Enhance Slot Evaluation and Scoring Mechanisms
Deploy Development / deploy (push) Successful in 46s
Test Suite / pytest-backend (push) Successful in 44s
Test Suite / lint-backend (push) Successful in 0s
Test Suite / build-frontend (push) Successful in 14s
Test Suite / k6 /health Baseline (push) Successful in 33s
Test Suite / playwright-tests (push) Successful in 1m26s

- Introduced new functions `_off_topic_semantic_scores_by_slot` and `_score_exercise_stage_fit_for_spec` to improve the evaluation of off-topic steps and exercise stage fit, enhancing the quality assessment process.
- Updated `_run_unified_slot_improvement_review` to incorporate off-topic scores and exercise stage fit scoring, refining the decision-making process for slot suggestions.
- Enhanced existing logic to streamline the handling of slot scores and improve the overall robustness of slot management in path evaluations.
This commit is contained in:
Lars committed 2026-06-13 12:33:16 +02:00
1 parent e9bf5bd1a5
commit cd457e3ea0
5 files changed
+654 -437

No files matched your search

@@ -501,53 +501,75 @@ export default function ProgressionGraphEditor({ graphId, embedded = false, onSa
const mergedAfterEval = mergeGapOffersForDraft(evaluated, baselineRes)
setGapFillOffers(mergedAfterEval.length > 0 ? mergedAfterEval : remainingOffers)
setMatchNotice('Schritt 2/2: Verbesserungsvorschläge für gemeldete Schachstellen…')
const reviewRes = await api.suggestProgressionPath({
...buildEvaluateRequest(synced),
evaluate_only: false,
unified_slot_review: true,
baseline_evaluate_steps: slotsToEvaluateSteps(synced),
baseline_path_qa_snapshot: baselineRes?.path_qa || null,
baseline_quality_score:
baselineRes?.path_qa?.quality_score != null
? Number(baselineRes.path_qa.quality_score)
: null,
include_llm_intent: false,
auto_rematch_after_qa: false,
})
if (!reviewRes?.unified_slot_review) {
throw new Error(
'Match-Review nicht verfügbar — Backend-Stand prüfen (unified_slot_review fehlt in der Antwort).',
)
setMatchNotice('Schritt 2/2: Slot-Alternativen prüfen…')
let compareRes
let reviewError = null
try {
const reviewRes = await api.suggestProgressionPath({
...buildEvaluateRequest(synced),
evaluate_only: false,
unified_slot_review: true,
baseline_evaluate_steps: slotsToEvaluateSteps(synced),
baseline_path_qa_snapshot: baselineRes?.path_qa || null,
baseline_quality_score:
baselineRes?.path_qa?.quality_score != null
? Number(baselineRes.path_qa.quality_score)
: null,
include_llm_intent: false,
auto_rematch_after_qa: false,
})
if (!reviewRes?.unified_slot_review) {
reviewError =
'Slot-Review nicht verfügbar — Backend neu starten/deployen (unified_slot_review fehlt).'
