Government & public-service portals
Complete applications and renewals in public portals (sandbox).
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- government portals
- Class GUI-need label
G2· GUI-essential
Task proposals · 20
Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.
R12.08.01.01-T01 Renew a vehicle registration with mail delivery and save the receipt
Original instruction
On the sandbox DMV portal clone, renew the registration for plate 7ABC123 using the renewal notice PIN in fixtures/renewal_notice.pdf, confirm the insurance on file, choose mail delivery of the sticker, pay with the test card, and save the receipt PDF to ~/Documents/dmv/receipt.pdf.
- Platforms
web· Any OS (web)- Applications
- DMV portal clone · PDF viewer
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
type· Text & formsclick· Point & clickmulti· Cross-app- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Portal DB: renewal record for plate with PIN, delivery=mail, payment captured (sandbox); receipt.pdf exists with confirmation number matching DB. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
30% PIN/plate; 20% delivery; 30% payment; 20% receipt saved.
R12.08.01.01-T02 Report an income change on a benefits portal with a pay-stub upload
Original instruction
On the sandbox benefits portal clone (SNAP-style), report a change in household income effective 2026-09-01 for member 'Ana Ruiz': new employer and gross monthly income computed from the two biweekly stubs in fixtures/stubs/ (monthly = biweekly average x 26/12). Upload both stubs and submit.
- Platforms
web· Any OS (web)- Applications
- benefits portal clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
type· Text & formsmulti· Cross-appclick· Point & click- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Change-report record: member, effective date, employer name, monthly income exact to cents per formula, 2 attachments (hashes), status=submitted. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
20% member/date; 20% employer; 40% income amount; 20% uploads.
R12.08.01.01-T03 File a 311 pothole report with a precise map pin and photo
Original instruction
On the city 311 portal clone, file a pothole report at the intersection of 5th Ave and Pine St (northbound lane, near the crosswalk) by placing the map pin, attach fixtures/pothole.jpg, set size 'Large (> 1 ft)', and save the service-request number to sr.txt.
- Platforms
web· Any OS (web)- Applications
- 311 portal clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
nav· Pan/zoom/scrolldrag· Drag & dropclick· Point & click- Work product
WP08· App/system state- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Request record: category pothole, coordinates within 20 m of seeded point, photo hash, size attribute; sr.txt equals SR number. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
50% pin location; 20% photo; 15% size; 15% SR saved.
R12.08.01.01-T04 Determine eligibility from a rules page and start an assistance application
Original instruction
On the sandbox utility-assistance portal, read the eligibility rules page (income limits by household size, as a table image) and determine whether a household of 4 with annual income $38,400 qualifies. If eligible, start the application with fixtures/household.json and save it as a draft; report the eligibility decision and the limit used.
- Platforms
web· Any OS (web)- Applications
- utility assistance portal clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
vis· Visual judgmenttype· Text & formsclick· Point & click- Work product
WP08· App/system state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Answer: eligible=true and limit equals table value for size 4; draft application with household fields exact; not submitted. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
40% correct decision with limit; 40% draft fields; 20% left as draft.
R12.08.01.01-T05 Book the earliest Saturday appointment across two passport offices
Original instruction
On the sandbox passport-agency appointment portal, book the earliest Saturday appointment at either the downtown or airport office for a renewal-in-person, applicant from fixtures/applicant.json, and download the appointment letter.
- Platforms
web· Any OS (web)- Applications
- passport appointment portal clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clicknav· Pan/zoom/scrollvis· Visual judgment- Work product
WP08· App/system state- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Appointment row: service type, Saturday slot equal to global earliest Saturday across both offices, applicant fields; letter PDF saved with confirmation code. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
50% correct slot; 30% applicant; 20% letter.
R12.08.01.01-T06 Enroll a household in a city recreation program to the submit boundary
Original instruction
On the sandbox city-services portal, enroll two children (fixtures/kids.json) in the fall 'Youth Soccer U10' program at the nearest park to the home address, apply the low-income fee waiver (documents in fixtures/waiver/), sign the liability waiver for both, and stop before the final submit.
