Use this static pack before a financial agent publishes, promotes, or restates a claim that could overstate evidence, safety, suitability, approval, or expected outcomes.

Example input

A user asks whether the phrase 'this strategy is verified and safe for users' can be published based on limited static sample material.

Expected static output

A source-bound claim-safety note that separates exact claim text, evidence links, retrieval time, blocked language, safe restatement, and human-escalation requirements.

Best used when

The output will be public, promotional, user-facing, or likely to be reused by another agent.
The agent needs to preserve source scope and limitation language before repeating a financial claim.

Not used for

Live screening, live compliance clearance, or request-time approval.
Trading, backtesting, execution, investment advice, or legal advice.

Source freshness

This pack preserves a static source snapshot and citation trail; it does not fetch fresh third-party data at request time.

Agent reading hint

Read the HTML page first for boundaries, then use the JSON artifact for structured retrieval.

Why this pack exists

This is one of the most reusable reliability tasks because many financial-agent failures come from overclaiming, not from missing raw data.

Linked service: Reliability Corpus

Pack id: claim_safety_pack

What this is not

Not a live review queue.

Not a request-time API.

Not a buy/sell/execute system.

Not a legal, sanctions, or compliance verdict.

Inputs / outputs

Stable contract for this scenario pack

Each pack stays readable for both humans and crawlers by using the same stable, bounded sections.

Inputs

domain: optional backtest_assumptions, runner_difference, claim_safety, or related category
severity: optional low, medium, or high
source: optional source-ref or text search
updatedSince: optional date filter for changed entries

Outputs

failure-pattern hits in the claim-safety domain
usage guidance for deeper review
do-not-upgrade boundaries for public language

How to use this pack

Read the scenario boundary before you reuse the sample output

This page explains a single review scenario. Use it to understand what evidence the scenario needs, what a bounded output looks like, and when the reader should stop, escalate, or move to a linked artifact.

Use this pack when

Use this pack when the task already matches this scenario: An agent is about to restate marketing, evidence, or strategy claims in public-facing language and needs a warning-layer review first.
Use it when the reader needs a bounded checklist, sample evidence records, and an example of safe versus blocked language.
Treat Claim-safety review before publication as a scenario-specific public review guide that sits on top of the linked Reliability Corpus service.

Stop at this pack when

Stop at this pack when you need a public, static explanation of what should be checked and how the review should be framed.
Stop here when the goal is to teach a crawler, agent, or human reviewer the shape of the evidence rather than produce a live decision.
Escalate beyond this pack when the case becomes ambiguous, private, high-value, or legally sensitive.

Move to the linked artifact when

Open the linked JSON artifact when another system needs the stable machine-readable pack payload.
Move to the related service page when the task needs the broader service contract instead of only this scenario.
Treat the related artifacts as public references and handoff surfaces, not as live approval records.

Pack JSON artifact: /reliability/public-materials/claim-safety-pack.json

Linked service page: Reliability Corpus

This pack does not do

A corpus hit is a warning shape, not proof that every related strategy is invalid
The corpus does not replace runner verification or human judgment
Source behavior, platform documentation, and assumptions can change over time
The pack does not issue a legal verdict or certify marketing language as compliant.

Usage fit

Where this pack helps and where it stops

Suitable for

AI agents drafting public financial summaries or promotional copy
Human reviewers checking whether evidence was upgraded into unsafe claims
Agent builders who need a static warning reference instead of a full scoring engine

Not suitable for

A corpus hit is a warning shape, not proof that every related strategy is invalid
The corpus does not replace runner verification or human judgment
Source behavior, platform documentation, and assumptions can change over time
The pack does not issue a legal verdict or certify marketing language as compliant.

Source / limitation policy

https://www.sec.gov/oiea/investor-alerts-and-bulletins
https://www.finra.org/investors/insights/artificial-intelligence-and-investment-fraud
https://www.finra.org/investors/insights/investment-group-imposter-scams
https://consumer.ftc.gov/articles/what-know-about-cryptocurrency-scams

Delivery shape

What a real static review package needs to collect and return

These sections make the pack more than a scenario description. They define the minimum evidence set, the bounded outputs the review should return, and the citation rules that keep the result trustworthy.

