Registry

Research system status

Every analytical component of the Macro Intelligence Platform is registered here with its current lifecycle stage, its question, and its validation state. ML models are one class of component among several: statistical engines, event studies, retrieval systems, simulation layers and decision layers are tracked separately because they are validated against different questions.

Components 30 across 7 layers Last updated 2026-10-08 Reviewed Quarterly Owner Tickflow Capital Limited
Validation is split into independent tracks. A component is only ever validated against the question it was built to answer. No component's validation supports another's, and no track gates the others.
Track A — Strategy validation

Whether a trading rule has positive expected edge on unseen data. Evidence: walk-forward profit factor, forward returns, drawdown, Sharpe, trade count. Consumed by the trading account.

Track B — Analytical validation

Whether an engine, model, retrieval system or decision layer answers its defined question correctly. Evidence depends on component class. Consumed by the platform and by downstream research.

How to read this page

Status labels

Every component carries one status label. Labels describe the current state of the work, not the ambition. A component marked architecture has a defined role in the eventual system but no implementation yet.

Live
Producing outputs in the current dashboard or on the trading account.
Shadow
Running against live data without affecting decisions. Outputs reviewed against expectations.
Validated
All applicable gates for its class passed. Eligible for shadow deployment. No live decisions informed.
Research
In active development. No external output. Not yet fit, not yet validated.
Designed
Specification complete. Implementation has not begun.
Planned
Acknowledged in the roadmap. Design work has not begun.
Architecture
Its role in the eventual system is defined; the component itself does not exist.
Blocked
Waiting on an upstream dependency that has not cleared its own gates.
Promotion
Stage transitions require the evidence defined in the validation methodology. No automatic promotion.

System view

Macro Intelligence Platform

A layered analytical system. Data and evidence enter at the top, flow through state and macro understanding, feed into asset research, pass through simulation and context, and arrive at a decision layer that produces a thesis state, a position state, and a risk state. A natural-language question interface sits over the whole system.

Reading this diagram. Each component is listed with the question it answers and its current status. Components in live or shadow are producing outputs today. Components in research or designed are being built. Components in planned or architecture have a defined role but no implementation. This page is deliberately architecture-first: it says what the system is meant to become, with the current status of each part stated plainly.
Layer 01

Data & evidence

Point-in-time integrity, quality control, features, and the macro state vector that everything downstream consumes.
Point-in-Time Data Layer
What did the system know at a given historical moment?
Research
Data Quality Engine
Is the data complete, fresh, and internally consistent?
Live
Feature Engine
What are the derived values of every tracked series?
Live
Macro State Engine
What is the current macro environment?
Live
Layer 02

Macro understanding

What changed, how dimensions relate, how unusual today is, and what regime the environment resembles.
Change Engine
What changed over 1D, 5D, 20D, 60D, and in which direction?
Research
Relationship Engine
Are historical cross-asset relationships behaving normally?
Designed
Statistical Laboratory
What variables statistically explain the observed variation?
Designed
Macro Regime Model
What macro state are we in, with what probability?
Research
Regime Transition Engine
Is the macro state itself changing?
Blocked
Layer 03

XAU research

Gold-specific analytical components: direction, distribution, horizon, history, event reaction, and attribution.
XAU Direction Model
Where is Gold biased across 1D–60D horizons?
Designed
XAU Return Distribution
What range of outcomes is probable, with what tails?
Blocked
XAU Horizon / Decay
How long does the current edge typically persist?
Designed
XAU Historical Analogues
When have we seen a comparable state before?
Planned
XAU Event Reaction
How does Gold respond to this type of surprise?
Research
XAU Attribution
Why did Gold move?
Designed
Layer 04

XAG research

Silver-specific components with their own transmission mechanism. Not a copy of the gold stack.
XAG Direction Model
Where is Silver biased, given its own drivers?
Designed
XAG Return Distribution
What range of Silver outcomes is probable?
Blocked
XAG Horizon / Decay
How long do Silver-specific edges persist?
Designed
XAG Historical Analogues
When have Silver conditions looked like this before?
Planned
XAG Event Reaction
How does Silver respond to this type of event?
Blocked
XAG Attribution
Why did Silver move?
Designed
Layer 05

