When a requested proxy country, observed egress, geolocation database, account locale, and website-rendered market disagree, do not let a majority vote decide the trial. Predeclare which evidence has authority for the task, then accept only that bounded sample.
This guide is for an authorized team checking one regional website journey or permitted data task. It does not locate a person, certify a city, diagnose a provider or website fault, or authorize collection. Prefer an official, licensed, manual, or no-proxy route whenever it answers the decision.
Name the decision and its authority first
“Does the approved product page render Canada?” and “Which country does a named database assign to an address?” are different questions. For the first, the recorded target-rendered market has authority. For the second, a named database field, version, and timestamp can have authority. Browser language and account locale are context, not location verdicts.
| Bounded question | Predeclared authority | Supporting context | Do not infer |
|---|---|---|---|
| Does the approved journey render its intended market? | Recorded target-rendered market at the named step. | Observed egress, account state, language, time and release. | Physical user location or provider fault. |
| Does a named database classify observed egress as expected? | Named database export or response, version and timestamp. | Observed egress and source schema. | What a target must render. |
| Does a browser feature have location permission? | Recorded permission/result for that feature. | Secure context, policy and device state. | IP-geolocation agreement. |
Before any route is used, record owner, permitted domain or source, country, one journey step, time window, acceptance authority, and stop condition. Without a named authority, pause the decision. For a page-market requirement, define how market is identified: an owner-confirmed country selector or documented rendered market field, not language or currency alone. If purchase also requires a named database to classify the observed address as CA, make that a separate mandatory rule. Every mandatory rule must pass; do not select the easiest authority after seeing results.
Keep the signals separate
An observed egress is the address reference seen by a particular approved observer for a correlated request. It establishes the address seen by that observer. A named classifier can answer a contract about database classification; neither observation proves the physical location of a person or endpoint. Keep an opaque job/trace reference in a shared report, with the actual address-to-request mapping in the authorized private record; the reference itself is not proof that the mapping is correct. An IP database is a dated vendor classification under its own schema. Account locale and navigator.language are preference or state; MDN describes the latter as the user’s preferred language, usually the browser UI language. A target-rendered market is the application output in one recorded context. None substitutes for another.
MaxMind explains that IP-geolocation accuracy varies by country, distance, IP type and ISP practice, and it cannot guarantee agreement with competitors. Retain a source’s field name, confidence, and accuracy radius where provided; do not turn a city label into physical precision. Browser geolocation is another signal: it needs a secure context, may be blocked by Permissions-Policy, and needs explicit user permission. Missing or denied browser location is missing feature evidence, not proof of a wrong exit.
Pass a comparability gate before interpreting a conflict
The following gate applies to evidence required by the declared rule. Missing or inconsistent auxiliary evidence stays in an investigation note; it does not silently become a new rejection rule. Conversely, a favorable auxiliary label cannot replace missing authoritative evidence.
Compare only records from the same authorized job or journey, bounded time window, and known schema. Keep job/trace correlation, UTC time, target step, observed-egress method, database name/version, and whether a value is country, registered country, city/radius, locale, or rendered market. A differently dated database or differently defined field is not a tie-breaker.
| Field | Why it matters | Missing or incomparable outcome |
|---|---|---|
| Job/journey and target step | Prevents a checkout result being compared with a landing-page result. | HOLD; collect same-step evidence. |
| UTC time and window | Routes, releases and account state may change. | HOLD; do not merge stale evidence. |
| Source schema/version | Country, registered country, and city-radius are not interchangeable. | HOLD; preserve raw field names. |
| Task authority | Stops majority voting from replacing acceptance. | HOLD; ask the named owner. |
A worked synthetic conflict ledger
All values below are invented teaching data. This is a market-label check, not a complete price-display or provider trial. The base contract is: job-7, product step, expected market CA, source sample-store, field schema market-country-v1, release-A, collected between 02:00 and 02:05 UTC on November 18. The source must be recorded before the test. Each independent case uses observation 02:02 UTC and matching trace-a references unless changed below. Trace references represent private evidence links, not countries.