Evidence to collect

The original claim text exactly as the agent plans to repeat it, including headline, body copy, and any guarantee or safety language.
Every supporting source URL, citation, or evidence fragment the claim relies on, with retrieval timestamp and source scope.
Audience, distribution context, and surrounding marketing or product language that could change how the claim is understood.
Missing evidence, conflicting evidence, and unsupported leaps from source facts into suitability, compliance, safety, or outcome claims.

Delivery outputs

A claim-by-claim mapping that separates supported wording, unsupported wording, and escalation-required wording.
A safe restatement that keeps source, retrieval time, scope, audience, and limitation language intact.
Blocked statements that must not be repeated downstream because they imply guarantees, approval, compliance, safety, suitability, or outcome certainty.
A limitation note explaining what the evidence does not establish about safety, suitability, legality, compliance, or future outcomes.

Decision matrix

Exact claim text row
Status: warning
Evidence fields: claim_text, claim_category
Source requirement: Record the exact claim text, headline, surrounding copy, audience, channel, and claim category before rewriting or publishing.
Allowed static use: Use the exact claim text as the source-bound evidence anchor for bounded static review and limitation-preserving rewrite work.
Blocked upgrade: Do not call a paraphrased or softened version verified, safe, suitable, compliant, approved, guaranteed, or protected if the original claim overreaches.
Agent action: Preserve the original wording, rewrite only inside the cited source boundary, and escalate if unsafe terms remain.
Supporting source and evidence row
Status: unresolved
Evidence fields: source_links, claim_evidence
Source requirement: Record every supporting source URL, evidence fragment, artifact version, source owner, and whether the source actually supports the claim category.
Allowed static use: Use supporting source evidence only for bounded static statements tied to the cited source and its limitations.
Blocked upgrade: Do not turn thin evidence into verified safety, compliant publication, suitable advice, approved use, guaranteed results, or user protection.
Agent action: Escalate unsupported source jumps and do not publish claims whose evidence does not match the claim category.
Retrieval time and source scope row
Status: warning
Evidence fields: source_links, limitation_language
Source requirement: Record retrieval time, source scope, static snapshot boundary, audience context, and limitation language that must travel with the output.
Allowed static use: Use the source scope and limitation language as the bounded static frame for any safe restatement.
Blocked upgrade: Do not publish the claim as safe, compliant, suitable, approved, protected, or verified when retrieval time or source scope is missing.
Agent action: Preserve limitation language in the final note and escalate or do not publish when scope or retrieval time is absent.
Guarantee, suitability, compliance, approval, and safety term row
Status: blocked
Evidence fields: claim_category, limitation_language
Source requirement: Identify whether the claim uses guarantee, low-risk, compliance, legality, suitability, approval, safety, protection, or outcome-certainty language.
Allowed static use: Use only a bounded source-backed restatement that keeps evidence limits and static-review boundaries visible.
Blocked upgrade: Block language that says the claim is guaranteed, compliant, suitable, approved, safe, verified, or protection for users unless a qualified human review explicitly owns that decision.
Agent action: Rewrite or block the claim, preserve the unsafe term as a blocked statement, and escalate before publication.

Citation rules

Keep every public restatement tied to a named source, retrieval time, and explicit scope.
Do not turn sample evidence, one source, or absence of negative evidence into guarantees or suitability claims.
Preserve limitation language whenever a claim touches returns, legality, compliance, safety, or user fit.
If support is missing or mixed, say the claim remains unresolved instead of smoothing it into marketing copy.

Static sample deliverable

What a bounded review output can look like

These sample fields make the pack feel closer to a real deliverable: a sample input summary, concrete evidence records, and a bounded output that stays inside static-review limits.

Sample input summary

Draft publication line: 'This strategy is verified and safe for users' based on static sample materials and partial supporting notes.

