Cangler Quant · Whole-of-ASX research intelligence

Research intelligence for the whole ASX, built on point-in-time evidence.

Cangler Quant helps analysts, funds, and data teams explore ASX companies, announcements, and signals through point-in-time-controlled data, source lineage, and governed AI research workflows — so research is faster, reproducible, and trustworthy.

Research infrastructure only. Not financial advice. No performance claims.

01 / The problem

ASX research is fragmented, hard to reproduce, and increasingly hard to trust.

Analysts and investors stitch together screens, announcements, spreadsheets, broker notes, and manual judgement. The work is slowed by stale data, ambiguous sources, point-in-time leakage, inconsistent entity mapping, and backtests that can’t be reproduced.

AI summaries make it worse when they aren’t tied back to evidence. Cangler Quant is being built to solve that research-infrastructure problem — coverage of the whole market, with the controls that make an output worth trusting.

02 / The edge

In active development

Structural edges, drawn from data unique to the ASX.

Beyond standard factors, we’re engineering signals from Australian-market disclosures that are rarely packaged cleanly for point-in-time quantitative research — corporate lifecycle, spatial geology, and the relationships between the people and companies behind each listing.

01

Cash-runway intelligence

In progress

Reads quarterly cash-flow filings to estimate how many quarters of funding a company has left — and flags the dilution risk that tends to arrive before a raise.

02

Geological nearology

In progress

A point-in-time map of mining tenements that links explorers to nearby discoveries and the moves of same-basin peers.

03

Director & promoter networks

In progress

Entity-resolved track records across boards and substantial holders — surfacing insider conviction and governance risk.

04

Narrative-drift detection

In progress

Detects when management quietly shifts the metrics they emphasise — a documented tell that the story is moving faster than the business.

Glass-box AI

We don’t make the model the source of truth.

You can’t make a frontier model’s weights transparent — so we solve the black box architecturally, by not making the model the authority. The LLM can draft, route and explain; but contracts, semantic metadata, validation gates, review cycles and an evidence vault decide what is true. The model is wrapped in a governed evidence architecture — a glass-box surround.

Questionplain EnglishGroundingVerifierReconcileReviewGolden evalsAbstentionObservabilityContractsLLMnarration onlyGrounded answercited · provenance · lineageHonest abstentionevidence missing — says soASKGOVERNED EVIDENCE SURROUNDTRUSTED OUTCOME

The model drafts in plain English — but every claim has to clear the governed evidence surround. The answer comes back grounded and cited, or the system abstains and says exactly what’s missing. The model narrates; the surround decides what’s true.

Model as interface, not authority

The LLM drafts, routes and narrates. Source-owned contracts, tests and the evidence vault decide what is true — it can’t invent facts, run its own queries, or answer from memory.

In progress

Grounding & provenance

Every accepted value points back to the exact source page, section or cell. No provenance, no trusted answer — unsupported fields are rejected, quarantined or flagged unresolved.

In progress

Response verifier

Before an answer is returned, a verifier checks every factual claim against the retrieved evidence bundle and strips anything a citation can’t support.

Planned

Fail-closed abstention

When the evidence can’t prove a claim, the system says so and names what’s missing — never “probably”, “usually”, or “based on the docs”.

In progress

Adversarial review

Separate reviewer agents and deterministic gates attack each output for hallucination, look-ahead leakage, overclaiming, privacy and source-rights before it’s trusted.

In progress

Golden-question evals

Extraction routes are benchmarked against human-adjudicated “golden” datasets. AI reviewers can flag issues — they can’t promote their own conclusions to ground truth.

In progress

Numeric reconciliation

Figures read from filings are cross-footed and checked for sign, period, unit-scale and magnitude, so a misread number can’t flow silently into a feature or signal.

Planned

Runtime observability

Every answer logs its input hashes, retrieved evidence, model version, tool calls, validation and reviewer results — an evidence trail you can audit after the fact.

In progress

Ask the glass box

An explainer engine over a semantic metadata graph.

Ask in plain English what a value means, how it’s calculated, or why a stock ranked — and get an answer traced to deterministic evidence: the formula and inputs, the source citations you can click, a lineage subgraph, and an explicit “why this can’t be answered” when the graph can’t prove it. Every meaning has exactly one owner in a version-controlled contract; generated prose and dashboards are read-only views, never the authority.

Planned

Keeping the quant models honest, too

The black box isn’t only the LLM. Complex ML can stay complex — the decision process around it has to be inspectable and governed.

Interpretable diagnostics

Feature attribution read with care — alongside the model as evidence, never instead of it.

Calibration & uncertainty

Confidence is calibrated and labelled as model confidence — never passed off as a real-world probability.

Drift & decay monitoring

Features and models are watched for drift, with champion–challenger evidence required on every promotion.

PIT-safe reproducibility

Frozen fold boundaries and point-in-time manifests, so a result can be re-derived exactly as first produced.

The glass box isn’t a single model choice — it’s a governed evidence architecture around the model: a semantic metadata graph, source-owned contracts, provenance, validation gates, review cycles and an evidence vault. Described here as designed — much of it is in active development, and the platform claims evidence-grounded, provenance-backed and benchmarked outputs, never “hallucination-free” AI.

