How to Build an Auditable Competitor Evidence Record

Plain-language definition

An evidence record preserves the path from source to permitted decision use

The simplest useful way to think about the record is as a replayable decision trail. A reviewer should be able to start with the exact claim, locate the referenced evidence, see the original observation, distinguish it from the analyst's interpretation and understand the boundary placed around any action.

Auditability does not require a huge compliance system. It requires enough identity, provenance, context, reasoning and control information to answer five questions without guessing: what was claimed, what was examined, what was actually observed, why the observation mattered and who allowed the claim to be used?

Why auditability matters in competitor analysis

Competitor evidence is unusually easy to detach from its original context. A pricing page changes, a third-party metric is refreshed, a product tier is renamed, an analyst leaves or a conclusion is copied into a presentation without its limitation. The later reader sees a polished statement but cannot tell whether it was a direct observation, a bounded inference or an unresolved assumption.

An auditable record reduces that handoff failure. It makes disagreement more productive because a reviewer can challenge a specific source, scope, interpretation or permission boundary instead of rejecting the whole analysis. It also makes updates faster: the team can reopen the affected record rather than repeating the entire competitor study.

• For beginners: it prevents notes from becoming unsupported claims.
• For practitioners: it makes evidence reusable without losing context.
• For managers: it shows which conclusions have review ownership.
• For founders: it separates decision-ready support from attractive but unsafe certainty.
• For reviewers: it provides a stable route back to the evidence, reasoning and change history.

The evidence record is not the evidence itself

A URL alone is not an auditable record. A screenshot alone is not an auditable record. A spreadsheet row containing a conclusion and a confidence label is also not enough. The record must preserve the relationship between the referenced evidence and the decision claim.

Semantic visual

The auditable record has four connected layers

The record is strongest when each layer can be inspected separately and the links between layers remain explicit.

Relationship model

Keep Evidence Items separate and link them to bounded Claim Records

A Claim Record contains the claim, interpretation and decision controls. An Evidence Item preserves one source-level item and its capture context. A Claim–Evidence Link connects them with a role: Supports, Contradicts or Context only.

Contradicting evidence remains attached. A later capture may supersede an older Evidence Item, but the older inspected state and its relationships remain preserved.

The minimum viable record

The minimum viable record is the smallest version another qualified reviewer can reconstruct without inventing missing identity, evidence or decision context. Advanced controls remain conditional; they should be added only when the source, workflow or decision risk requires them.

Examples that fail closed include Draft + Approved, Approved + Rejected and Retired + Pending review.

Use the full governed record when the claim is material, reusable, contested, restricted, AI-assisted, difficult to reverse or likely to be handed to another team or client.

Identity and change control

Freeze the record identity before adding more evidence

Duplicate identity is the normalized Record ID plus normalized Version. The same Record ID can have several valid versions, but one exact pair must identify only one Claim Record version.

The full governed record

The governed record adds the fields needed to manage uncertainty, disagreement, approval and change over time. More fields are not automatically better. Each field should either improve reconstruction, control decision use or make accountability visible.

Conditional controls

Capture mutable sources, restricted evidence and AI activity only where they apply

Mutable sources and visual evidence

When a page, dashboard, feed or visual can change, record the source version or represented period, a Capture ID or Capture Location and the Capture Date. A screenshot without page identity, observed date or capture location cannot establish what was inspected.

Access and handling

Restricted or confidential evidence needs an Access Restriction and Handling Condition. The record should also state whether the reviewer can access the evidence: Yes, No or Conditional. Conditional access can support review only under the documented handling rule; inaccessible evidence without remediation leaves provenance incomplete.

AI-assisted work

Record human verification on the same object as the AI activity. Do not duplicate the same activity across both objects. A shared AI Run Reference may be reused, but each affected object still states what that run created or transformed.

Field-by-field guidance

1. Exact claim

Write one statement that could be accepted, narrowed, contradicted or left unresolved. “Competitor onboarding” is a topic. “The competitor's enterprise onboarding page names a dedicated implementation owner for customers in the reviewed tier” is a recordable claim.

2. Decision supported

Name the decision that gives the record a purpose. Examples include whether to revise sales enablement, investigate a product gap, change a launch claim or defer a recommendation. A record without a decision context becomes a warehouse of facts with no controlled use.

3. Entity, product, market, geography, segment and time boundary

Use the boundaries that materially affect meaning. Do not fill every field mechanically. A global policy may not need a city; a pricing claim may require product tier, billing period, market and date. The goal is to prevent a true observation in one context from becoming a false generalization in another.

4. Source and provenance

Record the source producer, title, type, location and how the material was collected or generated. Provenance should be detailed enough to distinguish a first-party pricing page from a third-party summary, an archived export from a live page and a modelled metric from a direct count. A stable identifier is useful even when access to the underlying evidence is restricted.

5. Access date and observed date

The access date is when the analyst retrieved the material. The observed date is when the condition represented by the evidence applied. They can be the same, but they are not interchangeable. A report accessed today may describe a market from last year.

6. Exact observation

Describe only what the source directly shows. Avoid evaluative words such as “strong,” “leading” or “better” unless those words are themselves part of the source. Preserve a short excerpt, measurement, screenshot reference or stable internal evidence location where lawful and proportionate.

7. Interpretation and assumptions

The interpretation explains why the observation matters. Assumptions identify what must be true for that interpretation to remain usable. Keeping both separate from the observation lets a reviewer accept the source while challenging the reasoning bridge.

8. Evidence state, confidence boundary and contradiction status

These are recorded outputs from adjacent methods, not classifications owned here. Use Observed, Inferred and Unresolved Evidence for evidence state, How to Assign Confidence Without Fake Precision for confidence wording and How to Handle Contradictory Competitor Evidence for conflict classification and disposition.

