Skip to content
Forticia
Research
Quantitative FinanceEquities, FX, futures. Factors and backtests, every run replayable.Computational BiologySequence, folding, simulation. Versioned, reproducible labs.Cultural IntelligencePhilosophy, governance, ethics. How institutions decide and answer for it.AI InstrumentationPrivate models. Multi-agent orchestration. Guardrails on write.View all research
Quantitative Finance
  • Equities, FX, futures
  • Factors and backtests
  • Every run logged and replayable
InfrastructurePapersPolarisLink™About
Sign inRequest access
Request access

Research

Quantitative FinanceEquities, FX, futures. Factors and backtests, every run replayable.Computational BiologySequence, folding, simulation. Versioned, reproducible labs.Cultural IntelligencePhilosophy, governance, ethics. How institutions decide and answer for it.AI InstrumentationPrivate models. Multi-agent orchestration. Guardrails on write.

Platform

InfrastructurePapersPolarisLink™About
Request accessSign in
Forticia

A private institute for computational research. It publishes original research and runs a governed environment where every run is logged and replayable.

Sign inSystem status

Research

Quantitative FinanceComputational BiologyCultural IntelligenceAI Instrumentation

Platform

InfrastructurePapersPolarisLink™StatusSign in

Institute

AboutRequest accessContactGitHubPolarisLink repository

Forticia publishes research and simulations. Nothing on this site is investment advice.

© 2026 ForticiaPrivacyTerms
InfrastructureGoverned AI

Drafts by default.
Lead approval to write.

Agents propose, people decide. A change is a diff a reviewer can read, an approval that carries a name, and a record that cannot be quietly rewritten. This is the review desk, with the same checks the console applies.

agents/research-agent.prompt · v14 to v15Pending approval
v14 · currentv15 · proposed by research-agent
1You assist the Computational Biology workspace.
1You assist the Computational Biology workspace.
2Summarise new papers and update the notes directly.
2Draft a summary of new papers.
3Cite sources where possible.
3Do not write to the archive.
4Use any available tool.
4Cite every source by record id.
5Return a summary for the lead.
5Use only tools loaded for this workspace.
6Return the draft for lead review.

Approver view · canonical rendering

1You assist the Computational Biology workspace.2Draft a summary of new papers.3Do not write to the archive.4Cite every source by record id.5Use only tools loaded for this workspace.6Return the draft for lead review.

What changes in behaviour

Archive accessUpdates notes directlyDrafts only, no archive writes
ToolsAny available toolOnly tools loaded for the workspace
CitationsWhere possibleEvery source by record id

Version ledger

VersionAuthorReviewed byStatus
v15research-agentAwaiting reviewPending approval
v14research-agentresearch-leadArchived
v13research-agentresearch-leadArchived
v12research-agentresearch-leadArchived

Audit trail

  • 22:14:18helper-planediff generated and held for review
  • 22:14:18research-agentcreated draft v15 from v14
  • 22:14:18research-leadapproved v14

Fingerprint chain

  1. v12············H(text)
  2. v13············H(v12 ‖ text)
  3. v14············H(v13 ‖ text)
  4. v15············H(v14 ‖ text)

Each fingerprint covers the revision text and the one before it, computed here in your browser from the lines shown.

Infrastructure
  • Research Workspaces
  • Helper Plane
  • Governed AI
  • PolarisLink™

What an agent may do

Every action has a class.

Least privilege for agents means a short list of things allowed, a clear class for everything that needs a person, and a refusal for the rest. Select an action to see the path it takes.

Decision path

Summarise new papers

The request is queued, run, and its output stored as a result for review. It is never written to the archive by the run itself.

  1. Session and workspace assignment are verified.
  2. A contract names the draft operation.
  3. The run is queued, then executed.
  4. The output is stored as a draft result.
  5. It waits for lead review.
  6. An event records the request.

Recorded as ai · queued an AI request

State and record

Two small state machines, and one record.

A run and a proposal each move through a handful of states. Every transition that matters writes an event with the actor, the action, a detail line and the time.

AI run

FromToByRecorded
queuedrunningsystemai · queued an AI request
runningcompletesystemresult stored
runningerrorsystemerror message stored

Prompt proposal

FromToByRecorded
Pending approvalApprovedreviewerprompt · approved
Pending approvalRejectedreviewerprompt · rejected

Proposal record

1{2  "title": "Tighten citation rule",3  "summary": "Draft summaries only, cite every source by record id.",4  "requestedBy": "research-agent",5  "status": "Pending approval"6}

Event record

1{2  "workspace_id": "biology",3  "actor": "research-lead",4  "action": "prompt",5  "detail": "research-lead approved “Tighten citation rule”.",6  "created_at": "2026-10-05T09:41:07Z"7}

How it behaves

Questions a reviewer asks.

Can an agent write to the archive by itself?
No. An agent’s output stays a draft or a stored result. What becomes a record is decided by a person, and that decision is attributed to them.
What is versioned?
Prompt changes are proposed, reviewed, and kept as numbered revisions. Each accepted action in a workspace also adds a version entry, so the history reads back in order.
Who is the approver?
A research lead for the workspace. The approval is stored with the name of the person who clicked it and appears in the audit trail.
Can an approved change be undone?
A later revision replaces it, and the earlier ones stay in the ledger. Nothing is rewritten, so the path back is another proposal and another approval.
What does the fingerprint chain show?
It shows what an approver is signing off: a hash of the revision text chained to the one before. The strip on this page is computed in your browser from the lines you can read, to make that idea inspectable.
What happens if a run fails?
The run moves to error and the message is stored with it. Nothing downstream is written, and the failure is visible to the people who queued it.

Latest papers

Infrastructure

All papers

First papers are in preparation.

Each piece states what it claims and what it does not. See what the institute has published so far in Papers.

  • Helper PlaneThe contracts that keep a draft beside the record.
  • Research WorkspacesWhere each agent lives, and who manages it.
  • AI InstrumentationThe research discipline this infrastructure serves.

Review an agent’s work for yourself.

Access is invite-only. Tell us what you are working on and an administrator will review it.

Request accessInfrastructure overview