The research factory

The Agentic Alpha Lab — hypotheses proposed, tested, and falsified on the record.

A governed research factory — not a prompt-driven strategy generator. Agents turn academic literature and filings into feature candidates, search the approved hyperparameter space, and stress every result through walk-forward validation, negative controls, and leakage probes. Failed trials stay on the record, identities resolve point-in-time, and nothing promotes itself — discovery and promotion are separate gates.

· Literature → feature research· Hyperparameter optimization· Point-in-time entity resolution· Falsification-gated promotion

In active development. Console is illustrative.

Agentic Alpha Lab
illustrative · in active development
Campaign
Hypothesis
Experiment
Falsification
Result card
Promotion
01Hyperparameter optimization
trial 1,204 · running
objtrials →
best so far ↑pruned trials retainedwalk-forward objective
02Search space
parallel coordinates
learningratetreedepthfeaturewindowL2 regn_estlabelhorizonbest configuration
approved variables onlyhard exclusions enforced
03Literature → feature research
research primitives

Cross-sectional momentum under thin liquidity

arXiv · q-finlicence okconf med

Order-flow imbalance & short-horizon drift

SSRNlicence okconf high

Post-announcement drift in small caps

journallicence okconf med
↓ extractedfeature candidate ×7hypothesis bound

advisory evidence only · never implementation authority

04Entity resolution
ticker reuse · PIT-bounded
ticker “AQX”reused
’08–’14 · delisted
entity_id 0x8f2a…
’21–present
entity_id 0xb41c…
canonical entity_idroot/alias mapsurvivorship-safe1 ambiguous → blocked
05Result card— evidence bundle, falsification, blockers
1,204 trials · 1,198 rejected · failures retained
PIT-safe ✓entity-resolved ✓source-rights ✓walk-forward ✓negative controls ✓leakage probes ✓promotion: independent review required

No result promotes itself. Discovery and promotion are separate gates — independent review + domain-engine verdicts required.

03 / Evidence-backed

Every output carries its own evidence.

Not a black-box score — a research card. Each one is designed to show its sources, its point-in-time boundary, its data-quality state, the model’s status, and what it doesn’t know — so you can judge whether to trust it in seconds.

  • Source lineage on every value
  • Point-in-time boundaries, enforced
  • Data-quality and model status, shown
  • Assumptions and limits, stated
ASX: XMPL·Example Resources Ltd

Research card

Illustrative
Cash runway (est.)~3 quarters
Funding pressure (current)Elevated
Narrative focusShifted ×2 / 18mo
Liquidity (20d)Thin

Evidence trail

Source
Appendix 4C · 2026-Q1 · filed 28 Apr
Point-in-time
as of 2026-05-15
Data quality
verified
Model
candidate · not promoted
PIT-safeno advice · no forecast

04 / Lineage & proof

Every insight traces back to its source.

No value stands alone. Each one is designed to trace the whole way back — model, features, joins, the original filing — as an append-only evidence chain, cryptographically signed at every link. So a result can be re-derived and verified, not taken on faith.

Signed with a post-quantum signature profile (FIPS 205 / SLH-DSA). In active development.

  1. 01Insightsha-256 · signed

    the value shown on a research card

  2. 02Modelsha-256 · signed

    candidate · version · calibration

  3. 03Featuressha-256 · signed

    the engineered signals it used

  4. 04Joinssha-256 · signed

    the point-in-time-safe spine

  5. 05Sourcesha-256 · signed

    the original ASX filing · vintage · rights

Designed against

We map our controls to the standards institutions expect.

The platform is designed against — and maps control objectives to — the Australian regulatory, security, AI-governance and risk-data frameworks an institutional reviewer expects, so readiness can be evidenced as we build.

Australian regulatory & prudential

8 frameworks

ACSC Essential Eight

maturity level 2+

APRA CPS 234

information security

APRA CPS 230

operational risk

APRA CPG 235

data risk management

ASIC Market Integrity Rules

market integrity

ASIC CP 386

planned

automated-trading controls

AU Voluntary AI Safety Standard

responsible AI

Privacy Act 1988 & APPs

data protection

Security & zero-trust

8 frameworks

NIST 800-207

zero-trust architecture

ACSC Modern Defensible Architecture

secure-by-design

ISO/IEC 27001

information security

ISO/IEC 27701

privacy information management

SOC 2 (TSC)

service controls

NIST CSF 2.0

cyber risk

CIS Controls

baseline security

ACSC Secure AI Guidelines

secure AI development

AI governance, risk-data & assurance

4 frameworks

ISO/IEC 42001

AI management system

NIST AI RMF

AI risk

BCBS 239

risk data aggregation

ISO 9000

quality management

Alignment and control-mapping only — not certification, audit, or regulatory approval. Framework alignment is tracked as readiness mapping and evidenced where controls are in place.

05 / The platform & roadmap

In active development

Cangler Quant is built on Cangler OS.

Cangler OS is a horizontal institutional operating system for governed data analytics and AI. Cangler Quant is the first vertical product built on it — and at its core is UICE, the engine that routes every workflow through policy, compliance, assurance and point-in-time gates.