A second, separate contract asks whether sample-db classifies that correlated address as CA, using country-code-v1 and db-20261118 in the same collection window. This exact version is a hypothetical agreed snapshot, not a universal freshness requirement. Keep dataset version separate from collection time. Requested route, account locale and a secondary lookup remain contextual fields.
| Case / authority | Requested route / egress evidence | Authoritative observation | Other context or changed condition | Decision |
|---|---|---|---|---|
| G-01 / page market | CA / trace-a linked to job-7 | CA; sample-store, release-A | Secondary lookup CA; account en-GB. | ACCEPT_SAMPLE_ONLY |
| G-02 / page market | CA / trace-a linked to job-7 | CA; same valid source and window | Secondary lookup US; preserve its source/version for investigation. | ACCEPT_SAMPLE_ONLY_WITH_SECONDARY_REVIEW |
| G-03 / page market | US / trace-a linked to job-7 | CA; same valid source and window | Secondary lookup US; account en-US; expected page market remains CA. | ACCEPT_SAMPLE_ONLY for the page rule; no CA-route certification. |
| G-04 / page market | CA / trace-a linked to job-7 | Missing | Secondary lookup CA cannot replace the page observation. | HOLD_MISSING_EVIDENCE |
| G-05 / database classification | CA / trace-a linked to job-7 | US; sample-db, db-20261118 | Expected database classification is CA. | HOLD_REQUIREMENT_MISMATCH |
| G-06 / database classification | CA / trace-a linked to job-7 | CA; sample-db, db-20261001 | Snapshot differs from the predeclared db-20261118 contract. | HOLD_NOT_COMPARABLE |
| G-07 / page market | CA / trace-a | CA, but belongs to different-job | A matching label does not repair job correlation. | HOLD_NOT_COMPARABLE |
| G-08 / page market | CA / trace-a linked to job-7 | CA, collected at 02:06 UTC | Outside this demonstration's 02:00–02:05 window. | HOLD_NOT_COMPARABLE |
G-02 passes only the page-market rule and retains the unresolved secondary disagreement. G-05 does not pass its database rule, even if the page looks correct. If both rules are mandatory for a proposed purchase, a pass on one cannot erase a hold on the other. G-06 says the supplied evidence does not match this contract; it does not prove the older database is inaccurate. None of the eight cases identifies a fault owner, measures IPHTML, or establishes city precision, future behavior or overall plan suitability.
Run a small acceptance trial, then stop
- Use an approved direct, official, licensed, manual, or no-proxy route when sufficient.
- When a documented authorized gap remains, record authority and a minimal same-job control.
- Capture one harmless permitted journey: trace ID, UTC time, target step, observed-egress method, account state, language, source/version, and rendered result. Redact sessions and credentials.
- Accept only the stated sample if the predeclared authority meets its rule. Hold missing, incomparable, or contradictory authority evidence for the named owner. Do not expand one sample into a city, plan, or physical-location claim.
For a controlled journey, see the regional website QA test matrix. For bounded test scope, see the representative proxy trial guide. For target-side address gates, see the egress allowlist check. For permitted regional-report evidence, see regional search evidence delivery.
Replay the synthetic ledger offline
Download the Python 3.9 offline worksheet and run python3 proxy-geolocation-evidence-conflicts.py. It reproduces all eight independent cases and checks extra counterexamples for missing sources, unrelated traces, malformed labels and UTC timestamps. The worksheet verifies the structure of supplied synthetic evidence; a human must still confirm authorization, source meaning, actual trace mapping and the observation itself. It sends no requests and reads no credentials.
G-01: ACCEPT_SAMPLE_ONLY
G-02: ACCEPT_SAMPLE_ONLY_WITH_SECONDARY_REVIEW
G-03: ACCEPT_SAMPLE_ONLY
G-04: HOLD_MISSING_EVIDENCE
G-05: HOLD_REQUIREMENT_MISMATCH
G-06: HOLD_NOT_COMPARABLE
G-07: HOLD_NOT_COMPARABLE
G-08: HOLD_NOT_COMPARABLE
A HOLD keeps the trial or purchase decision open: obtain the named source, recover comparable evidence, or ask the owner to revise the requirement before a new test. Preserve the original result and rule version. Do not relabel the current record as accepted, rotate through more routes to search for a favorable label, or bypass a target restriction.
Only when the simpler approved route cannot answer a documented task: review IPHTML residential proxy information for conditional suitability, then contact IPHTML with a redacted scope: authorized domain/source, country-level task, one journey step, acceptance authority, evidence fields, owner, frequency, and stop condition. Confirm current terms and suitability before testing; do not send credentials or use a route to bypass controls.
Sources checked November 18, 2026. MaxMind’s geolocation accuracy guidance covers variability, limits, and accuracy radius; its IP geolocation data guide distinguishes returned fields and confidence. MDN documents browser language preference and browser-geolocation permissions. These sources do not establish provider performance, city precision, demand, paid conversion, or legal permission.