Define deep research inside Auto Research #46

Merged

lost-rob0t merged 3 commits from research/deep-research-execution-adapters into main

2026-07-27 07:50:35 +00:00

lost-rob0t commented

2026-07-27 01:13:29 +00:00

(Migrated from github.com)

What changed

adds STAR-RESEARCH-002 Deep Research Execution Adapter Analysis
compares the official architectures of LangChain Open Deep Research, GPT Researcher, STORM / Co-STORM, and PaperQA2
keeps ADARD as the research-run supervisor, branch DAG, ledger owner, validator, reviewer, and promotion gate
defines capability-based routing instead of selecting one universal deep-research engine
adds STAR-RESEARCH-PIPELINE-002 Deep Research Execution Adapters in REVIEW status
defines versioned request, event, and normalized result envelopes
defines claim, evidence, source, artifact, cancellation, timeout, replay, and failure boundaries
treats browser JavaScript actors as Auto Research worker adapters rather than Quasar-owned research semantics
updates the Auto Research index and dependency order

Architecture

ADARD supervisor / branch DAG
        |
        v
capability route + adapter registry
        |
        +--> Open Deep Research
        +--> GPT Researcher
        +--> STORM / Co-STORM
        +--> PaperQA2
        +--> browser JavaScript worker
        |
        v
normalized sources + evidence + claims + artifacts
        |
        v
ADARD deterministic validation, review, and promotion

Quasar may later start, pause, and display runs. It does not own the deep-research control plane or durable state.

Initial routing recommendation

general-web → Open Deep Research
recursive-web → GPT Researcher
perspective-plan → STORM
collaborative-curation → Co-STORM
scientific-literature → PaperQA2
browser-session → bounded browser JavaScript worker

Scope boundary

no Quasar implementation
no canonical Org writes by workers
no automatic promotion
no second run ledger or state machine
no implementation-slot change
no full browser crawler yet

Sources

Research uses the official project repositories retrieved on 2026-07-27.

Closes #45

## What changed - adds `STAR-RESEARCH-002 Deep Research Execution Adapter Analysis` - compares the official architectures of LangChain Open Deep Research, GPT Researcher, STORM / Co-STORM, and PaperQA2 - keeps ADARD as the research-run supervisor, branch DAG, ledger owner, validator, reviewer, and promotion gate - defines capability-based routing instead of selecting one universal deep-research engine - adds `STAR-RESEARCH-PIPELINE-002 Deep Research Execution Adapters` in `REVIEW` status - defines versioned request, event, and normalized result envelopes - defines claim, evidence, source, artifact, cancellation, timeout, replay, and failure boundaries - treats browser JavaScript actors as Auto Research worker adapters rather than Quasar-owned research semantics - updates the Auto Research index and dependency order ## Architecture ```text ADARD supervisor / branch DAG | v capability route + adapter registry | +--> Open Deep Research +--> GPT Researcher +--> STORM / Co-STORM +--> PaperQA2 +--> browser JavaScript worker | v normalized sources + evidence + claims + artifacts | v ADARD deterministic validation, review, and promotion ``` Quasar may later start, pause, and display runs. It does not own the deep-research control plane or durable state. ## Initial routing recommendation - `general-web` → Open Deep Research - `recursive-web` → GPT Researcher - `perspective-plan` → STORM - `collaborative-curation` → Co-STORM - `scientific-literature` → PaperQA2 - `browser-session` → bounded browser JavaScript worker ## Scope boundary - no Quasar implementation - no canonical Org writes by workers - no automatic promotion - no second run ledger or state machine - no implementation-slot change - no full browser crawler yet ## Sources Research uses the official project repositories retrieved on 2026-07-27. Closes #45