Cangler Quant

quant-finance vertical

Future verticals

reviewed strategy only

Cangler OS

core engine · UICE

A horizontal institutional operating system — governed data, assurance, research, simulation, evidence and command/control.

built by Cangler Labs

UICE — the Unified Intelligent Coordinating Engine

Not a free-form AI orchestrator — a deterministic control kernel. It’s designed to coordinate the platform’s governance engines so that no output is trusted without the evidence to back it.

UICE · coordinating kernel
illustrative concept

Policy engine

Operating policy — assignment, review, access, capacity, promotion and publication.

Compliance engine

Applicable regimes, obligations and controls — with the gaps, blockers and evidence behind them.

Assurance kernel

Classifies the strength and limits of the evidence behind every decision.

Evidence vault

Append-only custody of run snapshots, audit chains and proofs.

UICE

Unified Intelligent Coordinating Engine

The deterministic kernel that routes every workflow through these engines — and holds a blocker when the evidence is missing, stale or insufficient.

Zero-Trust engine

Who — or what — may access each dataset, model, artifact and data-room surface.

Point-in-time engine

Verifies what was knowable on the date, and holds back look-ahead leakage.

Data-quality engine

Freshness, schema drift, nulls, referential integrity and lineage completeness.

Model-risk engine

Eligibility, validation depth, drift, calibration, promotion and rollback.

What it powers

Cangler Nexus

Live visual observability

A live, spatial observability layer that maps the platform's state — portfolio risk, model confidence, point-in-time safety, entity graphs and research campaigns — into scenes you can read in seconds.

Agentic Alpha Lab

Governed research factory

Where alpha hypotheses are proposed, sandboxed and tested against falsification controls. Failed trials stay on the record; promotion is gated by independent review and a full evidence bundle.

ASX Feature Store

Pre-engineered · PIT-by-design

600+ engineered features across the whole ASX — engineered to be survivorship-bias-free and point-in-time-correct, and drawn from sources distinctive to the Australian market. Coverage and validation depth expand as the platform matures.

On the roadmap

Sequenced with design partners and shipped under the same point-in-time and evidence controls as everything else.

API & export layer

Planned

Structured ASX intelligence out — a programmatic API, exports, and integrations into existing research stacks.

Historical-filing intelligence

Planned

Searchable, point-in-time intelligence across historical announcements and filings, with every result traced to source.

Resource & JORC intelligence

Planned

Structured resource and reserve signals from JORC-coded reports — added depth for the ASX resource sector.

Governed end-to-end

From raw ASX data to a scene you can trust.

ASX market data, announcements and alternative data are coordinated by UICE through a governed data-engineering layer — graded by the assurance kernel, checked against compliance controls, point-in-time-bounded and signed with lineage. The Alpha Lab then coordinates the feature store and models built on that governed data — all rendered as a scene you can read in Cangler Nexus.

DATA SOURCESGOVERNED DATA ENGINEERINGFEATURE & MODEL FACTORYVISUALISEDASX Market DataASX AnnouncementsFundamentalsAlt DataUICEcoordinates data engineeringAssuranceevidence gradedCompliancecontrols checkedPoint-in-timeno look-aheadLineagetracked + signedAlphaLabcoordinates store + modelsFeature Store600+ PIT featuresModelscandidate · gatedCangler Nexusvisualises live state

Illustrative · governed data flow · in active development

Cangler Nexus

Live visual observability of the whole platform.

A spatial observability layer that maps the platform’s live state — entity graphs, model confidence, point-in-time safety, evidence and research campaigns — into scenes you can read in seconds, with every node tracing back to its source.

In active development.

Cangler Nexus
live visual observability · illustrative

06 / Use cases

Built for ASX-focused research workflows.

  • 01ASX-wide opportunity screening
  • 02Small- and mid-cap discovery
  • 03Announcement & catalyst monitoring
  • 04Watchlist & portfolio intelligence
  • 05Feature & signal research
  • 06Data-quality & source-lineage review

07 / Approach

Evidence over hype — by construction.

Point-in-time by default

Survivorship handling, entity resolution across ticker reuse, and date-correct joins — enforced, not bolted on.

Source-owned & reproducible

Outputs trace to sources with explicit assumptions and limitations, so results can be re-derived and audited.

Governed AI

Models accelerate research within claim controls and review gates — no unsupported certainty, no advice.

Detailed architecture, methodology, and signal logic are shared selectively with design partners under NDA.

Private design-partner beta

Help shape a whole-of-ASX research platform before public launch.

We’re partnering with a small number of ASX-focused analysts, funds and data teams. As a design partner you get early access, a direct hand in shaping the evidence cards and data products we build, founding-partner terms, and a direct line to the team — in exchange for candid feedback on what’s actually useful.

08 / About

Built by Cangler Labs.

Cangler Labs builds Cangler OS — a horizontal operating layer for governed, evidence-backed research. Cangler Quant is its first vertical: quantitative research intelligence for the Australian market.

The work is engineering-led and evidence-first: point-in-time discipline, source lineage, and an honest line between what’s live today and what we’re building next.