Forticia
PolarisLink™
One protocol and one gateway for agents, research systems and human approvers: streaming events, scoped keys, deliberation, telemetry and an audit trail. It is the shared layer every Forticia branch connects through, built to work in any domain.
- Protocol
- v3.0.0
- Transport
- REST + SSE
- License
- Apache-2.0
- SDKs
- C++20 · Python 3
System architecture
Dual-rail protocol topology.
Dual-rail means two channels into one gateway: a REST rail for requests and an SSE rail (Server-Sent Events) for live updates. Together they connect domain adapters, an autonomous multi-agent swarm and immutable provenance storage.
Unified gateway serving dual interfaces: REST-compatible endpoints for programmatic querying and persistent Server-Sent Events (/stream) delivering real-time agent coordination, turn-taking deliberations, cluster heartbeat synchronization, and audit vault notifications with zero polling overhead.
Guarantees & standards
The canonical invariants.
Mathematical truth and fail-closed protocols governing all quantitative research streams and model lifecycle events.
Zero Synthetic Drift Standard
In plain terms: no result is accepted without a recorded source. PolarisLink™ unconditionally rejects unverified or fabricated model telemetry. Backtests without immutable git provenance and deterministic run logs return explicit rejection markers rather than flattering ungrounded claims. The rule is domain-agnostic: a pipeline run or an evaluation with no recorded provenance is refused in exactly the same way.
Run intakeProvenance gateRunGit SHARun logVerdictrun-a419e3779bdeterministicVerifyingBacktests without immutable git provenance and deterministic run logs return explicit rejection markers.
Duplex Agent Telemetry
Persistent Server-Sent Events (SSE) multiplexer delivering instantaneous run/task lifecycle broadcasts, execution telemetry, and cluster alerts directly to listening threads. Listeners wake on events instead of polling. Any adapter, in any domain, subscribes to the same stream.
GET /stream HTTP/2Accept: text/event-streamLast-Event-ID: 131id: 128event: task.finisheddata: {"node":"autoheal"}id: 129event: cluster.heartbeatdata: {"cluster":"ok"}id: 130event: audit.sealeddata: {"vault":"append"}id: 131event: cluster.alertdata: {"severity":"notice"}Simulated streamFrames are generated in your browser
Autonomous Closed-Loop Healing
Integrated
polaris-autohealwatchdog automatically diagnoses infrastructure anomalies, resolves database connection pool starvation, repairs execution state gaps, and enforces rigorous git worktree hygiene. The same watchdog loop applies to any service the gateway coordinates.polaris-autohealWatchdog- Detect
- Diagnose
- Repair
- Verify
detectAnomaly: Connection pool starvation
diagnoseRoot cause recorded
repairResolved: connection pool starvation
verifyChecks passed · anomaly closed
Anomalies are diagnosed, repaired and verified without operator intervention, and worktrees are kept clean.
Swarm Consensus Deliberations
Formal multi-agent debate and deliberation protocol governing hypothesis promotion, parameter vetting, adversarial stress-testing, and strategy validation before deployment. Outside finance the same protocol records a governance review with its dissent alongside the decision.
Turn-based consensusDecision gatesHypothesis promotion, parameter vetting, adversarial stress-testing and strategy validation, one turn at a time.
Cryptographic Run Provenance
Every research and execution run is immutably audited with its full parameters and the exact git commit SHA hash, preventing model recycling or historical curve-fitting. For quantitative runs the record includes the Deflated Sharpe Ratio (DSR), Calmar ratio and trade counts; a pipeline run in another branch carries its own chain from inputs to outputs.
git logAudited runscommit 9f3c2a1parent 4be71d0Pending- Inputs
- Parameters
- Outputs
- Metrics
commit c07a5e9parent 9f3c2a1Pending- Inputs
- Parameters
- Outputs
- Metrics
commit 1d8f3b6parent c07a5e9Pending- Inputs
- Parameters
- Outputs
- Metrics
commit a2e90c4parent 1d8f3b6Pending- Inputs
- Parameters
- Outputs
- Metrics
Institutional Parity & Packaging
Designed for direct compatibility with institutional risk management standards. In quantitative finance that means seamless securitization across Swiss Actively Managed Certificates (AMCs with ISIN) and institutional Separately Managed Accounts (SMAs); other branches package outputs to their own institutional standards.
One run record, institutional wrappersQuantitative exampleSecuritization across Swiss Actively Managed Certificates (AMCs with ISIN).
At a glance
Facts.
Details you can check against the repository.
- Wire version
- v3.0.0
- Licence
- Apache-2.0
- Transport
- REST + SSE
- SDKs
- C++20 · Python 3
Sent in the X-PolarisLink-Protocol header
Protocol and SDKs
REST for requests, Server-Sent Events for live updates
Single-header C++20, zero-dependency Python
Case study
Blitz, the first reference implementation.
Blitz is Forticia's C++20 execution engine for quantitative finance: one project in one branch. It shows the protocol's mechanisms working end to end. The same mechanisms apply to any domain.
Execution telemetry
The engine emits execution telemetry through the single-header client over the SSE rail, so every listener sees dispatch events as they happen, without polling.
Protocol mechanism: Duplex agent telemetrySwarm deliberation
Research agents debate a hypothesis turn by turn through the institutional decision gates before it is promoted.
Protocol mechanism: Swarm consensus deliberationsVault provenance
Each run is sealed in the research vault with its parameters and git commit SHA, alongside its deterministic walk-forward evaluation record.
Protocol mechanism: Cryptographic run provenanceHuman approval
Drafts by default. A lead approves before anything is written, and the approval is part of the audit trail.
Protocol mechanism: Governed writes
Beyond Forticia
An open protocol.
Bring your own adapter.
The protocol and SDKs are published under Apache-2.0, so anyone can implement an adapter for their own systems. The hosted Forticia gateway is separate: connecting to it needs an approved key.
Designed for every Forticia branch
- Quantitative finance
- Computational biology
- Cultural intelligence and governance
- AI instrumentation
- Infrastructure
Developer integration
Client SDK studio.
Integrate any system. Single-header C++20 and zero-dependency Python 3 clients for programmatic agent coordination and execution integration. Open source under Apache-2.0.
#include <polarislink/polarislink.hpp>
#include <iostream>
int main() {
auto client = forticia::PolarisLinkClient::connect("https://forticia.uk/api/polarislink", {
.api_key = std::getenv("POLARISLINK_API_KEY"),
.workspace_id = "blitz",
.stream_channel = "swarm"
});
client.subscribe_swarm_events([](const forticia::SwarmEvent& event) {
std::cout << "[SWARM DELIBERATION] Node: " << event.node_id
<< " | Stage: " << event.stage
<< " | Hypothesis: " << event.hypothesis_id
<< " | Gate 12 Consensus: " << (event.passed_gates ? "PASSED" : "REJECTED") << "\n";
});
client.await_stream_events();
return 0;
}Protocol specification
Specification
The protocol and the SDKs are open: the full specification lives in the repository under Apache-2.0, and anyone can implement against it. The hosted Forticia gateway is different: it needs an approved key, granted to authorised desks, researchers and counterparties.
- Institutional Keying
Timing-safe HMAC credentials with per-workspace rate ceilings and permission scoping.
- Wire Protocol
HTTP/2 Server-Sent Events multiplexer with persistent reconnect handshake and Last-Event-ID replay.
- Swarm Coordination
Real-time turn alternation, deliberation state trees, and automated escalation protocol.
- Audit & Provenance
Cryptographic run verification with SHA-linked git commits and zero synthetic data validation.