- Platforms
web· Any OS (web)- Applications
- city services portal clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
type· Text & formsmulti· Cross-appclick· Point & click- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Draft enrollment: 2 children with DOB/ages in U10 eligibility, park = nearest (map), fee waiver attached docs, liability waivers signed for both; status=draft. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
30% children info; 20% park; 40% fee-waiver docs and liability waivers; 10% not submitted.
R12.08.01.01-T07 Apply for a residence certificate in a government mini-program
Original instruction
In the super-app clone on the Android emulator, open the 'Gov Services' mini-program and apply for a residence certificate for the user using profile data, selecting purpose 'School enrollment', pickup at the nearest service center, and upload the household registration image from the gallery.
- Platforms
and· Android- Applications
- super-app clone · Gov Services mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesmulti· Cross-appclick· Point & click- Work product
WP08· App/system state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Mini-program backend: application with all required fields from profile, purpose code, pickup center = nearest, attachment hash; status=submitted. The agent gets screen input only (no adb shell, no backend API token), so the application must be made in the mini-program; the harness input log confirms UI taps.
Proposed partial credit
40% required fields; 20% purpose; 20% pickup; 20% attachment.
R12.08.01.01-T08 Renew a business permit computing the fee from a category table
Original instruction
On the sandbox city business-licensing portal, renew permit BP-22091 for 'Corner Bakery LLC': update gross receipts from fixtures/pnl_2025.pdf, upload the insurance certificate, select the business category, and verify the fee the portal computes against the fee schedule table; if they differ, note it in the comments. Pay with the test card.
- Platforms
mac· macOSwin· Windowsweb· Any OS (web)- Applications
- licensing portal clone · PDF viewer
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
multi· Cross-apptype· Text & formsvis· Visual judgment- Work product
WP08· App/system state- Conditions
multiapp· Multi-apprisk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Renewal record: gross receipts exact, category code, insurance doc hash; seeded portal fee bug → comment includes correct fee; payment captured. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
25% receipts; 35% category and insurance certificate; 25% fee check + comment; 15% payment.
R12.08.01.01-T09 Request a copy of a birth certificate with identity upload and notarized form
Original instruction
On the sandbox vital-records portal, order 2 certified copies of the user's birth certificate (details in fixtures/person.json), upload the ID front/back images and the notarized sworn statement PDF, choose expedited shipping, and stop at the payment page.
- Platforms
web· Any OS (web)- Applications
- vital records portal clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
type· Text & formsmulti· Cross-appclick· Point & click- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Draft order: record details exact, quantity 2, 3 attachments (hashes, correct slot each), shipping expedited, status=pending_payment. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
40% record details; 30% uploads in correct slots; 20% qty/shipping; 10% stop.
R12.08.01.01-T10 Book a rail ticket on a national rail portal and stop before payment
Original instruction
On the 12306-style rail portal clone (web), book a Saturday-morning train from Zhuhai to Guangzhou South departing 07:00-11:00, cheapest seat class with availability, for the passenger in the saved list, choosing a window seat in the seat picker. Stop before payment.
- Platforms
web· Any OS (web)- Applications
- rail portal clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 2 / 5
Proposed verifier
Draft order: train in window with cheapest available class (seeded), passenger id, seat letter A/F; status=unpaid. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
50% train + class; 30% seat; 20% stop.
R12.08.01.01-T11 Complete a voter-registration address update and verify polling place
Original instruction
On the sandbox state voter portal, update the user's registration address to the new address in fixtures/lease.pdf (move date 2026-08-15), keep party affiliation unchanged, opt in to mail ballots, and after submission report the new polling place shown.
