How to Turn an ICP into Data Requirements, Evidence States, and Research Rules

Who this template is for

This template is for strategy, research, revenue operations, and data teams converting an approved ICP into repeatable account work. It is useful when the current specification contains phrases such as "mid-sized manufacturer," "high growth," "needs automation," or "likely buyer" without defining how those claims can be observed and what happens when evidence is missing.

Contents

1 Convert strategy language into observable requirements 2 Define the field contract 3 Establish evidence states and source roles 4 Write ordered research rules 5 Set freshness, conflicts, and provenance 6 Separate automation from decision authority 7 Use the ICP research template 8 Illustrative example 9 Common mistakes 10 What this does not mean 11 Common questions 12 Sources and methodology

requirements

Begin with one ICP statement at a time: Target regional manufacturers with complex washdown processes, multiple production lines, local maintenance capability, and an active replacement need. Decompose it:

• What does "regional" mean for delivery and service?

• Does "manufacturer" describe the facility, legal entity, or parent?

• What observable conditions establish "complex washdown process"?

• Does line count need a threshold, range, or qualitative state?

• What evidence confirms local maintenance capability?

• What qualifies as replacement need, and over what period? Delete any field that has no decision consequence. Add any gate that changes eligibility even if it is difficult to collect. The result should explain account, application, relationship type, and commercial conditions, not merely replicate a vendor database.

2. Define the field contract

Every required field needs: Contract element Required question Field ID and name What stable identifier and plain-English label are used? Business purpose Which ICP decision changes because of this field? Entity level Parent, legal entity, facility, buying location, product, or relationship? Definition What exactly is included and excluded? Contract element Required question Data type and unit Boolean, category, count, currency, date, range, or text? Pass/fail rule What evidence satisfies or disproves the criterion? Accepted evidence Which official or first-party records are sufficient? Discovery sources Which sources may generate a hypothesis but not confirm it? Freshness How recent must the observation be? Conflict rule What happens when credible sources disagree? Missing rule What state and research action follow? Provenance Which URL, document, page, quote or field, period, and retrieval date? Owner Who maintains the definition and resolves exceptions? GAO's data-reliability guide emphasizes accuracy, completeness, and applicability for the intended purpose. GAO, Assessing Data Reliability Apply that purpose test at field level. An official aggregate can be accurate but inapplicable to a named facility; a current first-party plant page may be applicable but incomplete.

3. Establish evidence states and source roles

Use controlled states:

• confirmed_fit: accepted credible evidence satisfies the written rule for the defined unit and date;

• confirmed_non_fit: accepted credible evidence fails the written rule for the defined unit and date;

• contradicted: two or more credible sources materially conflict and scope, unit, or date review has not resolved the difference;

• unknown: the required claim has not been established;

• not_applicable: optional; the approved ICP logic makes the field out of scope for this record. not_applicable requires a reason and should be excluded before eligible compensating weights are renormalized. It is not a convenient replacement for unknown. Hard gates remain outside weighted-score denominators and use the same states to derive pass, fail, or unresolved outcomes. Assign source roles: Role Use Decision authority Authoritative definition Defines code, unit, requirement, or official scope Supports the definition, not account fit automatically Account confirmation First-party or official evidence about the named account/unit Can set a state within its scope and date Discovery Finds possible accounts or claims Creates a research task only Conflict evidence Credible evidence that materially conflicts with another credible claim Sets contradicted and triggers resolution or hold The Census developer portal provides access and guidance for publicly released Census datasets and APIs. U.S. Census Bureau Developers An API response should retain dataset, vintage, variables, geography, query, extraction date, and relevant flags. It must not be stripped into an unexplained value.

4. Write ordered research rules

Rules should be readable before they are automated: 1 If legal entity and facility are unresolved, then stop fit scoring and create an identity task. 2 If a hard gate is confirmed_non_fit, then exclude from that motion and retain the reason. 3 If a hard gate is unknown or contradicted, then hold and assign its evidence action. 4 If application fit is confirmed_fit, then evaluate relationship and serviceability. 5 If relationship type differs from the primary ICP but fits an approved secondary ICP, then route with a new assignment; do not overwrite the observation. 6 If all applicable gates are confirmed_fit and minimum evidence coverage is met, then calculate the assessable fit result using compensating criteria only. 7 If timing is unknown, then preserve fit and route to evergreen validation or monitoring rather than inventing urgency. Write negative ICP rules with the same precision. "Too small" is not enough; state the unit, measure, threshold, source, exception, and action. For public filers, SEC EDGAR APIs expose submissions histories and XBRL data with documented structures and update behavior. SEC EDGAR APIs They can support entity, filing, and financial fields within scope, but should not be treated as universal private-company coverage.

5. Set freshness, conflicts, and provenance

Freshness follows volatility and consequence:

• legal name and ownership: event-driven plus periodic review;

• facility operating status: event-driven and periodic;

• procurement notice: deadline-driven;

• application and installed base: based on lifecycle and change risk;

• service coverage: review when partner, territory, or capability changes;

• classification: review at source revision or material business change. Do not use one account-level last_updated timestamp to imply every field is current. Store observed_at, retrieved_at, and, where relevant, valid_from and valid_to for each claim. Conflict handling should specify source scope, not a universal ranking. A current facility document may outweigh an older consolidated description for plant activity; an official legal filing may outweigh a brand page for legal identity. Preserve both observations and the resolution rationale.

