"""Source-bound academic roles, meeting records and explicit completion events. Neither authorship nor an academic heading proves supervision or graduation. This path needs no generative model: every delivered statement cites literal evidence from the same profile and work. Unknown is never treated as ongoing. """ from __future__ import annotations from collections import defaultdict from datetime import date import re from tools.answer_evaluation.context_policy import normalized from tools.knowledge.sources import advisor_name from tools.retrieval.person_names import resolve_person_names from tools.retrieval.work_selector import parse_work_selector def academic_intent(question): text = normalized(question) completion = re.search(r'\b(?:ukonc\w*|skoncil\w*|dokonc\w*|obhaj\w*|neobhaj\w*|absolvov\w*|' r'prerusil\w*|studuje|studium|stale prebieha)\b', text) subtask = re.search(r'\b(?:uloh\w*|experiment\w*|implement\w*|trenovan\w*)\b',text) academic_object = re.search(r'\b(?:obhaj\w*|studium|studia|studuje|ukoncenie prace)\b',text) if completion and (not subtask or academic_object): return 'completion' if re.search(r'\b(?:stretnut\w*|konzultaci\w*)\b', text) and re.search( r'\b(?:naposledy|posledn\w*|najnov\w*|kedy|o com)\b', text): return 'meeting' if re.search(r'\bkto\b.*\b(?:vedie|viedol|viedla)\b',text): return 'advisors' if re.search(r'\b(?:veduc\w*|skolitel\w*|konzultant\w*)\b', text): return 'advisors' return None def completion_event(unit): """Recognize narrow, explicit outcomes, never completed subtasks or plans.""" text = normalized(unit['text']) if unit.get('phase') in {'assignment','planned'} or unit.get('completion_basis') == 'checked_task': return None if re.search(r'\b(?:planuj\w*|plan |ciel\w*|mal by|ma byt|bude|budu|treba|chce|chcel|' r'nepotvrd\w*|nie je potvrden\w*|nie je dolozen\w*|zatial neukoncil\w*)\b',text): return None # Avoid interpreting a nested quote about somebody else as this student's outcome. if re.search(r'\b(?:iny student|inych studentov|kolega|kolegy\w*)\b',text): return None study = bool(re.search(r'\b(?:studium|studia|studuje|absolvov\w*)\b',text)) work = bool(re.search(r'\b(?:prac\w*|projekt\w*|obhaj\w*|neobhaj\w*)\b',text)) if study and work: return None # Two different event scopes in one sentence need separate evidence. status = None if re.search(r'\b(?:ukoncil\w*|skoncil\w*) predcasne\b|\bprerusil\w* studium\b',text): status = 'interrupted' if 'prerusil' in text else 'premature' elif re.search(r'\bneobhajil\w*\b|\bneuspesne (?:ukoncil\w*|obhajil\w*)\b',text): status = 'unsuccessful' elif re.search(r'\b(?:uspesne (?:ukoncil\w*|obhajil\w*|absolvoval\w*)|obhajil\w*|absolvoval\w*)\b',text): if not re.search(r'\b(?:neobhajil\w*|nie|neuspesne)\b',text): status = 'successful' elif re.search(r'\b(?:dokonc(?:il|ila|ene|ena)\w*|ukoncil\w*|odovzdal\w*)\b',text) and (study or work): status = 'submitted' if 'odovzdal' in text else 'completed_unspecified' elif study and re.search(r'\b(?:stale studuje|studium pokracuje|studium prebieha)\b',text): status = 'ongoing' elif work and re.search(r'\b(?:praca|projekt) (?:stale )?(?:prebieha|pokracuje)\b',text): status = 'ongoing' if not status or not (study or work or status in {'premature','interrupted'}): return None # A model, database, experiment or text being finished isn't a defended thesis. if re.search(r'\b(?:text prace|pisanie prace|minimov\w*|pracu na|ukoncil\w* trenovan\w*|dokonc\w* implement\w*)\b',text): return None scope = 'studies' if study else 'work' if work else 'unspecified' if scope == 'work': from tools.answer_evaluation.rag_evidence import academic_scope mentioned=academic_scope(unit['text']) if mentioned and unit.get('work_type') and mentioned[0]!=unit['work_type']: return None # A dated note is NOT automatically the date the event happened. years = set(re.findall(r'\b(?:v roku|roku|dna|v)\s+([12]\d{3})\b',text)) years.update(re.findall(r'\b\d{1,2}\.\s*\d{1,2}\.\s*([12]\d{3})\b',unit['text'])) event_year = int(next(iter(years))) if len(years)==1 else None event_date=None dates=re.findall(r'\b(\d{1,2})\.\s*(\d{1,2})\.