Releases: lowtidebuild/public-equity-research
Release list
v2.1.0 — Earnings season analysis suite (May 7, 2026)
v2.1.0 — Earnings season analysis suite
실적 프리뷰, 실적 리뷰, 그리고 지난 분석과의 차이를 한 눈에 — 어닝 발표일 전후 7일을 위한 분석 도구 셋.
TL;DR
English:
- Run a dedicated earnings-window analysis with Mode E. Within seven days of an earnings print, the agent decides whether you need a Preview (consensus, options-implied move, watchpoints) or a Review (actual vs. consensus, segment breakdown, post-print action plan).
- Re-run any analysis and see exactly what changed. Every report opens with a delta banner: R/R score change, target price change, verdict change, and the diff on the risk list — measured against your most recent snapshot.
- Use a Mode C report as a real reconciled valuation document. Peer tables populated from real fetched numbers, a 12-month catalyst timeline, and a Valuation Bridge widget reconciling DCF / peer multiples / analyst consensus / base scenario into one weighted fair value.
한국어:
- Mode E로 어닝 발표일 전후 7일 전용 분석을 받습니다. 발표일까지 7일 이내라면 에이전트가 Preview(컨센서스, 옵션시장 implied move, watchpoint) 또는 Review(실제 vs 컨센서스, 사업부별 분해, 사후 액션 플랜) 중 적합한 보고서를 자동 작성합니다.
- 같은 종목을 다시 분석하면 무엇이 달라졌는지가 보고서 최상단에 자동으로 뜹니다. R/R 점수, 목표가, Verdict, 리스크 리스트 diff — 마지막 스냅샷과 비교한 결과를 별도 명령 없이 받습니다.
- Mode C 보고서를 valuation reconcile 문서로 사용할 수 있습니다. Peer 표가 실제 fetch된 숫자로 채워지고, 12개월 카탈리스트 타임라인이 이벤트 cluster를 시각화하며, Valuation Bridge 위젯이 DCF / peer 멀티플 / 애널리스트 컨센서스 / base 시나리오 4개 anchor를 가중평균 적정가로 reconcile합니다.
What's in v2.1.0 / 포함된 기능
| Capability | What it does |
|---|---|
| Mode E — Earnings Preview/Review | Auto-detected earnings window (D-7 ~ D+3) analysis. Preview covers consensus/options/watchpoints; Review covers actual-vs-consensus + post-print action plan. |
| Auto Delta Banner | Re-running any analysis (Mode A/B/C/D/E) automatically prepends a banner showing R/R, target, verdict, and risk-list deltas vs. last snapshot. Toggle with --no-delta. |
| Mode C — Valuation Bridge | Single-section reconciliation of 4 valuation anchors (DCF / peer multiples / analyst consensus / base scenario) into one weighted fair value. |
| Mode C — Peer mini-pipeline | Peer comparison table populated from real yfinance fetched data (5-7 metrics, 24h cache) instead of placeholder estimates. |
| Mode C — Catalyst Timeline | 12-month horizontal-bar timeline with 5 categories (earnings/regulatory/product/macro/other) showing event clusters. |
| Mode B — Macro Context (light) | Peer comparison includes 3-5 macro series + per-peer macro sensitivity narrative (e.g., Beta-driven rate exposure). |
| Mode E — Accessibility Layer | Every Mode E report includes TL;DR (3 bullets), per-segment breakdown table, "왜 이 숫자가 중요한가" callouts per section, and a glossary footer (≥5 terms). |
Tested workflow / 검증된 사용 시나리오
AMD Q1 2026 Review (real D+2 test, 2026-05-07):
- Earnings released 2026-05-05 after close
- Stock reaction D+1: +18.61% to $421.39 (15× larger than 4Q historical average +1.24%)
- Print: EPS $1.37 vs $1.28 (+7.0% beat), Revenue $10.25B vs $9.89B (+3.6% beat)
- Segments: Data Center $5.8B (+57% YoY) / Client $2.88B (+26%) / Gaming $720M (+11%) / Embedded $873M (+6%)
- Q2 guide: $11.2B (+6.7% above consensus) + GM 56% + Server CPU TAM doubled from $60B to $120B
- Output: 33KB Korean HTML report with TL;DR + segment table + glossary + post-print action plan
Quick stats
- 37 commits since v1.1.0
- 81 files changed (+16,935 / −2,159 lines)
- Tests: 135 → 262 (+127)
- 4 sessions, 6 phases (A–F)
Read the full notes / 전체 릴리즈 노트
Disclaimer: This is not investment advice. For informational purposes only. / 본 문서는 투자 자문이 아닙니다. 정보 제공 목적으로만 작성되었습니다.
