VAIDH-AIकानून की चाबी, आपके हाथों में

Pension research methodology

How the pension dashboards are derived

This page brings across the Flask pension methodology for both Supreme Court and High Court analytics: what counts as a pension case, how government and public bodies are identified, how outcomes are inferred, and when a record is pushed for manual review.

Supreme Court statsHigh Court statsSupreme methodologyHigh Court methodology

At a glance

1. Case filteringFind pension disputesJudgments are parsed, court/source is checked, and pension or retiral-benefit language is detected.
2. Party labelingMap government involvementParties are classified as central government, state government, public body, non-government or unknown.
3. Outcome detectionInfer likely resultConclusion and judgment-tail phrases help infer winner, dismissal, remand, partial success and government result.
4. Review controlFlag weak recordsAmbiguous or incomplete records are marked for manual verification instead of being treated as clean data.

Supreme Court pension methodology

How Supreme Court pension cases are derived

The Supreme Court methodology keeps Supreme Court matters, detects pension language through weighted keyword logic, maps government/public-body roles, infers winner and government result, detects possible High Court to Supreme Court route signals, and flags weak cases for review.

  1. 1

    Read raw judgment

    HTML judgment files are parsed and important fields are extracted: title, court, facts, issues, conclusion, citations and party names.

  2. 2

    Court filtering

    The court/source field is checked so that only Supreme Court matters continue into the Supreme Court pension analysis pipeline.

    Supreme Court filtering
  3. 3

    Pension relevance

    Pension-related legal terms are scored using strong and medium keyword signals. Strong terms carry more weight; multiple medium terms can also mark a case as pension-related.

    keyword scoringpension terms
  4. 4

    Party classification

    Parties are classified as central government, state government, public body, or non-government entities using government and public-body lexicons.

    central govtstate govtpublic body
  5. 5

    Winner inference

    Outcome phrases such as appeal allowed, appeal dismissed, set aside, affirmed, modified and remanded are used to infer the likely winning side.

    high confidencemediumweak fallback
  6. 6

    Government result

    The pipeline checks whether the winning side is government/public body and detects possible High Court to Supreme Court route signals.

    government win/lossHC → SC signals
  7. 7

    Manual review

    Missing parties, unclear winners, weak extraction and borderline cases are flagged instead of being silently mixed with trusted results.

    review flagsconfidence control

Pension detection keywords

Strong and medium pension terms used to decide whether a Supreme Court judgment belongs in the pension dashboard.

Strong terms

pensionfamily pensionpensionary benefitsretiral benefitsretirement benefitsgratuity

Medium terms

provident fundgpfepfsuperannuationretired from servicequalifying servicecompulsory retirementretiralretirement

Government and public-body detection

Party names are matched against central government, state government and public-body terms to determine government involvement.

Central government

union of indiagovernment of indiagovt of indiacentral governmentministry ofdepartment ofrailwayrailwaysincome taxcentral boarddirectorate generalepfoemployees provident fundprovident fund commissionerregional provident fund commissionerassistant solicitor generaladditional solicitor generalsolicitor generalunion public service commissionfood corporation of indialic of indialife insurance corporation of indiaemployee state insuranceemployees state insuranceesicnational capital territory of delhi

State government

state ofthe state ofstate governmentgovernment ofgovt ofdistrict magistratedistrict collectorcollectortehsildartahsildardeputy commissionerprincipal secretarychief secretarysecretary to governmentcommissioner ofsub divisional officersub-divisional officersdodirector of pensiondirectorate of pensiontreasury officeraccountant generalprincipal accountant generaleducation officerblock education officerbasic education officerdirector of educationdirectorate of educationstate pension departmentdirector generaladditional directorjoint directordeputy directorassistant directordirector of treasuriesdirectorate of treasuriesdistrict treasury officersub treasury officersenior treasury officertreasury departmentstate electricity boardstate road transportstate transportstate of andhra pradeshstate of arunachal pradeshstate of assamstate of biharstate of chhattisgarhstate of goastate of gujaratstate of haryanastate of himachal pradeshstate of jharkhandstate of karnatakastate of keralastate of madhya pradeshstate of maharashtrastate of manipurstate of meghalayastate of mizoramstate of nagalandstate of odishastate of orissastate of punjabstate of rajasthanstate of sikkimstate of tamil nadustate of telanganastate of tripurastate of uttar pradeshstate of uttarakhandstate of west bengalstate of jammu and kashmirstate of delhistate of nct of delhistate of pondicherrystate of puducherrystate of chandigarhstate of andaman and nicobarstate of dadra and nagar havelistate of daman and diustate of lakshadweepstate of ladakh