Sample evidence records

Underlying source only supports a static sample artifact, not live user safety or suitability claims
Status: warning
Source: TRADING4AI Evidence Cards sample
Retrieved: 2026-05-25T00:00:00.000Z
The cited artifact is descriptive and bounded; it is not a live approval record.
No cited source provides a legal, compliance, or safety approval for the publication claim
Status: unresolved
Source: SEC investor alerts and bulletins
Retrieved: 2026-05-25T00:00:00.000Z
The source warns about financial promotion risk, but it does not certify this claim.
FINRA artificial intelligence and investment fraud guidance warns that AI trading claims can be used to sell unrealistic promises
Status: warning
Source: FINRA artificial intelligence and investment fraud
Retrieved: 2026-05-25T00:00:00.000Z
The source supports warning language around AI investment fraud risk; it does not support saying an AI strategy is safe, guaranteed, or suitable.
FINRA investment group imposter scams guidance is relevant when claims are distributed through social or private groups
Status: warning
Source: FINRA investment group imposter scams
Retrieved: 2026-05-25T00:00:00.000Z
The source supports a public-warning boundary for social investment-group claims, not an approval or safety clearance for the group.
Retrieval timestamp and limitation language are missing from the publication draft
Status: warning
Source: Claim-safety pack draft review
Retrieved: 2026-05-25T00:00:00.000Z
Without retrieval time and scope, readers cannot tell what the evidence actually covers.
The draft upgrades absence of negative evidence into a guarantee of protection or safety
Status: warning
Source: FTC cryptocurrency scam guidance
Retrieved: 2026-05-25T00:00:00.000Z
FTC cryptocurrency scam guidance supports caution around crypto payment and promotion claims; it does not turn limited evidence into a broad safety promise.

Evidence Card sample

Claim-safety review before publication sample evidence card

Review verdict: escalate

The cited material supports only a bounded, source-scoped description of the sample artifact at the listed retrieval time, with explicit limitations; it is not advice, not approval, and not a promise of user safety or outcomes.

Citation block

artifactVersion=public-material-claim-safety-pack-2026-05-25
generatedAt=2026-05-25T00:00:00.000Z
staticSnapshotAt=2026-05-25T00:00:00.000Z
sampleEvidenceRetrievedAt=2026-05-25T00:00:00.000Z
pagePath=/agent-verification/materials/claim-safety-pack
artifactPath=/reliability/public-materials/claim-safety-pack.json

Key findings

warning: Underlying source only supports a static sample artifact, not live user safety or suitability claims (The cited artifact is descriptive and bounded; it is not a live approval record.)
unresolved: No cited source provides a legal, compliance, or safety approval for the publication claim (The source warns about financial promotion risk, but it does not certify this claim.)
warning: FINRA artificial intelligence and investment fraud guidance warns that AI trading claims can be used to sell unrealistic promises (The source supports warning language around AI investment fraud risk; it does not support saying an AI strategy is safe, guaranteed, or suitable.)
warning: FINRA investment group imposter scams guidance is relevant when claims are distributed through social or private groups (The source supports a public-warning boundary for social investment-group claims, not an approval or safety clearance for the group.)
warning: Retrieval timestamp and limitation language are missing from the publication draft (Without retrieval time and scope, readers cannot tell what the evidence actually covers.)
warning: The draft upgrades absence of negative evidence into a guarantee of protection or safety (FTC cryptocurrency scam guidance supports caution around crypto payment and promotion claims; it does not turn limited evidence into a broad safety promise.)

Required disclosures

Static public sample only; not a live review, approval, compliance clearance, or request-time judgment.
A corpus hit is a warning shape, not proof that every related strategy is invalid
The corpus does not replace runner verification or human judgment
Source behavior, platform documentation, and assumptions can change over time
The pack does not issue a legal verdict or certify marketing language as compliant.
No cited source proves safety, suitability, or guaranteed protection for users.
The claim is not anchored to retrieval time, scope, or dataset limits.
The wording converts static sample material into a universal trust statement.

Do not claim

This strategy is verified and safe for users.
This result proves the asset will go up.
The public material certifies compliance, legality, or guaranteed protection.

Action routing

actionRouting type: financial_claim_publication_review

Preflight stage: before_publication_or_distribution

Default posture: escalate

Primary action risk: A financial claim may be published as safety, approval, guarantee, compliance, or suitability language without evidence that supports it.

Minimum input field ids: claim_text, source_links, claim_category, claim_evidence, limitation_language

Minimum required inputs

Exact claim text, audience, distribution channel, and whether the claim is public, marketing, support, or internal copy.
Evidence links, retrieval timestamp, artifact version, and source scope supporting the safest possible restatement.
Claim category, including returns, risk, legality, compliance, safety, suitability, approval, or protection language.
Required limitation language and the blocked statements that must not be restated.

Missing input fallback

Escalate or do not publish; collect missing source, audience, scope, and limitation evidence before restating the financial claim.

Route when

An agent is preparing public, marketing, product, or support language about a financial tool, strategy, subscription, or review.
The claim touches returns, low risk, safety, compliance, legality, suitability, approval, or user protection.

