-
Notifications
You must be signed in to change notification settings - Fork 6
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.
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. |
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.
The G7 Trust Model calculates Trust Magnitude (
| 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 |
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).
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.
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.
Assigned to implementations that successfully document their functionality. Requires a Trust Grade of C or above.
Requires a Trust Grade of B or above. Represents robust, reproducible open-source work with full execution instructions.
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.
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.
The pinnacle of contribution. Eligibility is restricted. The contributor must clear the G7 Apex Gate by simultaneously satisfying 6 active predicates:
-
aGradedOriginsGte5: Fusion closure contains$\ge$ 5 A-or-S graded origins. -
sourceTenureDaysGte180AorS: Evidence baseline age is$\ge$ 180 days. -
directNestedSuiteGte1: Direct component contains a nested suite. -
depth2OnlyReachableGte1: Fusion graph contains depth-2 reachable nodes. -
overallGradeS: Evaluated at Grade S. -
apexPromotionPrSigned: PR labeledapex-promotionsigned off by$\ge$ 2 cross-org 4★+ verifiers.
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 (◆) must satisfy rigorous requirements:
- Must specify
$\ge$ 5 prerequisites. - Requires Origin Fusion (the proposer must hold
origin: trueon at least 1 prerequisite). - Proposer must compile
$\ge$ 3 evidence sources of Grade A or B. - Named implementations are graded at level
5★or6★. - Requires 2 distinct verifier approvals.
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.
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.
To upgrade the level of a named skill:
- Update the YAML frontmatter inside the named skill file (
registry/named/{contributor}/{skill}.md) with new evidence entries. - Run local validation and recalculation commands:
gaia dev verify-tier # or gaia dev calibrate - Open a pull request against the registry repository. Reviewers will check the computed Trust Magnitude and update the status from
awakenedtonamedonce requirements are satisfied.