Public body

municipal corporationmunicipalityzila parishadpanchayatdevelopment authorityhousing boardelectricity boardtransport corporationroad transport corporationuniversityboardcorporationauthoritypublic sectorpsustatutory bodylocal bodyautonomous bodyaided institutiongovernment aidedcooperative societybank

Outcome and winner inference

Procedural phrases are checked in the conclusion, judgment tail and full text. Strong patterns are trusted more than fallback disposal phrases.

High-confidence patterns

appeal allowedcivil appeal allowedwe allow the appealpetition allowedwrit petition allowedspecial leave petition allowedappeal dismissedcivil appeal dismissedthe appeal stands dismissedwe dismiss the appealpetition dismissedwrit petition dismissedspecial leave petition dismissed

Medium-confidence patterns

appellant succeedspetitioner succeedsimpugned order set asidejudgment is set asideorder is quashedno interference is called forrespondent succeedsappeal must failimpugned order upheldjudgment is affirmedappeal partly allowedpetition partly allowedmatter is remandedcase is sent backrule is made absolute

Low and weak fallback patterns

entitled to pensionrelief is grantedrespondents are directed torelief is declineddisposed of accordinglydisposed of in the above termsdisposed ofno order as to costs

Facts, issues and route support

Extra context signals help extract facts/issues and identify possible High Court to Supreme Court movement.

Fact hints

employeeserviceretiredretirementpensionappointmentdepartmentbenefitclaimdisputegratuityprovident fundretiralsuperannuation

Issue starters

the question is whetherthe short issue isthe point for determination isthe controversy iswhether

Argument/noise words

arguedcontendedsubmittedurgedprayedlearned counselcounsel foraccording to the appellantaccording to the respondentleave grantedjudgmentorder

HC → SC inference signals

High Courtimpugned judgmentimpugned orderwrit petitionwrit appealletters patent appealdivision benchsingle judgeSLPcivil appeal

Manual review flags

The dashboard separates cleanly inferred records from ambiguous ones so the analytics remain honest.

High-priority flags

not_supreme_courtmissing_respondentmissing_appellant_or_petitionerwinner_unknownparse_error

Medium-priority flags

facts_and_issues_missingwinner_govt_unclear

Low-priority flags

winner_unclearfacts_missingissues_missingparty_confidence_lowroles_from_title_onlyhc_signal_borderlinegovt_confidence_low
Supreme Court summary: Raw judgments → extract court/title/facts/issues/conclusion → keep Supreme Court matters → detect pension language → classify parties using government/public-body lexicons → infer likely winner from procedural outcome phrases → infer government success and possible HC-to-SC route → flag weak cases for manual review → produce dashboard-ready analytics.

High Court pension methodology

How High Court pension cases are detected, classified and reviewed

The High Court methodology processes raw judgment HTML, identifies pension-related matters, classifies parties into government, public-body and non-government categories, infers procedural outcomes, and highlights uncertain records for manual verification.

  1. 1

    Scan all High Court HTML files

    The backend recursively searches High Court judgment HTML files, groups files by year using folder paths, and processes them year by year.

  2. 2

    Parse HTML and clean text

    Readable judgment text is extracted, and extra spaces and line breaks are normalized before analysis.

  3. 3

    Extract basic case metadata

    The system extracts document ID, state folder, case folder, folder year, judgment year, title and court/source field.

