Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
35 changes: 35 additions & 0 deletions README.ko.md
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@

<p align="center">
<a href="#빠른-시작">시작하기</a> ·
<a href="#실제-화면">실제 화면</a> ·
<a href="#무엇을-할-수-있나요">주요 기능</a> ·
<a href="docs/VERIFICATION.md">검증 기록</a> ·
<a href="CONTRIBUTING.md">기여하기</a><br>
Expand All @@ -37,8 +38,42 @@
> **개인·로컬 사용을 위한 실험적 소스 공개본입니다.** UI는 한국어 중심입니다.
> 집중 기능 검사는 통과했지만 최신 전체 회귀 검사는 통과하지 못했습니다. 아래 검증 범위를 확인하세요.
> **2026-09-29 반영:** 모델별 전용 인증과 재시작·저장 복구 수정을 포함합니다. 실제 모델 분석과 현재 소스의 전체 검사 통과는 미확인이고, Claude 로그인은 공식 Claude Code 실행 파일을 거치며 실계정으로는 아직 확인하지 않았습니다.
> **2026-10-11:** 합성 레이더 데모로 실제 모델 조사를 처음부터 끝까지 한 번 실행했습니다([실제 화면](#실제-화면)). 답의 품질은 평가하지 않았습니다.
> 결과를 사용하기 전에 [현재 상태와 한계](#현재-상태)를 확인하세요.

## 실제 화면

함께 들어 있는 [합성 레이더 데모](examples/synthetic-radar-demo-v1)로 실제 모델을 한 번 돌린 화면입니다. 자료는 지어낸 값이며
측정값이 아닙니다. 조사 한 번에 약 9분 40초가 걸렸습니다. 2026-10-11 개발 빌드 화면이라 일부 문구가 `main`과 조금 다를 수 있습니다.

<table>
<tr>
<td width="50%"><img src="docs/screenshots/01-trace-verdict.png" alt="비 오는 날 결과를 들여온 뒤 Detection rate in rain 판정이 보류에서 기준 미달로 바뀐 추적표"></td>
<td width="50%"><img src="docs/screenshots/02-question.png" alt="미달 줄에서 원인 조사를 눌러 질문이 자동으로 채워진 입력창"></td>
</tr>
<tr>
<td><b>1 · 결과가 기준에 못 미칩니다.</b> 시험 결과를 추적표로 들여옵니다. 판정은 모델이 아니라 규칙으로 계산합니다. 0.80은 "0.90 이상"에 못 미쳐 <i>기준 미달</i>이 됩니다.</td>
<td><b>2 · 그 줄에서 원인을 묻습니다.</b> 미달 줄의 "원인 조사"를 누르면 측정값과 기준이 담긴 질문이 만들어집니다.</td>
</tr>
<tr>
<td><img src="docs/screenshots/03-progress.png" alt="연결한 파일의 어느 줄을 확인하고 있는지 보여 주는 진행 화면"></td>
<td><img src="docs/screenshots/04-answer.png" alt="확인된 사실과 보류된 판단을 나누고 원인 후보를 가를 시험을 제시하는 답"></td>
</tr>
<tr>
<td><b>3 · 자료를 읽는 과정이 보입니다.</b> 지금 어느 연결 파일의 몇 번째 줄을 확인하는지 보여 줍니다.</td>
<td><b>4 · 추측 대신 보류를 밝힙니다.</b> 확인된 사실과 아직 열린 판단을 나누고, 원인 후보마다 그것을 가를 시험을 붙입니다.</td>
</tr>
<tr>
<td><img src="docs/screenshots/05-hypotheses.png" alt="시험마다 각 가설이 맞다면 나올 결과를 AI 제안으로 표시한 가설 비교"></td>
<td><img src="docs/screenshots/06-evidence.png" alt="원문 근거에서 0.80을 검색해 연결 결과 파일 15번째 줄을 찾은 화면"></td>
</tr>
<tr>
<td><b>5 · 가설을 나란히 비교합니다.</b> 시험마다 각 가설이 맞다면 어떤 결과가 나올지 봅니다. 사람이 확인하기 전까지는 미확정 AI 제안으로 표시됩니다.</td>
<td><b>6 · 원문 줄을 확인합니다.</b> 원문 근거에서 "0.80"을 검색하면 연결한 파일의 정확한 줄로 갑니다.</td>
</tr>
</table>


## 무엇을 할 수 있나요

| 근거를 따라갑니다 | 판단을 검토합니다 | 변화를 남깁니다 |
Expand Down
36 changes: 36 additions & 0 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@

<p align="center">
<a href="#quick-start">Get started</a> ·
<a href="#see-it-work">Screens</a> ·
<a href="#what-you-can-do">Features</a> ·
<a href="docs/VERIFICATION.md">Verification</a> ·
<a href="CONTRIBUTING.md">Contribute</a><br>
Expand All @@ -38,8 +39,43 @@ see what remains unresolved, and compare two saved results before deciding what
> **Experimental source preview for personal/local use.** The interface is Korean-first.
> Focused feature checks passed; the latest repository-wide verification is not passing. See the verification limits below.
> **2026-09-29 update:** isolated model authentication and restart recovery are included. Real-provider analysis and a current-source FULL pass remain unverified. Claude sign-in goes through the official Claude Code executable and has not been checked with a real account.
> **2026-10-11:** one real model investigation on the synthetic radar demo ran end to end ([screens](#see-it-work)). Answer quality has not been evaluated.
> See [status and limits](#status) before relying on a result.

