Skip to main content
Developer platform · LQM Builder

Build governed quantitative models on FinanceGPT.

Register, train, evaluate, evidence, publish and promote quantitative/generative LQM modules through API v2. Language remains a separate QLM-selected runtime.

OpenAPI YAML Sign in to build
Overview Quickstart API reference Guides LQM Builder SDKs Auth & webhooks Changes Versioning CI Support Certification Programme Ecosystem Solutions References Evidence exchange Status Marketplace API plans
Canonical definition

An LQM is quantitative, not linguistic.

LQM Builder produces quantitative/generative modules for forecasting, valuation, risk, simulation, optimization and scenario generation. LQMs accept governed numeric, categorical and time-series features and emit typed versioned evidence. They do not receive raw document text and they do not generate language.

Evidence is the interface. A QLM language runtime may explain a verified financegpt.lqm.evidence.v1 packet. It does not receive LQM latent state and does not become the source of numeric authority.
Lifecycle
1Register
2Seal data contract
3Seal training spec
4Train
5Evaluate
6Issue evidence
7Operate
8Publish
9Promote
10Compose in qlm

API reference

Workspace-scoped endpoints with explicit least-privilege scopes.
Available
MethodEndpointScopePurpose
GET/api/v2/lqm/developer-manifestlqm:build:readMachine-readable LQM developer contract, capability profiles and lifecycle boundaries.
GET/api/v2/lqm/models/{model}/developer-manifestlqm:build:readModel-specific lifecycle state and currently eligible developer operations.
POST/api/v2/lqm/modelslqm:build:writeRegister a quantitative LQM identity before training.
GET/api/v2/lqm/models/{model}lqm:build:readRead registry, training, evaluation, evidence, operations and promotion state.
POST/api/v2/lqm/datasets/{dataset}/contractslqm:build:writeSeal a workspace-scoped numeric/time-series training-data contract.
POST/api/v2/lqm/models/{model}/training-specslqm:build:writeSeal a reproducible VAE-GAN training specification.
POST/api/v2/lqm/models/{model}/training-runslqm:build:runQueue governed VAE-GAN training through the private numeric runtime.
POST/api/v2/lqm/models/{model}/evaluatelqm:build:runRun statistical-fidelity and adversarial/data-poisoning evaluation gates.
GET/api/v2/lqm/models/{model}/evidence-schemalqm:build:readRead the versioned Evidence Packet schema that language runtimes may explain.
POST/api/v2/lqm/models/{model}/evidence-packetslqm:build:runGenerate and seal bounded numeric inference as an Evidence Packet.
POST/api/v2/lqm/evidence-packets/{packet}/verifylqm:build:runIndependently verify packet, schema, checkpoint, evaluation, inference and ledger evidence.
PUT/api/v2/lqm/models/{model}/operations-policylqm:build:writeConfigure drift/retraining cadence without automatic production replacement.
POST/api/v2/lqm/models/{model}/drift-assessmentslqm:build:runRun deterministic drift assessment against governed numeric data.
POST/api/v2/lqm/models/{model}/retraining-requestslqm:build:writeCreate a human-review retraining request.
POST/api/v2/lqm/retraining-requests/{request}/decisionlqm:build:writeRecord explicit approve/reject/cancel decision for candidate retraining.
POST/api/v2/lqm/retraining-requests/{request}/advancelqm:build:runAdvance an approved candidate through training and evaluation.
GET/api/v2/lqm/models/{model}/publicationslqm:build:readList model-hub publication records and immutable supply-chain hashes.
POST/api/v2/lqm/models/{model}/publishlqm:publish:writeHuman-confirm publication of an exact evaluated, evidence-verified checkpoint to the configured model hub.
GET/api/v2/lqm/publications/{publication}lqm:build:readRead publication revision, artifact hashes, attribution and supply-chain evidence.
POST/api/v2/lqm/models/{model}/promotelqm:promote:writeHuman-promote an exact eligible checkpoint into the workspace Quant Models catalog.
POST/api/v2/lqm/promotions/{promotion}/retirelqm:promote:writeRetire a promoted checkpoint; existing QLM bindings fail closed.
Quick start

Read the developer manifest

curl -H "Authorization: Bearer $FINANCEGPT_API_KEY" \
  -H "Accept: application/json" \
  https://financegpt.dev/api/v2/lqm/developer-manifest

The manifest is machine-readable and exposes lifecycle operations, required scopes, capability profiles and non-negotiable execution boundaries.

Evaluation

Evaluate an exact checkpoint

curl -X POST -H "Authorization: Bearer $FINANCEGPT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"training_run_uid":"lqmrun_..."}' \
  https://financegpt.dev/api/v2/lqm/models/abc123/evaluate

Evaluation runs statistical-fidelity, temporal, cross-series, tail, diversity and robustness gates. Passing evaluation does not itself activate the checkpoint.

Least privilege

Choose the smallest capability profile your application needs.

Observe

Read LQM registry, training, evaluation, Evidence Packet, operations and promotion evidence.

lqm:build:read

Build

Register LQMs, seal quantitative data contracts and training specifications, plus read evidence.

lqm:build:readlqm:build:write

Build & operate

Build, train, evaluate, generate/verify Evidence Packets and run governed drift/retraining operations.

lqm:build:readlqm:build:writelqm:build:run

Build, operate & publish

Adds governed model-hub publication of an exact evaluated, evidence-verified checkpoint using the builder's connected publishing identity.

lqm:build:readlqm:build:writelqm:build:runlqm:publish:write

Build, operate & promote

Adds explicit human-governed promotion/retirement of eligible checkpoints into Quant Models.

lqm:build:readlqm:build:writelqm:build:runlqm:promote:write

Publish an evaluated checkpoint

A separate model-supply-chain action using an attributable builder identity.
Governed
curl -X POST -H "Authorization: Bearer $FINANCEGPT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model_version_id":123,"license":"apache-2.0","visibility":"public","human_confirmation":true}' \
  https://financegpt.dev/api/v2/lqm/models/abc123/publish

The API application needs lqm:publish:write, its attributable human creator needs an active publishing connection in the workspace, and the exact checkpoint must still pass training, evaluation, Evidence Packet and drift re-attestation at execution time.

The repository disclosure bundle includes the model card, governance addendum, evaluation evidence, Evidence Packet schema, ML-BOM and a publication manifest. Repository publication does not activate the model inside FinanceGPT.

Promotion boundary

Internal promotion is not public publishing.

Publishing and promotion are independent governed actions. Publishing uses the builder’s connected model-hub identity and lqm:publish:write; promotion uses lqm:promote:write and changes the internal Quant Models projection only after its own human decision. Neither action implies the other.

Execution boundary

Model access never creates financial authority.

LQM scopes do not grant Financial Actions authority, transaction credentials, automatic production promotion or silent QLM rebinding. Existing approval, policy, reconciliation and execution controls remain separate.

Public LQM ecosystem

Publicly published LQMs can appear in the FinanceGPT model gallery with evaluation, Evidence Packet and model-supply-chain evidence. The public leaderboard uses transparent evidence-first ordering rather than an opaque cross-task score.

Explore the LQM ecosystem · Read the specification

Usage & commercial evidence

LQM training, evaluation and promoted inference can be measured through the existing FinanceGPT usage control plane. Pricing is versioned from measured infrastructure-cost evidence rather than architecture assumptions, while experimental published LQMs remain uncallable until separately promoted.

GET /api/v2/lqm/commercial-model
GET /api/v2/lqm/usage-evidence
POST /api/v2/lqm/workflow-outcomes