6. Separate automation from decision authority

Automation can:

• retrieve permitted public records;

• normalize names and units;

• detect missing required fields;

• propose entity matches;

• flag expired evidence;

• calculate transparent formulas;

• route exceptions to review. Automation should not silently:

• convert missing to false;

• promote discovery evidence to confirmation;

• merge entities on domain alone;

• infer application from an industry label;

• declare a hard-gate pass without required evidence;

• treat activity as intent or qualification;

• overwrite contradictions. The SAM.gov Opportunities API documents fields, pagination, update cadence, and access conditions for published U.S. federal opportunities. SAM.gov Get Opportunities Public API The USAspending API provides public access to U.S. government spending data and documented endpoints. USAspending API Opportunity notices and award history serve different research purposes; neither should be recoded as present account intent without the appropriate event and scope.

7. Use the ICP research template

A. ICP header

• ICP ID and version;

• product and application;

• relationship type;

• geography;

• unit of analysis;

• owner and approver;

• effective and review dates. B. Field contract Use one row per requirement with every element from Section 2. C. Gate register Gate ID Required node Pass evidence Fail evidence unknown action Specialist owner [ID] [Unit] [Rule] [Rule] [Hold/research] [Role] D. Research procedure Order sources, maximum effort, stop conditions, contradiction review, and account decision. E. Quality measures

• field completeness;

• evidence coverage;

• unknown exposure;

• contradicted share;

• reviewer agreement;

• median research time;

• hard-gate resolution rate. These diagnose the research system. They do not forecast response or revenue.

8. Illustrative example

Illustrative example - fictional. A fictional exporter tests a 12-field research contract on 25 fictional facilities. The expected field observations are 12 x 25 = 300. After the bounded test:

• confirmed_fit observations = 198;

• confirmed_non_fit observations = 32;

• contradicted observations = 10;

• unknown observations = 60. Check: 198 + 32 + 10 + 60 = 300. Assessable observations = 198 + 32 = 230. Evidence coverage = 230 / 300 x 100 = 76.7%. Contradiction share = 10 / 300 x 100 = 3.3%. Unknown exposure = 60 / 300 x 100 = 20.0%. Forty of the 60 unknown observations concern one facility-level field, "documented washdown zone," so that field alone accounts for 40 / 300 = 13.3% of all expected observations. The team does not delete the criterion or assume fit. It expands accepted evidence to include a verified sanitation specification and assigns specialist review, then tests a fresh sample. The figures evaluate fictional researchability, not market size or sales potential.

9. Common mistakes

• Naming fields without definitions: researchers interpret the same label differently.

• Using one source tier for every claim: legal identity, application, and timing need different evidence.

• Making commercial data self-confirming: a vendor field becomes proof without provenance.

• Treating unknown as false: coverage problems become negative ICP decisions.

• Storing one source per account: claim-level lineage disappears.

• Using one freshness date: volatile and stable fields age together.

• Automating ambiguity: speed multiplies classification and entity errors.

• Collecting fields without actions: research cost grows without changing decisions.

10. What this does not mean

A data contract does not make all evidence objective or complete. It makes definitions, judgments, and uncertainty visible. It does not require collecting every possible company field. It does not authorize scraping, messaging, or data use beyond applicable rules. An official dataset can define a frame but cannot prove named-account application or intent by itself.

11. Common questions

How many required fields should an ICP have? Use the minimum set that changes eligibility, prioritization, routing, or exclusion. A field without a decision purpose belongs in optional context or should be removed. Can commercial databases be used? They can support discovery or provide fields under their terms, but critical claims should follow the ICP's accepted-evidence rule and retain provenance, date, scope, and uncertainty. What is the difference between unknown and contradicted? unknown means acceptable evidence has not established the claim. contradicted means two or more credible sources materially conflict after checking scope, date, and unit. The latter requires a preserved conflict record and an explicit resolution or hold. Should every hard gate require manual review? Not necessarily. Automation may confirm a gate when the rule and authoritative evidence are deterministic and auditable. High-consequence, ambiguous, or specialist-dependent gates should route to review. When is the data contract ready to scale? When a bounded sample demonstrates adequate researchability, consistent decisions, transparent unknown records, and controlled exceptions under predeclared thresholds.

12. Sources and methodology

Official sources provide data-reliability principles and documentation for Census, SEC, SAM.gov, and USAspending data systems. The field contract, evidence states, source roles, research rules, automation boundaries, quality measures, and worksheet are SellersGPT editorial tools. They require adaptation to the product, jurisdiction, and decision risk. 1 U.S. Government Accountability Office, Assessing Data Reliability (GAO-20-283G). https://www.gao.gov/products/gao-20-283g. Accessed August 24, 2026. 2 U.S. Census Bureau, Developers. https://www.census.gov/data/developers.html. Accessed August 24, 2026. 3 U.S. Securities and Exchange Commission, EDGAR Application Programming Interfaces. https://www.sec.gov/search-filings/edgar-application-programming-interfaces. Accessed August 24, 2026. 4 U.S. General Services Administration, SAM.gov Get Opportunities Public API. https://open.gsa.gov/api/get-opportunities-public-api/. Accessed August 24, 2026. 5 USAspending.gov, USAspending Application Programming Interface. https://api.usaspending.gov/. Accessed August 24, 2026.