9. Limitations

State what the record does not establish. Useful limitation statements identify missing coverage, source constraints, model dependence, sampling restrictions or the absence of a causal link. “More research needed” is usually too vague.

10. Materiality, reversibility and permitted decision use

Materiality describes how strongly the missing or uncertain information could affect the decision. Reversibility describes the practical cost of changing course. Permitted use is the explicit action boundary: exploratory only, supports a bounded decision, review required before use or do not use.

11. Approval, owner and reviewer

The owner maintains the record and coordinates reopening. The reviewer challenges the evidence-to-claim connection and the proposed decision use. Approval should state whether the record is approved, approved with conditions, rejected or reopened; it should never be inferred from silence.

12. Review date, reopen condition, change history and final disposition

The review date records the completed governance event. The reopen condition specifies a trigger such as a pricing change, a new product tier, a conflicting source, expiry of the review window or a decision moving into a less reversible stage. Change history preserves what changed, why and who authorized it. Final disposition states whether the record is retained, narrowed, superseded, retired or reopened.

Complete worked example

A reconstructable competitor onboarding record

The following fictional example shows the complete path from observation to controlled decision use. It does not claim that the example company or result is real.

Deliberately incomplete example

A record that looks useful but cannot support a material decision

This entry is not auditable because “better” is undefined, the source cannot be located, the direct observation is missing, scope and dates are absent, confidence has no visible basis, limitations are hidden and the proposed action has no approval or reopening control.

Change history should preserve decisions, not just timestamps

A useful change-history entry records the previous version, the field changed, the reason, the author, the reviewer and the effect on permitted use. Do not overwrite a disputed or superseded conclusion as if it never existed.

The old version remains part of the audit trail. The current version should point to the superseded record and explain whether earlier decisions remain valid.

Review and reopening keep the record useful after publication

Review is not a ceremonial signature. The reviewer should test whether the claim is bounded, whether the evidence can be located, whether the observation is accurately described, whether the interpretation is proportionate and whether the permitted use matches the uncertainty and consequence.

Example review outcome

Good reopen conditions are observable

• A named page, product tier, price or policy changes.
• A new source introduces a material contradiction.
• An assumption fails or is replaced by direct evidence.
• The decision moves into a more costly or difficult-to-reverse stage.
• The dedicated freshness review says the evidence is no longer eligible.

Evidence Freshness and Reliability Standard owns the dedicated freshness method. This record stores the relevant dates and reopening trigger without duplicating that standard.

An implementation sequence that works from one analyst to a larger team

1. 1
   Define the claim and decision first.

   Name the bounded decision before collecting more evidence.

2. 2
   Create and validate Record ID, Version and Record Status.

   Normalize the Record ID, validate the version grammar and keep lifecycle separate from approval.

3. 3
   Create Evidence Items before compressing conclusions.

   Preserve source identity, version or period, observed date and capture context when the source is mutable or visual.

4. 4
   Link claims and evidence without duplication.

   Use Claim–Evidence Links so one claim can use several items and one item can support several claims.

5. 5
   Write the observation independently.

   A reviewer should be able to check it without accepting the interpretation.

6. 6
   Add interpretation, assumptions and adjacent-method outputs.

   Keep direct observation separate from Evidence State, confidence and contradiction controls.

7. 7
   Set the decision and access boundaries.

   Record limitations, materiality, reversibility, permitted use, restrictions and handling conditions.

8. 8
   Record AI activity on the object it transformed.

   Keep source-level activity on Evidence Items, claim-level activity on Claim Records and human verification on the same object.

9. 9
   Assign owner, reviewer and current approval state.

   Store Approver, Approval Date and Approval Conditions on the relevant Claim Record version.

10. 10
    Reopen and supersede non-destructively.

    Create a Change History event and governed next version rather than overwriting the earlier record.

11. 11
    Export the right handoff.

    Use the workbook for repeated records and the PDF for one static review or decision handoff.

Approval and ownership guidance

One person may hold more than one role in a small organization, but the record should still make each responsibility explicit. For a consequential decision, independent review is preferable where practical.

The Evidence Register version stores the current Approval Status, Approver, Approval Date and Approval Conditions. Change History records the version transition and reopening event; it does not point to a separate approval-event object.

Common failure modes

Use the workbook for repeated records and the PDF for one portable handoff

Connected evidence methods

Use the complete record without duplicating adjacent methods

This method owns the reconstructable record connecting claim, evidence, interpretation, controls, decision use, approval and reopening. It records the outputs of adjacent methods but does not replace their tests.

Frequently asked questions

Sources and method basis

This framework is an original competitor-analysis method informed by established provenance, records-management, research-data and analytical-quality principles. The sources below support the general need for identifiable provenance, metadata, responsibility, versioning and fit-for-purpose review; they do not certify this specific method.

• W3C PROV-DM: The PROV Data Model — separates entities, activities, agents and their relationships.
• W3C PROV-O: The PROV Ontology — provides interoperable provenance concepts and relations.
• ISO 15489 Records management — establishes principles for creating, capturing and managing records.
• ISO 23081 Metadata for records — addresses metadata used to support record authenticity and reliability.
• NIST Research Data Framework — includes metadata, provenance and data lifecycle themes.
• The AQuA Book — guidance for robust, fit-for-purpose analysis and quality assurance.
• WCAG 2.2: Error Identification and Labels or Instructions — accessibility basis for the interactive check.

Make the claim replayable before asking anyone to trust it

A useful competitor evidence record does not bury uncertainty under more fields. It preserves the exact claim, the evidence path, the reasoning boundary, the allowed decision use and the people responsible for review. When another qualified reviewer can reconstruct those elements without guessing, the record is auditable. When the record is also reliable enough for the intended use, it can support a controlled decision.