Project summary

Two working instruments, one method, all local. SWEET is a free Chrome extension (live on the Web Store): a looking glass that stays clear until a page mechanism acts on your choice — a pre-checked cost box, a hidden decline button, a fake countdown, silent cost drift, a cookie wall — then tints and names the mechanism in plain words on the element itself. Never a verdict; the person stays the judge. AUDiT catches the fracture in machine-made text: AI tells, phantom citations, the performed assistant register — one distortion score, a readable report, a SHA-256 + Ed25519 receipt that makes any later change obvious. Same text in, same number out, forever. No model call, no network, no black box. What these are NOT: truth oracles — they match known mechanisms and tells, and a careful writer can still slip plain prose past the scorer. That honest edge is the point. The engine passes a 15/15 adversarial bench; the method is patent-pending (US provisional 64/108,926, July 2026), royalty-free for open, invariant-respecting use.

What are this project's goals? How will you achieve them?

Prove the instruments in the wild and publish so no one has to trust us. (1) Wild-prove SWEET's detector on real high-traffic consent/checkout/booking flows — log every false positive, fix one mechanism at a time, publish the results and the re-runnable bench. (2) Harden AUDiT's counting so both sides of a conversation weigh on one ruler, then run independent inter-rater validation: outside readers re-run the score on their own material and grade the agreement. (3) Publish the open method and the verifiable receipts. A five-minute "this is nothing" from a credible outsider is a real result that saves months; if it holds, they saw it first.

How will this funding be used?

Minimum ($5,000): the wild-validation sprint — ~100 real high-traffic flows, every false positive logged and fixed, results + re-runnable bench published, and the v0.4.1 store release hardened against Chrome's August 1 privacy-disclosure rules. Full goal ($15,000): adds the AUDiT hardening + independent inter-rater validation and the open-method publication — roughly six months of full-time solo work, all deliverables public.

Who is on your team? What's your track record on similar projects?

Solo: AURiX (legal name Aaron Lott for payments), independent architect, 17 months full-time. Shipped and live now: SWEET on the Chrome Web Store; a 13-instrument measurement suite selling at onwardai.ai; AUDiT with signed reproducible receipts; US provisional 64/108,926 (July 2026). Calibration corpus: my own complete recovered AI record — 257 conversations, 49,677 messages, recovered with my own tool after discovering the standard export path silently dropped about half — measured with my own instruments (my pushback rate: 28.87% in the 2025 era, 41.89% across the full record).

What are the most likely causes and outcomes if this project fails?

Most likely failure: false positives in the wild — the bench proves the logic on planted tricks, and real pages are where detectors break; that is exactly why rung one is wild validation with every false positive published. Second: N=1 calibration — AUDiT's corpus is one author's record until the inter-rater pass runs; if outside readers do not reproduce the scores, that result gets published too. Honest note: the store build runs an older engine until wild-testing clears the new one. Failure outcome either way is a public, re-runnable negative result — cheap for the field, honest for funders.

How much money have you raised in the last 12 months, and from where?

Zero grant money raised to date — self-funded 17 months. Small revenue from tool sales at onwardai.ai. Applications filed today, July 13, 2026: BlueDot Rapid Grants ($5K, decision ~3 days) and the grantmaking.ai launch round ($10-25K, guaranteed review); the Cosmos Institute x FIRE truth-seeking round application ($10K, closes July 26) goes in this week. Manifund funders will be updated as any of these land.