Skip to content

Stars and Ranks

Jump to bottom
mbtiongson1 edited this page Jun 22, 2026 · 2 revisions

Stars and Ranks

Skills progress through rigorous evidence validation, not arbitrary declarations. Under the G7 framework, a clear distinction is drawn between starless generic capabilities and leveled named contributor implementations.

The Seven Ranks

Under G7, all generic skill nodes (registry/nodes/**/*.json) are starless (rank-less) and represent abstract capabilities. Named skill implementations (registry/named/**/*.md) are assigned star ratings (levels 2★ to 6★, with 1★ as a demote penalty) based on their evaluated Trust Grade.

Stars Rank Label Trust Grade Floor Progression Meaning
0★ Unawakened — Refers to an unawakened generic capability, or a draft named skill.
1★ Awakened — Penalty demote state for named skills that lose verified blob links (META §2.4).
2★ Named Grade C or higher First confirmed real-world demonstration.
3★ Evolved Grade B or higher Reproducible, fully documented open-source implementation.
4★ Hardened Grade B or higher Battle-tested implementation with documented failure modes.
5★ Transcendent Grade A or higher Complex composable capstone with multi-source validation.
6★ Apex Grade S The absolute peak. Demands S-grade validation plus G7 Apex Gate clearance.

Evidence Classes

Important

The single-axis Class A/B/C evidence classification is deprecated and replaced by the G7 orthogonal Evidence Type (provenance) and Evidence Grade (quality) axes.

Evidence Types & Weights

The G7 Trust Model calculates Trust Magnitude ($TM$) by evaluating and summing 10 distinct evidence types:

Evidence Type ID Description Weight Max Cap
fusion-recipe Emergence from graded prior fusions 1.5 None (Soft-capped)
github-stars-own Stars on the skill's own repository 1.0 200
proxy-containment Artifact bundled in an external project 1.0 160
verifier-attestation Attestation signed by a 4★+ verifier 1.5 None
benchmark-result Reproducible percentile on public datasets 1.4 100
arxiv Academic citation count of associated paper 1.0 100
peer-review Structured walkthrough by 4★+ reviewers 1.2 None
repo-own Contributor repo volume (commits + authors) 0.6 60
self-attestation Self-declared claim of capability 0.5 10
social-signal Video views and external engagement 1.0 80

Overall Trust Grades

The recomputed Trust Magnitude determines the overall grade, subject to a diversity gate check:

  • Grade S (TM $\ge$ 250): Requires $\ge$ 3 distinct evidence types and at least 1 non-self-producible type.
  • Grade A (TM $\ge$ 100): Requires $\ge$ 1 distinct evidence type.
  • Grade B (TM $\ge$ 50): Requires $\ge$ 1 distinct evidence type.
  • Grade C (TM $\ge$ 20).
  • Ungraded (TM < 20).

How Ranks Work in Practice

0★ — Unawakened

An entry in registry/nodes/ exists to define a generic taxonomy node but has not yet been implemented by any contributor. Draft named skill PRs also begin in the unawakened/awakened intake state.

1★ — Awakened

A named skill falls back to 1★ if its active proof links break or fail automated integrity audits. It remains in the graph but acts as a progression penalty state.

2★ — Named

Assigned to implementations that successfully document their functionality. Requires a Trust Grade of C or above.

3★ — Evolved

Requires a Trust Grade of B or above. Represents robust, reproducible open-source work with full execution instructions.

4★ — Hardened

Requires a Trust Grade of B or above, plus documented failure modes and edge-case behaviors. Contributors can also qualify via the Origin Fusion Pathway: combining 3+ evolved skills with at least one origin contribution.

5★ — Transcendent

Requires a Trust Grade of A or above. Represents highly-polished, widely adopted codebases. Alternatively unlocked via the Ultimate Pathway for capstones requiring 5+ named prerequisites, Origin Fusion, and 10k+ stars.

6★ — Apex

The pinnacle of contribution. Eligibility is restricted. The contributor must clear the G7 Apex Gate by simultaneously satisfying 6 active predicates:

  1. aGradedOriginsGte5: Fusion closure contains $\ge$ 5 A-or-S graded origins.
  2. sourceTenureDaysGte180AorS: Evidence baseline age is $\ge$ 180 days.
  3. directNestedSuiteGte1: Direct component contains a nested suite.
  4. depth2OnlyReachableGte1: Fusion graph contains depth-2 reachable nodes.
  5. overallGradeS: Evaluated at Grade S.
  6. apexPromotionPrSigned: PR labeled apex-promotion signed off by $\ge$ 2 cross-org 4★+ verifiers.

Demerit System

Demerits lower a named skill's effective level (affecting its progression/TUI status) without modifying its base level. Demerits are specified on the generic capability definition.

Demerit ID Penalty Description
broken-evidence −1★ Active evidence links are dead or unreachable.
niche-integration −1★ Integration is highly proprietary or vendor-coupled.
experimental-feature −1★ Feature is experimental and subject to breaking changes.
heavyweight-dependency −1★ Requires massive external platforms or heavy runtime environments.
  • Demerit penalties apply to named implementations at 2★ and above.
  • The effective level is floored at 0★.

Ultimate Skills and Ranks

Ultimate skills (◆) must satisfy rigorous requirements:

  • Must specify $\ge$ 5 prerequisites.
  • Requires Origin Fusion (the proposer must hold origin: true on at least 1 prerequisite).
  • Proposer must compile $\ge$ 3 evidence sources of Grade A or B.
  • Named implementations are graded at level 5★ or 6★.
  • Requires 2 distinct verifier approvals.

Unique Skills and Ranks

Unique skills (◉) are specialized basic capabilities:

  • Must have empty prerequisites.
  • Must be graph-isolated (0 derivatives referenced by other nodes).
  • Generic unique nodes are starless. Named implementations require level $\ge$ 4★ to unlock.

Rarity

Warning

Rarity (common, uncommon, rare, epic, legendary) is deprecated in G7. It is completely removed from the JSON schemas and has been replaced by the dynamic, evidence-driven overallTrustGrade and trustMagnitude variables.

Upgrading a Skill's Rank

To upgrade the level of a named skill:

  1. Update the YAML frontmatter inside the named skill file (registry/named/{contributor}/{skill}.md) with new evidence entries.
  2. Run local validation and recalculation commands: 
    gaia dev verify-tier
# or
gaia dev calibrate
  3. Open a pull request against the registry repository. Reviewers will check the computed Trust Magnitude and update the status from awakened to named once requirements are satisfied.

Clone this wiki locally