DIGITAL EMPLOYEE WORKS · CURRENT CAPABILITY RELEASE

Research Report Production Engine V1.3

A Digital Researcher production system built on ChatGPT, textual workflows, and GitHub First.

V1.32026-08-04 · Current Capability Release简体中文 →

Download and entry points

• Download the current source ZIP
• Open the GitHub repository
• Read the V1.3 Quick Start
• Open the Digital Researcher Operations Center

The project does not require an APK, EXE, or conventional installer. The download contains the research skills, source-intelligence registry, textual workflows, Runtime specifications, scheduling configuration, website source, validators, and operating evidence. Actual research execution uses the operator’s own ChatGPT and GitHub environment.

One-sentence definition

Research Report Production Engine V1.3 is a Digital Researcher application implemented through ChatGPT. It is not a one-shot article-generation prompt: it organizes a position, source intelligence, three-column topic decisions, governed skills, a production shift, release gates, and GitHub evidence as a continuously operating textual production line.

Research Analyst position
        ↓
Research Intelligence System
        ↓
Three-column research plan
        ↓
Research Skills 01–08
        ↓
15:00 complete-report production
        ↓
20:00 GitHub and website release
        ↓
Runtime Record + Commit Verify

Why the current version is V1.3

V1.3 is not a cosmetic rename of V1.0. It formally records three successive engineering upgrades that are already present in the repository.

Version	Core capability
V1.0	Completed the first real Production Test, validating bilingual notes, visuals, PR, CI, repair, merge, and commit verification.
V1.1	Added Runtime Records, per-task work-outcome reporting, and the Digital Researcher Operations Center.
V1.2	Added independent decisions for three columns, the 15:00 Production shift, and a release-only 20:00 Publication shift.
V1.3	Added Research Intelligence System: AI Platform, GitHub Engineering, and Published Research intelligence pipelines.

Complete V1.3 product structure

1. The Digital Researcher position

position: Research Analyst
worker: Digital Research Employee
platform: ChatGPT
work_system: Research Operating System
control_plane: Research Runtime Center V4
scheduler: Research Runtime Scheduler V2.0
skills: Research Skills V2.0
source_layer: Research Intelligence System V1.0
system_of_record: GitHub

The formal work is not “chatting.” The worker receives governed work, discovers sources, makes topic decisions, advances research, produces complete reports, releases them, and leaves inspectable and reconstructable records.

2. Three research-intelligence pipelines

Research Skills V2.0 turns Skill 01 into an intelligence dispatcher with three profiles:

1. AI Platform Change Intelligence continuously checks official updates, documentation, forums, GitHub, and status channels for OpenAI, Claude, Gemini, Cursor, GitHub Copilot, and Microsoft Copilot Platform;
2. GitHub Engineering Intelligence checks controlled watchlists, query matrices, and incremental Releases, tags, merged PRs, high-value issues, discussions, security advisories, architecture files, and benchmarks;
3. Published Research Intelligence scans papers, preprints, technical reports, benchmarks, datasets, system cards, model cards, standards, associated repositories, and evaluation assets.

AI Platform Change Intelligence ─┐
GitHub Engineering Intelligence ─┼→ Unified signals → Deduplication → Three-column triage
Published Research Intelligence ─┘

Source pipelines and research columns are separate dimensions. All three pipelines serve all three columns.

3. Three permanent research columns

At 10:00, Research Runtime Queue must make a Selected or No Selection decision for each column:

• Digital Employee — positions, responsibilities, workflows, runtime, waiting, recovery, approval, delivery, and evaluation;
• Industry Architecture — major AI platforms’ agent products, workspaces, runtimes, permissions, connectors, enterprise controls, and product boundaries;
• Open-source Engineering — agent runtimes, protocols, SDKs, tools, benchmarks, recovery, tests, and observability engineering.

One change object has one primary column, while secondary impact may be recorded for the other columns.

4. Eight Research Skills

01 Research Intelligence Discovery
02 Three-Column Research Triage
03 Deep Reading
04 Research Analysis
05 Research Writing
06 Research Visualization
07 Evidence & Citation
08 Publication Editing

The article is not the execution unit. The Skill is the execution unit.

5. Daily operating rhythm

Time	Runtime	Work outcome
09:00	Research Runtime Engine	Advance one governed lifecycle transition.
10:00	Research Runtime Queue	Complete three intelligence pipelines, candidate scoring, and three-column decisions.
11:00	Research Runtime Knowledge	Admit evidence-validated Research Notes into Knowledge.
Monday 12:00	Research Runtime Architecture	Make architecture, specification, and lifecycle dispositions.
15:00	Research Runtime Production	Complete bilingual reports, visuals, citations, and publication editing; create Publication Candidates.
20:00	Research Runtime Publication	Consume complete candidates only; update public notes, indexes, the website, GitHub commit, and commit verification.
Sunday 20:30	Research Runtime Weekly	Produce genuinely new cross-topic synthesis.
Wednesday 10:00	Research Runtime Academic	Perform focused work on papers, benchmarks, standards, conferences, and institutions.

