> ## Documentation Index
> Fetch the complete documentation index at: https://meganharrisonconsulting.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Process & Strategy

> Visual explainer for how Alleato turns raw project evidence into one current intelligence packet and why the packet-first model exists.

This page explains the two ideas that matter most:

1. The **process** that turns raw source material into a current project packet.
2. The **strategy** behind the packet-first model so the product does not collapse into generic chat or fragment spam.

<img src="https://mintcdn.com/meganharrisonconsulting/RLVFnIUL26ufLIlX/images/ai-features/project-intelligence/intelligence-packet-process-and-strategy.png?fit=max&auto=format&n=RLVFnIUL26ufLIlX&q=85&s=fff1454b18a2fcec16739617a73830de" alt="Alleato Intelligence Packet process and strategy" width="1536" height="1024" data-path="images/ai-features/project-intelligence/intelligence-packet-process-and-strategy.png" />

## The Process

The current intended packet writer is the rolling-state synthesis path in:

```text theme={null}
backend/src/services/intelligence/project_intelligence.py
compiler_version = project_intelligence_synthesis_v1
```

The real flow is:

```mermaid theme={null}
flowchart LR
    A["Raw Sources<br/>emails, Teams, transcripts, docs"] --> B["Attribution<br/>map source to project and target"]
    B --> C["Signal Extraction<br/>decisions, risks, tasks, schedule, money"]
    C --> D["Insight Cards<br/>evidence-backed promoted state"]
    D --> E["Rolling Synthesis<br/>prior packet + new raw delta + structured facts"]
    E --> F["Packet Output<br/>executive read, what changed, risks, decisions, actions, timeline"]
    F --> G["Feedback Loop<br/>wrong, stale, useful, corrected"]
    G --> E
```

### Step-by-step

| Step              | What happens                                                                                                     | Why it exists                                                                           |
| ----------------- | ---------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------- |
| Raw sources       | The system reads meetings, emails, Teams, documents, and project records.                                        | Packet quality starts with real evidence, not prompt creativity.                        |
| Attribution       | Sources are linked to the right project and intelligence target.                                                 | Wrong-project evidence is one of the fastest ways to destroy trust.                     |
| Signal extraction | Communications compilers extract decisions, risks, tasks, flags, and exposures.                                  | This converts messy text into reviewable operating signals.                             |
| Insight cards     | High-confidence signals promote into `insight_cards` with linked evidence.                                       | Cards are the durable evidence layer, not just temporary model output.                  |
| Rolling synthesis | The packet writer loads the prior packet, reads only the new source delta, and adds deterministic project facts. | This creates one current read instead of recomputing disconnected fragments every time. |
| Packet output     | The packet writes the exact sections the product renders.                                                        | The UI needs a stable strategic shape, not free-form prose.                             |
| Feedback loop     | Users mark output stale, wrong, or useful and the system can refresh again.                                      | A packet that cannot be corrected will drift.                                           |

## What The Packet Must Answer

The current packet shape is designed to answer the operating questions first:

* What changed
* What matters now
* Where risk is building
* What needs a decision
* What to do next

In code, the current summary shape maps to:

* `currentExecutiveRead`
* `immediateAttention`
* `currentFocus`
* `risks`
* `openDecisions`
* `recommendedActions`
* `whatChanged`
* `timeline`

## The Strategy

The process only makes sense if the layers stay separate.

```mermaid theme={null}
flowchart TD
    A["Source Truth<br/>document_metadata, document_chunks"] --> B["Project Intelligence<br/>intelligence_targets, insight_cards, intelligence_packets"]
    B --> C["Assistant and Briefing Surface<br/>project page, advisor, executive brief"]
    C --> D["Feedback and Learning<br/>reviews, retrieval feedback, promotions"]
```

### Strategic rules

| Rule                                       | Meaning                                                                                        |
| ------------------------------------------ | ---------------------------------------------------------------------------------------------- |
| Packet-first, not chat-first               | For project status questions, start from the current packet before broad retrieval.            |
| Keep evidence separate from interpretation | Raw source truth stays in source tables; interpreted project state lives in cards and packets. |
| One coherent read beats fragment spam      | The packet should feel like a sharp PM wrote one current operating summary.                    |
| Never invent numbers                       | Budget, RFI, and change-event facts come from deterministic reads, not model guesses.          |
| Corrections must improve future packets    | Feedback should not die in a chat thread; it should feed the packet-quality loop.              |

## Why This Exists

Without packet-first intelligence, the assistant tends to feel like random RAG and tool execution. The packet layer exists to give the product a durable project memory that is:

* strategic instead of chatty
* current instead of purely historical
* source-backed instead of speculative
* correctable instead of frozen

## Related Articles

* [Project Intelligence](/help/articles/project-intelligence)
* [Current State](/ai-features/project-intelligence/current-state)
* [Activation Runbook](/ai-features/project-intelligence/activation-runbook)
* [Verification](/ai-features/project-intelligence/verification)
* [Self-Learning Intelligence Architecture](/help/articles/self-learning-intelligence-architecture)