compareRes = buildProgressionComparePayload(baselineRes, {
...reviewRes,
unified_slot_review: true,
slot_reviews: [],
review_error: reviewError,
})
} else {
compareRes = buildProgressionComparePayload(baselineRes, reviewRes)
}
setGapFillOffers(mergeGapOffersForDraft(evaluated, baselineRes, reviewRes))
} catch (e) {
reviewError = e.message || 'Slot-Review fehlgeschlagen'
compareRes = buildProgressionComparePayload(baselineRes, {
unified_slot_review: true,
slot_reviews: [],
review_error: reviewError,
path_qa: baselineRes?.path_qa,
})
}
const compareRes = buildProgressionComparePayload(baselineRes, reviewRes)
setGapFillOffers(mergeGapOffersForDraft(evaluated, baselineRes, reviewRes))
presentMatchCompare(compareRes, { source })
presentMatchCompare(compareRes, { source, reviewError })
return compareRes
}
const presentMatchCompare = (res, { source = 'manual' } = {}) => {
const presentMatchCompare = (res, { source = 'manual', reviewError = null } = {}) => {
setSemanticBrief(res?.semantic_brief_summary || null)
setTargetSummary(res?.target_profile_summary || null)
setComparePayload(res)
setComparePayload(reviewError ? { ...res, review_error: reviewError } : res)
setCompareSource(source)
setProposedPathQa(res?.proposed_path_qa_pipeline || null)
setCompareOpen(true)
const baselineQa = res?.baseline_path_qa || null
const diffCount = res?.slot_diff_count ?? compareDiffsForDialog(res).length
const slotReviews = res?.slot_reviews || []
const autoCount = slotReviews.filter((r) => r?.library_alternative?.auto_select).length
const diffCount = autoCount || res?.slot_diff_count || 0
const rejectedCount = res?.slot_diff_count_rejected ?? rejectedCompareDiffs(res).length
const problemCount = res?.match_summary?.problem_slot_count
?? (res?.problem_slots ? Object.keys(res.problem_slots).length : 0)
const bPct = pathQaQualityPercent(baselineQa)
let notice =
diffCount > 0
? `Match: ${diffCount} Verbesserung(en) für gemeldete Schachstellen.`
: problemCount > 0
? `Match: ${problemCount} Schachstelle(n) erkannt, aber kein Bibliotheks-Ersatz mit Gewinn — KI-Angebote im Panel prüfen.`
: 'Match: Keine Schachstellen — Pfad wirkt konsistent.'
let notice = reviewError
? `Match: Dialog geöffnet — ${reviewError}`
: slotReviews.length > 0
? `Match: ${slotReviews.length} Slot(s) geprüft, ${autoCount} Empfehlung(en) vorausgewählt.`
: diffCount > 0
? `Match: ${diffCount} Verbesserung(en).`
: problemCount > 0
? `Match: ${problemCount} Schachstelle(n), keine bessere Bibliotheks-Alternative.`
: 'Match: Pfad geprüft — siehe Dialog.'
if (rejectedCount > 0) {
notice += ` ${rejectedCount} Vorschlag/Vorschläge verworfen (Verschlechterung oder neutral).`
}
@@ -1,13 +1,16 @@
/**
* Gegenüberstellung: Verbesserungsvorschläge mit Slot-Bewertung (Pro/Contra).
* Slot-Match-Dialog: je Slot Bewertung, Bibliotheks-Alternative, optional KI.
*/
import React, { useMemo, useState } from 'react'
import FormModalOverlay from './FormModalOverlay'
import {
compareDiffsForDialog,
compareSlotReviews,
defaultSelectedCompareDiffs,
pathQaQualityPercent,
qualityDeltaPercent,
rejectedCompareDiffs,
slotFitScorePercent,
slotReviewSelectionKey,
} from '../utils/progressionGraphDraft'
function qaLabel(pathQa) {
@@ -17,12 +20,10 @@ function qaLabel(pathQa) {
return ok ? 'OK' : 'Hinweise'
}
function deltaLabel(diff) {
const pct = qualityDeltaPercent(diff)
if (pct == null) return null