Simulation & context

Hypothetical questions, event classification, cross-asset divergence, and research triggers.
Scenario Engine
What happens if X changes?
Designed
News & Event Intelligence
What happened, and does it matter for XAU/XAG?
Designed
Cross-Asset Divergence
Are other markets confirming or contradicting the current state?
Designed
Alert / Condition Engine
Has a validated condition been triggered?
Research
Layer 06

Decision synthesis

Thesis, position, and risk. Three separate layers because the thesis can be bullish while the position is wait, and the risk layer has final authority.
Thesis Engine
What does the combined evidence imply?
Architecture
Position Engine
Entry, add, hold, reduce, exit, wait, or sit out?
Architecture
Risk Engine
How much exposure is permitted?
Architecture
Layer 07

Interface

Natural-language routing over the analytical system. The model routes and synthesises; the numbers come from the engines below.
Question Engine
What does the user want to know?
Architecture

Coverage

Questions the platform is built to answer

The system is defined by the questions it is designed to answer. Each question maps to a specific component or chain of components in the architecture above. The column at the bottom of each cell shows whether the question is currently answerable, or awaiting the components that would answer it.

State
What is happening right now?
Answerable · live
Change
What changed over the last 5, 20, 60 days?
Answerable · live (partial)
Explain
Why did Gold move?
Designed · not built
Compare
How does today compare with a historical state?
Planned
Relationship
Is Gold behaving normally relative to real yields?
Designed · not built
Regime
What macro regime are we in?
Research · fit pending
Forecast
What is the expected direction over 1D–60D?
Designed · not built
Distribution
What range of outcomes is probable?
Shadow · dependency pending
History
When have we seen this before?
Planned
Event
How does Gold react to this type of surprise?
Research
Scenario
What happens if real yields rise another 50bp?
Designed · not built
Attribution
What factors explain today's move?
Designed · not built
Horizon
How long does this thesis normally matter?
Designed · not built
Thesis
Is the macro thesis strengthening or weakening?
Architecture
Position
Is this a long, short, wait or sit-out environment?
Architecture
Risk
How much exposure is justified?
Architecture

Registry view

Tracked artifacts

The architecture above describes the system. The registry below describes the artifacts that exist or are being built. Entries marked internal carry version strings that have not been independently verified.

Registry snapshot — 2026-10-06. Artifact identifiers are stable; dataset versions increment with material changes.
Artifact Class Version Stage Dataset Source Last evaluated
macro_regime_gmm ML model 0.1.0 Research regime_ds_v4 design doc In design
regime_transition Statistical model 0.3.0 Blocked regime_ds_v4 internal 2026-05-18
xau_conditional_dist Forecast model 0.2.1 Shadow xau_macro_v2 internal 2026-05-20
xag_conditional_dist Forecast model 0.2.1 Blocked xag_macro_v2 internal 2026-05-20
event_reaction Event study 0.1.4 Research events_v1 internal 2026-05-15
historical_analogues Retrieval — Planned — planned —

Regime classification

macro_regime_gmm : 0.1.0 · design locked, fit pending

Research

Classifies the current macro regime from a point-in-time composite of four canonical dimensions — growth, inflation, liquidity and risk — and will publish posterior probabilities across a set of empirically derived macro states once the model is fit. The model does not output a position, size or buy/sell signal. It is an upstream feature layer for XAU/XAG research.

Feature space
GROWTH · INFLATION · LIQUIDITY · RISK
Dataset
regime_ds_v4
Cluster count (K)
Under selection (candidates: 3, 4, 5)
Walk-forward folds
Not yet run
Leakage checks
Pending vintage validation
Access
Waitlist open
Current status and known limitations
  • Point-in-time vintage handling is unresolved (blocking). The composites are built from macro series subject to revision. The "no look-ahead" property is contingent on using first-release or otherwise point-in-time vintages for every input. Until this is specified and tested, no fit is run.
  • Design locked; model not yet fit. Cluster-count selection, parameter fitting, and walk-forward evaluation have not yet been performed.
  • LABOR and HOUSING are excluded from V1. Both are being redesigned with explicit directional semantic contracts before they enter the regime feature set. They remain available to the broader Macro State layer used downstream.
  • Cluster count is not assumed. K will be selected from {3, 4, 5} on the basis of BIC, AIC, stability under resampling, persistence over time, and economic coherence.
  • No separation metric, walk-forward result, or leakage check has been published. Any earlier figure attributed to this model should be disregarded.
Join the waitlist Validation methodology

Regime transition

regime_transition : 0.3.0 internal

Blocked

Estimates the probability of transition between the macro regimes published by the upstream regime model. Blocked until the upstream model has a fitted, stable cluster assignment.

