← Selected work01 / Independent IC simulation / Public version

Physical AI · Neurodiversity

Consumer AI Hardware

Investment question

Can a narrow behavioral wedge become a venture-scale platform?

DecisionPass on evidence quality—not price.

The return model can support venture outcomes across the tested entry range. What it cannot yet support is the probability of reaching them.

FormatIndependent analysis

EvidencePublic sources

Decision modelOwnership + fund return

Evidence cutoff27 July 2026

Early demandVisible / company-reported

Retained behaviorUnknown

Production economicsUnknown

Fund fitScenario-dependent

01 / Thesis

Decision in one line

Early sell-through proves desire. It does not yet prove durable behavior, scalable margins, or fund-level return.

The initial thesis was that a focused physical interface could turn an underserved executive-function problem into a broader recurring software relationship.

The revised view separates the quality of the problem from the investability of the company. A credible wedge can still be too early when retention, production economics, software attach, financing requirements, and exit paths remain largely inferred.

Wedge → system test

Physical presence may reduce the friction of remembering to use a support tool. The unresolved question is whether that advantage creates retained usage and a software relationship—or remains a compelling hardware interaction.

02 / Evidence

Evidence ladder

What the available signal can—and cannot—support.

Company-reported

100-unit early batch

Supports willingness to try and product salience. Does not establish week-8 use, returns, or retained engagement.

Company-reported

400+ discovery interviews

Suggests intensive problem discovery. Interview quality, participant mix, and decision impact remain unverified.

Inference

Hardware + software relationship

Creates a plausible recurring layer. Pricing, attach, churn, inference cost, and gross retention are not public.

Unknown

Production margin

BOM, landed cost, defects, returns, warranty reserve, MOQ, and working capital require verification.

Company-reported

Local voice activation

May improve privacy and latency. Full data flows, retention, deletion, and model-training policies still require diligence.

Unknown

Defensible software moat

A distinctive interface does not by itself establish proprietary behavior data, workflow lock-in, or low churn.

03 / Fund fit

Illustrative return math

My model shows that price is not why I pass.

At an AUD 3m post-money entry, the illustrative outcome is 45.0× gross MOIC and 0.63× of a AUD 5m fund. Even at AUD 8m, the model produces 16.8× and 0.24× of the fund. Those outputs are too strong to support a conclusion that entry price is the binding constraint.

The pass therefore rests on evidence quality and fund construction: the model assumes a AUD 300m exit without underwriting its probability, has no verified retention or margin evidence, and does not establish the capital required to preserve ownership through later rounds.

AUD 70k cheque55% dilutionAUD 300m exitAUD 5m fund

Same cheque, dilution, exit, and fund assumption; entry valuation changes
Post-money entry	Initial ownership	Diluted ownership	Exit proceeds	Gross MOIC	Fund returned
AUD 3m	2.33%	1.05%	AUD 3.15m	45.0×	0.63×
AUD 5m	1.40%	0.63%	AUD 1.89m	27.0×	0.38×
AUD 8m	0.88%	0.39%	AUD 1.18m	16.8×	0.24×

Post-money entryIllustrative fund returned

AUD 3m

0.63×

AUD 5m

0.38×

AUD 8m

0.24×

Dilution sensitivity at AUD 5m post-money and AUD 300m exit
Cumulative dilution	Exit ownership	Exit proceeds	Gross MOIC	Fund returned
55% · base case	0.63%	AUD 1.89m	27.0×	0.38×
80% · hardware downside	0.28%	AUD 0.84m	12.0×	0.17×

Dilution reference—not a measured sector benchmark

I use 55–70% cumulative dilution as a planning range for capital-efficient software and 65–80% for hardware from pre-seed to exit. These are analyst ranges derived from repeated venture rounds, not published hardware-versus-software outcome data. Carta’s stage data shows roughly mid-to-high-teens dilution per primary round; hardware may require more rounds, inventory funding, and bridge capital. The 80% row is therefore the more decision-useful downside case.

Initial concentration

1.4% of fund

A AUD 70k first cheque is manageable in a AUD 5m fund. Initial cheque concentration alone is not a pass reason.

Follow-on reserve

Not specified

Preserving ownership could require several multiples of the initial cheque. Without a reserve policy, the model cannot show total capital at risk.

Pro-rata ability

Unproven

Pro-rata rights, round sizes, cap-table capacity, and whether the fund can finance later hardware milestones are unknown.

Probability weighting

Headline return ≠ expected return

At AUD 5m entry, a 27.0× outcome weighted at a 5% probability becomes 1.35× expected gross MOIC and about 0.019× expected fund return before time, fees, and carry.

04 / Update

How the view changed

The quality of the language should match the quality of evidence.

Initial belief

Focused hardware could become the interface for a recurring executive-function platform.

Disconfirming evidence

Early demand was visible, while retention, subscription attach, unit margin, privacy architecture, and terms remained unresolved.

Decision impact

Replace “promising PMF” with “strong early demand signal” and pass until evidence quality catches up with the narrative.

What would change my mind

Week-8 retained use and cohort-level engagement
Verified hardware gross-margin bridge
Subscription attach, churn, and AI serving cost
Clear data retention and deletion architecture
Financing plan, follow-on needs, and pro-rata requirements
Evidence supporting the probability of a venture-scale exit

05 / Sources

Source ledger

Every external claim is separated from the model.

S01Company-reported / secondary

Startmate founder profile

Used for reported interviews, 20-then-80 unit sell-through, 150 second-round deposits, local voice activation, and production ambitions. Treated as company narrative.

Open source

S02Company primary source

HeySTU product website

Used for current product positioning, feature claims, and the company-reported statement that 100 hand-built STUs are on desks.

Open source

M01Analyst assumption

Illustrative fund model

AUD 70k cheque, AUD 3–8m post-money scenarios, 55% dilution, 80% downside dilution, AUD 300m exit, and AUD 5m fund are analytical inputs—not company or fund terms.

Model disclosure

S03Market benchmark

Carta dilution data

Used only to anchor per-round dilution. The cumulative software and hardware ranges remain analyst scenarios, not observed sector medians.

Open source

U01Unknown / diligence request

Retention, economics, privacy, and terms

No public cohort retention, complete cost bridge, subscription data, data-flow architecture, cap table, or executable financing terms were used.

Decision boundary

Sources last checked 27 July 2026. Independent public-source analysis; no affiliation with HeySTU, Startmate, or any fund.

Research output

Attractive return math does not compensate for unpriced evidence risk.

Request the full memo