if (pct > 0) return `+${pct} % Pfad-QS`
if (pct === 0) return '±0 % Pfad-QS'
return `${pct} % Pfad-QS`
function slotScoreLabel(score) {
const pct = slotFitScorePercent(score)
if (pct == null) return '—'
return `${pct} % Stufen-Fit`
}
function ProContraList({ title, items, tone = 'neutral' }) {
@@ -41,89 +42,157 @@ function ProContraList({ title, items, tone = 'neutral' }) {
)
}
function DiffRow({ diff, checked, onToggle, applying }) {
const midx = Number(diff.roadmap_major_step_index)
const delta = deltaLabel(diff)
const pc = diff.pro_contra || {}
const isAi = diff.suggestion_type === 'ai_gap' || diff.proposed_is_ai_proposal
const isFill = diff.baseline_exercise_id == null && !isAi
function SlotReviewRow({ review, selected, onToggle, applying }) {
const midx = Number(review.roadmap_major_step_index)
const lib = review.library_alternative
const ai = review.ai_alternative
const libKey = slotReviewSelectionKey(midx, 'library')
const aiKey = slotReviewSelectionKey(midx, 'ai')
const pc = lib?.pro_contra || {}
const pathDelta = qualityDeltaPercent({ quality_delta: lib?.quality_delta })
const slotDelta = lib?.slot_score_delta
return (
<li
style={{
padding: '10px 12px',
padding: '12px',
borderRadius: '8px',
border: '1px solid var(--border)',
background: checked ? 'var(--surface2)' : 'var(--surface)',
border: `1px solid ${review.slot_problem ? 'var(--danger)' : 'var(--border)'}`,
background: 'var(--surface)',
fontSize: '12px',
}}
>
<label style={{ display: 'flex', gap: '8px', alignItems: 'flex-start', cursor: 'pointer' }}>
<input
type="checkbox"
checked={checked}
onChange={() => onToggle(midx)}
disabled={applying}
style={{ marginTop: '3px' }}
/>
<span style={{ flex: 1 }}>
<div style={{ display: 'flex', flexWrap: 'wrap', gap: '6px', alignItems: 'center' }}>
<strong>Slot {midx + 1}</strong>
{isFill ? (
<span style={{ fontSize: '10px', color: 'var(--accent-dark)' }}>Lücke füllen</span>
) : isAi ? (
<span style={{ fontSize: '10px', color: 'var(--accent-dark)' }}>KI-Alternative</span>
) : diff.off_topic ? (
<span style={{ fontSize: '10px', color: 'var(--danger)' }}>Passt nicht — Ersatz</span>
) : (
<span style={{ fontSize: '10px', color: 'var(--text2)' }}>Bessere Übung</span>
)}
{delta ? (
<span
style={{
fontSize: '10px',
fontWeight: 600,
color: qualityDeltaPercent(diff) > 0 ? 'var(--accent-dark)' : 'var(--text2)',
}}
>
{delta}
</span>
) : null}
</div>
<div style={{ display: 'flex', flexWrap: 'wrap', gap: '6px', alignItems: 'center', marginBottom: '8px' }}>
<strong>Slot {midx + 1}</strong>
{review.slot_problem ? (
<span style={{ fontSize: '10px', color: 'var(--danger)' }}>Schachstelle</span>
) : review.off_topic ? (
<span style={{ fontSize: '10px', color: 'var(--danger)' }}>Passt nicht</span>
) : (
<span style={{ fontSize: '10px', color: 'var(--text2)' }}>OK</span>
)}
{review.roadmap_learning_goal ? (
<span style={{ fontSize: '10px', color: 'var(--text3)', flex: '1 1 100%' }}>
{review.roadmap_learning_goal}
</span>
) : null}
</div>
<div
style={{
display: 'grid',
gridTemplateColumns: '1fr 1fr',
gap: '10px',
marginTop: '8px',
}}
>
<div>
<div style={{ fontSize: '10px', fontWeight: 600, color: 'var(--text3)', marginBottom: '4px' }}>
Aktuell
</div>
<div style={{ color: 'var(--text2)' }}>
{diff.baseline_title || '— leer —'}
{diff.baseline_exercise_id != null ? ` (#${diff.baseline_exercise_id})` : ''}
</div>