  4. 4

    Extract facts, issues and conclusion

    Structured tags are preferred. If unavailable, fallback logic searches paragraphs containing factual hints or issue-style phrases such as “whether” and “the question is whether”.

    fact hintsissue patterns
  5. 5

    Detect pension relevance

    The extraction system combines title, facts, issues and full text, then checks pension, retiral-benefit, gratuity, leave-encashment and provident-fund phrases.

    pension termstopic terms
  6. 6

    Extract and classify parties

    Party names are extracted from the structured pre block where possible. If not available, the title is split around “v”, “vs”, or “versus”.

    pre blocktitle fallbackparty roles
  7. 7

    Detect government/public-body involvement

    Each party is classified as state government, central government, public body, non-government, or unknown.

    state variantsstate govtcentral govtpublic body
  8. 8

    Infer winner and government result

    Conclusion, judgment tail and full text are scanned for outcome patterns. The winning side is mapped back to party type to decide whether government won, lost, partially succeeded, or needs review.

    petitioner successrespondent successpartial/remand
  9. 9

    Flag weak records

    Records are flagged when parties are missing, winner is unclear, facts/issues are missing, party confidence is low, government confidence is low, or parsing failed.

    manual reviewconfidence model
  10. 10

    Rebuild pension-only final CSV

    At the end of each year and at the end of the run, the final CSV is rebuilt from the debug CSV while keeping only pension cases.

Pension detection terms

High Court detection uses the full pension, retiral, gratuity, leave encashment and provident-fund signal library.

Pension terms

pensionpensionarypensionary benefitspensionary benefitfamily pensionold pension schemenew pension schemenational pension systemnational pension schemecontributory pension schemenon-contributory pensionwidow pensiondisability pensioninvalid pensioncompassionate pensionprovisional pensioncommuted pensioncommutation of pensionrevision of pensionrevised pensionpension arrearsarrears of pensionpension payment orderppo

Retiral, gratuity and fund terms

retiral benefitsretiral duespost retiral benefitspost-retiral benefitsretirement benefitsretirement benefitretirement duesretiremental benefitsterminal benefitsterminal duessuperannuationsuperannuatedretired from serviceretired employeeretired teacherretired government servantqualifying servicepast serviceservice pensiongratuitydeath-cum-retirement gratuitydeath cum retirement gratuitydcrgleave encashmentearned leave encashmentencashment of leaveprovident fundgeneral provident fundemployees provident fundemployee provident fundcontributory provident fundpublic provident fundgpfepfcpfpfprovident fund commissionerregional provident fund commissioneremployee pension schemeemployees pension schemeepseps-95eps 95employees' pension schemeemployees pension scheme 1995

State, central government and public-body signals

The High Court pipeline expands government detection with state/UT variants and department-style administrative terms.

Generated state and UT variants

andhra pradesharunachal pradeshassambiharchhattisgarhgoagujaratharyanahimachal pradeshjharkhandkarnatakakeralamadhya pradeshmaharashtramanipurmeghalayamizoramnagalandodishaorissapunjabrajasthansikkimtamil nadutelanganatripurauttar pradeshuttarakhandwest bengaljammu and kashmirdelhinct of delhipondicherrypuducherrychandigarhandaman and nicobardadra and nagar havelidaman and diulakshadweepladakh

State government and departments

state ofthe state ofstate governmentgovernment ofgovt ofdistrict magistratedistrict collectorcollectortehsildartahsildardeputy commissionerprincipal secretarychief secretarysecretary to governmentcommissioner ofsub divisional officersub-divisional officersdodirector of pensiondirectorate of pensiontreasury officeraccountant generalprincipal accountant generaleducation officerblock education officerbasic education officerdirector of educationdirectorate of educationstate pension departmentdirector generaladditional directorjoint directordeputy directorassistant directordirector of treasuriesdirectorate of treasuriesdistrict treasury officersub treasury officersenior treasury officertreasury departmentstate electricity boardstate road transportstate transport