\s*([12]\d{3})\b',unit['text']) if len(dates)==1: try:event_date=date(int(dates[0][2]),int(dates[0][1]),int(dates[0][0])).isoformat() except ValueError:return None return {'status':status,'scope':scope,'event_year':event_year,'event_date':event_date} def answer_academic(question, index, *, engine, limit=5, units=None, mode=None): mode = mode or academic_intent(question) rows = units if units is not None else index['units'] catalog = [{'name':name,'kind':'Person','source_document':path} for name,path in sorted({(r['person'],r['source_path']) for r in rows})] role_names = {normalized(name):name for r in rows for role in ('supervisor','consultant') if (name:=advisor_name(r,role))} known = {normalized(p['name']) for p in catalog} catalog += [{'name':name,'kind':'Person'} for key,name in role_names.items() if key not in known] resolution = resolve_person_names(question,catalog) base = {'engine':engine,'index_version':index['index_version'],'knowledge_version':index['version'], 'accepted_for_delivery':True,'claims':[],'complete':False,'truncated':False, 'truncation_reasons':[],'name_resolution':resolution} def respond(text, evidence=(), *, status='ok', reason=None, truncated=False): unique = list({r['id']:r for r in evidence}.values()) urls = list(dict.fromkeys(r['source_url'] for r in unique)) text = '\n\n'.join([*resolution.get('notices',[]),text]) if urls:text += '\n\nZdroje: '+', '.join(urls) return {**base,'status':status,'reason':reason,'answer':text,'source_urls':urls, 'truncated':truncated,'truncation_reasons':['academic_result_limit'] if truncated else [], 'evidence':unique,'answer_coverage':{ 'scope':'list' if mode=='advisors' else 'overview', 'status':'not_found' if status=='no_answer' else 'unavailable' if status=='error' else 'partial' if reason or truncated else 'supported', 'warnings':[],'limitations':[reason] if reason else []}} if resolution['status']!='ok': return respond(resolution['message'],status=resolution['status'],reason=resolution['reason']) if index.get('academic_schema')!=1: return respond('Index rolí, stretnutí a ukončení ešte nie je pripravený pre túto verziu dát. ' 'Je potrebná obnova obsahového indexu.',status='error',reason='content_index_unavailable') resolved=resolution.get('resolved_query',question) text=normalized(resolved) selection=parse_work_selector(resolved) from tools.retrieval.question_constraints import unsupported_constraint,constraint_message unsupported=unsupported_constraint(resolved) if unsupported not in {None,'work_event_date','year_range'}: return respond(constraint_message(unsupported),status='ambiguous',reason='unsupported_constraint') if re.search(r'\b(?:tag\w*|klucov\w* slov\w*|autor dokumentu|semest\w*)\b',text): return respond('Túto kombinovanú podmienku neviem spoľahlivo overiť. Uveď osobu, konkrétnu prácu ' 'a požadovanú rolu, stretnutie alebo ukončenie.',status='ambiguous',reason='unsupported_academic_join') matches=resolution.get('matches',[]) staff={normalized(m['name']) for m in matches if normalized(m['name']) in role_names} students=[m for m in matches if m.get('source_document')] # A known mentor can also have a student profile: only treat it as mentor # when the question explicitly filters by leadership/consultation. role_filter=bool(staff and re.search(r'\b(?:pod vedenim|vedie|viedol|veden\w*|veduc\w*|skolitel\w*|konzultant\w*)\b',text)) if role_filter:students=[m for m in students if normalized(m['name']) not in staff] global_question=bool(re.search(r'\b(?:kto|ktori|ktore)\b',text) and mode=='completion') if not students and not role_filter and not global_question: return respond('Uveď celé meno študenta alebo meno vedúceho, podľa ktorého mám práce vybrať.', status='ambiguous',reason='missing_subject') if len(staff)>1 and role_filter: return respond('Uveď jedného vedúceho alebo konzultanta pre tento výber prác.', status='ambiguous',reason='ambiguous_advisor_filter') from tools.answer_evaluation.rag_evidence import requested_types types=requested_types(resolved,[m['name'] for m in students]) # Ordinary wording of the same work-type constraint must not silently # broaden "diplomovku" to every bachelor thesis and project on a profile. informal={kind for pattern,kind in [(r'\bdiplomovk\w*\b','diploma_thesis'), (r'\bbakalark\w*\b','bachelor_thesis')] if re.search(pattern,text)} if len(informal)>1 or informal and types and not set(types.values())<=informal: return respond('Upresni, či sa pýtaš na bakalársku alebo diplomovú prácu.', status='ambiguous',reason='ambiguous_work_type') if informal: types={m['name']:next(iter(informal)) for m in students} global_types=set(types.values()) if not students: from tools.answer_evaluation.rag_evidence import work_type kind=work_type(resolved) if kind:global_types={kind} elif informal:global_types=informal # Completion-year filters apply to the EVENT, not the work heading. from tools.knowledge.query import _period try:period=_period(resolved) except ValueError: return respond('Uveď platný rok alebo interval dátumov.',status='ambiguous',reason='invalid_period') work_year_match=re.search(r'\b(?:diplomov\w*|bakalarsk\w*|timov\w*)\s+(?:prac\w*|projekt\w*)' r'(?:\s+\d{1,2})?