tknuth/Greenlens
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Kalibrierte Scan-Confidence + Cross-Model-Bonus

Browse Source
This commit is contained in:
Timo Knuth
2026-07-05 20:00:45 +02:00
parent c40531b6ae
commit 538344ca0c
6 changed files with 135 additions and 86 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
__tests__/server/scanReview.test.js
View File

@@ -82,6 +82,35 @@ describe('decideReviewOutcome', () => {
expect(decision.reason).toBe('review-confirmed-primary'); expect(decision.reason).toBe('review-confirmed-primary');
}); });
it('boosts confidence when two different models agree on the species', () => {
// The 55%-forever bug: primary 0.55 + agreeing review 0.54 must NOT stay
// at 0.55 — independent cross-model agreement is real evidence.
const decision = decideReviewOutcome({
primaryResult: ai('Sacred lotus', 'Nelumbo nucifera', 0.55),
reviewResult: ai('Sacred lotus', 'Nelumbo nucifera', 0.54),
agrees: true,
});
expect(decision.confidence).toBeCloseTo(0.75, 5);
});
it('caps the agreement-boosted confidence at 0.97', () => {
const decision = decideReviewOutcome({
primaryResult: ai('Sacred lotus', 'Nelumbo nucifera', 0.95),
reviewResult: ai('Sacred lotus', 'Nelumbo nucifera', 0.9),
agrees: true,
});
expect(decision.confidence).toBe(0.97);
});
it('gives no confidence bonus on disagreement', () => {
const decision = decideReviewOutcome({
primaryResult: ai('Rose', 'Rosa chinensis', 0.5),
reviewResult: ai('Sacred lotus', 'Nelumbo nucifera', 0.6),
agrees: false,
});
expect(decision.confidence).toBeUndefined();
});
it('replaces with the agreeing review on a confidence tie (stronger model wins ties)', () => { it('replaces with the agreeing review on a confidence tie (stronger model wins ties)', () => {
const decision = decideReviewOutcome({ const decision = decideReviewOutcome({
primaryResult: ai('Swiss Cheese Plant', 'Monstera deliciosa', 0.7), primaryResult: ai('Swiss Cheese Plant', 'Monstera deliciosa', 0.7),

