Forticia
Request access
Request access

Forticia Research Institute

Research held to
an institutional
standard.

Forticia Research Institute is a private institute for computational research across quantitative finance, computational biology, philosophy and governance, and AI instrumentation. It publishes original research and runs a governed environment where researchers and agents work together, with every run logged and replayable.

Request accessExplore the research
Forticia/Quantitative Finance/factor-research
L
Proposed changeAI draft
1name: cross_sectional_momentum
2universe: liquid_equities
3signal:
4 lookback: 12m_skip_1m
−rebalance: monthly
5+rebalance: weekly
6+turnover_cap: enabled
7costs: venue_aware
8validation:
9 scheme: walk_forward
10 holdout: untouched_until_signoff
11reporting: run_logged
Audit trailRecording
  • SystemWorkspace opened
  • SystemPrompt recorded
  • AgentDraft created
Writes require lead approvalAI draft
Disciplines
Four
Access
Invite-only
AI writes
Lead-approved
Record
Every run logged

Research

Four disciplines. One operating layer.

Isolated workspaces across empirical fields, each with its own tooling, data access, and archive.

Quantitative Finance

Equities, FX, futures. Factors and backtests, every run replayable.

Enter the workspace
Notebook/hypothesis-draft/Simulated series

Does the lagged feature carry information about the next day?

In [1]
series = simulate(seed=13, regime="clustered")
spec = Spec("x_lag1", horizon="1d", costs=Costs.linear(bps=7))
gates = Gates(placebo_p=0.05, holdout_ratio=0.50)
freeze(spec, gates)
Out [1]

Specification frozen62ec1d3e

seed 13
In [2]
explore, holdout = split(series, at=672)
curve = backtest(spec, explore)
verdict = gates.check(curve, holdout=None)
Out [2]
0.751.001.251.50peak0240480672960ExplorationSealed
Simulated series, log scaleSignal seriesReference series
exploration.test_statistic = 0.86
holdout.test_statistic = None
In [3]
null = [backtest(spec.shift(k), explore).statistic
        for k in rng.integers(40, 632, size=240)]
p = (sum(v >= curve.statistic for v in null) + 1) / (len(null) + 1)
Out [3]
95th percentileObserved 0.86−3−2−10123

240 shifted-signal controls. 8 at or above the observed value, p = 0.037.

Computational Biology

Sequence, folding, simulation. Versioned, reproducible labs.

Enter the workspace
Structure workbench/target-0418/104 residues, one chain
Run v3 pinned to 8acd4722
ChainTicks under the ribbon pack against the selection
  • Hydrophobic
  • Polar
  • Charged
  • Gly, Pro
SS5090Conf1102030405060708090100Y40SHDAD45KVFSA50MEQVF55QRVES60HHET64
Backbone
  • Helix
  • Strand
  • Loop
Aligned error
0 Å30 Å

Drag along the matrix to select a range

Selection44–60Helix α2

ADKVFSAMEQVFQRVES

Confidence
89.1lowest 83 at 60
Error within
1.2 Å
Error to rest
5.5 Å
Packs against
13, 17, 21, 25
Runs

Cultural Intelligence

Philosophy, governance, ethics. How institutions decide and answer for it.

Enter the workspace
Governance workbench/Decision record/reopening a verdict

Reopening a closed verdict

Ratified. Owner: lead. Reviewed by someone other than the author.

Proposal

Allow a rejected hypothesis to be tried again after a data or procedure fix.

Options considered
  1. ANever reopen a verdictBuries procedure errors too.Rejected
  2. BReopen when the author asksInvites tuning until it passes.Rejected
  3. CReopen only by improving the testChanges the court, not the defendant.Chosen
Dissent, preserved unedited
Ratified, with reasons. Each clause traces to a principle
Argument map
P1A verdict is only as good as its procedure
P2Any candidate can be tuned until it passes
P3Burying every failure also buries procedure errors
R1Write the change first, retry every case, have it reviewed independently
R2State the cost and decline cases the data cannot answer
ClaimReopen a verdict only by improving the test, never by coaching the candidate
O1Improving the test and rescuing the candidate can look alike
O2Retrying every case may cost more than it is worth
Trace

Clause 1 traces to §4 declare the question first, which grounds the reply.

Changing the test after seeing the result is how a pass gets manufactured. Writing it first means the retry cannot steer it.

History
  • Step 1Proposed by the lead
  • Step 2Options reviewed, two dissents recorded
  • Step 3Ratified with reasons

AI Instrumentation

Private models. Multi-agent orchestration. Guardrails on write.