Stop or escalate when

The claim implies guaranteed returns, low risk, verified safety, legal compliance, suitability, or user protection.
Retrieval time, source scope, audience, limitation language, or evidence boundary is missing.

Search intent

Intent id: claim_safety_pack_search_intent

User task: Check whether financial wording can be safely published without implying guarantees, suitability, or clearance.

Route reason: Route to Claim-safety review before publication when the user task matches financial_claim_publication_review and needs A source-bound claim-safety note that separates exact claim text, evidence links, retrieval time, blocked language, safe restatement, and human-escalation requirements.

queryExamples: safe wording for investment claim, can I say guaranteed returns, financial claim safety review, can agent say strategy is verified, avoid guaranteed profit claim, safe financial marketing wording

Use when

An agent is preparing public, marketing, product, or support language about a financial tool, strategy, subscription, or review.
The output will be public, promotional, user-facing, or likely to be reused by another agent.
Read the HTML page first for boundaries, then use the JSON artifact for structured retrieval.

Do not use as

live approval
investment advice
compliance clearance
real-time screening
backtesting or execution advice

Escalate to qualified human review when required inputs are missing, source confidence is unresolved, payment or transfer risk is present, or the user asks for approval, clearance, execution, suitability, or compliance guarantees.

Input field glossary

claim_text: The exact financial claim before it is restated, published, or reused. Missing input risk: The agent may miss overclaiming, suitability language, or unsupported safety language.

source_links: Official registries, warning pages, filings, releases, or cited public pages used as evidence. Missing input risk: The agent cannot separate source-bound evidence from unsourced assertion.

claim_category: The risk class of the claim, such as returns, risk, legality, compliance, suitability, approval, or protection. Missing input risk: The agent may apply generic wording where a stricter blocked-claim rule is needed.

claim_evidence: Source material supporting performance, risk, registration, approval, compliance, or safety claims. Missing input risk: The agent may repeat promotional language as if it were verified evidence.

limitation_language: The caveats, source scope, retrieval time, and blocked statements that must travel with the output. Missing input risk: The agent may publish a bounded evidence note as if it were live clearance or final advice.

Preflight questionnaire

Answer every question with source-bound evidence before upgrading the review posture; if any required input is missing, apply the missing-input fallback and do not treat the static pack as approval.

claim_text: What evidence identifies the claim text for this financial_claim_publication_review review? Acceptable evidence: guaranteed return sentence, low-risk phrase, compliance claim, approval claim. If missing: The agent may miss overclaiming, suitability language, or unsupported safety language. Escalate or do not publish; collect missing source, audience, scope, and limitation evidence before restating the financial claim.

source_links: What evidence identifies the source links for this financial_claim_publication_review review? Acceptable evidence: OFAC URL, SEC filing URL, FRED series URL, public warning page. If missing: The agent cannot separate source-bound evidence from unsourced assertion. Escalate or do not publish; collect missing source, audience, scope, and limitation evidence before restating the financial claim.

claim_category: What evidence identifies the claim category for this financial_claim_publication_review review? Acceptable evidence: returns, risk, legality, compliance, suitability. If missing: The agent may apply generic wording where a stricter blocked-claim rule is needed. Escalate or do not publish; collect missing source, audience, scope, and limitation evidence before restating the financial claim.

claim_evidence: What evidence identifies the claim evidence for this financial_claim_publication_review review? Acceptable evidence: performance report, registration page, risk disclosure, audit statement. If missing: The agent may repeat promotional language as if it were verified evidence. Escalate or do not publish; collect missing source, audience, scope, and limitation evidence before restating the financial claim.

limitation_language: What evidence identifies the limitation language for this financial_claim_publication_review review? Acceptable evidence: static snapshot, not legal advice, not live screening, evidence gap. If missing: The agent may publish a bounded evidence note as if it were live clearance or final advice. Escalate or do not publish; collect missing source, audience, scope, and limitation evidence before restating the financial claim.

Decision policy

Default posture: escalate

Allowed static output: The cited material supports only a bounded, source-scoped description of the sample artifact at the listed retrieval time, with explicit limitations; it is not advice, not approval, and not a promise of user safety or outcomes.

Blocked static output: This strategy is verified and safe for users, guaranteed to protect them, compliant for publication, approved for use, and suitable for the audience.