Stage
Blocked
Dataset
regime_ds_v4
Upstream dependency
macro_regime_gmm (unfit)
Last evaluated
2026-05-18
Known limitations
  • Cannot be validated until the upstream regime model is fit and its clusters are stable.
  • Transition estimation is heavily sample-constrained: only a small number of full regime cycles exist in the historical record.
  • Results are highly sensitive to the definition of regime boundaries.

XAU conditional distribution

xau_conditional_dist : 0.2.1 internal

Shadow

Conditional return distribution for XAUUSD. Intended to expose a full quantile set, probability thresholds, expected volatility, MAE/MFE expectations, and tail-risk estimates — conditional on the macro state, regime, and asset-level inputs. Currently running in shadow mode without a fitted regime layer.

Stage
Shadow
Dataset
xau_macro_v2
Conditioned on
Macro state (interim) · regime (pending)
Upstream dependency
macro_regime_gmm (unfit)
Known limitations
  • Distribution is currently conditioned on an unfitted regime state; calibration cannot be completed until the upstream model is fit.
  • Historical sample size is limited for tail regimes; tail estimates carry wide uncertainty.
  • Not offered as a public output in the current phase.

XAG conditional distribution

xag_conditional_dist : 0.2.1 internal

Blocked

Conditional return distribution for XAGUSD. Silver has its own transmission mechanism — China growth, industrial metals, copper, global manufacturing, the gold/silver ratio — and is validated independently. Blocked pending that work.

Stage
Blocked
Dataset
xag_macro_v2
Reason
Independent validation required
Last evaluated
2026-05-20
Known limitations
  • Contract specifications differ materially from gold; parameters do not carry across instruments.
  • Requires its own walk-forward validation before any live use.
  • Silver-specific transmission model is designed but not yet implemented.

Event reaction

event_reaction : 0.1.4 internal

Research

Studies XAUUSD behaviour around scheduled macro releases — inflation prints, employment data, central-bank decisions — and conditions historical responses on the surprise direction, magnitude, and macro state.

Stage
Research
Dataset
events_v1
Last evaluated
2026-05-15
Publication
None pending
Known limitations
  • Sample sizes per event type are small; results are noisy at the tails.
  • Overlaps with existing calendar-blocking logic in the trading engine are being reviewed.

Historical analogues

historical_analogues : —

Planned

Retrieval of historical periods that resemble the current macro state, with the subsequent forward behaviour of XAU and XAG. The output is a distribution of outcomes from comparable historical states, along with similarity scores and stated differences. Not a probability; a conditional historical comparison.

Stage
Planned
Expected dependency
Macro State Engine, Regime Model

Separate system

Internal trading systems

The trading engine is not part of the Macro Intelligence Platform. It is validated against a different question (does the strategy make money on unseen data?) and is not consumed by any analytical component above. It is listed here for completeness and because its operational status matters for the firm's capital.

TIES — systematic trading engine

ties_engine : run012

Validated

The execution engine built to run the firm's own capital on XAUUSD. Combines a signal generation pipeline, a sequential pre-trade risk framework, an execution bridge with preflight validation and fill reconciliation, and an append-only audit log with trace-ID correlation. Walk-forward validation is complete. The strategy has not been deployed to live capital, and no live track record exists. Walk-forward and simulated-holdout results are published on the main site.

Status and known limitations
  • Validated, not deployed. The strategy has passed its walk-forward gates but has not been run against live capital. The registry does not treat walk-forward or simulated-holdout results as a live track record — see the evidence-ladder definition in the validation methodology.
  • Forward-period figures are gross of commission and swap; slippage and queue position are not modelled in the backtest.
  • Single instrument and one directional regime in the current phase. Other configurations require their own validation.
  • Trade count is small relative to institutional samples; Sharpe and t-statistics carry wide uncertainty.
  • Not offered for external operation in the current phase.
  • Validation of TIES is independent of every component on this page. It is not evidence for, and is not supported by, any analytical component above.
See the full validation record Evidence-ladder definition

Open source & previous research

Independent projects

Released under permissive licences and outside the Macro Intelligence Platform. Listed here as engineering references, not as part of the analytical stack above.

