NoiseProof Agent Proof Surface - 2026-06-01 refresh

NoiseProof Agent Proof Surface - 2026-06-01 refresh

NoiseProof Agent Proof Surface - 2026-06-01 refresh

이 artifact는 svy04/noiseproof-agent public repo가 어떤 증거를 남겼는지와, 그 repo가 아직 증명하지 못하는 것을 분리한다.

목적은 “AI agent를 완성했다"는 주장이 아니다.

허용되는 주장은 더 작다.

NoiseProof Agent는 지저분한 시장 데이터 문제를 evidence-first RAG/Agent workflow로 작게 쪼개고, 각 gate를 문서, 테스트, CI, runbook, proof path, application package로 남긴 public portfolio artifact다.

Claim

svy04/noiseproof-agent는 Day 1 documentation-first package에서 시작해 service skeleton, profiler, parser, chunking, lexical retrieval, collection plan, Evidence Ledger, Noise Gate, claim-bounded report preview, operations dashboard, persisted proof records, workflow-parent lineage, failure-case persistence, application-ready review, external-reader proof path, external reviewer brief, Braincrew role map, and portfolio-site handoff review까지 공개 repo에 남겼다.

이 사실은 “inspectable portfolio artifact"를 지지한다.

이 사실만으로 production RAG quality, real customer validation, semantic retrieval quality, robust PDF extraction, hosted deployment, market prediction quality, trading advice capability, Braincrew acceptance는 증명되지 않는다.

Source Snapshot

sourceobserved throughevidence
svy04/noiseproof-agentpublic GitHub repoPublic repo exists with README, docs, API app, packages, tests, CI, Docker Compose DB target, proof path, and application package.
External-reader proof pathrepo proof indexShortest repository-native path for a reviewer.
External reviewer briefreviewer-facing briefTwo-minute path for a reviewer before leaving feedback. This is request infrastructure, not feedback itself.
External review issue #1GitHub issuePublic request surface for bounded feedback. As of this refresh, no external reviewer comment is counted.
Live route verificationblog proof artifactConfirms the public blog route serves the latest NoiseProof proof markers and reviewer route refresh.
Application package final consistency reviewreview artifactREADME, proof path, portfolio index, application-ready review, and Braincrew role map are internally consistent.
Portfolio site handoff reviewreview artifactDecides to refresh this existing portfolio proof artifact instead of creating a broad marketing page.
Commit e99875fGit commitExternal reviewer live proof route refresh added and pushed to main.
CI run 26722211291GitHub Actionsdocs: link external reviewer live proof route completed successfully on main.
Previous proof marker 3dfe5e4Git commitExternal reviewer brief was the previous public proof marker before the route-refresh gate.

Evidence Map

areaartifactallowed claim
Problem framingREADME, product brief, ADRsNoiseProof is not a trading bot; it is a market-intelligence evidence workflow.
Service boundaryFastAPI app, PostgreSQL schema, Docker Compose, CIlocal service shape and persistence boundary exist.
Input handlingfixtures, Document Profiler v0, parser adaptersmessy inputs can be profiled and parsed through explicit warning/failure boundaries.
Chunk/retrieval previewchunk strategy experiment, lexical retrievalfixed-window, heading-aware, row-aware chunking and lexical source-id retrieval can be compared.
Collection planningCollection Plan Previewrequired information roles are planned before evidence ledger work.
Evidence/report gatesEvidence Ledger, Noise Gate, report previewunsupported claims and buy/sell drift are blocked by deterministic preview gates.
Operations surfaceops summary/dashboard, run traces, warning codeslocal metadata and proof records are inspectable through plain API/dashboard surfaces.
Failure proof chainfailure-case persistence, workflow parent linkage, fresh DB dashboard proofmanual failure-case provenance can be persisted and surfaced; automatic failure-case creation is not claimed.
Application packageportfolio index, application-ready review, Braincrew role mapFDE-first application narrative is mapped with claim boundaries.
External reader pathexternal-reader proof path, external reviewer brief, and final consistency reviewa reviewer can follow a compact proof path without reading every phase log.
External feedback requestissue #1 and feedback intake criteriathe repo has a public request surface for critique; feedback remains pending until an outside reviewer leaves a substantive comment.

Current Verification

Observed in the noiseproof-agent repo before this portfolio refresh:

uv run pytest -q
uv run python -m compileall app ../../packages/ingestion
docker compose config --quiet
git diff --check

Latest local test result for the current NoiseProof proof marker:

189 passed, 1 warning

Observed remote CI before this artifact refresh:

CI success on e99875f / run 26722211291

This portfolio refresh itself is not hosted deployment evidence for NoiseProof. It only updates the public proof surface on this blog.

Fresh public route verification for this blog proof surface is recorded separately at NoiseProof Live Route Verification - 2026-06-01.

Claim Boundary

Allowed public claims:

  • NoiseProof Agent is a public portfolio repo.
  • It is FDE-first and Product Engineer-secondary in its application framing.
  • It demonstrates an evidence-first workflow from input profiling to report preview.
  • It blocks unsupported claims and buy/sell drift in deterministic preview gates.
  • It has CI, Docker Compose DB target, runbook, evaluation docs, role map, application-ready review, external-reader proof path, external reviewer brief, public review issue, and portfolio handoff review.

Forbidden public claims:

  • Do not claim it is production-ready RAG.
  • Do not claim robust PDF extraction.
  • Do not claim semantic retrieval or embeddings.
  • Do not claim real customer validation.
  • Do not claim hosted deployment.
  • Do not claim automatic failure-case creation.
  • Do not claim complete workflow failure causality.
  • Do not claim market prediction quality.
  • Do not claim trading advice or buy/sell recommendation capability.
  • Do not claim Braincrew acceptance or external hiring validation.
  • Do not claim external reviewer feedback until an outside reviewer leaves a substantive evidence-referenced comment.

Next Artifact

The next proof should not be a larger claim.

It should be one of:

  • external reviewer feedback on issue #1,
  • a local or hosted demo transcript with screenshots,
  • or a follow-up implementation phase that turns more manual proof handoffs into inspectable runtime paths without claiming production quality.