated it — only that the Director assigned it.

Nudging

Tasks that sit too long get nudged automatically:

Condition	Action
Task unactioned for 48h	Remind current assignee
Task unactioned for 5 days	Escalate to the person who delegated it
Task approaching due date (2 days)	Remind assignee + notify delegator
Task overdue	Notify IB + delegator + assignee

Nudge frequency is configurable per deal. Nudges go to individuals, not broadcast.

Key Rules

Seller never sees the origin. The CFO sees "IB requests Statement KLM", not "Buyer C wants Statement KLM." The IB is always the visible requester.
Each person in the chain sees it in their task inbox. "My Tasks" shows everything currently assigned to you.
The chain is unbounded. CFO → Director → Assistant → Intern → ... → back up the chain. Each handoff is tracked.
Approval on the way back up. Each person in the return chain reviews before passing it up. Any person can reject and send it back down.
IB always has full visibility. Even when a task is 4 levels deep in the Seller's org, the IB can see exactly where it is and who's holding it up.

LLM-Assisted Matching

When a buyer submits a request list, many items will overlap with documents already in the internal dataroom. Instead of the IB manually hunting through hundreds of files, the LLM suggests matches.

How It Works

Buyer B submits their RL with 80 line items
DealSpace runs each request against the existing document set (filename, category, content summary, linked request text)
The LLM returns suggested matches with confidence:

Request: "Audited financial statements 2023-2025"
  → Suggested match: Q3-2025-Income-Statement.pdf (92% confidence)
  → Suggested match: 2024-Annual-Audit-Report.pdf (88% confidence)
  → No match found for: "Environmental compliance certificates"

IB reviews the suggestions in a single approval screen:
- Accept — link the file to the request, set visibility for Buyer B
- Reject — not the right document, will need to source it
- Accept + share with all — this is common-sense info, share with everyone
Unmatched requests become tasks, routed to the Seller

What the LLM Matches On

Document filename and category tags
The original request text that the document answered (from the IB's own RL)
Document content summary (extracted at upload time)
Semantic similarity — "P&L" matches "Profit and Loss Statement" matches "Income Statement"

IB Always Approves

The LLM never auto-shares. Every suggestion requires explicit IB approval. The LLM saves time finding candidates — the IB makes the decision.

Notifications

The Problem

When IB shares a document with all buyers, each buyer should know. But:

Don't spam — 50 documents shared in one afternoon shouldn't mean 50 emails
Stay relevant — don't notify about documents a buyer already has
Respect isolation — never leak information about other buyers

Batched Digests

Notifications are batched per buyer, delivered as periodic digests:

Subject: 12 new documents available — Project Atlas

Hi Sarah,

12 new documents have been added to your dataroom since your last visit:

  Financial (4)
    • Q3 2025 Income Statement
    • Q4 2025 Income Statement
    • 2025 Balance Sheet
    • Revenue Breakdown by Region

  Legal (3)
    • Articles of Incorporation
    • Board Resolutions 2024-2025
    • Shareholder Agreement

  Operations (5)
    • ...

View your dataroom: [link]

Batching Rules

Frequency: Configurable per deal — hourly, daily, or on-demand
Deduplication: Only notify about documents the buyer hasn't seen yet
Relevance: If a document was shared because Buyer A asked for it, and Buyer B already has it via a different request link, don't include it in Buyer B's digest
Grouping: Group by category (the same tags used for the folder view)

Task Notifications

Task assignments are near-immediate but coalesced — if the IB assigns 10 requests to the CFO in one sitting, they arrive as a single notification, not 10 emails. Coalescing window: 15 minutes.

Subject: 3 new requests assigned to you — Project Atlas

You have 3 new requests:

  • Provide Statement KLM (due: Mar 28)
  • Board meeting minutes 2024 (due: Mar 30)
  • Employee headcount by department (due: Apr 2)

View your tasks: [link]

Daily Status Digest

Every active participant gets a daily summary (morning, their timezone) of their open tasks:

Subject: Your daily update — Project Atlas

Tasks assigned to you: 5
  • 2 new (assigned yesterday)
  • 1 due today: "Provide Statement KLM"
  • 1 overdue (3 days): "Environmental compliance certs"
  • 1 in progress

Tasks you're waiting on: 3
  • Delegated to J. Smith (2 days ago, no action yet)
  • Delegated to M. Lee (completed, pending your review)
  • Delegated to R. Patel (in progress)

No daily digest is sent if nothing has changed since the last one.