Enter the workspace
Trace workbench/Update the research note/pinned to config v14
SpanDurationStatus
Policy gatewrite gateNeeds approval
Needs approval
Rule
writes.require-approval
Reason
The archive keeps every version, but a write still changes what readers see. A named person approves the exact payload by its hash.
Enforced by
Deterministic policy outside the model
Approval inboxWrite to research note

Base v14. Requested by orchestrator. Needs lead.

Canonical view of the pending write
{
  "action": "archive.write",
  "patch": {
    "add_sources": ["src-31", "src-32", "src-35"],
    "edit_claims": 2,
    "remove_citations": 1
  },
  "target": "research-note@v14"
}

sha256:e8d5b72c8311145 bytes, keys sorted

Approval binds to this hash. If the payload changes, it no longer applies.

Infrastructure

Built for institutional rigour.

Governance, reproducibility, and controlled access are the baseline, not an add-on.

Quantitative, Biology, Cultural, AI workspaces sit above the Helper Plane, the layer that lets applications reach data only through named contracts. Every read and write crosses it. Governed AI sits on the write path. The perimeter admits invited, role-gated members only.

Governed AI

Drafts by default.
Lead approval to write.

Every change is versioned and attributable. This is the path a single prompt revision takes, from draft to the archive.

biology/literature-summary.promptWaiting for a draft
Prompt revisionv14 to v15
1You assist the Computational Biology workspace.
2-Summarise new papers and update the notes directly.
3+Draft a summary of new papers.
4+Do not write to the archive.
5+Cite every source by record id.
6Return the draft for lead review.
7-Use any available tool.
8+Use only tools loaded for this workspace.
9Keep the draft inside the workspace.

The agent has not proposed a change.

Draft only. This agent has no write access to the archive.

Prompt versions

  • v15research-agentNot started
  • v14research-agentArchived
  • v13research-agentArchived
  • v12research-agentArchived

Audit trail

  • No entries yet.

Approach

A lean core.
Modules on demand.

Forticia stays small by design. A small core handles sign-in, permissions and storage, and everything else attaches to a workspace only when the work requires it.

  1. Lean core

    Sign-in, permissions and storage form the always-loaded core, the kernel. It stays small.

  2. Modular load

    Capabilities attach on demand, per workspace.

  3. Accountable AI

    Human-in-the-loop. Versioned prompts. Full audit trail.

  4. Honest limits

    Findings are written up with their limits, and every run is logged and replayable.

Parts list
No.PartAttaches viaState
1StorageKernel busMounting
2PermissionsKernel busMounting
3AuthKernel busMounting
4Quant toolsHelper Plane contractStaged
5Biology toolsHelper Plane contractStaged
6Cultural dataHelper Plane contractStaged
7AI runtimeHelper Plane contractStaged
8ArchiveHelper Plane contractStaged
9CommsHelper Plane contractStaged
10Lead approvalWrite path gateStaged
11PerimeterRole-gated entryOpen
The kernel, a core of three layers: storage, permissions and sign-in. Six modules mount through Helper Plane contracts, the scoped read and write rules between applications and data. A lead-approval gate guards the write path, and a role-gated perimeter surrounds the whole. Modules attach per workspace.

Ahead

Special purpose vehicles.

A research programme can be isolated as its own vehicle — its own perimeter, its own credentials, the same audit trail. Forticia remains the institute. The SPV sits under it.

Selected
Forticia, the institute
Perimeter
Holds models, archives and instrumentation
Credentials
Issued per vehicle
Audit trail
One trail across every vehicle
Licence
Intellectual property stays with Forticia
ForticiaThe institute

Models

Archives

Instrumentation

Domain
vehicles

Governed
extensions

Audit trail

  • licence issuedQuantitative Financea41f9c
  • access reviewedComputational Biologyb72e10
  • record versionedCultural Intelligencec19d84
  • approval recordedAI Instrumentationd03a5e
  • credential rotatedFund researche58b21
  • licence issuedStructured vehiclesf6c740
  • record versionedQuantitative Financea8e3d2
  • access reviewedAI Instrumentationb14c97
  • licence issuedQuantitative Financea41f9c
  • access reviewedComputational Biologyb72e10
  • record versionedCultural Intelligencec19d84
  • approval recordedAI Instrumentationd03a5e
  • credential rotatedFund researche58b21
  • licence issuedStructured vehiclesf6c740
  • record versionedQuantitative Financea8e3d2
  • access reviewedAI Instrumentationb14c97
Licence flowAudit pathPerimeter

Access

Request research
access.

By invitation or application, reviewed on a rolling basis. The application takes about two minutes.

  1. Apply

    Seven short steps. Nothing is stored until you submit.

  2. Committee review

    Compliance and fit are assessed before any credentials are issued.

  3. Provisioning

    A scoped workspace, role-gated, with a full audit trail from the first session.

Contact

Speak with the
research office.