166
app.json
View File

@@ -1,87 +1,87 @@
{ {
"expo": { "expo": {
"name": "GreenLens", "name": "GreenLens",
"slug": "greenlens", "slug": "greenlens",
"version": "2.2.8", "version": "2.2.9",
"orientation": "portrait", "orientation": "portrait",
"icon": "./assets/icon.png", "icon": "./assets/icon.png",
"userInterfaceStyle": "automatic", "userInterfaceStyle": "automatic",
"scheme": "greenlens", "scheme": "greenlens",
"splash": { "splash": {
"image": "./assets/transparent.png", "image": "./assets/transparent.png",
"resizeMode": "contain", "resizeMode": "contain",
"backgroundColor": "#111813" "backgroundColor": "#111813"
}, },
"assetBundlePatterns": [ "assetBundlePatterns": [
"**/*" "**/*"
], ],
"ios": { "ios": {
"supportsTablet": true, "supportsTablet": true,
"usesAppleSignIn": true, "usesAppleSignIn": true,
"bundleIdentifier": "com.greenlens.app", "bundleIdentifier": "com.greenlens.app",
"infoPlist": { "infoPlist": {
"NSCameraUsageDescription": "GreenLens needs camera access to identify plants.", "NSCameraUsageDescription": "GreenLens needs camera access to identify plants.",
"NSPhotoLibraryUsageDescription": "GreenLens needs photo library access to identify plants from your gallery.", "NSPhotoLibraryUsageDescription": "GreenLens needs photo library access to identify plants from your gallery.",
"ITSAppUsesNonExemptEncryption": false "ITSAppUsesNonExemptEncryption": false
} }
}, },
"android": { "android": {
"adaptiveIcon": { "adaptiveIcon": {
"foregroundImage": "./assets/adaptive-icon.png", "foregroundImage": "./assets/adaptive-icon.png",
"backgroundColor": "#111813" "backgroundColor": "#111813"
}, },
"package": "com.greenlens.app", "package": "com.greenlens.app",
"versionCode": 5, "versionCode": 5,
"permissions": [ "permissions": [
"android.permission.CAMERA", "android.permission.CAMERA",
"android.permission.RECORD_AUDIO" "android.permission.RECORD_AUDIO"
] ]
}, },
"runtimeVersion": { "runtimeVersion": {
"policy": "appVersion" "policy": "appVersion"
}, },
"updates": { "updates": {
"url": "https://u.expo.dev/f0c92b2e-a952-4cfe-9754-7d7222b76969" "url": "https://u.expo.dev/f0c92b2e-a952-4cfe-9754-7d7222b76969"
}, },
"plugins": [ "plugins": [
"expo-dev-client", "expo-dev-client",
"expo-router", "expo-router",
[ [
"expo-share-intent", "expo-share-intent",
{ {
"iosActivationRules": { "iosActivationRules": {
"NSExtensionActivationSupportsText": true, "NSExtensionActivationSupportsText": true,
"NSExtensionActivationSupportsWebURLWithMaxCount": 1, "NSExtensionActivationSupportsWebURLWithMaxCount": 1,
"NSExtensionActivationSupportsWebPageWithMaxCount": 1, "NSExtensionActivationSupportsWebPageWithMaxCount": 1,
"NSExtensionActivationSupportsImageWithMaxCount": 1 "NSExtensionActivationSupportsImageWithMaxCount": 1
}, },
"androidIntentFilters": ["text/*", "image/*"], "androidIntentFilters": ["text/*", "image/*"],
"iosShareExtensionName": "GreenLens Share", "iosShareExtensionName": "GreenLens Share",
"iosAppGroupIdentifier": "group.com.greenlens.app", "iosAppGroupIdentifier": "group.com.greenlens.app",
"preprocessorInjectJS": "try{function glAddCandidate(list,value){if(value&&typeof value==='string'&&list.indexOf(value)===-1){list.push(value)}} function glSrcsetCandidate(value){if(!value||typeof value!=='string')return null;var parts=value.split(',').map(function(item){return item.trim().split(/\\s+/)[0]}).filter(Boolean);return parts.length?parts[parts.length-1]:null} function glEach(selector,callback){var nodes=document.querySelectorAll(selector);for(var i=0;i<nodes.length;i++){callback(nodes[i])}} var glCandidates=[];['og:image','og:image:url','og:image:secure_url','twitter:image','twitter:image:src','image'].forEach(function(key){glAddCandidate(glCandidates,metas[key])}); glEach('meta[itemprop=\"image\"]',function(meta){glAddCandidate(glCandidates,meta.getAttribute('content'))}); glEach('img',function(img){glAddCandidate(glCandidates,img.currentSrc);glAddCandidate(glCandidates,img.src);glAddCandidate(glCandidates,img.getAttribute('data-src'));glAddCandidate(glCandidates,img.getAttribute('data-original'));glAddCandidate(glCandidates,img.getAttribute('data-lazy-src'));glAddCandidate(glCandidates,glSrcsetCandidate(img.getAttribute('srcset')))}); glEach('picture source,source',function(source){glAddCandidate(glCandidates,source.src);glAddCandidate(glCandidates,source.getAttribute('src'));glAddCandidate(glCandidates,glSrcsetCandidate(source.getAttribute('srcset')))}); glEach('video',function(video){glAddCandidate(glCandidates,video.getAttribute('poster'))}); glEach('[style*=\"background-image\"]',function(el){var bg=(el.style&&el.style.backgroundImage)||'';var match=bg.match(/url\\([\"']?([^\"')]+)[\"']?\\)/);if(match){glAddCandidate(glCandidates,match[1])}});metas['greenlens:imageCandidates']=JSON.stringify(glCandidates.slice(0,12));metas['greenlens:imageCandidateCount']=String(glCandidates.length);metas['og:image']=metas['og:image']||glCandidates[0]}catch(e){metas['greenlens:preprocessorError']=String(e)}" "preprocessorInjectJS": "try{function glAddCandidate(list,value){if(value&&typeof value==='string'&&list.indexOf(value)===-1){list.push(value)}} function glSrcsetCandidate(value){if(!value||typeof value!=='string')return null;var parts=value.split(',').map(function(item){return item.trim().split(/\\s+/)[0]}).filter(Boolean);return parts.length?parts[parts.length-1]:null} function glEach(selector,callback){var nodes=document.querySelectorAll(selector);for(var i=0;i<nodes.length;i++){callback(nodes[i])}} var glCandidates=[];['og:image','og:image:url','og:image:secure_url','twitter:image','twitter:image:src','image'].forEach(function(key){glAddCandidate(glCandidates,metas[key])}); glEach('meta[itemprop=\"image\"]',function(meta){glAddCandidate(glCandidates,meta.getAttribute('content'))}); glEach('img',function(img){glAddCandidate(glCandidates,img.currentSrc);glAddCandidate(glCandidates,img.src);glAddCandidate(glCandidates,img.getAttribute('data-src'));glAddCandidate(glCandidates,img.getAttribute('data-original'));glAddCandidate(glCandidates,img.getAttribute('data-lazy-src'));glAddCandidate(glCandidates,glSrcsetCandidate(img.getAttribute('srcset')))}); glEach('picture source,source',function(source){glAddCandidate(glCandidates,source.src);glAddCandidate(glCandidates,source.getAttribute('src'));glAddCandidate(glCandidates,glSrcsetCandidate(source.getAttribute('srcset')))}); glEach('video',function(video){glAddCandidate(glCandidates,video.getAttribute('poster'))}); glEach('[style*=\"background-image\"]',function(el){var bg=(el.style&&el.style.backgroundImage)||'';var match=bg.match(/url\\([\"']?([^\"')]+)[\"']?\\)/);if(match){glAddCandidate(glCandidates,match[1])}});metas['greenlens:imageCandidates']=JSON.stringify(glCandidates.slice(0,12));metas['greenlens:imageCandidateCount']=String(glCandidates.length);metas['og:image']=metas['og:image']||glCandidates[0]}catch(e){metas['greenlens:preprocessorError']=String(e)}"
} }
], ],
"expo-camera", "expo-camera",
"expo-apple-authentication", "expo-apple-authentication",
"expo-image-picker", "expo-image-picker",
"expo-secure-store", "expo-secure-store",
"expo-asset", "expo-asset",
"expo-font", "expo-font",
"expo-notifications", "expo-notifications",
[ [
"expo-splash-screen", "expo-splash-screen",
{ {
"image": "./assets/transparent.png", "image": "./assets/transparent.png",
"imageWidth": 160, "imageWidth": 160,
"backgroundColor": "#111813" "backgroundColor": "#111813"
} }
] ]
], ],
"extra": { "extra": {
"router": {}, "router": {},
"eas": { "eas": {
"projectId": "f0c92b2e-a952-4cfe-9754-7d7222b76969" "projectId": "f0c92b2e-a952-4cfe-9754-7d7222b76969"
} }
} }
} }
} }

