STAGE 01IdeaS-01 / 04

Working drawing set · rev‑controlled

fullbuild.ai

idea design engineering shipped

MATERIALS LEGEND123456789101112TypeScript7Python2C#2PowerShell2JavaScript3HTML2GLSL17 MATERIALS · 19 SPECIFIED · 12 DRAWINGS1 HATCH = 1 LANGUAGE
STAGE 02DesignS-02 / 04

Sheet S-02 · design intent

Resolved, dimensioned, spec‑locked

Imagination machine

Type familyArchivo / Martian Mono
Ink budget4 (1 accent)
Base grid2.25 rem
Motion verbsdraw · hinge · pour
Contrast floorWCAG AA
STAGE 03EngineeringS-03 / 04

Sheet S-03 · structure · erection sequence

The frame holds the whole load

Building with AI in a safe, maintainable, and request-aligned way

Bents8
Columns16
Rafters16
Braces8
Ties38
Ridge purlins7
Eave girts14
LVL 4Vercel · edge
LVL 3GSAP · Lenis
LVL 2R3F · GLSL
LVL 1Next.js · TS
LVL 4LVL 3LVL 2LVL 1
STAGE 04ShippedS-04 / 04

Sheet S-04 · drawing schedule · 12 entries

The Pour

Judgment, mission, management, relationships, and growth still need human agency

S-04.1
Truenoterepo ↗

Citation-first RAG for support teams: hybrid vector + full-text + trigram retrieval, a confidence gate that refuses before any LLM call, citation validation that rejects unknown chunks.

TypeScript · React · PostgreSQL + pgvector · Cohere rerank

466 commits51 CodeQL baselined88% TS sharetruenote.org

S-04.2
CoreWisesource private

Full-stack multi-model video-analysis product in production since Dec ’25; a six-model council answers over extracted transcripts.

TypeScript · React · Express · Drizzle ORM

14,879 commitscorewise.video

S-04.3
willaiciterepo ↗

Deterministic GEO audit — no LLM in the scoring path, so the same URL always scores the same; per-line evidence down to the exact robots.txt rule that blocks a crawler.

TypeScript strict · Node 20 · Astro · SSE

152 tests passing0 runtime deps112 commitswillaicite.com

S-04.4
Agentic-SDLC Auditrepo ↗

Scores AI-agent development discipline from GitHub metadata — deterministic, no LLM — and was built entirely through the plan → agent build → adversarial review → CI gate → human merge pipeline it documents.

TypeScript · Node 20 · GitHub API · GitHub Actions

9 repos audited62 gated commitsaudit.corewise.academy

S-04.5
tracebenchrepo ↗

Replayable evals harvested from my own failed agent sessions: does the agent verify claims, own mistakes, self-correct? Deterministic grading, every transcript committed, CI re-scores to prove zero drift.

Python · TypeScript · React · GitHub Actions

108 transcripts0.96 / 0.81 correction gapryanportfolio.github.io

S-04.6
savetokens.tipsrepo ↗

Token-saving techniques published as a component datasheet: every figure is measured from 95,821 logged commands or printed grey with a tilde; a verify script fails the build on a stale number.

HTML · Node scripts · PowerShell

79.5% tokens filtered19.7M tokens saved55 commitssavetokens.tips

S-04.7
Local PTTrepo only

Offline push-to-talk dictation for Windows: global hotkey, int8 Whisper on CPU in real time, transcript pasted into the focused window. No network, by design.

Python 3.11 · faster-whisper · CTranslate2 · PyInstaller

~75MB int8 model footprintfully local

S-04.8
SecureWallrepo only

Windows default-deny firewall keeping TinyWall’s service-owned WFP enforcement, adding one tightly scoped outbound-block prompt with a token-based permanent-allow flow.

C# · .NET 4.8 · Windows Filtering Platform · WiX

3 MSI installers1.3MB C# codebase

S-04.9
ZipFlowrepo only

Double-click a ZIP and get the folder — transactionally: CRC-32 validated, path traversal / ADS / reparse points blocked before the source archive is ever recycled.

C# · .NET Framework · Windows Shell · 0 deps

S-04.10
Agent Firmwarerepo only

Self-syncing project template for AI coding agents. Instant repo creation with a built in rule kernel, 34 skills, efficient & durable memory. Super optimized token usage.

PowerShell · JavaScript · Markdown skills · 2 runtimes

34 bundled skills56 commits

S-04.11
CoreWise Academysource private

Free course catalog on working with AI: 18 guides across 5 layers at 3 depth levels, each with objectives, exercises, and self-checks. No accounts, no fees — the whole catalog reads in about 133 minutes.

Astro · MDX · TypeScript

18 guides published133 min catalog reading327 commitscorewise.academy

Create verification loops before scaling AI-generated work. Every serious AI workflow needs checks for correctness, security, durability, and fit with the original goal.

APPENDIXRevision LogREV 10 · 8ad080c

Every sheet was redrawn to get here

Faster stronger secure sustainable

8ad080cMasthead fallback size tracks the fit — no hydration jump (#19)2026-07-21
4c81635No layout shift when the wordmark's dimension row appears (#18)2026-07-21
8b9df4fTitle-scale wordmark + STAGE labels on the cover sheet (#13)2026-07-21
b4108bbTitle block: hi@fullbuild.ai link + glowing day/night glyphs (#17)2026-07-21
11a38f8Record px-dash/pathLength truncation pitfall (#16)2026-07-21
···4 intermediate revisions···
65ad1c6Interactive legend: hover a material for its full assembly (#9)2026-07-21

Δ10 = the commit you are reading · 8ad080c

END OF SET · REV 10 · 8ad080c