6. Production in the afternoon, release in the evening

15:00 does not create an unfinished draft. It creates a complete Publication Candidate:

Research Writing
→ Visualization
→ Evidence & Citation
→ Publication Editing
→ Publication Candidate

A candidate must contain:

• complete Chinese Markdown;
• complete English Markdown;
• valid frontmatter, metadata, and column assignment;
• a completed visual or an explicit no-visual decision;
• verified evidence and citations;
• completed publication editing.

20:00 does not restart research or writing. It performs:

Publication Candidate
→ public bilingual articles
→ metadata / indexes / website
→ GitHub Commit
→ Commit Verify
→ Release

7. Work outcomes, not status slogans

Every scheduled task must report:

Input
→ Work Outcome
→ Durable Output
→ Next Governed Action
→ Metrics
→ Artifacts and GitHub Evidence

The Operations Center therefore reports the signals found, selected topics, object transitions, files produced, next actions, and commit evidence—not merely “Completed.”

Authoritative operating artifacts

research/intelligence/REGISTRY.json

The fixed platforms, repositories, research sources, topics, and evidence levels for all three intelligence pipelines.

research/intelligence/runs/YYYY/MM/YYYY-MM-DD-intelligence.json

Daily source coverage, restricted channels, failed checks, signals, candidates, and three-column decisions.

research/runtime/YYYY/MM/YYYY-MM-DD-runtime.md

Each shift’s status, input, outcome, output, next action, log, and GitHub verification.

research/runtime/plans/YYYY/MM/YYYY-MM-DD-plan.json

The day’s topic decision for all three research columns.

research/runtime/candidates/YYYY/MM/YYYY-MM-DD-candidates.json

The complete Publication Candidates created at 15:00 and consumed at 20:00.

What a user can do after downloading

A user can:

1. read or reuse Research Skills V2.0;
2. change the source Registry for the three intelligence pipelines;
3. replace the three columns with their own research directions;
4. create the corresponding workers in their own ChatGPT environment;
5. connect their own GitHub repository;
6. use GitHub Actions to open Runtime execution slots;
7. run Runtime, intelligence, column-plan, and candidate validation;
8. publish their own VitePress research website.

Minimal commands

git clone https://github.com/joinwell52-AI/joinwell52.git
cd joinwell52
npm install
npm run runtime:validate
npm run docs:build

Operating boundaries

ChatGPT performs the actual worker execution

GitHub Actions open scheduled execution slots, initialize records, and run validators. Source reading, judgment, writing, visualization, citation work, and editing are performed by the ChatGPT Digital Researcher worker.

Therefore:

GitHub cron triggered
≠
research work completed

Without worker execution, a task remains Waiting, Blocked, or Failed; the system must not manufacture Completed.

The repository is public and downloadable; usage follows the license

The repository can be browsed, cloned, and downloaded publicly. Reproduction, modification, redistribution, or commercial use of papers, specifications, reports, diagrams, website content, code, and scripts is governed by the current LICENSE.md.

Boundary with TMPA

The engine applies a single-writer lifecycle-governance subset of TMPA: explicit states, gates, persistent evidence, Git commits, and Reader reconstruction. It is not a complete validation of TMPA’s multi-writer role separation, and a single-writer production record cannot establish every TMPA claim.

Production evidence

The V1.0 Production Test actually produced:

• three Daily Research objects;
• three Academic Observation objects;
• twelve Chinese and English Markdown files;
• six independent SVG covers;
• a GitHub branch, PR, CI failure, repair, merge, and commit verification;
• a real YAML frontmatter defect detected on the first build and corrected before the second build passed.

V1.1–V1.3 then added Runtime Center V4, Scheduler V2.0, three-column topic decisions, afternoon production, evening release, and Research Intelligence System.

Formal evidence and documents

• Research Intelligence System V1.0
• Research Runtime Center V4
• Research Skills V2.0
• Runtime Worker Contracts V2
• Production Test V1
• V1.3 Release Record

Formal position

Research Report Production Engine V1.3 is a downloadable, text-driven Digital Researcher production system built on ChatGPT and GitHub First. It organizes source intelligence, three-column topic decisions, governed research skills, complete-report production, formal release, and verifiable operating records as one continuing work line.