Research / inspect the design

Make the method visible.

Read the draft, inspect the prototype’s rule definitions, and distinguish demonstration checks from measured real-world accuracy.

Versioned document · 12 pages

TFA white paper, draft v0.2

The original supplied document, unchanged. It describes planned capabilities and open questions, not a completed network.

Open the PDF

SHA-256: 418a73761a52067267d6bcbbe308695f3b1003be4f2fb17c9038d56da336215f

Implementation note: after this draft, the owner chose to plan the private pilot on the existing server with separate identities, restricted files, resource limits and overload pauses. This supersedes the draft’s separate-server assumption; the shared-host risks remain.

Prototype rule cards

These are research summaries of the implemented fixture behavior. They are not active public detection rules, approved live seeds or findings about real projects.

FLOW-001 / 0.2.0 · Bounded transfer allocation

Starts with an explicitly selected event and amount, then follows supported transfers within one chain/network and asset. Prototype limits: 3 hops, 24 hours, 100 accounts and 5,000 scanned events.

Amounts use integer base units. Source-first allocation through mixed accounts is a hypothesis; conservative mode stops at observed mixing. Missing history, ambiguous ordering, unsupported effects and resource limits stop a branch. A transfer or split is not an accusation.

SOL-008 / 0.2.0 · Launch relationships

The synthetic prototype distinguishes reuse of an evidenced address in the same role from a bounded funding-path hypothesis. Proposed fixture limits: 30 days and up to 2 funding hops, with known matching network identity.

Shared fee sponsors, services and exchange funding do not prove common human control. A withdrawn seed removes dependent links. Real launch-role decoders and historical evaluation remain required.

SEED-FIXTURE-001 / 1.0.0 · Fixture-only review input

Accepts a deliberately authored synthetic review assertion. It does not create flags from a transfer, liquidity withdrawal, news story, marketing copy, model response or missing data. Fixture seeds must never become live allegations.

A historical seed design is still inactive. Real past-case memory requires sourced evidence, review of the specific role and working correction support.

What has actually been checked?

The Stage 1 review dated 27 September 2026 records 45 passing automated checks, 6 synthetic S10 demonstrations with 14 checks, and 15 passing author-visible regression cases.

These are development checks, not production accuracy. No held-out false-alert rate, missed-event rate or live detection claim has been established.

Four historical reference cases remain unassessable as chain-rule tests because the required historical transaction bodies and decoders are missing. They are not included as successful detections or measured negative controls. The public site does not identify these projects or wallets.

Stage 3 must fix evaluation targets in advance, use held-out cases and report sample size, decision-time evidence, coverage, errors, unassessable cases, delays and costs. Public accuracy reports wait for Stage 4.