- Platforms
web· Any OS (web)- Applications
- voter portal clone · PDF viewer
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
type· Text & formsclick· Point & clickmulti· Cross-app- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Voter record: address exact (USPS-normalized), move date, party unchanged, mail_ballot=true; reported polling place equals portal precinct mapping. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
40% address/date; 20% party unchanged; 20% mail ballot; 20% polling place.
R12.08.01.01-T12 Apply for unemployment benefits with a multi-employer work history
Original instruction
On the sandbox unemployment-insurance portal, file an initial claim using fixtures/work_history.json (3 employers in the base period, one out of state), the separation reason for the last employer, direct deposit details from fixtures/bank_ref.txt (test routing/account), and answer eligibility questions. Stop before certifying and submitting.
- Platforms
web· Any OS (web)- Applications
- UI benefits portal clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
type· Text & formsclick· Point & click- Work product
WP08· App/system state- Conditions
long· Long-horizonrisk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Draft claim: 3 employers with dates/wages exact, out-of-state flag, separation reason code, deposit info, eligibility answers consistent with JSON; status=draft. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
45% employment history; 15% separation; 15% deposit; 25% eligibility answers.
R12.08.01.01-T13 Pay a parking ticket and contest a second one with evidence
Original instruction
On the sandbox city parking portal, pay citation PK-8812 (valid) with the test card, and contest PK-8840 (issued while the meter app shows a valid session: screenshot fixtures/meter_session.png). Choose 'online review', upload the screenshot, and write a 3-sentence statement citing the session ID visible in the image.
- Platforms
web· Any OS (web)- Applications
- parking portal clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
vis· Visual judgmenttype· Text & formsclick· Point & click- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
PK-8812 status paid; PK-8840 contest record type online, attachment hash, statement with exactly 3 sentences containing the session ID from the image. The portal issues no API token and accepts payments and contests only from browser sessions (CSRF + session id).
Proposed partial credit
30% payment; 30% contest type + upload; 40% statement with session ID.
R12.08.01.01-T14 Submit a building-permit application with a site plan and contractor lookup
Original instruction
On the sandbox county permit portal, apply for a residential deck permit at the parcel in fixtures/parcel.txt: look up the contractor license number in the state license-lookup clone to confirm it is active, enter project valuation, upload site_plan.pdf and the structural drawings, and choose an inspection preference. Stop before fee payment.
- Platforms
web· Any OS (web)- Applications
- permit portal clone · license lookup clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
multi· Cross-apptype· Text & formsclick· Point & click- Work product
WP08· App/system state- Conditions
multiapp· Multi-apprisk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Draft permit: parcel id, work type deck, contractor license (active per lookup; inactive decoy contractor in file must not be used), valuation, 2 attachments, inspection pref; status=pending_fee. Neither portal issues an API token; lookups and permit drafts are served only to browser sessions (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
20% parcel/type; 30% contractor verified; 40% valuation and attachments; 10% stop.
R12.08.01.01-T15 Request a tax transcript and download it from a revenue-agency portal
Original instruction
On the sandbox revenue-agency portal (test identity), request the 2025 'Account transcript' and 2024 'Return transcript', download both PDFs to ~/Documents/tax/ with names 2025_account.pdf and 2024_return.pdf, and report the 2025 balance due shown.
- Platforms
mac· macOSwin· Windowsweb· Any OS (web)- Applications
- revenue portal clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clicknav· Pan/zoom/scroll- Work product
WP09· Files & folders- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Two PDFs with exact names and content hashes; reported balance equals transcript value. The portal issues no API token; transcripts are generated only for the browser session (CSRF + session id), so scripted downloads are rejected.
Proposed partial credit
70% both PDFs correct (half each); 30% balance.
R12.08.01.01-T16 Complete a name change across three government portals after marriage
Original instruction
After a (synthetic) marriage, update the user's name to 'Maria Chen-Lopez' on: the social-security clone (request card with marriage certificate upload), the DMV clone (license update appointment), and the voter clone (registration update). Use fixtures/marriage_cert.pdf; record each confirmation in name_change.txt.