\s+(?:z\s+roku\s+)?([12]\d{3})\b',text) work_year=int(work_year_match[1]) if work_year_match else None if mode=='completion' and period and work_year: return respond('Máš na mysli rok evidovaný pri práci, alebo rok jej skutočného ukončenia? ' 'Uveď výslovne požadovaný dátum ukončenia; tieto roky nemusia byť rovnaké.', status='ambiguous',reason='completion_year_scope') selected_paths={m['source_document'] for m in students} groups=defaultdict(list) for r in rows: if selected_paths and r['source_path'] not in selected_paths:continue if selection.specific and not selection.matches(r.get('work_label') or ''):continue kind=types.get(r['person']) if kind and r.get('work_type')!=kind:continue if not students and global_types and r.get('work_type') not in global_types:continue if mode=='advisors' and period and not (r.get('work_year') and int(period[0][:4])<=r['work_year']<=int(period[1][:4])):continue if mode=='meeting' and work_year and r.get('work_year')!=work_year:continue groups[(r['source_path'],r.get('work_label'))].append(r) requested_role='consultant' if re.search(r'\bkonzultant\w*\b',text) and not re.search(r'\bveduc\w*|\bveden\w*',text) else 'supervisor' if role_filter: groups={key:part for key,part in groups.items() if key[1] and any( normalized(advisor_name(r,requested_role) or '') in staff for r in part)} evidence=[];lines=[];omitted=False status_labels={'successful':'úspešné ukončenie / obhajoba','unsuccessful':'neúspešné ukončenie / obhajoba', 'premature':'predčasné ukončenie','interrupted':'prerušenie štúdia', 'completed_unspecified':'ukončenie bez doloženého výsledku obhajoby', 'submitted':'odovzdanie; úspešná obhajoba nepotvrdená','ongoing':'výslovne zaznamenané pokračovanie'} study_question=bool(re.search(r'\bstudium\b|\bstudia\b|\bstuduje\b',text)) wants_success=bool(re.search(r'\buspesn\w*\b',text)) wants_failure=bool(re.search(r'\bneuspesn\w*\b|\bneobhaj\w*\b',text)) if wants_success and wants_failure:wants_success=wants_failure=False outcomes=defaultdict(list) conflicts=set() if mode=='completion': for key,part in groups.items(): all_events=[(r,event) for r in part if (event:=completion_event(r))] # Unresolved contradictory outcomes of the same work must not # produce an unqualified success list. A later dated re-defense # can be queried explicitly by date instead. statuses={event['status'] for r,event in all_events if event['scope']=='work'} if {'successful','unsuccessful'}<=statuses and any(not event['event_date'] for r,event in all_events): conflicts.add(key) continue for r,event in all_events: if study_question and event['scope']!='studies':continue if not study_question and key[1] and event['scope']=='studies':continue if role_filter and event['scope']!='work':continue if wants_success and event['status']!='successful':continue if wants_failure and event['status']!='unsuccessful':continue if period: if event['event_date']: if not period[0]<=event['event_date']<=period[1]:continue elif not (period[0].endswith('-01-01') and period[1].endswith('-12-31') and event['event_year'] and int(period[0][:4])<=event['event_year']<=int(period[1][:4])):continue outcomes[key].append((r,event)) # Unscoped introductory metadata must not become a fictional extra work. keys=sorted((key for key,part in groups.items() if key[1] or any( completion_event(r) or advisor_name(r,'supervisor') or advisor_name(r,'consultant') or r['kind']=='meeting' for r in part)),key=lambda k:(groups[k][0]['person'],k[0],k[1] or '')) if mode=='completion' and (role_filter or not students):keys=[k for k in keys if outcomes[k] or k in conflicts] if mode=='meeting': by_person=defaultdict(list) for key in keys:by_person[key[0]].extend(groups[key]) for path,part in by_person.items(): markers=[r for r in part if r['kind']=='meeting' and (not period or work_year or r.get('date') and period[0]<=r['date']<=period[1])] dated=[r for r in markers if r.get('date')] if not markers: lines.append(f'**{part[0]["person"]}**: v tomto výbere nie je doložený záznam stretnutia.');continue latest=max((r['date'] for r in dated),default=None) chosen=[r for r in markers