Research Runtime Center V4

Definition

Research Runtime Center V4 is the Digital Researcher’s operations center: three research-intelligence pipelines jointly discover evidence for the three columns, morning shifts advance research, the afternoon shift produces complete reports, and the evening shift releases them through GitHub and the website.

Research Intelligence System

Where evidence is discovered and what the Research Center studies are separate dimensions.

Three source-intelligence pipelines

Skill 01-P — AI Platform Change Intelligence
Skill 01-G — GitHub Engineering Intelligence
Skill 01-R — Published Research Intelligence

The pipelines produce one unified, deduplicated signal pool before Skill 02 performs three-column research triage.

P0 AI platform coverage

The daily platform set is fixed:

• OpenAI / ChatGPT / Codex;
• Anthropic / Claude / Claude Code;
• Google / Gemini;
• Cursor;
• GitHub Copilot;
• Microsoft Copilot Platform.

The worker checks official release notes, documentation, forums or communities, official GitHub sources, status pages, enterprise controls, permissions, and product boundaries. Authentication-required channels must be recorded as restricted rather than reported as checked with no update.

GitHub engineering radar

GitHub is not scanned platform-wide. The governed radar uses:

70% fixed organization and repository watchlist
20% versioned topic × change-type queries
10% bounded new-project exploration

It inspects incremental Releases, Tags, merged PRs, high-value Issues, Discussions, Security Advisories, and architecture, test, and benchmark files.

Published research radar

The system observes papers, preprints, technical reports, benchmarks, datasets, system and model cards, specifications, and associated code or evaluation artifacts. An abstract or vendor summary alone cannot establish an engineering conclusion.

Read Research Intelligence System V1.0

Three research columns

At 10:00, Research Runtime Queue must make a separate decision for each column:

1. Digital Employee — positions, responsibilities, workflows, runtime, governance, recovery, delivery, and evaluation;
2. Industry Architecture — major AI platform products, workspaces, Agent operating models, control planes, permissions, connectors, enterprise controls, and product boundaries;
3. Open-source Engineering — open-source runtimes, protocols, SDKs, tools, benchmarks, recovery, testing, and observability engineering.

Every column must show:

• today’s selected topic, or No Selection;
• the reason for selection or non-selection;
• the primary source;
• priority and lifecycle state;
• the next governed action.

Daily shifts

Time	Shift	Work outcome
09:00	Research Runtime Engine	Advance exactly one governed lifecycle transition.
10:00	Research Runtime Queue	Execute the three intelligence pipelines, deduplicate evidence, score candidates, and publish the three-column research plan.
11:00	Research Runtime Knowledge	Admit qualified Research Notes into Knowledge and maintain architecture candidates.
15:00	Research Runtime Production	Complete bilingual reports, visualization, citations, and publication editing, producing Publication Candidates.
20:00	Research Runtime Publication	Release complete candidates, update indexes and the website, commit to GitHub, and verify the commit.

Monday also includes Architecture at 12:00; Sunday includes Weekly at 20:30; Wednesday includes Academic at 10:00.

The 10:00 intelligence and selection shift

Queue completion no longer means that one news item was found. It requires:

• a coverage result for all three intelligence pipelines;
• every due source checked, explicitly restricted, or failed with a reason;
• cross-source deduplication of the same platform, GitHub, and research change;
• exactly one primary column for every selected object;
• a Selected or No Selection decision for every column;
• matching selected IDs between the Intelligence run and the Daily Research Plan.

The 15:00 Production shift

The afternoon shift invokes the existing Research Skills:

Skill 05 — Research Writing
→ Skill 06 — Visualization
→ Skill 07 — Evidence & Citation
→ Skill 08 — Publication Editing

Its output is not a partial draft. It is a complete Publication Candidate:

• complete Chinese Markdown;
• complete English Markdown;
• valid metadata and column assignment;
• completed visualization, or an explicit decision that no visual is required;
• verified evidence and citations;
• completed publication editing;
• not yet placed in the public article directory.

The 20:00 release shift

The evening shift releases only; it does not restart research or writing:

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

A candidate that fails the release gate returns to Production or an earlier research stage.

Four authoritative runtime artifacts

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

Records each shift’s status, input, work outcome, output, next action, and GitHub evidence.

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

Records coverage, failures, signals, candidates, and three-column triage decisions for all intelligence pipelines.

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

Records the daily topic decision for all three columns and must agree with the Intelligence run.

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

Records the complete Publication Candidates produced at 15:00 and consumed at 20:00.

Completion rate

Only Completed counts toward completion:

completion rate = Completed scheduled tasks ÷ all scheduled tasks

Waiting and Running are unfinished. Skipped, Blocked, and Failed are terminal but not complete.

Governing principles

All three intelligence pipelines serve all three research columns.

Checked with no important change and not checked are different facts.

Every scheduled task must report its actual work outcome, not merely that it ran.

The afternoon shift produces the complete report; the evening shift performs release only.

A publication without a Runtime Record and Commit Verify is not an official runtime output.