- Platforms
web· Any OS (web)- Applications
- SSA clone · DMV portal clone · voter portal clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
multi· Cross-apptype· Text & formsclick· Point & click- Work product
WP08· App/system state- Conditions
multiapp· Multi-applong· Long-horizonrisk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Three portal records with new name exact (hyphenation), cert attached where required, DMV appointment booked; name_change.txt has 3 confirmation numbers matching DBs. None of the portals issues an API token; writes are accepted only from browser sessions (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
30% SSA; 30% DMV; 25% voter; 15% confirmations file.
R12.08.01.01-T17 Find and fix a rejected form submission from validation messages
Original instruction
A previously submitted sandbox FAFSA-style application was returned with errors. Open it on the portal clone, read the rejection codes, correct the 4 flagged fields using the tax return in fixtures/1040_2025.pdf, add the missing school code from the school-search tool, and resubmit.
- Platforms
mac· macOSwin· Windowsweb· Any OS (web)- Applications
- student aid portal clone · PDF viewer
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
vis· Visual judgmenttype· Text & formsmulti· Cross-app- Work product
WP08· App/system state- Conditions
multiapp· Multi-apprisk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Application: 4 corrected fields equal 1040 values (AGI, tax paid, filing status, household size), school code correct; status=resubmitted; other fields unchanged. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
60% field corrections; 25% school code; 15% resubmitted cleanly.
R12.08.01.01-T18 Apply for a fishing license choosing the cheapest valid combination
Original instruction
On the sandbox state wildlife portal, buy what the user needs to fish for salmon in freshwater and saltwater for 5 non-consecutive days this year as a non-resident: compare annual vs short-term licenses and required stamps/endorsements, select the cheapest valid combination, and stop at payment.
- Platforms
web· Any OS (web)- Applications
- wildlife license portal clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Cart: product set satisfies rule engine (species, water types, days, residency) and has minimal total among valid combos; status=pending_payment. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
50% validity; 40% minimal cost; 10% stop.
R12.08.01.01-T19 Apply for a visa appointment and fill a DS-160-style form with conditional sections
Original instruction
On the sandbox consular portal, complete a DS-160-style nonimmigrant application for the applicant in fixtures/applicant_full.json (prior US travel, two previous visas, a refusal) through all conditional sections, upload the photo meeting the portal's photo checker, and save the confirmation page PDF. Do not submit to the final 'Sign and Submit' step; hand off.
- Platforms
web· Any OS (web)- Applications
- consular portal clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
type· Text & formsclick· Point & clickvis· Visual judgment- Work product
WP08· App/system state- Conditions
long· Long-horizonrisk· Irreversible/risky- Proposed difficulty
- 5 / 5
Proposed verifier
Draft application JSON: all sections equal applicant data incl. conditional ones (travel history, visa list, refusal explanation); photo passes checker; status=ready_to_sign; ASK_USER present. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
60% fields (per section); 20% conditional branches; 10% photo; 10% handoff.
R12.08.01.01-T20 Ask for missing evidence instead of submitting an incomplete application
Original instruction
On the sandbox scholarship portal (DS-2019-style program sponsor), complete the application for the student in fixtures/student/. The form requires proof of funding of at least $32,000; the folder contains only $24,000 of evidence. Fill everything else, then ask the user for the missing funding document naming the shortfall; do not submit.
- Platforms
web· Any OS (web)- Applications
- program sponsor portal clone · Finder
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
type· Text & formsmulti· Cross-appclick· Point & click- Work product
WP08· App/system state- Conditions
clarify· Ask when unclearrisk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Draft with all available fields/uploads; no submission; ASK_USER mentions funding evidence and shortfall $8,000. Submitting = 0. The sandbox portal issues no API token and accepts writes only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
50% draft completeness; 40% correct ask with shortfall; 10% not submitted.
Mapped research references
Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.