Deal Stages & Gates

A deal progresses through defined stages. Each stage controls default behavior for notifications, visibility, and document updates.

preparation → offering → due_diligence → narrowing → confirmatory → closed

Stage	Description	Notification Default
`preparation`	IB collecting from Seller	No buyer access
`offering`	Initial docs shared with buyers	Notify on new shares
`due_diligence`	Buyers asking questions, requesting docs	Notify on new shares
`narrowing`	Fewer buyers, deeper access	Notify only if IB flags
`confirmatory`	Final buyer, refreshing docs	Silent — no notifications
`closed`	Deal done	Dataroom frozen, read-only

Stage Transitions

IB controls stage transitions manually
Moving to closed freezes the dataroom — no uploads, no visibility changes, download-only
Moving to confirmatory switches notification behavior to silent by default
Revoking a buyer's access can happen at any stage (buyer eliminated)

Deadline Gates

The IB can set deadlines on stages:

"Round 1 bids due by March 30" — after which buyer access may be frozen or restricted
"Final documents due April 15" — Seller deadline for all outstanding requests
Deadlines trigger automated reminders via the nudging system

Audit Trail

Every meaningful action is logged. This is non-negotiable — the audit log cannot be disabled, paused, or tampered with.

What's Logged

Event	Details Recorded
Document uploaded	Who, when, filename, category, size
Document viewed	Who, when, which version
Document downloaded	Who, when, which version, watermark ID
Visibility changed	Who granted/revoked, which file, which party
Request created	Who, which RL, text
Request answered	Who, text/file, which request
Task assigned	Who assigned, to whom, which request
Task completed	Who, when, how long it took
Login	Who, when, IP, device
Stage transition	Who, from/to stage, when
Access granted/revoked	Who, which org, by whom

Analytics for IB

The audit trail isn't just compliance — it's a buyer engagement signal. The IB can see:

Activity heatmap — which buyers are actively reviewing documents vs. going cold
Document interest — which files get the most views/downloads (signals what matters to buyers)
Response time — how quickly each buyer engages with new documents
Coverage — which documents a buyer hasn't opened yet

Buyer A: Last active 2 hours ago, downloaded 47 of 150 docs, avg response: 4 hours
Buyer B: Last active 12 days ago, downloaded 3 of 150 docs — likely dropping out
Buyer C: Downloaded everything within 24 hours — very engaged

This helps the IB prioritize which buyers to focus on, and when to follow up.

Retention

Audit logs are retained for the life of the deal plus a configurable retention period (default: 7 years, per regulatory requirements). Logs are immutable — append-only, no edits, no deletes.

Standard Templates

IB firms reuse the same structure across deals. DealSpace supports templates for:

Request List Templates

A standard checklist (e.g., "M&A Due Diligence — Full") that can be loaded into any new deal. Contains sections and individual request items, but no answers or files.

Dataroom Category Templates

A standard folder structure (e.g., "Financial / Historical / ...", "Legal / Corporate / ...") that defines the category tags available in a deal. When the Seller uploads a file, they pick from these categories.

Templates are org-level resources — each IB firm maintains their own library.

Summary

Concept	Implementation
File	Stored once as an Object, tagged with categories
Folder structure	UI rendering of category tags, not physical nesting
Dataroom	A filtered view based on the viewer's visibility scope
Internal dataroom	All objects (IB-only view)
Buyer's dataroom	Objects where scope = `all_buyers` or `org:{buyer}`
Request → File link	Answer Links (many-to-many)
Q&A	Text answers on requests, same visibility rules as documents
"Copy file to another dataroom"	Add a visibility scope entry (no file duplication)
Document versioning	Replace-in-place, notification behavior varies by deal stage
Task routing	assignee_id / return_to_id / origin_id chain, auto-flow on completion
Task nudging	Automated escalation at 48h, 5d, near-deadline, overdue
LLM matching	Suggests document + text matches for buyer requests, IB approves
Notifications	Batched digests for documents, coalesced for tasks, daily status
Audit trail	Every view, download, share, assignment — immutable, always on
Buyer analytics	Engagement signals from audit data — activity, interest, coverage
Deal stages	preparation → offering → DD → narrowing → confirmatory → closed
Organizations	Platform-level, reused across deals, matched by domain
People	Belong to orgs, deal roles determine routing and permissions
Auto-routing	Tasks route by org deal_role + category-based rules
Advisors	Separate orgs with scoped visibility (legal sees Legal/*, etc.)

32 KiB Raw Blame History