datastrat System Manifest

v1.6.0last_updated: 2026-06-11
Exit to Human View

# 1. System Context

## SITE_MAP / ENTRY_POINTS

Root Manifest (JSON):/agent.json
System Manifest (this):/agent
LLM Guide (TXT):/.well-known/llms.txt
Manifest (Markdown):/agent.md
Human Manifesto:/manifesto
Sectors Index:/sectores
The Method:/explicado
Antigrid Deep Agent:/antigrid
Evidence Deep Agent:/evidence
Experimental Index:/experimental

datastrat builds agent work systems that automate a company's differentiated judgment. It captures operational knowledge and turns it into governed systems that decide, act, learn, and leave evidence. The category is judgment automation: frontier models are available to everyone; a company's operational judgment is not.

## CORE_OBJECTIVE

Automate differentiated judgment.

Operate decisions inside the company.

Leave evidence for every critical workflow.

"We do not do consulting. We do not bill for reports. We do not sell hours. We build decision infrastructure."

Deep Agents is the conceptual frame. The system is composed of two specialized Deep Agents:

  • Antigrid: Deep Agent for project delivery and capital allocation. Orchestrates cost, time, contracts, and cash.
  • Evidence: Deep Agent for verifiable knowledge. Retrieves documentary information, cites sources, and declares insufficient evidence.

Target Business Sectors

Infrastructure and constructionPublic utilitiesAgribusinessGovernments

# 2. What datastrat Does

Core Problem Being Solved

In essential industries, critical decisions are still distributed across documents, spreadsheets, meetings, and tacit judgment.

datastrat solves this by building agent systems that capture operational context, encode business judgment, and turn decisions into executable infrastructure.

Two Operating Models

datastrat does not prescribe a single solution. Each company picks the model it needs — or both together.

01 · FORWARD DEPLOYMENT

We deploy AI agents to solve critical problems and automate core operations.

Helm · Edges · Atlas

02 · ACADEMY

We train your team to lead the transformation from within.

Edges · Helm

The Focused aOS

Four functional layers assemble around enterprise judgment and are enclosed as an agentic operating system (aOS) — operated by the company's people, connected to operations, and executing real work.

01 Enterprise judgment — rules, priorities, exceptions, operational judgment
02 Frontier models — reasoning, language, analysis, execution
03 Harness — reliable operational workflows
04 Data & domain capabilities — sources + specialized agents

The Deep Agents Frame

Think of datastrat as decision infrastructure with specialized Deep Agents:

  • Embedded context: Agents operate inside the company's real workflows, data, catalogs, documents, and approval paths.
  • Governed action: Changes are routed through scenarios or approved workflows instead of direct production mutation.
  • Evidence trail: Every critical output should preserve citations, lineage, or audit evidence.
  • Installed capability: The outcome is a working system in production, not a report.

Antigrid — Deep Agent for projects and capital

Orchestrates cost, time, contracts, and cash as an operable reality.

  • • Hierarchical Financial Models (Program → APU)
  • • Catalog Management with regional pricing
  • • Budget Integrity enforcement
  • • Scenario analysis for decision workflows
  • • Referential integrity and change control

Evidence — Deep Agent for verifiable knowledge

Searches, reads, extracts, and cites. If support is insufficient, it says so.

  • • Deep documentary retrieval
  • • Evidence-backed answers with citations
  • • Audit trails and document lineage
  • • Context engineering for large document sets
  • • Explicit uncertainty when evidence is insufficient

The Workflow: Decide → Simulate → Execute → Verify

  1. Decide: A user or agent identifies a decision point (e.g., "Should we switch from Material A to Material B?").
  2. Simulate: A scenario is created. The agent applies the material substitution, recalculates all affected APUs, propagates cost changes up the WBS hierarchy, and forecasts the new NPV/IRR.
  3. Execute: If the scenario is approved, it's merged into the base state. The change is logged, and all downstream systems are notified.
  4. Verify: Evidence (photos of Material B delivery, field reports) is linked to the corresponding WBS node, creating an audit trail.

# 3. Making Better Decisions

datastrat exists to help organizations make better decisions, faster. Traditional project management tools track what happened, but they don't help you decide what should happen next or simulate alternative futures.

With Antigrid, a project manager can literally chat with their project. Instead of navigating through 10 screens to answer "What happens to our NPV if steel costs increase by 8%?", the manager asks the question in natural language and receives an instant, calculated response backed by the system's own financial models.

"The agent doesn't invent numbers. When it needs real data, it queries. When it needs calculations, it computes using the same formulas as the UI."

Why This Matters

Speed

Reduces analysis time from hours to seconds. No navigating through multiple screens to answer a single question.

Exploration

Enables reproducible What‑If scenarios. Test hypotheses without fear of breaking production data.

Consistency

Uses the same rules and calculations as the rest of the system. No spreadsheet drift or formula errors.

Traceability

Every decision, every calculation, every scenario is logged and auditable. Full evidence lineage.