Proceed only when

Collected and cited: Exact claim text, audience, distribution channel, and whether the claim is public, marketing, support, or internal copy.
Collected and cited: Evidence links, retrieval timestamp, artifact version, and source scope supporting the safest possible restatement.
Collected and cited: Claim category, including returns, risk, legality, compliance, safety, suitability, approval, or protection language.
Collected and cited: Required limitation language and the blocked statements that must not be restated.

Fallback when missing inputs

Escalate or do not publish; collect missing source, audience, scope, and limitation evidence before restating the financial claim.

Stop or escalate when

The claim implies guaranteed returns, low risk, verified safety, legal compliance, suitability, or user protection.
Retrieval time, source scope, audience, limitation language, or evidence boundary is missing.
The statement promises returns, low risk, safety, or suitability.
The claim converts a static sample or one source into a universal conclusion.
The language turns an absence of evidence into a safety or compliance clearance claim.

Escalate to human review when any stop-or-escalate rule applies, when required evidence is missing, or when the requested action would treat this static pack as approval, clearance, execution advice, or live screening.

Decision guardrails

escalate: One or more minimum inputs are missing or uncited: Exact claim text, audience, distribution channel, and whether the claim is public, marketing, support, or internal copy.; Evidence links, retrieval timestamp, artifact version, and source scope supporting the safest possible restatement.; Claim category, including returns, risk, legality, compliance, safety, suitability, approval, or protection language.; Required limitation language and the blocked statements that must not be restated.. Required action: Escalate or do not publish; collect missing source, audience, scope, and limitation evidence before restating the financial claim. Allowed output: The cited material supports only a bounded, source-scoped description of the sample artifact at the listed retrieval time, with explicit limitations; it is not advice, not approval, and not a promise of user safety or outcomes. Blocked upgrade: This strategy is verified and safe for users, guaranteed to protect them, compliant for publication, approved for use, and suitable for the audience. Human review required: yes.

escalate: Any stop-or-escalate trigger applies: The claim implies guaranteed returns, low risk, verified safety, legal compliance, suitability, or user protection.; Retrieval time, source scope, audience, limitation language, or evidence boundary is missing.; The statement promises returns, low risk, safety, or suitability.; The claim converts a static sample or one source into a universal conclusion.; The language turns an absence of evidence into a safety or compliance clearance claim.. Required action: Stop automatic action, keep the unresolved risk visible, and route the item to human review before payment, publication, execution, or downstream trust transfer. Allowed output: The evidence supports a bounded observation with stated limitations and retrieval time. This material can describe what was observed, but it does not guarantee future outcomes or user suitability. The claim should stay tied to the cited source rather than being upgraded into approval language. Blocked upgrade: This strategy is verified and safe for users. This result proves the asset will go up. The public material certifies compliance, legality, or guaranteed protection. Human review required: yes.

allow_with_limits: All minimum inputs are collected, cited, and no stop-or-escalate trigger applies; the output still remains a bounded static material summary. Required action: Emit only source-bound, timestamped, limitation-preserving static output and carry unresolved items into the final note. Allowed output: The cited material supports only a bounded, source-scoped description of the sample artifact at the listed retrieval time, with explicit limitations; it is not advice, not approval, and not a promise of user safety or outcomes. Blocked upgrade: This strategy is verified and safe for users, guaranteed to protect them, compliant for publication, approved for use, and suitable for the audience. Human review required: no.

Misuse patterns

block: Evidence upgraded into verified-and-safe guarantee. False signal: A cited artifact or warning source exists, so the agent says the strategy, claim, or product is verified and safe. Why misleading: Source presence does not establish future outcomes, user suitability, compliance, legal approval, safety, or guaranteed protection. Safe alternative: Restate only what the cited source supports, with retrieval time, scope, audience, and limitation language intact. Blocked agent action: Do not publish guarantee, protection, safety, approval, compliance, legality, suitability, or verified-performance language. Evidence fields: claim_text, claim_category, claim_evidence, limitation_language.

escalate: Missing retrieval or source scope treated as publishable. False signal: The wording sounds conservative, so the agent publishes it without source scope, retrieval time, or limitation language. Why misleading: A conservative tone can still mislead when the audience, distribution channel, source scope, timestamp, or limitation boundary is missing. Safe alternative: Attach source URL, retrieval time, audience, channel, artifact version, and limitations before any public restatement. Blocked agent action: Do not publish or distribute the claim when retrieval time, source scope, or limitation language is absent. Evidence fields: claim_text, source_links, limitation_language.

warning: Softened paraphrase hides the unsafe original claim. False signal: The agent rewrites an unsafe claim into softer language and treats the rewrite as reviewed. Why misleading: A paraphrase can preserve the same unsupported implication while hiding guarantee, safety, suitability, compliance, or approval language from review. Safe alternative: Keep the exact original claim visible, map unsafe terms to blocked statements, and only publish a source-bound restatement. Blocked agent action: Do not use a softened rewrite to bypass review of the original claim text, audience, source scope, or limitation requirements. Evidence fields: claim_text, claim_category, claim_evidence.

