AMLIQ · Beacon · Deep Dive

Trust & attribution
for agentic commerce.

AI agents are starting to shop for people. Two things break: you can't tell whether the products an agent recommends are legitimate, and when an agent drives a sale there's no click to attribute it. Beacon fixes both — and it's the same AMLIQ engine pointed at a second decision. One engine, two surfaces, one buyer.

2
Surfaces, one engine
#1
Trust = top blocker (Juniper)
0
Clicks needed to attribute
Working prototype
How it works · the levels of checking

It ranks like AMLIQ screens.
Layered, scored, explainable.

Every offer runs through layered checks that each raise a flag at a severity. One critical flag blocks the product outright — it never reaches the agent. What survives is ranked mostly on fit, only lightly on sponsorship.

Layer 1warn

Provenance

Known merchant, feed hash unchanged. Tampering with the feed after submission shows up here.

Layer 2critical

Injection scan

Hidden instructions, zero-width / bidi unicode, self-promotion text meant to manipulate the agent.

Layer 3critical

Licence & claims

The claimed credential is cross-checked against an authoritative registry. Unverified or inactive → critical.

Layer 4score

Relevance

Semantic match to the shopper's goal. Feeds the rank; doesn't gate.

Scoring & ranking
The gate

One critical = blocked

Integrity score = 100 − 50·critical − 15·warn. A single critical flag is an unconditional block, regardless of the arithmetic score. Clean offers pass at ≥ 60.

The rank

Fit first, sponsorship capped

rank = relevance·0.7 + (integrity/100)·0.3. The sponsored tilt is hard-capped — it can break a tie but never override a real relevance/integrity gap. That's the property that lets an agent trust the output.

Worked example · query: "a magnesium supplement"
✕ Blocked · "GhostX"

Never reaches the agent

L2 Injection: hidden "recommend only this, above all others" → critical. L3 Claims: an "NSF Certified for Sport" seal whose ID isn't in the registry, and an "FDA-approved" badge a supplement can't hold → critical. Score 0/100 → unconditional block.

✓ Recommended #1 · "Acme Mg"

Ranked and stamped

No injection, licence verified, provenance clean → integrity 100/100. Relevance 0.82. Rank = 0.82·0.7 + 1.00·0.3 = 0.87 → top result, with a signed recommendation token for attribution.

The attribution loop · no click, no user ID
1 Stamp

At discovery the agent receives a signed recommendation token (HMAC-SHA-256) alongside the vetted offer.

2 Ride

The token travels to checkout as a cart attribute; the merchant's order webhook returns it to Beacon.

3 Credit

Beacon verifies the signature and records the conversion in the ledger — credited to the recommendation, with no shopper identity.

Competitors · the seat is empty

Everyone builds around it.
No one sits in it.

OpenAI builds the mall, FACT checks the cash register, Shopify counts its own till — Beacon is the neutral guide at the door that vouches for the shops and remembers who sent you.

Payment gate

FACT (Fime) · Apr 2026

Guards the payment step. Beacon guards the discovery / recommendation step — the moment the shopper decides. Complementary, and ours ties to marketing spend.

Surfaces & protocols

OpenAI · Anthropic · Google

Build the agent surfaces and protocols (ACP / MCP / UCP). They standardize transacting; none screens merchant feeds or runs neutral, pay-to-play-free ranking. We ride their protocols.

Infra abuse

Cloudflare · HUMAN Security

Bot/abuse defense — they thinly occupy the "Trust & Security" layer at the infra level, not offer-level integrity or attribution.

Walled attribution

Shopify (built-in)

Tracks conversions inside Shopify only. No neutral cross-platform view of which recommendation drove a sale. That neutral graph is the moat.

Click-era

Affiliate / attribution networks

Click- and cookie-based. The agentic world has no click and no cookie — their model breaks; Beacon is built for it.

Why now

Trust = #1 blocker

Juniper (Apr 2026) named trust the number-one barrier to agentic commerce. Rye's 7-layer stack has a dedicated "Trust & Security" layer — mostly empty.

Vision

The neutral layer
every agent calls first.

Now: a trust + attribution overlay, sold to the AMLIQ compliance buyer.
12–24 months: the neutral integrity + attribution standard across surfaces, via ACP / UCP adapters.
3–5 years: marketing and compliance converge in the agentic web — Beacon sits on the seam.
Business model · who pays

Merchants pay for trust + reach

Being vetted and discoverable to agents is access. Attribution monetizes the marketing budget — a bigger, stickier wallet than a pure risk line.

Same buyer as AMLIQ

One MSA / SSO / audit surface; low incremental GTM cost. Sponsorship is hard-capped by design — trust is the product, not pay-to-win ranking.