Central government

union of indiagovernment of indiagovt of indiacentral governmentministry ofdepartment ofrailwayrailwaysincome taxcentral boarddirectorate generalepfoemployees provident fundprovident fund commissionerregional provident fund commissionerassistant solicitor generaladditional solicitor generalsolicitor generalunion public service commissionfood corporation of indialic of indialife insurance corporation of indiaemployee state insuranceemployees state insuranceesicnational capital territory of delhi

Public-body signals

municipal corporationmunicipalityzila parishadpanchayatdevelopment authorityhousing boardelectricity boardtransport corporationroad transport corporationuniversityboardcorporationauthoritypublic sectorpsustatutory bodylocal bodyautonomous bodyaided institutiongovernment aidedcooperative societybank

Party extraction and role-source logic

Structured party roles are preferred; title fallback is used only when the structured block is unavailable.

Party role signals

pre blockcaption linesappellantpetitionerrespondenttitle fallbackvvsversusrole_sourcepre = high confidencetitle = medium/low confidence

High Court outcome patterns

High Court outcome detection separates petitioner success, respondent success, partial/remand outcomes and weak fallbacks.

Petitioner/appellant success

petition allowedwrit petition allowedappeal allowedrule made absolutepetitioner succeedsappellant succeedsrelief is grantedentitled to pensionrespondents are directedimpugned order set asideorder is quashed

Respondent/government success or dismissal

petition dismissedwrit petition dismissedappeal dismissedwrit petition rejectedpetition failsnot entitled to pensionno right to claim pensionno relief can be grantedwrit petition not maintainableclaim barred by delay and lachesimpugned order upheldjudgment is affirmedrelief is declined

Partial, remand and disposal

appeal partly allowedpetition partly allowedpartially allowedimpugned judgment modifiedmatter is remandedmatter is remittedcase is sent backauthority shall reconsiderdisposed of with directionsdisposed of with libertypetition is closednothing survives for adjudicationpetition has become infructuous

Argument-context filter

arguedcontendedsubmittedurgedprayedlearned counselaccording to the appellantaccording to the respondenton behalf of the appellanton behalf of the respondent

Fact, issue and topic signals

Fallback extractors look for factual background, legal issues and broad topic buckets after cleaning the text.

Fact hints

employeeserviceretiredretirementpensionappointmentdepartmentbenefitclaimdisputegratuityprovident fundretiralsuperannuation

Issue patterns

the question is whetherthe short issue isthe point for determination isthe controversy iswhether

Issue/fact noise

leave grantedjudgmentwe have heardheard learned counselfor conveniencelearned counsel appearingappellant submittedrespondent submittedorder

Other topic buckets

income taxindustrial disputearticle 226article 227constitutionservice rulesdismissalsuspensiongovernment servantcriminal proceedingspublic trust

Confidence model and manual review

Major outputs receive high, medium or low confidence labels. Pension records with serious uncertainty are marked for manual verification.

High confidence

structured facts/issuespre/caption party rolesclear conclusion outcomestrong government term

Medium confidence

fallback paragraph facts/issuestitle fallback partiesjudgment tail outcomeofficial designation government match

Low confidence

missing facts or issuesweak disposal phrasegeneral public-body inferenceroles from title only

Manual verification indicators

parse_errormissing_respondentmissing_appellant_or_petitionerwinner_unknownwinner_unclearfacts_and_issues_missingfacts_missingissues_missingparty_confidence_lowroles_from_title_onlygovt_confidence_low
High Court summary: Raw High Court judgment HTML files are processed year-wise. The system extracts case metadata, party names, facts, issues, conclusions, citation counts and cited document IDs. Pension relevance is identified using the complete pension, retiral-benefit, gratuity, leave-encashment and provident-fund signal library. Parties are classified using state-government, central-government, public-body and fallback government identifiers. Outcome phrases are scanned across the conclusion, judgment tail and full text to infer the likely procedural result. Weak outcomes, missing fields, low-confidence parties and unclear government results are marked for manual verification.