# 4. What‑If Scenarios Explained

A scenario is a logical copy or change layer applied on top of the base model. It allows users and agents to explore "what if?" questions without modifying the production state.

BASE

Production State

The real, current state of your project. This is the source of truth stored in the database. It represents "what is" right now.

SCENARIO

Variant State

A hypothetical variant (OPT, PES, custom) where you apply changes—cost adjustments, schedule shifts, revenue modifications—to see the impact.

Typical Scenario Workflow

  1. Create scenario — Give it a name (e.g., "Steel +8%" or "Pessimistic Q3")
  2. Apply targeted changes — Modify specific parameters: material costs, APU yields, global costs, schedule dates, revenue assumptions
  3. Recalculate — The system propagates changes through the entire financial model (APU → Budget → Cash Flow → KPIs)
  4. Compare vs BASE — View diff reports: ΔCAPEX, ΔNPV, ΔIRR, ΔDSCR, etc.
  5. Decide: Apply or Discard — If approved, merge changes into the base state with full audit trail. If rejected, discard without side effects.
The key insight: scenarios let you test hypotheses and make decisions without risk to production data. It's "Git branching" for project finance.

# 5. Agent Capabilities

The Agent Service is a microservice that enables conversational interaction with Antigrid. It can execute controlled actions over your data (APU, budget, schedule, revenue, financial KPIs) to accelerate analysis and simulations—without requiring users to manually navigate the UI.

What the Agent Enables

In practical terms, the agent enables three core capabilities:

1. Query Project State

Retrieve summaries, lists, and comparisons from the project/budget/model. Ask questions like "What are the top 10 budget items by cost?" or "Show me all APUs using steel."

2. Calculate / Recalculate Results

Compute APU totals, budget rollups, AIU/IDC, cash flows, NPV, IRR, DSCR, and other KPIs using the system's native logic. The agent uses the same calculation engine as the UI.

3. Simulate Changes (What‑If)

Create scenarios, apply parameter changes, recalculate the model, and compare impact against the base case—all without modifying production data.

How Tools Work

Tools are named functions that the agent can invoke. Think of them as internal endpoints with a stable contract. There are several types:

query
Read-only data retrieval
calculate
Recalculate results
scenario
What-if simulations
update
Controlled mutations

Execution Flow (Simplified)

  1. User types an instruction in natural language
  2. Agent determines if a tool is needed (and which one)
  3. Agent executes the tool with structured parameters
  4. Agent returns response + results/artifacts (summary, diff, metrics)

This ensures the agent never invents data: when it needs real numbers, it queries; when it needs calculations, it computes using the system's native formulas.

# 6. Neural Architecture

The following diagram outlines the high-level request flow between the Frontend, Backend services (Auth, Core, Business), and the AI Agent Service.

architecture_v1.ascii
+-----------------------------------------------------------------------------------+
|                                   FRONTEND                                        |
|                           React 19 + TypeScript                                   |
|                                Port: 10001                                        |
+-----------------------------------------------------------------------------------+
        |                    |                    |                    |
        | REST /api/v1       | REST /api/v1       | REST /api/v1       | WebSocket
        v                    v                    v                    v
+---------------+   +----------------+   +----------------+   +------------------+
|   BACKEND     |   |    BACKEND     |   |    BACKEND     |   |  AGENT SERVICE   |
|   AUTH (M1)   |   |   CORE (M2-M6) |   | BUSINESS (M7-8)|   |     (M5.1)       |
+---------------+   +----------------+   +----------------+   +------------------+
|               |   |                |   |                |   |                  |
| - Login/JWT   |   | M2: Catalogs   |   | M7: Revenue    |   | - AI Chat        |
| - Tenant      |   | M3: Projects   |   | M8: Financial  |   | - 48 Tools       |
| - Users       |   | M4: APU        |   |                |   | - Scenarios      |
| - Roles       |   | M5: Budget     |   |                |   | - What-If        |
|               |   | M6: Schedule   |   |                |   |                  |
+---------------+   +----------------+   +----------------+   +------------------+
        |                   |                    |                    |
        +-------------------+--------------------+--------------------+
                                    |
                                    v
+-----------------------------------------------------------------------------------+
|                               POSTGRESQL 15                                       |
|                          Port: 10002 | RLS Multi-Tenant                           |
+-----------------------------------------------------------------------------------+
                                    |
                                    v
+-----------------------------------------------------------------------------------+
|                                    REDIS                                          |
|                          Port: 10004 | Cache + Sessions                           |
+-----------------------------------------------------------------------------------+

# 7. Functional Ontology

Domain entities are organized into three distinct phases: Foundations (Auth/Catalogs), Business Core (Projects/Budgets), and Extensions (Schedule/Financials).