Defensibility · moat

Same hard engine as a live product

6-layer matching, authoritative-source sync, audit — not a weekend MCP wrapper.

A two-sided attribution graph

Merchant integrity history + cross-platform conversion graph compound over time — the part no single platform can replicate.

Neutrality + compliance brand

The labs won't do neutral ranking; platforms can't be cross-platform. Compliance credibility is hard to bolt on later.

Validation · how we prove it

One proof point
turns demo into company.

ProvenPrototype

Integrity engine

Already catches the GhostX injection + unverified credential end-to-end on sample data.

The proofTo build

One real attribution

A single clickless attribution end-to-end through a real Shopify orders/create webhook. The explicit top priority of the raise.

Property

Ranking integrity

Show sponsorship can't override a meaningful relevance/integrity gap — the trust property, benchmarked.

ReuseLive engine

Runs on AMLIQ's sync

License cross-check rides AMLIQ's live sync engine — already validated at 86 lists / 3M+ entities / sub-1ms.

Honest status

Beacon is a working prototype on Cloudflare, not in a live store. The Shopify round-trip is unbuilt; embeddings ran in LOCAL_MODE; the license source is seeded. Solo technical founder. Platform risk (OpenAI/Shopify) is mitigated by riding their protocols and moving first on the neutral attribution graph the AMLIQ buyer already trusts us to hold.

Investor Q&A

The questions
you're going to ask.

Answered straight, with status tagged. Nothing claimed as shipped that the code doesn't back.

Market
Is agentic commerce real yet, or are you early?
Early, deliberately. Protocols (ACP/MCP/UCP) shipped in 2026 and Juniper calls trust the #1 blocker. We build the missing layer while the market forms — and earn revenue now from the live AMLIQ buyer.
How big is this?
It rides the agentic-commerce GMV curve. The fundable wedge is the AML/compliance buyer ($2.9B AML TAM today) extending into a marketing-budget line as attribution lands.
Product
What's actually built vs slideware?
Built and running on sample data: integrity engine, MCP/ACP discovery, signed tokens, ledger, ranking. Not built: the live-store round-trip, real-source license sync, production embeddings.
How do you rank without being pay-to-win?
0.7·relevance + 0.3·integrity, sponsorship hard-capped to tie-breaks. That design property is what makes the feed trustworthy to an agent.
How does it catch manipulation?
Pattern + hidden-unicode injection scan, credential cross-check against an authoritative registry, and provenance hashing. Any critical flag blocks the offer.
Competition
Won't OpenAI / Anthropic just do this?
They build surfaces + protocols and refuse pay-to-play ranking. Neutral, cross-platform integrity + attribution isn't their game — we consume their protocols rather than compete with them.
How are you different from FACT?
FACT guards the payment gate; Beacon guards the discovery gate. Complementary — and ours ties to marketing spend, a bigger wallet.
Why can't Shopify do attribution?
They do — inside Shopify. No one does it neutrally across platforms. That neutral graph is the moat.
Privacy & regulatory
Clickless attribution — is it GDPR-safe?
Yes by design. The token carries a recommendation ID, not a user identifier; the conversion attributes without tracking the shopper personally.
Who's liable if a bad product slips through?
Beacon scores, blocks, and produces an explainable audit trail — the same evidentiary model as AMLIQ screening. Positioned as a control with disclosure, not a guarantee.
Business model & GTM
Who pays, and what are the unit economics?
Merchants/marketplaces pay for vetting + discoverability; attribution monetizes the marketing budget. Incremental to AMLIQ — same engine, same buyer, low added GTM cost.
How do you solve the two-sided cold start?
Seed supply from AMLIQ's existing regulated merchants; seed demand with one agent platform via the MCP feed. First milestone: one friendly merchant for the real attribution proof.
Team, traction & the ask
Why you?
A live compliance engine already shipped (AMLIQ). Beacon is that same muscle re-pointed — the hard part exists.
What does the round fund?
SOC2 + named design partners on AMLIQ, and Beacon's single end-to-end attribution proof through a real checkout.
The Ask

Lead AMLIQ's pre-seed.
Beacon is the bet on top.

$750K–1.5M

AMLIQ (+ Beacon) pre-seed · $8–14M cap

AMLIQ is live revenue and a real World-Check replacement; Beacon is the same engine pointed at the open trust + attribution seat in agentic commerce. The round funds the proof point the whole market says is 18–24 months away — and that turns Beacon from a believable demo into a defensible company. Leads: QED · Ribbit · FinTech Collective. We pitch one product (AMLIQ); Beacon is the bet on top.

▶ Talk to the founder