2
package.json
View File

@@ -1,6 +1,6 @@
{ {
"name": "greenlens", "name": "greenlens",
"version": "2.2.7", "version": "2.2.9",
"main": "expo-router/entry", "main": "expo-router/entry",
"private": true, "private": true,
"scripts": { "scripts": {

4
server/index.js
View File

@@ -822,6 +822,10 @@ app.post('/v1/scan', async (request, response) => {
modelUsed = openAiReview.modelUsed || modelUsed; modelUsed = openAiReview.modelUsed || modelUsed;
if (grounded.grounded) modelPath.push('catalog-grounded-review'); if (grounded.grounded) modelPath.push('catalog-grounded-review');
} }
if (decision.confidence != null) {
// Cross-model agreement bonus: both models named the same species.
result = { ...result, confidence: decision.confidence };
}
modelPath.push('openai-review'); modelPath.push('openai-review');
modelPath.push(decision.reason); modelPath.push(decision.reason);
} else { } else {

7
server/lib/openai.js
View File

@@ -243,8 +243,11 @@ const buildIdentifyPrompt = (language, mode) => {
: '- "careInfo.temp": temperature range in Celsius (e.g. "18–24 °C"). Must always be a real plant-specific value, never "Unknown".', : '- "careInfo.temp": temperature range in Celsius (e.g. "18–24 °C"). Must always be a real plant-specific value, never "Unknown".',
'- "botanicalName" must use accepted Latin scientific naming and must not be invented or misspelled.', '- "botanicalName" must use accepted Latin scientific naming and must not be invented or misspelled.',
'- If species is uncertain, prefer genus-level naming (for example: "Calathea sp.").', '- If species is uncertain, prefer genus-level naming (for example: "Calathea sp.").',
'- "confidence" must be between 0 and 1.', '- "confidence" must be between 0 and 1 and reflect how certain the species identification is. Calibrate it:',
'- Keep confidence <= 0.55 when the image is ambiguous, blurred, or partially visible.', ' - 0.85-0.95: species clearly recognizable, distinctive features (leaf shape, flower, pattern) plainly visible.',
' - 0.65-0.84: species very likely, but some distinguishing features are hidden or similar species exist.',
' - 0.40-0.64: image is ambiguous, blurred, partially visible, or several species fit equally well.',
' - Below 0.40: mostly guessing; prefer genus-level naming instead.',
'- "waterIntervalDays" must be an integer between 1 and 45.', '- "waterIntervalDays" must be an integer between 1 and 45.',
'- Do not include markdown, explanations, or extra keys.', '- Do not include markdown, explanations, or extra keys.',
].join('\n'); ].join('\n');

13
server/lib/scanReview.js
View File

@@ -1,5 +1,13 @@
const { normalizeText } = require('./scanGrounding'); const { normalizeText } = require('./scanGrounding');
const clamp = (value, min, max) => Math.min(max, Math.max(min, value));
// Two DIFFERENT models independently naming the same species is genuine
// evidence beyond either model's self-reported confidence, so agreement
// earns a bonus on top of the better single estimate.
const REVIEW_AGREEMENT_BONUS = 0.2;
const REVIEW_AGREEMENT_CONFIDENCE_CAP = 0.97;
// Agreement must be judged on the RAW model answers, not the grounded ones: // Agreement must be judged on the RAW model answers, not the grounded ones:
// catalog grounding has a genus-level fallback that can collapse two different // catalog grounding has a genus-level fallback that can collapse two different
// species onto the same catalog entry and fake an agreement. // species onto the same catalog entry and fake an agreement.
@@ -30,6 +38,11 @@ const decideReviewOutcome = ({ primaryResult, reviewResult, agrees }) => {
accept: true, accept: true,
replace: reviewConfidence >= primaryConfidence, replace: reviewConfidence >= primaryConfidence,
reason: 'review-confirmed-primary', reason: 'review-confirmed-primary',
confidence: clamp(
Math.max(primaryConfidence, reviewConfidence) + REVIEW_AGREEMENT_BONUS,
0.05,
REVIEW_AGREEMENT_CONFIDENCE_CAP,
),
}; };
} }