if r.get('date')==latest] if latest else markers for marker in chosen[:limit]: lines.append(f'**{marker["person"]} — {marker.get("work_label") or "profil"}**\n' f'Posledný záznam s úplným dátumom: {latest}.' if latest else f'**{marker["person"]}** — záznam: {marker["date_label"]}; rok nie je uvedený.') evidence.append(marker) notes=[r for r in part if r.get('meeting_id')==marker['meeting_id'] and r['kind'] not in {'meeting','work_record','reference','supervisor','consultant'}] from tools.answer_evaluation.content_selection import unit_score notes=[r for r in notes if not r['text'].rstrip().endswith(':')] notes.sort(key=lambda r:(-unit_score(r['text'],'progress' if r['phase'] in {'result','progress'} else 'task'),r['phase']!='result',r['line'])) actual=[r for r in notes if r['phase'] in {'result','progress'}] planned=[r for r in notes if r['phase']=='planned'] selected=actual[:4]+planned[:4] selected.extend(r for r in notes if r not in selected) notes=selected for note in notes[:8]: phase={'result':'Zaznamenaný výsledok','progress':'Priebežný stav','planned':'Plán','assignment':'Zadanie'}.get(note['phase'],'Záznam') quote=re.sub(r'^\s*(?:[-*+]|\d+[.)])\s+','',note['text']) lines.append(f'- {phase}: {quote}');evidence.append(note) if not notes:lines.append('Obsah tohto stretnutia nie je v profile uvedený.') if len(notes)>8:omitted=True if len(chosen)>limit:omitted=True partial=[r for r in part if r['kind']=='meeting' and not r.get('date')] if partial: lines.append('Profil obsahuje aj záznamy bez úplného dátumu; nemožno spoľahlivo určiť, ' 'či sú novšie. Rok nedopĺňam z roku práce.') evidence.extend(partial) else: for key in keys[:limit]: part=groups[key];person=part[0]['person'];label=key[1] or 'profil (bez priradenia ku konkrétnej práci)' lines.append(f'**{person} — {label}**') header=next((r for r in part if r['kind']=='work_record'),None) if header:evidence.append(header) title=next((r for r in part if r['kind']=='topic' and not r['text'].rstrip().endswith(':')),None) if title:lines.append('Názov / téma zo zdroja: '+title['text']);evidence.append(title) if mode=='advisors' or role_filter: for role,name_label in [('supervisor','Vedúci práce'),('consultant','Konzultant')]: roles=[r for r in part if advisor_name(r,role)] names=list(dict.fromkeys(advisor_name(r,role) for r in roles)) lines.append(name_label+': '+(', '.join(names) if names else 'v dostupných podkladoch nie je uvedený')+'.') evidence.extend(roles) if mode=='advisors': lines.append('Autor dokumentu nie je automaticky vedúci ani konzultant práce.') if mode=='completion': if key in conflicts: lines.append('Zdroj obsahuje rozporné záznamy o výsledku tejto práce; bez doloženého poradia ' 'obhajob nemožno potvrdiť úspešné ukončenie.') evidence.extend(r for r in part if completion_event(r)) continue found=outcomes[key] for r,event in found: scope={'studies':'štúdium','work':'práca','unspecified':'rozsah ukončenia nie je výslovne uvedený'}[event['scope']] lines.append(f'- {status_labels[event["status"]]} ({scope}): „{r["text"]}“') lines.append('Rok udalosti: '+str(event['event_year'])+'.' if event['event_year'] else 'Dátum ukončenia nie je doložený; rok práce ho nenahrádza.') evidence.append(r) if not found: lines.append('Úspešné ani neúspešné ukončenie tejto práce' + (' či celého štúdia' if study_question else '')+ ' nie je v dostupných podkladoch doložené. Nemožno z toho usúdiť, že stále prebieha.') notes=[r for r in part if completion_event(r)] for note in notes: lines.append('Doplňujúci záznam s odlišným alebo neurčeným rozsahom: „'+note['text']+'“.');evidence.append(note) omitted=len(keys)>limit if not lines: return respond('V dostupných profiloch nie je doložená práca, ktorá by súčasne spĺňala všetky podmienky ' 'otázky. Vedúci, konkrétna práca, výsledok ukončenia a rok udalosti musia byť doložené ' 'pri tom istom zázname. Rok práce, autor dokumentu ani odovzdanie textu nepotvrdzujú ' 'úspešnú obhajobu. To neznamená, že takí študenti neexistujú.', status='no_answer',reason='academic_conditions_not_documented') if omitted:lines.append('Zobrazený je obmedzený výber. Zúž otázku na študenta, konkrétnu prácu alebo obdobie, prípadne zvýš limit.') unknown=mode=='completion' and not any(outcomes.values()) return respond('\n\n'.join(lines),evidence,status='no_answer' if unknown else 'ok', reason='conflicting_completion_evidence' if conflicts else 'completion_not_documented' if unknown else None, truncated=omitted)