---
title: "Decision records, not decision engines"
description: "IBM and SAS sell you a machine that makes decisions. DecisionGateway publishes the record of decisions that already happened — with receipts. Why the difference matters, and why the public version barely exists yet."
canonical_url: "https://dgw.tmarket.io/blog/2026-11-18-decision-records-not-engines"
---

> Index of every page on this site: [llms.txt](https://dgw.tmarket.io/llms.txt)

There is a category of software now called "decision intelligence": IBM Decision
Intelligence, SAS Intelligent Decisioning, and a long tail of smaller players.
Read the fine print and a pattern holds everywhere. They sell you an engine —
something that evaluates your policies, prices your promotions, approves your
claims — inside *your* four walls, on *your* data, with governance you control.
The decision record they produce is yours, an internal artifact of your own
operations.

That is a real product category with a real buyer. It is not this one.

This post argues for a different artifact, mostly missing in public: the
**decision record** — a durable, citable, machine-readable note of *who decided
what, on what evidence, with what outcome*, drawn from the public record. We
build one. Here is what we mean by it, and what it is visibly not.

## Two different artifacts share one name

<table>
<thead>
  <tr>
    <th>
      
    </th>
    
    <th>
      Decision <strong>
        engine
      </strong>
      
       (IBM/SAS/Dataiku/Quantexa)
    </th>
    
    <th>
      Decision <strong>
        record
      </strong>
      
       (this site)
    </th>
  </tr>
</thead>

<tbody>
  <tr>
    <td>
      Input
    </td>
    
    <td>
      your policies, your data, your enterprise systems
    </td>
    
    <td>
      public speech: Telegram, news, filings — ingested continuously
    </td>
  </tr>
  
  <tr>
    <td>
      Output
    </td>
    
    <td>
      an executed decision (approved/rejected + audit log)
    </td>
    
    <td>
      an attributed claim: actor + what they decided + the verbatim quote + source
    </td>
  </tr>
  
  <tr>
    <td>
      User
    </td>
    
    <td>
      enterprise operations teams
    </td>
    
    <td>
      analysts, researchers, agents reading the world
    </td>
  </tr>
  
  <tr>
    <td>
      Governance
    </td>
    
    <td>
      yours, in your tenant
    </td>
    
    <td>
      public, citable, versioned
    </td>
  </tr>
  
  <tr>
    <td>
      Value loop
    </td>
    
    <td>
      "better calls inside your company"
    </td>
    
    <td>
      "better reads on the world outside it"
    </td>
  </tr>
</tbody>
</table>

An engine answers "should we?" A record answers "what did they do, and were they right?"

## Why the public version barely exists

Three things are hard, and all three are engineering, not money:

1. **Attribution.** "Company X announced Y" is not a decision record until a
specific actor, on a specific date, has been tied to a specific claim, with
the source sentence extracted verbatim. In our graph, 14,220 of 31,064
claims carry an attributed actor as of 2026-11-18 (read-only count, this
morning). The rest are real claims we have not finished attributing. Both
numbers are moving, which is the point.
2. **Outcomes.** A record is only decision-grade if it can be scored later.
That is why forecasts in this graph carry a condition, a deadline and a
resolution rule: 517 forecasts on the books as of 2026-11-18, 38 resolved.
The unglamorous discipline of resolving them is what separates a ledger
from a narrative.
3. **Verification.** Every datum we publish carries its source sentence —
most recently the market-datum surface (19 numbers, each with the exact
line it was read from, EN/RU/AR/DE/ES). If you cannot point at the sentence,
it is an opinion, not a record.

## What this is not

- Not investment advice, not a policy recommendation engine, not a trading
signal service. We do not tell you what to do; we publish what, observably,
happened and what was forecast about it.
- Not prediction markets. We do not host positions; we score falsifiable
claims after the world resolves them.
- Not "decision intelligence" in the IBM sense (policy automation), and not
agent memory (your agent's own traces). Both are real categories; neither
is a public ledger of the world's decisions.

## Who this is for

- **Analysts and researchers** who need "what did X actually do" with citations,
not summaries of summaries.
- **Agents** — the API and MCP server exist so software can ask the same
question the same way, with the receipts attached.
- **Anyone auditing a narrative** — a story is cheap; a resolved forecast is
expensive to fake.

## The honest current state

2,395 decisions, 10,685 events, 11,975 actors, 426 active situations, 517
forecasts — small next to IBM's marketing page, and structured differently on
purpose. We would rather publish a verifiable record than a large unverified
one. If you want the decision *engine*, buy one from the vendors above; they
are good at it. If you want the *record* — the citable trail of what actors
actually decided and how it turned out — that is the gap this project exists
to close, one claim at a time.

*Methodology: counts read read-only from the production graph on 2026-11-18
(~08:37 UTC). Attribution rate is a property of our pipeline, not a ceiling.
Nothing on this page is investment or policy advice.*

## Sitemap

See the full [sitemap](https://dgw.tmarket.io/sitemap.md) for all pages.