## See it work

A real run on the bundled [synthetic radar demo](examples/synthetic-radar-demo-v1): invented data, not measurements.
One model investigation took about 9 minutes 40 seconds. Screens are from a 2026-10-11 development build, so a few
labels may differ slightly from `main`.

<table>
<tr>
<td width="50%"><img src="docs/screenshots/01-trace-verdict.png" alt="Trace table after importing rain results: Detection rate in rain changes from held to below the criterion."></td>
<td width="50%"><img src="docs/screenshots/02-question.png" alt="The investigation question is filled in from the failing trace row."></td>
</tr>
<tr>
<td><b>1 · A result misses its criterion.</b> Import results into the trace table. Verdicts are computed by rules, not by the model: 0.80 against "0.90 or higher" becomes <i>below criterion</i>.</td>
<td><b>2 · Ask why, from that row.</b> "원인 조사" on the failing row writes the question with the measured value and the criterion.</td>
</tr>
<tr>
<td><img src="docs/screenshots/03-progress.png" alt="Live progress: the model reports which connected file and line it is checking."></td>
<td><img src="docs/screenshots/04-answer.png" alt="Answer separating confirmed facts from a held judgment and listing discriminating tests."></td>
</tr>
<tr>
<td><b>3 · Watch it read your material.</b> Progress names the connected file and line being checked.</td>
<td><b>4 · Get a held answer, not a guess.</b> Confirmed facts stay apart from the judgment that remains open; each candidate cause comes with a test that could tell it apart.</td>
</tr>
<tr>
<td><img src="docs/screenshots/05-hypotheses.png" alt="Hypothesis comparison showing what each hypothesis predicts for a test, marked as an AI suggestion."></td>
<td><img src="docs/screenshots/06-evidence.png" alt="Source evidence panel: searching 0.80 finds line 15 of the connected result file."></td>
</tr>
<tr>
<td><b>5 · Compare hypotheses.</b> For each test, see what every hypothesis would predict. These are marked as unconfirmed AI suggestions.</td>
<td><b>6 · Check the source line.</b> Search the evidence panel for "0.80" and land on the exact line of the connected file.</td>
</tr>
</table>


## What you can do

| Follow the evidence | Inspect the judgment | Keep the changes |
Expand Down
40 changes: 35 additions & 5 deletions SOURCE_MANIFEST.json
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
{
"schema_version": "thoth-source-preview-manifest-v1",
"created_at": "2026-10-10T08:13:11.279804+00:00",
"created_at": "2026-10-11T11:42:23.815981+00:00",
"license": "MIT",
"git_history_included": false,
"full_suite_verified": false,
Expand Down Expand Up @@ -53,13 +53,13 @@
},
{
"path": "README.ko.md",
"sha256": "e92c248d8605ecd6f502710e89293146ad4e63db88f20105fa91115e615b38e7",
"bytes": 11855
"sha256": "e81a1273aceec1a42828081c012a91b5632da162ffd49961fd2a8603f6305cc6",
"bytes": 14654
},
{
"path": "README.md",
"sha256": "706bb9d807e7b339b0f5d18ee52a8ec5133388fccc0950f1140230665d811535",
"bytes": 10429
"sha256": "57698aa7ba73bc7a4a1a381ce2be8cb089513e8a666850967c849a79085e09f5",
"bytes": 12895
},
{
"path": "THIRD_PARTY_NOTICES.md",
Expand Down Expand Up @@ -1736,6 +1736,36 @@
"sha256": "f997f3b91b0e785608fb7d801f1bbc3b397313e7677ae06ff1b3f018df1fe892",
"bytes": 2446
},
{
"path": "docs/screenshots/01-trace-verdict.png",
"sha256": "800800b0e2bcfdd290d2f911618aab8a8e60e8e803d2400df9829cd61a351989",
"bytes": 169161
},
{
"path": "docs/screenshots/02-question.png",
"sha256": "a41bf8421b96d72c7a97c8a663d248f118dde6bad78aee75945e9e7797d74b22",
"bytes": 80357
},
{
"path": "docs/screenshots/03-progress.png",
"sha256": "7214715cdcc606c1c0f12368f7c6395c188204314446962be551ec875ed2235f",
"bytes": 164664
},
{
"path": "docs/screenshots/04-answer.png",
"sha256": "04f3cb6002ccc7ed1b58702cbfa0ebc43081021f421a9c85be6a97b05554d8ad",
"bytes": 242801
},
{
"path": "docs/screenshots/05-hypotheses.png",
"sha256": "a78447e9e5f8a63c8dd2bfeca76e943588cfb9e4f613d603c22c543bf987a14c",
"bytes": 406289
},
{
"path": "docs/screenshots/06-evidence.png",
"sha256": "2cff40afec991c105b4813fd4b1bf1d9092f836b8cd55b1239511ef480a488c5",
"bytes": 287062
},
{
"path": "examples/projectpacks/6g-sandbox-hero/criteria.json",
"sha256": "747651e1fd5c2cd1d0481c890e13465371e8fa740ca5f0d551a94fd20510a819",
Expand Down
Binary file added docs/screenshots/01-trace-verdict.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added docs/screenshots/02-question.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added docs/screenshots/03-progress.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added docs/screenshots/04-answer.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added docs/screenshots/05-hypotheses.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added docs/screenshots/06-evidence.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Loading