# Consumer-portal access study — frozen initial protocol
Frozen before portal pilot: 2026-10-16. Publisher: FreePolicyTools.

## Question
Can a privacy-conscious visitor reach public reporting guidance and the first reporting screen of selected official US consumer-protection services through specific VPN and proxy routes?

## Sample
Purposive sample: FTC ReportFraud, FTC IdentityTheft, FBI IC3, CFPB, FCC, and consumer complaint entry pages for the ten most populous US states in the 2020 Census (CA, TX, FL, NY, PA, IL, OH, GA, NC, MI). State selection fixed before outcomes, not a representative sample. Official search discovery supplies exact URLs and source links. Separate FTC services are separate tasks, not independent agencies.

## Routes and observations
Normal connection (contextual, expected Kenya); one already purchased USA IPv4 Proxy-Seller endpoint; two available US Proton exits, distinct exit hashes required. No new subscriptions or purchases. Every route measured in two passes with reversed site order. Same Chromium build, fresh browser context per site per pass, en-US, UTC, 1360×900, default JavaScript, no extensions or interception, normal certificate validation. No stealth, challenge solving, retries designed to evade controls, logins, personal data, or report submissions. Max two concurrent pages; 35-second navigation budget plus fixed 3-second rendering grace. Record start/end times, navigation errors, main-document statuses, visible text, screenshot and request hostnames. HTTP status alone is not access success. Public artifacts omit credentials, raw IPs, query strings and cookies.

## Task boundary
Guidance/entry page then one identified official start/report link or button, stopping before any required personal information, declaration/terms acceptance, login, CAPTCHA completion or report submission. Terms/login/CAPTCHA boundaries are recorded separately and are not automatically network blocks. Exact safe actions must be verified in the pilot and frozen before main collection. If a portal requires additional navigation, amend protocol explicitly before the main run; do not improvise data-entry flows.

## Outcome adjudication
Manual evidence review: entry accessible; first reporting screen accessible; required terms/login boundary; challenge; explicit access denied/restriction; transport failure; inconclusive. Distinguish embedded CAPTCHA on otherwise readable form from a challenge that blocks reaching it. Repeat differences within the session and validate major apparent restrictions with a visible/full browser when possible. No claims of deliberate VPN blocking based only on failures, no causal attribution without matched direct US residential control. Record all failures and exclusions. Pilot is separate from main observations.

## Analysis
Report task-level route counts, repeat consistency and per-portal paired patterns. Repeated visits are not independent people or sites. No population estimates, significance tests or provider ranking. Compare entry vs reporting-stage outcomes. Intermediary hostnames are descriptive; their presence does not establish tracking, transmission of report contents or improper processing. Findings apply to tested exits, browser and date only.

## Disclosure and delivery
FreePolicyTools is a Proxy-Seller affiliate and received $10 in research credit. Paid proxy type is IPv4; do not describe it as a residential connection without evidence. Publish protocol, exact sample, route metadata with hashed exits, observation table, classifications, sanitized supporting screenshots, analysis code and article draft. No scheduling, publication or outreach in this task.

## Pilot note
Direct-entry pilot saved 14/15 pages. Ohio was not saved: the page navigated while its DOM links were being read, causing the pilot script to fail. This is an instrumentation failure, not a site-blocking result. The main collector catches snapshot errors and keeps incomplete evidence distinct. Separate proxy pilot follows.

## Pilot-driven amendment before main collection
- Browser-proxy pilot reached a US exit (IPinfo: Florida, Datacamp ASN) but all 14 .gov entry hosts failed the HTTP CONNECT tunnel. A two-repeat raw CONNECT diagnostic, plus publisher/IPinfo non-government controls and the .com Florida portal, records the proxy response status before any destination TLS. No restriction bypass attempted. Preserve these as proxy-side restrictions, not origin denials.
- Capture all reachable child-frame text and navigation status metadata. Permit up to three pilot-verified navigation actions when a guidance page precedes the reporting screen; no incident category/data entry or legal acceptance.
- Screenshots are saved privately for every reached stage, and reviewed examples will be copied to a public evidence gallery.
- User explicitly authorized dedicated Chromium/Playwright, and requested unused tabs be closed. Max two pages/contexts concurrently; account tab closed.

## Instrumentation correction before main runs
Pilot exact-role and absolute-href selectors missed ordinary controls on some pages. These failed actions are instrument errors, not access failures. Main navigation resolves the raw href from a matching DOM-resolved official URL, and FTC button text uses the observed button tag. Main visits have a 90-second whole-journey watchdog and at most 12 child-frame reads, preventing a dynamically navigating page (Ohio pilot) from leaving an indefinite browser task. Pilot runs remain excluded from main totals. Route identification uses IPinfo with Cloudflare trace fallback when IPinfo is rate-limited; raw responses stay private.

## Main freeze
Main collection: 15 services × 4 routes × 2 complete passes = 120 intended journeys. Snapshot grace increased to 8 seconds before every screenshot/text capture, allowing normal rendering through distant exits. Whole-visit watchdog 120 seconds. Popup targets are awaited and captured rather than accidentally re-capturing the opener. No custom browser user-agent or challenge evasion. Boston chosen after the old saved California server failed to connect; connection failure is setup metadata, not a portal result. Second US city will be New York if available. Proton's existing malware-only NetShield setting remains in place on both VPN routes and is disclosed as a route confound.

Task endpoint is the first reporting questionnaire/form, a mandatory CAPTCHA/terms gate, or a pre-reporting notice. These are separate categories; reaching a notice does not establish form access or successful submission. CFPB stops after the two verified notice links. Illinois has no pilot-identified online start control on the selected entry page; guidance-only/online form unverified is retained rather than excluding the service. NC's embedded form may be blank; do not equate its readable wrapper with form access. Main frozen sample/code hashes are in main-freeze.json.

## Publication processing and supplementary checks
Public exitHash values are keyed HMAC-SHA256 identifiers over the privately captured SHA256 IP digests. The unpublished key stays in the ignored private directory. Repeated observations retain matching identifiers, but the public bundle does not expose ordinary unsalted IP hashes. Original collection code/hashes remain frozen for audit.

After the main 120 visits, focused headed Chromium checks investigated NC/Illinois rendering, Ohio route differences, Texas DNS failures, and paid-proxy tunnel behavior. These are exploratory validation records, excluded from main counts. NC checks scroll the observed SAS iframe into view and wait 20 additional seconds. Headed mode, scroll and longer timing change together. Texas uses the original Boston exit with NetShield malware-only, off, then malware-only again, with a fresh browser process per visit; settings and exit identity are recorded. No report data, declarations, consent boxes or CAPTCHAs are completed. Restore disconnected VPN and malware-only NetShield afterward.