<ProContraList title="Pro" items={pc.current_pro} tone="pro" />
<ProContraList title="Contra" items={pc.current_contra} tone="contra" />
</div>
<div>
<div style={{ fontSize: '10px', fontWeight: 600, color: 'var(--text3)', marginBottom: '4px' }}>
Vorschlag
</div>
<div
style={{
display: 'grid',
gridTemplateColumns: '1fr 1fr',
gap: '10px',
marginBottom: lib || ai ? '10px' : 0,
}}
>
<div>
<div style={{ fontSize: '10px', fontWeight: 600, color: 'var(--text3)', marginBottom: '4px' }}>
Aktuell
</div>
<div style={{ color: 'var(--text2)' }}>
{review.baseline_title || '— leer —'}
{review.baseline_exercise_id != null ? ` (#${review.baseline_exercise_id})` : ''}
</div>
<div style={{ fontSize: '10px', color: 'var(--text3)', marginTop: '4px' }}>
{slotScoreLabel(review.baseline_slot_score)}
</div>
{(review.problem_reasons || []).slice(0, 3).map((text, i) => (
<p key={`pr-${i}`} style={{ margin: '4px 0 0', color: 'var(--danger)', fontSize: '11px' }}>
{text}
</p>
))}
</div>
<div>
<div style={{ fontSize: '10px', fontWeight: 600, color: 'var(--text3)', marginBottom: '4px' }}>
Beste Bibliotheks-Alternative
</div>
{lib ? (
<>
<div style={{ color: 'var(--accent-dark)' }}>
{diff.proposed_title || '—'}
{diff.proposed_exercise_id != null ? ` (#${diff.proposed_exercise_id})` : isAi ? ' (KI)' : ''}
{lib.title || '—'}
{lib.exercise_id != null ? ` (#${lib.exercise_id})` : ''}
</div>
<div style={{ fontSize: '10px', color: 'var(--text3)', marginTop: '4px' }}>
{slotScoreLabel(lib.slot_score)}
{slotDelta != null && Number(slotDelta) !== 0 ? (
<span style={{ marginLeft: '6px', color: Number(slotDelta) > 0 ? 'var(--accent-dark)' : 'var(--text2)' }}>
({Number(slotDelta) > 0 ? '+' : ''}{Math.round(Number(slotDelta) * 100)} PP)
</span>
) : null}
{pathDelta != null ? (
<span style={{ marginLeft: '6px' }}>· Pfad {pathDelta > 0 ? `+${pathDelta}` : pathDelta} %</span>
) : null}
</div>
<ProContraList title="Pro" items={pc.proposed_pro} tone="pro" />
<ProContraList title="Contra" items={pc.proposed_contra} tone="contra" />
</div>
</div>
</span>
</label>
</>
) : (
<div style={{ color: 'var(--text3)' }}>Kein passender Bibliotheks-Treffer</div>
)}
</div>
</div>
{lib ? (
<label
style={{
display: 'flex',
gap: '8px',
alignItems: 'flex-start',
cursor: 'pointer',
padding: '8px',
borderRadius: '6px',
background: selected.has(libKey) ? 'var(--surface2)' : 'transparent',
border: '1px solid var(--border)',
marginBottom: ai ? '8px' : 0,
}}
>
<input
type="checkbox"
checked={selected.has(libKey)}
onChange={() => onToggle(libKey, 'library')}
disabled={applying}
style={{ marginTop: '3px' }}
/>
<span>
<strong>Bibliothek übernehmen</strong>
<span style={{ display: 'block', fontSize: '11px', color: 'var(--text3)', marginTop: '2px' }}>
{lib.auto_select
? 'Empfohlen — Stufen-Fit besser als aktuell'
: 'Optional — nicht besser als aktuell bewertet'}
</span>
</span>
</label>
) : null}
{ai ? (
<label
style={{
display: 'flex',
gap: '8px',
alignItems: 'flex-start',
cursor: 'pointer',
padding: '8px',
borderRadius: '6px',
background: selected.has(aiKey) ? 'var(--surface2)' : 'transparent',
border: '1px solid var(--border)',
}}
>
<input
type="checkbox"
checked={selected.has(aiKey)}
onChange={() => onToggle(aiKey, 'ai')}
disabled={applying}
style={{ marginTop: '3px' }}
/>
<span>
<strong>KI-Vorschlag nutzen</strong>
<span style={{ display: 'block', fontSize: '11px', color: 'var(--text3)', marginTop: '2px' }}>