Tickflow Intelligence

tickflow_intelligence : 1.0.0-beta.1

Open source · MIT

AI stock intelligence platform combining LSTM neural networks for directional price prediction, FinBERT for financial-news sentiment analysis, and a hybrid model that weights both against technical indicators. Full REST API with authentication, rate limiting and Swagger documentation. Source, deployment and model space are public.

Licence
MIT
Stack
Django · React · PyTorch · FinBERT
Deployment
Vercel · Render · Hugging Face Spaces
Last evaluated
2026-04-12
Known limitations
  • Predictions are directional estimates, not forecasts. They are informational and do not constitute advice.
  • Accuracy figures are reported in the repository and are bounded by the evaluation sample described there.
  • Upstream data sources (news and price feeds) are third-party and subject to their own availability.
Live demo Source API docs

Transparency boundary

What this page does not publish

This registry is intended to be substantive and honest, but it is not exhaustive. The following are deliberately withheld from public pages, including this one:

Parameters, rules and thresholds

Specific numerical parameters that determine how components behave — including entry thresholds, stop distances, position-sizing inputs, session blocks, scoring weights, and the sign conventions or weight vectors of any composite — are not published. These are the substance of the strategies.

What is published and what is not

  • Published: the layered architecture, the question each component answers, current status of each component, the identity of the canonical composites used as features, walk-forward window counts and outcomes, forward-period summary statistics.
  • Not published: the specific value of any threshold, parameter, or time window; the sign convention or weight of any composite component; the exact composition of any dataset; specific model hyperparameters; internal artifact identifiers beyond the registry entries above.

Transparency about process and outcomes is compatible with discipline about parameters. Both are required to run this firm.

History

Recent registry updates

Material changes to any component are recorded here. Corrections to published entries are recorded as corrections rather than as silent edits.

2026-10-08

TIES — status and metrics corrected

The TIES card previously carried a "Production" badge and a full set of forward-period metrics — return, Sharpe, drawdown, trade count, Monte Carlo — that did not match the applied record on the homepage and that, on their own, implied a live capital track record the firm does not have. The card now carries a Validated status, matching the validation methodology's lifecycle vocabulary; states plainly that live deployment has not occurred; and links to the homepage's validation section and to the evidence-ladder definition for the underlying figures. The four metrics have been removed from this page so the record lives in one place and cannot drift.

2026-10-06

Page restructured around the analytical architecture

The page has been reorganised from an artifact list into an architecture-first document. Seven layers are now shown — data and evidence, macro understanding, XAU research, XAG research, simulation and context, decision synthesis, and interface — with the current status of each component. Component classes are now distinguished (ML models, statistical engines, event studies, retrieval, decision layers, interface). TIES and the open-source project have been moved out of the main registry to their own sections. A question-coverage table has been added.

2026-10-06

macro_regime_gmm — entry corrected

Prior entries overstated the artifact's state by attributing a completed 12-fold validation, a separation score of 0.79, and a Shadow deployment stage. Those claims were not supported and have been removed. The feature space is locked; fitting, cluster selection, walk-forward and leakage testing remain outstanding.

2026-05-25

TIES — forward-period metrics refreshed

Forward-period summary for the 2025 – Q1 2026 window re-evaluated from the trade log. No strategy parameters changed.

2026-05-24

Composite sign conventions reviewed

LABOR identified as having mixed directional semantics; HOUSING identified as conflating activity, price and financial-condition concepts. Both flagged for redesign. Regime feature space reduced to four coherent composites.

2026-05-20

XAU and XAG conditional distributions — shadow deployment

Both promoted to shadow. No external outputs. Downstream dependence on a fitted regime model remains outstanding.

2026-04-22

SILVER disabled pending independent validation

XAGUSD research paused. Contract specifications differ from gold; parameters do not carry across instruments. Requires its own walk-forward pass before reinstatement.

2026-04-12

Tickflow Intelligence — public beta

First public release under MIT licence. Source, API documentation and deployment pipelines published.

2026-01-13

Registry established

Tickflow Capital Limited incorporated in Kenya (PVT-6Y1YWM6V). Registry initialized with the TIES artifact.

Questions about a component

If you need to understand a specific validation result, request access to the Regime Module waitlist, or discuss research collaboration, use the contact form or email directly.

Contact form Email Validation methodology