domain_model.txtReadOnly
╔═══════════════════════════════════════════════════════════════════════════════════╗
║                            PHASE 1: FOUNDATIONS                                   ║
╠═══════════════════════════════════════════════════════════════════════════════════╣
║                                                                                   ║
║  ┌─────────────────────────────┐      ┌─────────────────────────────────────────┐ ║
║  │ M1: AUTH + TENANT           │      │ M2: CATALOGS                            │ ║
║  │ ─────────────────           │      │ ───────────                             │ ║
║  │ • tenant                    │      │ • material                              │ ║
║  │ • user                      │      │ • labor                                 │ ║
║  │ • user_tenant               │      │ • equipment                             │ ║
║  │ • role                      │      │ • transport                             │ ║
║  │ • permissions               │      │ • input_category                        │ ║
║  │ • session                   │      │ • regional_price                        │ ║
║  └─────────────────────────────┘      └─────────────────────────────────────────┘ ║
╠═══════════════════════════════════════════════════════════════════════════════════╣
║                         PHASE 2: BUSINESS CORE                                    ║
╠═══════════════════════════════════════════════════════════════════════════════════╣
║                                                                                   ║
║  ┌─────────────────────────────┐      ┌─────────────────────────────────────────┐ ║
║  │ M3: PROJECTS                │      │ M4: APU (Unit Price Analysis)           │ ║
║  │ ──────────                  │      │ ────                                    │ ║
║  │ • program                   │      │ • apu                                   │ ║
║  │ • macroproject              │      │ • project_phase                         │ ║
║  │ • project                   │      │ • apu_input                             │ ║
║  │ • project_parameter         │      │ • yield                                 │ ║
║  └─────────────────────────────┘      └─────────────────────────────────────────┘ ║
║  ┌─────────────────────────────┐                                                  ║
║  │ M5: BUDGET                  │                                                  ║
║  │ ────────────                │                                                  ║
║  │ • budget                    │                                                  ║
║  │ • wbs (Work Breakdown)      │                                                  ║
║  │ • budget_item               │                                                  ║
║  │ • global_cost               │                                                  ║
║  └─────────────────────────────┘                                                  ║
╠═══════════════════════════════════════════════════════════════════════════════════╣
║                            PHASE 3: EXTENSIONS                                    ║
╠═══════════════════════════════════════════════════════════════════════════════════╣
║  ┌─────────────────────────────┐      ┌─────────────────────────────────────────┐ ║
║  │ M6: SCHEDULE                │      │ M8: FINANCIAL                           │ ║
║  │ ───────────                 │      │ ─────────────                           │ ║
║  │ • schedule                  │      │ • cash_flow                             │ ║
║  │ • activity                  │      │ • npv / irr                             │ ║
║  │ • dependency                │      │ • financial_indicator                   │ ║
║  └─────────────────────────────┘      └─────────────────────────────────────────┘ ║
╚═══════════════════════════════════════════════════════════════════════════════════╝

# 8. Entity Hierarchy

entity_relationships.tree
PROGRAM (1)
└── MACROPROJECT (N)
    └── PROJECT (N) [Access Control Boundary]
        └── PHASE (N)
            └── BUDGET (N)
                └── BUDGET_ITEM (N) [WBS Nodes]
                    └── APU (1) [Analysis]
                        └── APU_DETAIL (N) [Resources]
                            ├── Material
                            ├── Equipment
                            ├── Labor
                            └── Transport

# 9. Agent Instructions

When operating within this environment, agents must adhere to the following PRIME_DIRECTIVES:

No Direct Mutation

Agents cannot mutate the Source of Truth directly. All changes must be proposed via Scenarios.

Uncertainty Declaration

Stochastic outputs must be explicitly flagged with confidence intervals. Do not hallucinate deterministic data.

Referential Integrity

When querying or proposing changes, always verify relationships across the entity hierarchy. Breaking foreign key constraints is forbidden.

Context Preservation

All operations must maintain audit trails. Every decision, calculation, or data transformation must be traceable to its origin.

# 10. Agent Toolbox (Runtime)

The AI Agent Service provides 48 specialized tools organized into the following categories:

Data Query Tools

  • query_budget_items — Retrieve WBS nodes with filters
  • query_apu_details — Fetch unit price analysis breakdowns
  • query_catalog_resources — Search materials, labor, equipment, transport

Scenario Tools

  • create_scenario — Fork the current state for what-if analysis
  • apply_scenario_delta — Modify parameters in a sandbox
  • compare_scenarios — Generate diff reports between base and scenario

Financial Tools

  • calculate_npv — Net Present Value given cash flow and discount rate
  • calculate_irr — Internal Rate of Return
  • forecast_cash_flow — Project monthly/quarterly cash requirements

Schedule Tools

  • calculate_critical_path — CPM algorithm on activity network
  • resource_leveling — Balance resource allocation over time
  • detect_conflicts — Identify scheduling constraint violations

Compliance Tools

  • validate_apu_formula — Verify unit price calculations
  • check_budget_integrity — Ensure WBS totals match project budget
  • audit_trail_query — Retrieve change history for any entity