{ai.title_hint || 'Neue Übung per KI entwerfen'}
</span>
</span>
</label>
) : null}
</li>
)
}
@@ -136,10 +205,10 @@ export default function ProgressionOptimizeCompareModal({
onApplySelected,
applying = false,
}) {
const dialogDiffs = useMemo(() => compareDiffsForDialog(comparison), [comparison])
const slotReviews = useMemo(() => compareSlotReviews(comparison), [comparison])
const rejected = useMemo(() => rejectedCompareDiffs(comparison), [comparison])
const defaultSelected = useMemo(
() => defaultSelectedCompareDiffs(comparison),
() => new Set(defaultSelectedCompareDiffs(comparison)),
[comparison],
)
@@ -155,48 +224,59 @@ export default function ProgressionOptimizeCompareModal({
const baselineQa = comparison.baseline_path_qa
const baselinePct = pathQaQualityPercent(baselineQa)
const rejectedCount = rejected.length
const reviewError = comparison.review_error || null
const toggle = (midx) => {
const toggle = (key, kind) => {
setSelected((prev) => {
const next = new Set(prev)
if (next.has(midx)) next.delete(midx)
else next.add(midx)
const parsed = String(key)
const midx = Number(parsed.split(':')[0])
const libKey = slotReviewSelectionKey(midx, 'library')
const aiKey = slotReviewSelectionKey(midx, 'ai')
if (next.has(parsed)) {
next.delete(parsed)
return next
}
if (kind === 'library') next.delete(aiKey)
if (kind === 'ai') next.delete(libKey)
next.add(parsed)
return next
})
}
const toggleAll = (on) => {
setSelected(on ? new Set(dialogDiffs.map((d) => Number(d.roadmap_major_step_index))) : new Set())
}
const title =
mode === 'match' ? 'Übungs-Match — Verbesserungen' : 'Optimierung vergleichen'
mode === 'match' ? 'Übungs-Match — Slot-Bewertung' : 'Optimierung vergleichen'
return (
<div
className="modal-overlay"
role="dialog"
aria-modal="true"
aria-labelledby="optimize-compare-title"
style={{ zIndex: 2200 }}
onClick={(e) => {
if (e.target === e.currentTarget && !applying) onClose()
}}
>
<FormModalOverlay open={open} raised onBackdropClick={applying ? undefined : onClose}>
<div
className="card modal-content"
style={{ maxWidth: '760px', width: '100%', maxHeight: '90vh', overflow: 'auto' }}
className="card modal-panel--form modal-panel--narrow"
style={{ maxHeight: '92vh', overflow: 'auto' }}
onClick={(e) => e.stopPropagation()}
>
<h3 id="optimize-compare-title" style={{ marginTop: 0 }}>
{title}
</h3>
<p style={{ fontSize: '12px', color: 'var(--text3)', marginTop: 0, lineHeight: 1.45 }}>
Bewertung und Vorschläge in einem Durchlauf: je Slot wird geprüft, ob eine passendere
Übung (Bibliothek oder KI) den Pfad verbessert. Nur messbare Verbesserungen erscheinen
hier — mit Pro- und Contra-Punkten auf Slot-Ebene.
Je Slot: aktuelle Bewertung, beste Bibliotheks-Alternative und optional KI. Haken nur
vorausgewählt, wenn die Alternative einen höheren Stufen-Fit hat.
</p>
{reviewError ? (
<p
style={{
fontSize: '12px',
color: 'var(--danger)',
padding: '8px 10px',
borderRadius: '8px',
border: '1px solid var(--danger)',
background: 'var(--surface2)',
}}
>
{reviewError}
</p>
) : null}
<div
style={{
padding: '10px 12px',
@@ -216,97 +296,37 @@ export default function ProgressionOptimizeCompareModal({
{rejectedCount > 0 ? (
<p style={{ fontSize: '11px', color: 'var(--text2)', margin: '0 0 14px', lineHeight: 1.45 }}>
{rejectedCount} Alternative(n) verworfen — kein QS-Gewinn gegenüber deinem Pfad
{baselinePct != null ? ` (${baselinePct} %)` : ''}.
