Task Lab · grounded iteration 02

From random walk to real work

A tool graph can tell us what may connect. It cannot tell us what a colleague would actually ask, which policy must come first, or what documents would be on their desk. This lab starts with a known real-world workflow, admits graph variations only when research supports them, then mutates correlated public records into one internally consistent mock input packet.

Paper baselineDAG → random walk → prompt

Executable, but the sequence and inputs can feel assembled.

Grounded variationWorkflow → evidence DAG → constrained mutation

Every added step changes the prompt, documents, or verifier for a reason.

8 primary receipts2 real records mutated6 guarded mutations3 correlated input documents
01Researchsales operations

Start with a real job, not a tool chain

The unit of research is one recognizable job: qualify and open a public-sector expansion deal. Primary guidance establishes the order and policy gates; captured tool contracts establish what is executable; public records establish the actual field shapes and values we are allowed to mutate.

World · sales operations

Qualify and open a public-sector expansion deal

A field rep has supplied discovery notes and a federal award extract for an enterprise prospect. The specialist must resolve the account identity, complete duplicate and ownership checks across internal and public records, and open a pending-approval deal only when the checks are clear.

Revenue operations specialist

Evidence ledger

paper methodGraph-Based Task Synthesis

The paper constructs weighted tool-dependency graphs, performs random walks to obtain tool-call sequences, and asks an LLM to draft task descriptions and answers.

Agent-World · baseline to improve
workflow guidanceDuplicate Management Overview

Before creating an account or opportunity, search existing records for the same company and its known aliases so duplicates and ownership conflicts are caught before new records exist.

Salesforce · workflow spine and required fields
policy constraintCPQ Advanced Approvals Overview

Quotes and discount terms route through an approval chain before they can be committed, so a new deal carries no pricing promise until the approver clears it.

Salesforce · pricing approval gate
deployed agent workflowAI Agents for Revenue Teams

Deployed revenue agents follow the same post-discovery spine: consolidate call facts, update the system of record, and hand structured summaries to the owning rep.

Gong · workflow evidence
captured tool contractPipedrive MCP tool inventory

The captured, verified MCP surface supplies the organization, deal, file, and note read/write schemas used to bind returned organization and deal identifiers across the executable sequence.

Pipedrive / Smithery · internal crm tool contract
captured tool contractThe Stall public-data MCP inventory

The captured finance and public-data MCP surface exposes a read-only full-text search across SEC EDGAR filings, used to verify a prospect's public identity and aliases without an API key.

The Stall / Smithery · public filings search contract
public data contractEDGAR Submissions API

The public JSON submissions endpoint provides filer identity metadata including current name, former names, incorporation state, and business address without an API key.

U.S. Securities and Exchange Commission · entity identity fields
public data contractUSAspending Award API

The public award endpoint returns recipient, obligated amount, period of performance, awarding agency, and industry codes for every federal contract action without an API key.

U.S. Department of the Treasury · contract award record fields
Captured benchmark and MCP inventory0 packs · 6 relevant servers
SourceReceiptCoverage
Apolloapollo
sha256:6684da6c15789… · composio-v3-registry
31 operations · 16 reads · 15 writes
Calendlycalendly
sha256:c8cdcda64020f… · smithery-registry
35 operations · 25 reads · 10 writes
Gonggong
sha256:2b7162038497b… · composio-v3-registry
53 operations · 28 reads · 25 writes
Pipedrivepipedrive
sha256:46563bb93ad8e… · smithery-registry
200 operations · 101 reads · 99 writes
Slackslack
sha256:b4837e941c5ce… · smithery-registry
11 operations · 8 reads · 3 writes
The STALLthebrierfox/the-stall
sha256:a60b7112d80ba… · smithery-registry
293 operations · 21 reads · 272 writes