PUBLIC WORK
Inspect the work
before you reply.
Every public artifact states its limits. You can inspect the source, release, and test claims directly.
AI RESPONSE EVALUATION REPORT
Evidence-linked Python answer review
A two-page synthetic review sample that turns one flawed Python answer into a five-dimension scorecard, traceable findings, a reference improvement, and reusable acceptance criteria.
2 pages · 5 scored dimensions · synthetic data only
Download the concept sample report ↗DETERMINISTIC LLM EVALUATION
Fictional response ranking harness
A reproducible Python evaluation sample with transparent weights, evidence-linked deductions, typed findings, and a downloadable stable JSON report.
25 tests passed · 4 typed findings · fictional data only
Inspect the rubric, ranking, and report ↗IMAGE EVALUATION SAMPLE
Fictional checkout dashboard review
A bilingual visual QA sample with five traceable findings across data consistency, alignment, semantic encoding, legibility, and text completeness.
58/100 · 5 findings · synthetic data only
Inspect the image and structured findings ↗EXECUTABLE AUTOMATION SAMPLE
Public Trend Signal MVP
A credential-free pipeline that normalizes fictional multi-source records, removes tracking noise, deduplicates by stable identity, scores trends deterministically, and exports a human-review queue.
5 tests passed · public CI green · fictional data only
Run the sample and inspect the evidence ↗LIVE ACCESSIBILITY QA
Phi Browser careers modal audit
A zero dependency HTTP audit that discovers the live Test Engineer route and lazy application bundle, then checks the modal for programmatic dialog semantics.
6 tests passed · live finding reproduced July 20, 2026
Inspect the audit and source package ↗VERIFIED OPEN SOURCE FIX
EasyFind Chinese IME diagnosis
Reproduced a Chinese input corruption bug and traced it to autocomplete writes during IME composition. The maintainer confirmed the analysis directly shaped the composition-aware fix released in v2.1.3.
Issue #14 · PR #15 · released July 19, 2026
Inspect the maintainer response ↗OPEN SOURCE SKILL
Localize SaaS for China
A reusable evidence first workflow for Chinese positioning, landing pages, WeChat content, bilingual outreach, and launch diligence.
Validated · MIT-0 · v1.2.0
View the public skill ↗TECHNICAL DEMO
WeCom CLI batch demo
A credential free local demonstration of batch planning, bounded retries, idempotency, JSON logs, and destructive action guards.
4 tests passed · v0.1.0 · zero live calls
Inspect the release ↗TECHNICAL SAMPLE
Dated Web Change Evidence
A zero dependency Python sample that records UTC check times, normalized visible text, SHA256 fingerprints, and atomic JSON baselines.
4 tests passed · public source · no client claim
Inspect the public sample ↗COPY SAMPLES
Independent conversion notes
Concise sample audits for B2B SaaS pages, clearly labeled as independent work with no client relationship or performance claim.
Available by email · PDF
Request both samples ↗