{rejectedCount} Alternative(n) ohne Pfad-Gewinn
{baselinePct != null ? ` (Basis ${baselinePct} %)` : ''}.
</p>
) : null}
{dialogDiffs.length === 0 ? (
<>
<p style={{ fontSize: '12px', color: 'var(--text2)' }}>
Keine Bibliotheks-Verbesserung mit messbarem Gewinn — Schachstellen siehe unten.
KI-Angebote im Bewertungs-Panel oder „Brücke / KI-Angebot“ nutzen.
</p>
{comparison?.problem_slots && Object.keys(comparison.problem_slots).length > 0 ? (
<ul
style={{
listStyle: 'none',
padding: 0,
margin: '12px 0 0',
display: 'flex',
flexDirection: 'column',
gap: '8px',
}}
>
{Object.entries(comparison.problem_slots).map(([midxRaw, reasons]) => {
const midx = Number(midxRaw)
const reasonList = Array.isArray(reasons) ? reasons : [reasons]
return (
<li
key={`problem-${midxRaw}`}
style={{
padding: '10px 12px',
borderRadius: '8px',
border: '1px solid var(--danger)',
background: 'var(--surface2)',
fontSize: '12px',
}}
>
<strong style={{ color: 'var(--danger)' }}>
Schachstelle Slot {midx + 1}
</strong>
<ul style={{ margin: '6px 0 0', paddingLeft: '16px', color: 'var(--text2)' }}>
{reasonList.filter(Boolean).map((text, i) => (
<li key={`${midxRaw}-r-${i}`}>{text}</li>
))}
</ul>
</li>
)
})}
</ul>
) : null}
</>
{slotReviews.length === 0 ? (
<p style={{ fontSize: '12px', color: 'var(--text2)' }}>
Keine Slot-Daten — Backend-Stand prüfen oder erneut „Graph bewerten“ und „Übungen
matchen“.
</p>
) : (
<>
<div style={{ display: 'flex', gap: '8px', marginBottom: '8px', flexWrap: 'wrap' }}>
<button
type="button"
className="btn btn-secondary"
style={{ fontSize: '11px' }}
onClick={() => toggleAll(true)}
>
Alle wählen
</button>
<button
type="button"
className="btn btn-secondary"
style={{ fontSize: '11px' }}
onClick={() => toggleAll(false)}
>
Keine
</button>
</div>
<ul
style={{
listStyle: 'none',
padding: 0,
margin: 0,
display: 'flex',
flexDirection: 'column',
gap: '8px',
}}
>
{dialogDiffs.map((diff) => (
<DiffRow
key={`improve-${diff.roadmap_major_step_index}-${diff.suggestion_type || 'lib'}`}
diff={diff}
checked={selected.has(Number(diff.roadmap_major_step_index))}
onToggle={toggle}
applying={applying}
/>
))}
</ul>
</>
<ul
style={{
listStyle: 'none',
padding: 0,
margin: 0,
display: 'flex',
flexDirection: 'column',
gap: '10px',
}}
>
{slotReviews.map((review) => (
<SlotReviewRow
key={`slot-review-${review.roadmap_major_step_index}`}
review={review}
selected={selected}
onToggle={toggle}
applying={applying}
/>
))}
</ul>
)}
<div
@@ -324,13 +344,13 @@ export default function ProgressionOptimizeCompareModal({
<button
type="button"
className="btn btn-primary"
disabled={applying || selected.size === 0 || dialogDiffs.length === 0}
disabled={applying || selected.size === 0}
onClick={() => onApplySelected([...selected])}
>
{applying ? 'Übernehmen …' : `Auswahl übernehmen (${selected.size})`}
</button>
</div>
</div>
</div>
</FormModalOverlay>
)
}