Private domain genes for better LLM artifacts.
Hosted runtime. Authorized activation. Structured, auditable outputs.
Try the live runtime demo · Star this repo · Evidence Summary · Marketplace Brief · Contracts · Synthetic Examples · Disclosure Boundary
JAM Gene Runtime is a hosted runtime for activating private domain genes through authorized API or agent-to-agent style calls. It returns structured artifacts, validation summaries, warnings, and usage metadata while keeping private gene content server-side.
This repository shows the safe public interface: contracts, synthetic examples, evidence summaries, community files, and disclosure boundaries. To see the product surface, open the live Gene Runtime page: jamstudio.tech/runtime.html.
No private technical bottom layer is published here: no raw gene packs, no prompt wrappers, no activation graphs, no runtime source code, no customer data, and no provider traces.
Run the public-safe local validation. It checks contracts and synthetic examples only; it makes no network calls.
node scripts/validate-public-examples.js
bash scripts/verify-public-release.shLLMs are strong generators, but domain work often fails in predictable ways:
- missing required constraints
- unsupported metrics or overclaimed evidence
- weak validation structure
- no repair path when an artifact is wrong
- outputs that are hard to audit or meter
JAM Gene Runtime treats private domain expertise as a runtime-activated layer:
public capability metadata
-> authorized runtime activation
-> private server-side gene pack
-> structured specialist artifact
-> validation, warnings, repair hints, metering
The goal is not to sell a prompt file. The goal is to give agents and applications a controlled way to call private domain expertise and receive an artifact they can inspect, validate, and route to human review.
A JAM gene is designed to shape LLM output into a reviewable domain artifact with explicit structure, checks, warnings, and repair hints.
| Output dimension | Bare LLM risk | Gene-runtime behavior |
|---|---|---|
| Structure | Free-form answers drift | Contracted JSON artifacts with required fields |
| Domain coverage | Missing variables, constraints, or evidence boundaries | Public validation themes and boundary ids |
| Repairability | Errors are hard to localize | Warnings, risks, and repair suggestions |
| Auditability | Hard to review or meter | Trace ids, activation ids, cost metadata, audit summaries |
| Safety | Private logic or sensitive input may leak | Public/private disclosure boundary and release blocker |
The private gene content is not published here. The public interface is the artifact contract.
These results are directional readiness evidence. They are useful for evaluating the runtime pattern, but they are not production-readiness, statistical-proof, guaranteed-correctness, solver-replacement, or employment-decision claims.
| Evidence stream | What improved | Current result | Claim-safe reading |
|---|---|---|---|
| EnergyOptGene preliminary run | Domain artifact quality over bare/generic prompting | Avg score +79% vs bare LLM; +49% vs generic expert prompt | Directional single-run evidence across 10 cases |
| EnergyOptGene preliminary run | Constraint coverage | 9.45x vs bare LLM; 2.04x vs generic expert prompt | Suggests stronger formulation coverage, not statistical proof |
| ORgene proof pack | Solver-adjacent artifact correctness on recorded cases | Objective matches: 13,836 = 13,836 and 4,430 = 4,430 | Recorded proof-pack evidence with customer-authorized solver boundary |
| HRgene synthetic run | Safety behavior | 10 / 10 safe-pass controls passed; 0 false positives in the first slice | Useful boundary evidence; tuning still required |
Read the details and caveats in docs/EVIDENCE_SUMMARY.md.
JAM Gene Runtime is designed as a hosted, contract-first runtime surface that can fit a SaaS marketplace motion without publishing the private gene implementation.
| Marketplace concern | Public-safe readiness signal |
|---|---|
| Discover | Clear product surface, social preview, topics, and live demo link |
| Try | Synthetic examples and local validation with no network calls |
| Buy | gene_activation as the public usage unit for commercial packaging |
| Use | Authorized activation request and structured artifact contracts |
| Trust | Disclosure boundary, security docs, claim discipline, and release blocker |
For the Google Cloud Marketplace conversation, see docs/GOOGLE_MARKETPLACE_BRIEF.md.
HRgene reviews candidate-facing and hiring-related text for evidence-boundary
risks. Its public artifact is SignalIntegrityReport.
It can flag:
- unsupported metrics
- ownership inflation
- JD keyword stuffing
- causal overclaims
- unsupported technical specificity
- missing evidence required for a safe rewrite
HRgene does not make hire, reject, rank, protected-class inference, background check, or legal-compliance decisions.
ORgene supports optimization-model formulation, validation, repair guidance, and
public-safe model specifications. Its public artifact is OptimizationModelSpec.
It can describe:
- decision variables
- objective intent
- constraint families
- feasibility checks
- numeric risks
- repair suggestions
- solver boundary warnings
ORgene does not replace Gurobi, CPLEX, OR-Tools, Pyomo, JuMP, or any solver. Solver execution requires a customer-authorized solver environment.
contracts/ Public activation and artifact schemas
agent-cards/ Draft public Agent Card metadata
examples/ Synthetic activation requests and example artifacts
docs/ Evidence, security, privacy, disclosure, release guidance
assets/ Public-safe brand asset
scripts/ Release verification
CONTRIBUTING.md Public-safe contribution boundary
SECURITY.md Security reporting and release check
SUPPORT.md Website and contact paths
Open the live product surface:
Inspect the synthetic examples:
- HRgene activation request
- HRgene SignalIntegrityReport
- ORgene activation request
- ORgene OptimizationModelSpec
Inspect the contracts:
Public by design:
- capability descriptions
- request and response schemas
- synthetic examples
- public boundary ids
- activation ids and trace ids
- warning and validation summaries
- evidence summaries with caveats
- pricing unit language such as
gene_activation
Private by design:
- raw
gene.json - complete private gene packs
- activation graphs
- compiled private constraints
- internal prompts and prompt wrappers
- raw LLM traces
- customer inputs and logs
- secrets, credentials, certificates, and deployment internals
See docs/DISCLOSURE_BOUNDARY.md.
For publication metadata, recommended topics, social preview guidance, and launch copy, see docs/GITHUB_DISCOVERY_CHECKLIST.md and docs/LAUNCH_PLAYBOOK.md.
Run the release check before every commit intended for GitHub:
node scripts/validate-public-examples.js
bash scripts/verify-public-release.shThe validation script checks only public contracts and synthetic examples. The
release script blocks common leakage classes such as .env files, certificates,
databases, generated model traces, build outputs, private implementation
directories, and secret-looking content.
Use:
- hosted runtime
- authorized activation
- structured artifact
- synthetic examples
- preliminary directional evidence
- human review required
- readiness package
Avoid:
- production-ready
- fully autonomous compliance engine
- downloadable gene
- prompt-pack sale
- solver replacement
- guaranteed correctness
- statistical outperformance claims without repeated public benchmarks
Try the public product surface at jamstudio.tech/runtime.html.
This repository is the safe public technical package. The hosted runtime, private gene packs, commercial terms, and production endpoints are separate product surfaces.
No open-source implementation license is granted by this readiness package. See LICENSE.md.