How to Handle Conflicting, Duplicate, or Stale Signals

Who this framework is for

This framework is for teams receiving events from web analytics, email providers, forms, enrichment, public research, and human review. It helps when automated workflows need a reliable current record but governance requires that original evidence and corrections remain traceable.

Contents

1 Separate raw observation from derived state 2 Name the exception precisely 3 Apply the SellersGPT CDS Resolution Ledger 4 Resolve current state with source and scope rules 5 Set evidence-specific freshness and review controls 6 Illustrative example 7 Common mistakes 8 What this does not mean 9 Common questions 10 Sources and methodology

1. Separate raw observation from derived state

Use four layers: 1 Raw event: what the source reported, with source ID, timestamps, payload reference, and integrity information. 2 Normalized event: a mapped internal class such as valid visit, direct reply, hard bounce, form, unsubscribe, or correction. 3 Evidence assertion: a claim supported by the event, such as “mailbox invalid,” “person requested a quote,” or “company relationship Unknown.” 4 Current operational state: lead, qualified lead, research, owner, task, suppression, existing customer, stop, or closed route under a policy version. HTTP and email standards explain technical messages and fields, not the entire commercial conclusion. IETF - RFC 9110, HTTP Semantics IETF - RFC 5322, Internet Message Format Retain enough raw and normalized evidence to reproduce why a derived state changed. A correction should append a new assertion and supersession link. Do not edit the old form, reply, or visit so it appears that the source always said the new value.

2. Name the exception precisely

Exception Definition Example Required response Exact duplicate Same source event delivered more than once Provider retries same event ID Link and count once operationally Semantic duplicate Different records represent one underlying action Double-click form submissions with same request Preserve both; consolidate action Corroboration Independent event supports the same fact Reply confirms application stated in form Increase confidence for that fact Conflict Credible assertions cannot both be current within same scope Form says end user; reply says distributor Mark field Conflicted; resolve Supersession Later or more authoritative evidence replaces a prior value Person corrects phone number Make new value current; retain old Stale Evidence is too old for the current decision use Employment last checked 18 months ago Refresh or set Unknown Out-of-scope difference Records concern different site, application, or episode Plant A replacement vs Plant B new build Keep separate, not conflicted Stop override Current restriction changes allowed action Later unsubscribe after earlier positive reply Apply stop immediately Duplicates and corroboration are not the same. Two webhook deliveries with one event ID provide one occurrence. A later direct reply independently confirming a form request can corroborate the application, but it still may belong to the same lead. Conflict also requires shared scope. Two different subsidiaries can have different industries or purchasing routes without contradiction. Define account/site, person, application, relationship, decision episode, and evidence date before marking conflict.

3. Apply the SellersGPT CDS Resolution Ledger

The CDS Resolution Ledger is an original SellersGPT editorial tool. CDS stands for Conflict, Duplicate, and Stale. For every exception, record: Field Purpose Exception ID and class Stable reference and precise category Affected object and field Event, person, account, episode, lead, or treatment field Shared decision scope Account/site, application, relationship, stage, and time Source A and Source B Exact evidence, source type, and dates Source integrity Valid, suspect, unknown, or conflicted Current-source rule Recency, directness, authority, confirmation, or policy priority Resolution Deduplicate, preserve both, supersede, split scope, refresh, stop, or review Current value Known, Inferred, Unknown, Contradicted, or stopped state Actor and time Rule/reviewer and decision timestamp Downstream effects Lead, qualification, owner, task, suppression, and reporting corrections Use hard precedence only where justified. Current unsubscribe outranks older positive engagement for communication treatment. An authenticated person correction can supersede an inferred enrichment value for their direct details. A current customer master can block prospect routing. For most commercial facts, no universal source hierarchy exists; evaluate scope, directness, date, and corroboration.

4. Resolve current state with source and scope rules

Use this sequence: 1 Contain: pause affected high-impact automation if a conflict could cause wrong contact, duplicate work, qualification error, or stop violation. 2 Validate sources: confirm event integrity and exact timestamps. 3 Define scope: determine whether records concern the same person, site, company, application, relationship, and episode. 4 Classify: duplicate, corroborating, conflicting, superseding, stale, or independent. 5 Apply precedence: use documented stop, customer, direct-correction, and policy rules. 6 Resolve or preserve Unknown: assign a current value only when evidence supports it. 7 Propagate: update tasks, lead state, owner, suppression, and reporting through traceable operations. 8 Review: set a date or condition for unresolved and volatile evidence. The W3C PROV overview provides a model for relating entities, activities, agents, derivations, and revisions. W3C - PROV Overview The CDS Ledger uses that general provenance discipline but is not a W3C standard. NIST's log-management guidance discusses processes for generating, transmitting, storing, analyzing, and disposing of logs. NIST SP 800-92 - Guide to Computer Security Log Management A GTM ledger is different from a security log, but event integrity, access, review, and retention remain useful bounded practices.

