Decision-Dependence Memory · Pre-Seed

What did this
decision actually
depend on?


Invariant does not record what people say they believed.
It records what their capital repeatedly depended on — and shows how exposed they are if that dependence fails.

02 — The Problem

Belief does not crystallise
at a single moment.

It hardens through reuse. By the time capital is meaningfully at risk, no one can trace how it got there.

03 — The Primitive

Dependence under reuse,
elevated by counterfactual
consequence.

Invariant is built on one locked primitive. Everything else follows from it.

The Load-Bearing Artefact

An artefact becomes load-bearing only when two conditions are met simultaneously:

It persists under reuse. The same dataset, signal, or parameter is relied upon repeatedly — not just once.
Removing it causes material consequences. If it were invalidated or substituted, a meaningful portion of exposure would be affected.

Artefacts that persist without being tested are explicitly labelled: persistent but untested. This is not failure — it is information about current exposure.

01

Reliance ledger

Passively records which components decisions actually used — not what was later claimed in a narrative.

02

Probe coverage map

Shows which load-bearing artefacts have been tested under removal, substitution, or retuning — and which have not.

03

Evidence replay

Any past decision can be reproduced exactly as it existed — not as it is remembered under scrutiny or success.

04 — Product

Exposure replay.
Not retrospective audit.

One concrete output. Produced automatically when capital exposure changes. Answers the question no post-mortem can.

EXPOSURE REPLAY · MOMENTUM_FACTOR_DESK · 2024-11-14 LOCKED
Observed reliance accrual
  • Dataset A reused across six independent signals
  • Static liquidity cost model applied in every sizing step
  • Signal decay parameter fixed at inception — never revisited
Component Probe type Coverage Status
Dataset A Removal / substitution None UNTESTED
Liquidity cost model Alternative cost simulation 1 of 3 regimes PARTIAL
Decay parameter Retuning sensitivity ±20% range only PARTIAL
Vol regime filter Regime switching test Full coverage TESTED
Exposure Statement

Current exposure relies most heavily on Dataset A. That dependence has never been tested under removal or substitution. If invalidated, approximately 40% of factor exposure is affected simultaneously. This is not a recommendation — it is a record of what is load-bearing and what is not.

No forms. No confessions. No acknowledgements. Just: what this depended on.

05 — Market & Timing

Outcomes move faster
than explanations.

The forces creating this gap are structural — and widening.

  • LP due diligence is intensifying post-2022. Reproducible explanations are no longer optional.
  • ML is now standard at quant and multi-strat desks. Model provenance is almost never tracked.
  • UK and EU regulatory pressure on algorithmic decision-making is growing.
  • No current product reconstructs causal reliance at decision-time. GRC is retrospective. Audit is narrative. Invariant is neither.

The buyer is specific.

Primary

CIOs and Heads of Risk at quant desks and multi-strat funds (AUM £500m–£10bn). They own the scrutiny problem and the LP relationship.

Champion

Quant researchers and model risk leads who live inside the causal chains. They understand the pain without explanation.

Economic buyer

COO or CFO where LP relationships are at risk. Invariant is auditable infrastructure — not a product line.

£4.2bn
UK hedge fund
addressable base (AUM)
~£80k
Target ACV,
enterprise SaaS model
0
Direct competitors solving
this specific problem
06 — Traction

Built, live,
and reachable.

The prototype is no longer a concept. Invariant generates Evidence Bundles today — across three realistic fund scenarios — with a public demo anyone can run.

01

Working demo live on nbviewer. Three fictional but mechanically realistic fund scenarios. Evidence Bundles, dependency graphs, counterfactual tests, and probe coverage tables — generated automatically. Publicly accessible, no setup required.

02

Full source code on GitHub. The engine, scenarios, and notebook are public. Clone it, run it, inspect the logic. github.com/wannabequantcmugz/Invariant

03

Product landing page live. axiomcapitalresearch.com/invariant.html — deck, demo, and GitHub accessible in one place. Built to the same standard as this deck.

04

Practitioner outreach underway. Initial contact made with quant researchers at systematic funds. The single question being validated: does the problem of untraceable decision dependence exist at your desk?

05

Zero direct competition confirmed. Compliance tools are retrospective. Audit workflows are narrative. Model risk platforms don't track dependence under reuse. The gap is real, unoccupied, and now demonstrable.

07 — The Ask

Pre-seed.
£150k–£250k.

"What assumption are we most exposed to right now?"

That question is rarely answered structurally. Invariant exists to make answering it a procedure — not a post-mortem.

Use of funds

3–5 paying design partners

Founder-led sales into UK quant desks. 60-day pilots with proof-of-value metrics agreed upfront. First live exposure replay is the singular milestone.

The claim

Procedural authority over evidence

We are not claiming epistemic authority over markets. We claim to know — precisely and reproducibly — what decisions depended on. That is a smaller, testable, expandable wedge.

The vision

The default way institutions answer hard questions

Invariant becomes the infrastructure layer that makes learning survive power, pressure, and success. Not to assign blame. Not to enforce virtue. To make reliance legible.

Why us

The primitive is locked

The core concept — dependence under reuse, elevated by counterfactual consequence — is precise enough to build on and defend. Everything else follows from it.

[email protected] axiomcapitalresearch.com