Concrete patterns that will be rejected on PR review, each mapped to the rubric dimension(s) it fails. See RUBRIC.md for scoring.
These are archetypes, not accusations of specific projects. Examples are illustrative.
Pattern. A thin SDK that calls a chat completion endpoint, with no first-class support for tool use, memory, multi-step control flow, or state.
Example (archetypal). “A Python package that exposes client.chat(prompt) and nothing else — no tool interface, no state, no orchestration.”
Rubric dimensions it fails. Agentic relevance (hard gate, likely 0–1), Uniqueness (low — dozens exist).
Pattern. Landing pages, launch announcements, or sponsored blog posts presented as reference material. Typically long on adjectives (“world-class”, “state-of-the-art”) and short on methodology, code, or failure modes.
Example (archetypal). “A vendor’s ‘Why our agent platform is the future of AI’ blog post submitted as an entry in Orchestration Frameworks.”
Rubric dimensions it fails. Evidence (hard gate, 0), Reliability (no operator signal).
Pattern. Performance or capability claims without a reproducible benchmark, without a methodology statement, or sourced only from a launch post. Includes cherry-picked metrics that omit the comparison baseline.
Example (archetypal). “‘Beats GPT-4 on agentic tasks’ cited from a single tweet thread with no eval harness, no dataset, and no baseline configuration.”
Rubric dimensions it fails. Evidence (hard gate, 0), Reliability (claims cannot be independently verified).
Pattern. A new entry restates an existing one — same project, same claims, same tier — without new evidence, a new angle, or an additional rubric-relevant distinction. Also includes mirrored entries across sections (e.g., listing the same framework in both Orchestration and Reference Architectures without justifying the split).
Example (archetypal). “Proposal to add LangGraph to Reference Architectures when it is already covered in Orchestration Frameworks, with identical bullet points.”
Rubric dimensions it fails. Uniqueness (0–1), Agentic relevance (no added value).
Pattern. A repo with no commits in 12+ months, a GitHub archived flag, an unresolved maintainer transition, or a single contributor and no external deployments — framed as production-grade.
Example (archetypal). “A research prototype from a 2023 paper, no releases since, submitted as a ‘production orchestration layer’.”
Rubric dimensions it fails. Maturity (0–1), Reliability (hard gate, likely 0).
Pattern. A submission whose primary justification is star count, trending status, or social-media virality — with no operator write-ups, no benchmarks, no stability signals.
Example (archetypal). “‘10k stars in a week’ cited as the rationale for inclusion, with no linked field report or docs review.”
Rubric dimensions it fails. Evidence (hard gate, 0), Reliability (popularity ≠ production-readiness). Community signal alone is tiebreaker-only by design.
Pattern. Generic ML infra, chat UIs without tool use, prompt-only utilities, or content-generation demos with no agent loop, no tool interface, and no multi-step control. Useful software — just not agentic.
Example (archetypal). “A prompt templating library, or a ChatGPT-style web UI with no MCP/tool-use layer.”
Rubric dimensions it fails. Agentic relevance (hard gate, 0).
Pattern. An entry that lists the rubric dimensions and declares the tool passes them, without citing evidence or naming specifics. Box-ticking instead of judgement.
Example (archetypal). “‘Reliability: High. Evidence: Strong. Uniqueness: Yes.’ — with no links, no examples, no tradeoffs.”
Rubric dimensions it fails. Evidence (hard gate, 0), Reliability (unsubstantiated).
| Pattern | Primary failing dimension(s) |
|---|---|
| 1. Generic LLM wrappers | Agentic relevance, Uniqueness |
| 2. Vendor marketing | Evidence, Reliability |
| 3. Unverified benchmarks | Evidence, Reliability |
| 4. Duplicates without added structure | Uniqueness, Agentic relevance |
| 5. Archived / single-author prototypes | Maturity, Reliability |
| 6. Stars-as-evidence | Evidence, Reliability |
| 7. Out-of-scope tooling | Agentic relevance |
| 8. Rubric-restatement | Evidence, Reliability |
If in doubt, open a suggestion issue instead of a PR.