Evidence audit

A reviewer checklist for the proof chain.

This audit turns each headline claim into a verifier task: inspect the artifact, check the stated evidence, and understand the trust boundary where private production context cannot be public.

Last reviewed: 2026-05-08
Audit method: Every headline proof item should resolve to a public or privacy-safe artifact, name what can be inspected, and state the trust boundary when private production work is involved.

01site page

Business security and release control at e-commerce scale

Evidence: 0 P0/P1 incidents, API and Web UI automation, release control loops

Source: Experience page

Verifier checks

Confirms the production domains without exposing employer-confidential implementation details.
Separates the measurable incident claim from the process claims around automation and release control.

Credibility note: This proof is intentionally privacy-safe: the public artifact can verify scope and operating model, not internal ticket data.

02site page

Digital Twin as a file-first engineering workflow

Evidence: On-site explainer, wiki-centered memory, autonomous upgrade cadence

Source: Digital Twin page

Verifier checks

Explains how durable files, Codex skills, and site updates connect into one workflow.
Shows the operating loop as a productized system rather than a loose prompt collection.

Credibility note: The artifact is strongest as a systems explanation; private vault content remains outside the public proof surface.

03technical writing

Claude Code Sourcemap deep dive

Evidence: 1,884 files analyzed, 8 tutorial chapters, bilingual learning asset

Source: Technical essay

Verifier checks

Publishes the research narrative and architecture explanation in a long-form technical write-up.
Connects the source-analysis count to a reusable learning artifact instead of a one-off summary.

Credibility note: The public write-up is the verifier entry point; the linked project provides the deeper artifact trail.

04repository

Agent Scorecard for token ROI and autonomy decisions

Evidence: Trace-first evals, ROI verdicts, deterministic Markdown reports

Source: Agent Scorecard repository

Verifier checks

Repository README states the concrete decision question: whether an agent deserves more tokens, permissions, and attention.
Docs and examples should show trace inputs, rule-based checks, score bands, and generated reports rather than a generic benchmark claim.

Credibility note: This is public code-and-standard proof for the autonomy evaluation layer; real production traces remain private until deliberately sanitized.

05repository

CLIProxyAPI for coding-tool provider routing

Evidence: Go + Gin, OAuth support, multi-provider routing

Source: CLIProxyAPI repository

Verifier checks

Repository can be inspected for implementation language, routing structure, and provider-facing API surface.
README and code should support the OAuth and multi-provider claims made in the portfolio copy.

Credibility note: This is direct code proof: the repository is the primary evidence artifact.

06paper

First-author IEEE paper

Evidence: First author, 87% parameter reduction, ICCECT 2023

Source: IEEE paper record

Verifier checks

Paper record verifies publication venue, authorship order, and bibliographic details.
Abstract and paper details support the segmentation research claim.

Credibility note: This proof depends on the publisher record for authorship and publication metadata.

07site page

Personal website as a proof-chain control layer

Evidence: Next.js 16, MDX + RSS + sitemap, Obsidian sync scripts

Source: Personal website

Verifier checks

Public pages verify the narrative layer, routing, RSS, sitemap, and proof-chain surfaces without exposing private repo history.
The site is the canonical public proof surface; implementation details can be promoted into public write-ups selectively.

Credibility note: This artifact is self-referential by design: the public site verifies the public proof layer while the source repo can remain private.

Back to proof chain