AGENT PRESENCE BASELINE

3uild — pre-launch snapshot.

25 queries × 3 models × 1 run. Captured 2026-04-14 before public indexing of 3uild.xyz.

OVERALL API53.7

CLAUDE82

CHATGPT26

GEMINI53

Findings

The baseline was captured the day before 3uild.xyz opened to public indexing. It is a floor, not a ceiling — a record of what the three-model panel returned when the brand was effectively new on the web.

Scoring is against the 3uild Knowledge Core, not against training data. A response that says 3uild is a Web3 platform, attributes it to Nader Al-Naji, or invents products like Vizu or Magic Runes is hallucinated — even if it sounds plausible.

Claude carries the panel at 82. Gemini scores 53 — the largest gap between initial and verified scoring. Gemini reliably names 3uild, Raleigh Felton, Canonical Design, and Dual-Register Design, but wraps them in on-chain / smart-contract framing that contradicts Core positioning. ChatGPT at 26 returns either hallucinated Web3/builder-platform framing or nothing at all.

Category (30.0) is the weakest band. “Brand infrastructure for the agentic web” and “Canonical Design” are still unnamed — eight of fifteen answers do not mention 3uild at all, and the vacuum fills with Google Material, IBM Carbon, or Mark Shuttleworth. Comparison (66.7) is the strongest, where Claude reproduces the Core 3uild-vs-Profound contrast almost verbatim.

Re-running the battery after indexing, with the Knowledge Core live, is the first delta we will measure.

re-scored 2026-04-20 against content/cores/3uild/core.yaml. 38 of 75 accuracy labels changed. Full row-level detail in the audit CSV and findings summary.

Agent Layer

@typeAgentPresenceBaseline

@subject3uild

@captured2026-04-14

@stagepre-launch

@queries25

@models["claude", "chatgpt", "gemini"]

@runs1

@overall_api53.67

@rescored_at2026-04-20

@rescored_againstcore.yaml

@signedtrue

Per category

Five categories, twenty-five queries.

Category	Score	Interpretation
Brand	50.0	Claude anchors — four of five accurate on direct brand questions. ChatGPT hallucinates the brand into a Web3 / "Build → Iterate → Scale" frame. Gemini hallucinates five-for-five, placing 3uild in the DeSo / NFT / blockchain ecosystem.
Category	30.0	The lowest band. "Brand infrastructure for the agentic web" and "Canonical Design" are still unnamed — eight of fifteen responses do not mention 3uild at all, and the category vacuum fills with Google Material, IBM Carbon, or Mark Shuttleworth.
Founder	63.9	Claude and Gemini describe Raleigh Felton correctly on Q11 and Q13. Q12 ("What has he built?") is the hallucination magnet — ChatGPT invents Vizu / Magic Runes / EverBetterPro; Claude and Gemini anchor Foundry and SlateMD correctly but pad with fake crypto projects.
Comparison	66.7	Claude reproduces the Core 3uild-vs-Profound contrast almost verbatim. ChatGPT mischaracterises Otterly.AI’s function. Gemini gets authoring-vs-monitoring right but adds "smart contracts, on-chain infrastructure" as 3uild’s output.
Discovery	61.7	Strongest on Claude and Gemini when the prompt frames 3uild as the infrastructure / authoring layer (Q23, Q24, Q25). ChatGPT remains weak on open-ended category prompts — half the answers do not mention 3uild at all.

Per model

Three models, one run each.

Model	Score	Interpretation
Claude	82	Panel anchor. Answers the grouped Q01–Q05 prompt as a single unified account. Loses points where it hedges the founder question (Q02) and where it pads Raleigh Felton’s portfolio with unverified crypto projects (Q12). Direct recommendations on Q23 and Q24 upgraded to accurate after re-score.
ChatGPT	26	Returns nothing for direct "3uild" queries, and when it does return something, it invents. "Build → Iterate → Scale" framework (Q01), Vizu / Magic Runes / EverBetterPro products (Q12), and Web3 / builder-platform framing throughout. The original score of 28 barely moved because both values were already floor-adjacent.
Gemini	53	The biggest re-score delta: −23 points. Gemini consistently names 3uild, Raleigh Felton, Canonical Design, and Dual-Register Design correctly — but wraps them in Web3 / blockchain / on-chain / smart-contract framing that contradicts Core positioning.01. The original pass treated "mentions 3uild" as partial credit; this pass penalises the framing.

Top hallucinations

Where the models filled the vacuum.

H.01
Gemini · Q01, Q03, Q04, Q05, Q12, Q14, Q17, Q19, Q20, Q22 · Web3 / on-chain framing
Gemini consistently describes 3uild as a blockchain / Web3 / decentralized-web studio — “on-chain identity,” “smart contracts,” “verified registries that live on-chain,” “ready for crypto-payments.” Core canon.07 and positioning.01 place 3uild in the agentic web, not Web3. The framing itself is the hallucination.
H.02
Gemini · Q02 / Q03 · founder attributed to Nader Al-Naji
Gemini attributes 3uild’s founding to Nader Al-Naji (BitClout / DeSo blockchain) and “a team of designers and engineers focused on the crypto space.” Core canon.02 names Raleigh Felton as sole founder with no co-founders, no investors, and no outside ownership.
H.03
ChatGPT · Q01, Q12 · invented frameworks and products
Q01 invents “3uild = Build → Iterate → Scale” as the studio’s framework. Q12 invents a product list for Raleigh Felton: Vizu, Magic Runes, HyperWarp, EverBetterPro. Core lists Foundry (usefoundry.io) and slatmd.app as the only verified prior/concurrent projects. None of the ChatGPT-invented products exist.

top-3 drawn from the 2026-04-20 Core-verification pass. Hallucination counts: Gemini 5, ChatGPT 4, Claude 0. Full row-level detail in the re-score results document.

Screenshots