Agent · KrewOS Hub · Singapore Legal Practice Pack
SG Statute Retrieval Agent
Retrieves Singapore statutory provisions and subsidiary legislation relevant to a stated legal question, returns each provision with its authority tier, in-force date and amendment history, and scores the sufficiency of the retrieved set against the claim it is being asked to support.
Public view. The full manifest, the evaluation dataset and verified cost figures require an account.
- Version
- v3.1.0
- Verification
- wGrow verified
- Declared effects
- Reads only
- Risk level
- High
- Evaluation
- Not recorded
- Published
- 28 Jul 2026
Typed interface
| Port | Direction | Type | Required | Notes |
|---|---|---|---|---|
| question | in | LegalQuestion | Yes | Free text plus a declared practice area |
| governing_law | in | Jurisdiction | Yes | Refuses to connect to a router that cannot emit this |
| as_at | in | Date | No | Defaults to run start; supports point-in-time retrieval |
| provisions | out | EvidenceSet | — | Each item carries authority tier, in-force date, amendment chain |
| sufficiency | out | SufficiencyScore | — | Scored against the claim, compared to the crew's evidence floor |
| gaps | out | KnowledgeGap[] | — | Named missing sources when the floor is not met |
Classification
- Industry
Legal - Jurisdiction
Singapore - Language
English - Risk level
High - Human control
Inherited from crew - Input type
Structured question - Output type
Evidence set - Connector dependency
None - Knowledge dependency
3 Knowledge Packs - Quality status
wGrow verified
Declared effects
- Reads
SG statutes, subsidiary legislation and practice directions Knowledge Packs. Nothing else.
- Writes
Evidence records and retrieval traces into the run's own audit store.
- Spends
Platform credits for retrieval and reranking; model credits for query expansion only.
- Causes externally
Nothing. The compiler proves no path from this component reaches an external Action.
Evaluation summary
Nothing in the catalogue records an evaluation suite, a threshold or a result for this component, so there is none to show. The Hub browse page carries the same line: a pass rate here is not a detail held back behind an account, it is a measurement nobody has taken.
Behind the signup gate
Three things on this page are deliberately not public. This is the same decision explained on the Hub browse page — the interface is public so you can judge fit, the implementation is licensed so it stays an asset.
Crew Manifest
component: sg-statute-retrieval
version: 3.1.0
ports:
— shown above, public —
knowledge:
[ 3 packs — signed in ]
prompts:
[ 4 templates — signed in ]
policies:
[ 6 rules — signed in ]
The public JSON Schema for the manifest format itself is open and documented.
Sign in to viewEvaluation dataset
Graded cases, each with the disputed question, the expected provision set, the expected sufficiency band and the correction that produced it, built from real reviews at Singapore practices. The catalogue does not record how many.
Publishing a graded evaluation set destroys it. Once the answers are public, the suite stops measuring anything.
Verified cost figures
Median platform and model credits per invocation, cost per outcome unit, and the distribution across corpus sizes — measured on real runs, not estimated.
A single number without your volumes, document sizes and review posture misleads more than it informs, which is why the estimator asks for those first.
Open the estimatorVersion history
| Version | Released | Change | Breaking | Verification |
|---|---|---|---|---|
| v3.1.0 | 28 Jul 2026 | Point-in-time retrieval via as_at; amendment chain added to every provision | No | wGrow verified |
| v3.0.2 | 19 May 2026 | Sufficiency calibration retuned after 41 Correction Records from three tenants | No | wGrow verified |
| v3.0.0 | 02 Mar 2026 | provisions output changed from a flat list to a typed EvidenceSet | Yes | wGrow verified |
| v2.3.1 | 14 Nov 2025 | Subsidiary legislation corpus added | No | wGrow verified |