5. Set evidence-specific freshness and review controls

Freshness is not one global number. Set evidence validity according to volatility and decision consequence. Evidence Volatility Review trigger example Email delivery event Historical event does not expire; contactability conclusion can New delivery result or address change Employment and role High New reply, departure notice, scheduled refresh Company legal identity Lower but not permanent Filing, acquisition, dissolution, or annual review Application and project timing High Milestone passes, explicit correction, inactivity window ICP version Governed New version activation or draining policy Unsubscribe and company stop Persistent treatment state Valid revocation or authorized policy process only Website behavior Short-lived for action Episode window closes or new explicit signal arrives Stale does not mean false. It means unsuitable for the current decision without refresh. Keep the historical fact with its date while current state becomes Unknown or Review Required. NIST's Information Quality Standards emphasize utility, objectivity, integrity, and correction. NIST - Information Quality Standards SellersGPT uses these as bounded editorial criteria: evidence useful for one historical report may be too stale for person-level outreach today.

6. Illustrative example

Illustrative example. The company, person, events, dates, and calculations below are fictional. They are not a SellersGPT customer result or benchmark. A fictional supplier has these records for GreenVale Foods: 1 August 1: enrichment says Morgan is Maintenance Manager. 2 August 20: a unique-link visit is attributed to Morgan's campaign context. 3 August 20: the provider delivers the same visit webhook three times with one event ID. 4 August 21: Morgan replies, “I left GreenVale in May. Remove me.” 5 August 22: a natural-traffic form from Alex at GreenVale requests a quote and identifies Alex as a contractor managing the retrofit. Resolution:

• The three webhooks are exact duplicates: preserve three receipts, normalize to one visit.

• Morgan's departure contradicts the current employment assertion and supersedes it with direct dated evidence.

• Morgan's unsubscribe creates a person stop and blocks any later round.

• Alex's form is not a duplicate of Morgan's episode evidence. It is a new person and possibly the same project.

• The company relationship is now Conflicted: the active ICP expects end user, while Alex states contractor. Research must determine whether GreenVale remains the end user and Alex is an external buying-group member. There are seven raw source records if the three webhook deliveries are counted separately. They resolve to five distinct evidence occurrences: enrichment, one visit, reply, form, and the processing correction. The lead does not receive three visit counts. Qualification pauses on the relationship hard gate while the quote request routes to an owner.

7. Common mistakes

Deleting duplicate rows Link and exclude them operationally while retaining enough trace to explain provider retries and reconciliation. Selecting the most favorable source Positive activity does not outrank a current stop, correction, or failed criterion. Treating every difference as conflict Different sites, applications, people, or decision episodes may both be true. Using newest as automatically correct Recency matters, but source integrity, directness, scope, and authority also matter. Expiring suppression like behavioral data Stop states require their own governing rules and must not disappear because a generic activity window closed. Fixing the current view without downstream correction Resolve tasks, qualification, owner, notifications, reporting, and any other derived consequence through traceable changes.

8. What this does not mean

• It does not require retaining every payload forever.

• It does not make the newest source universally authoritative.

• It does not average contradictions into a middle score.

• It does not permit stale engagement to restart a stopped sequence.

• It does not define commercial dispute, billing, or refund mechanics.

9. Common questions

How should provider webhook retries be counted? Use the provider event ID and event semantics to link retries. Preserve receipt history, but count one underlying event in operational metrics. When does old data become stale? When its age, volatility, source, and decision use no longer support the current action. Define review triggers by field rather than one blanket number. What if a reply contradicts a form from the same person? Preserve both. Determine whether the reply corrects, narrows, or concerns a different scope. Apply direct supersession only to the affected field with evidence. Can a stale qualified lead remain in history? Yes. Historical transition and current eligibility are separate. Preserve what was valid then, mark what requires refresh now, and do not silently rewrite history.

10. Sources and methodology

Sources 1 W3C - PROV Overview. Used for provenance, derivation, revision, and responsible-agent concepts. Accessed August 25, 2026. 2 NIST SP 800-92 - Guide to Computer Security Log Management. Used as bounded guidance for event integrity, access, analysis, retention, and disposal. Accessed August 25, 2026. 3 NIST - Information Quality Standards. Used for bounded information utility, objectivity, integrity, and correction concepts. Accessed August 25, 2026. 4 IETF - RFC 9110, HTTP Semantics. Used to bound the technical meaning of web requests and retries. Accessed August 25, 2026. 5 IETF - RFC 5322, Internet Message Format. Used for email message and thread evidence when resolving corrections and conflicts. Accessed August 25, 2026. SellersGPT analysis and framework The exception taxonomy, CDS Resolution Ledger, resolution sequence, freshness examples, and fictional scenario are original SellersGPT editorial tools. They are not universal source-precedence rules or purchase-probability models. Update policy Review this framework when event schemas, identity rules, evidence retention, ICP versions, suppression handling, customer routing, or correction processes change. Test duplicate, conflict, split-scope, and stale scenarios before release.