v1.1.0 - More reliable, safer stock analysis
v1.1.0 Release Notes
English
v1.1.0 makes the stock analysis agent more reliable, safer, and easier to trust in everyday use.
What Improved
-
More dependable answers
The agent is better at checking whether numbers, dates, sources, and conclusions actually fit together before presenting an investment view. This reduces polished-looking reports that contain hidden inconsistencies. -
Less guessing when data is weak
When a number cannot be verified well enough, the agent now prefers leaving it blank or flagging the uncertainty instead of filling the gap with a confident-looking estimate. -
Clearer source confidence
Reports now make it easier to understand whether the analysis is based on strong primary data, structured market data, fallback sources, or limited web evidence. -
Safer use of web and fetched content
Content collected from the web is treated more carefully, so untrusted text is less likely to steer the agent’s reasoning or sneak instructions into the workflow. -
Better handling of Korean and U.S. data paths
The workflow is more consistent across U.S. and Korean stocks, including improved handling of fallback data sources and Korean disclosure data. -
More reliable saved analysis history
Saved analysis snapshots are easier to reuse later without mixing old run files with current results. -
Cleaner review flow before delivery
The agent now separates mechanical checks from judgment-based critique, so reports can catch basic issues faster while keeping human-style investment reasoning focused where it matters. -
Faster and leaner analysis context
The agent passes a more compact evidence package into the main analysis step, helping reduce unnecessary context while keeping the facts needed for the final view. -
Improved quick briefings
Lightweight Mode A outputs still keep essential safeguards such as source confidence, date visibility, disclaimer checks, and blank-over-wrong behavior.
Why It Matters
This release is focused on trust. The agent should now be less likely to overstate weak data, less likely to produce internally inconsistent reports, and better at explaining uncertainty in a way users can act on.
한국어
v1.1.0은 주식 분석 에이전트를 더 믿고 쓰기 쉽도록 안정성, 안전성, 결과 품질을 개선한 릴리즈입니다.
개선된 점
-
답변 신뢰도 향상
투자 의견을 내기 전에 숫자, 날짜, 출처, 결론이 서로 맞는지 더 꼼꼼히 확인합니다. 그럴듯해 보이지만 내부적으로 앞뒤가 맞지 않는 보고서를 줄이는 데 초점을 맞췄습니다. -
불확실한 데이터에 대한 추측 감소
충분히 검증되지 않은 숫자는 억지로 채우지 않고, 비워 두거나 불확실성을 표시하도록 개선했습니다. -
출처 신뢰도 이해가 쉬워짐
분석이 공시 같은 강한 1차 자료에 기반한 것인지, 구조화된 시장 데이터인지, 보조 출처나 제한적인 웹 근거인지 더 명확히 구분할 수 있습니다. -
웹에서 가져온 내용 처리 안전성 강화
외부에서 수집한 텍스트가 에이전트의 판단이나 작업 지시를 오염시키지 않도록 더 조심스럽게 다룹니다. -
미국/한국 종목 워크플로 일관성 개선
미국 종목과 한국 종목 모두에서 fallback 데이터, 한국 공시 데이터, 실행 경로 처리가 더 안정적으로 맞춰졌습니다. -
저장된 분석 기록 재사용 안정성 향상
과거 분석 스냅샷을 다시 불러올 때 이전 실행 파일과 현재 결과가 섞일 가능성을 줄였습니다. -
보고서 전달 전 검토 흐름 개선
단순한 형식/계산 검사는 자동으로 더 빠르게 잡고, 투자 판단이 필요한 비평은 별도로 다루도록 정리했습니다. -
분석 컨텍스트 경량화
최종 분석 단계에 필요한 핵심 근거만 더 작고 정리된 형태로 전달해, 불필요한 정보로 판단이 흐려지는 문제를 줄였습니다. -
빠른 브리핑 품질 보강
짧은 Mode A 브리핑에서도 출처 신뢰도, 기준일, 고지문, 검증 불가 데이터 비움 처리 같은 기본 안전장치를 유지합니다.
왜 중요한가
이번 릴리즈의 핵심은 신뢰입니다. 에이전트가 약한 데이터를 과장하지 않고, 내부적으로 모순된 보고서를 줄이며, 사용자가 실제로 판단할 수 있는 방식으로 불확실성을 드러내도록 개선했습니다.
Update Notes (v1.0.0.)
Release Notes — 2026-04-18
This is a bilingual companion note for the larger update published on main at commit a4c7970.
Action Required
English
- Please pull the latest changes from
main. - The remote
masterbranch has been removed. If you still have a localmaster, switch tomain.
git checkout main
git pull origin main한국어
- 최신 변경 사항을 받으려면
main브랜치에서 pull 해주세요. - 원격
master브랜치는 제거되었습니다. 아직 로컬에서master를 쓰고 있다면main으로 전환해 주세요.
git checkout main
git pull origin mainEnglish Summary
- Trust-boundary hardening is now in place for fetched artifacts. We added prompt-injection sanitization,
_sanitizationmetadata, and validator enforcement for unsanitized raw inputs. - Run-local artifact contracts were added for
research-plan,validated-data,analysis-result,quality-report,patch-plan,analysis-patch, andpatch-loop-result. - The analyst/critic patch loop is now part of the workflow, including patch-plan validation, patch application, critic recheck merge, delivery gating, and eval coverage.
- Output and workflow tooling were refreshed: scripted Mode B comparison rendering, updated Mode A/C/D renderer docs, and better run-local artifact handling.
- CI/local verification is easier now because
tools/contract_checks.pycan materialize missing runtime fixtures from in-repo eval samples. - Repository hygiene was improved: sensitive historical material was removed from git history, the remote now uses
mainonly, and the repo is aligned around the clean active working copy.
한국어 요약
- 수집 아티팩트에 대한 신뢰 경계 보강이 적용되었습니다. 프롬프트 인젝션 sanitization,
_sanitization메타데이터, 그리고 미정제(raw) 입력에 대한 validator 강등 처리까지 포함됩니다. research-plan,validated-data,analysis-result,quality-report,patch-plan,analysis-patch,patch-loop-result에 대한 run-local 아티팩트 계약(schema/contract)이 추가되었습니다.- 이제 analyst/critic patch loop가 정식 워크플로에 포함됩니다. patch-plan 검증, patch 적용, critic 재검토 병합, 전달 게이트, eval 커버리지가 함께 들어갔습니다.
- 출력/워크플로 도구도 정비했습니다. 스크립트 기반 Mode B 비교 렌더링, Mode A/C/D 렌더러 문서 업데이트, run-local 아티팩트 처리 개선이 포함됩니다.
tools/contract_checks.py가 저장소 내 eval fixture를 활용해 누락된 런타임 샘플을 보완할 수 있게 되어, CI와 로컬 검증이 더 쉬워졌습니다.- 저장소 위생도 개선했습니다. 민감한 과거 이력은 git history에서 제거했고, 원격은
main만 사용하도록 정리했으며, 활성 작업 레포 기준으로 구조를 맞췄습니다.
Validation
python3 -m unittest discover tests -vpython3 tools/contract_checks.py
References
- Detailed note:
docs/releases/2026-04-17-security-and-contracts.md - Published head:
a4c7970