Static action note template

Use only as a static, source-bound Reliability Corpus note after completing action routing, the preflight questionnaire, and the decision policy.

This is a static action note template, not approval, not live screening, not backtesting, not execution advice, and not a clearance decision.

Required sections

action_context: State the action type, preflight stage, default posture, and collected minimum inputs: Exact claim text, audience, distribution channel, and whether the claim is public, marketing, support, or internal copy.; Evidence links, retrieval timestamp, artifact version, and source scope supporting the safest possible restatement.; Claim category, including returns, risk, legality, compliance, safety, suitability, approval, or protection language.; Required limitation language and the blocked statements that must not be restated..
source_bound_evidence: List only cited evidence collected for this pack: The original claim text exactly as the agent plans to repeat it, including headline, body copy, and any guarantee or safety language.; Every supporting source URL, citation, or evidence fragment the claim relies on, with retrieval timestamp and source scope.; Audience, distribution context, and surrounding marketing or product language that could change how the claim is understood.; Missing evidence, conflicting evidence, and unsupported leaps from source facts into suitability, compliance, safety, or outcome claims..
decision_policy: Apply fallback 'Escalate or do not publish; collect missing source, audience, scope, and limitation evidence before restating the financial claim.' and stop/escalate when any policy trigger applies.
safe_restatement: Use bounded language no stronger than: The cited material supports only a bounded, source-scoped description of the sample artifact at the listed retrieval time, with explicit limitations; it is not advice, not approval, and not a promise of user safety or outcomes.
blocked_language: Do not restate or imply: This strategy is verified and safe for users, guaranteed to protect them, compliant for publication, approved for use, and suitable for the audience.
citation_trail: Preserve citation requirements: Keep every public restatement tied to a named source, retrieval time, and explicit scope.; Do not turn sample evidence, one source, or absence of negative evidence into guarantees or suitability claims.; Preserve limitation language whenever a claim touches returns, legality, compliance, safety, or user fit.; If support is missing or mixed, say the claim remains unresolved instead of smoothing it into marketing copy..

Static action note example

Example id: claim_safety_pack_static_action_note_example

Note status: escalate

Source snapshot: 2026-05-25T00:00:00.000Z

This is a static action note template, not approval, not live screening, not backtesting, not execution advice, and not a clearance decision.

Filled sections

action_context: Draft publication line: 'This strategy is verified and safe for users' based on static sample materials and partial supporting notes. Action type: financial_claim_publication_review. Default posture: escalate. Minimum inputs: Exact claim text, audience, distribution channel, and whether the claim is public, marketing, support, or internal copy.; Evidence links, retrieval timestamp, artifact version, and source scope supporting the safest possible restatement.; Claim category, including returns, risk, legality, compliance, safety, suitability, approval, or protection language.; Required limitation language and the blocked statements that must not be restated..
source_bound_evidence: warning: Underlying source only supports a static sample artifact, not live user safety or suitability claims. The cited artifact is descriptive and bounded; it is not a live approval record. unresolved: No cited source provides a legal, compliance, or safety approval for the publication claim. The source warns about financial promotion risk, but it does not certify this claim. warning: FINRA artificial intelligence and investment fraud guidance warns that AI trading claims can be used to sell unrealistic promises. The source supports warning language around AI investment fraud risk; it does not support saying an AI strategy is safe, guaranteed, or suitable. warning: FINRA investment group imposter scams guidance is relevant when claims are distributed through social or private groups. The source supports a public-warning boundary for social investment-group claims, not an approval or safety clearance for the group. warning: Retrieval timestamp and limitation language are missing from the publication draft. Without retrieval time and scope, readers cannot tell what the evidence actually covers. warning: The draft upgrades absence of negative evidence into a guarantee of protection or safety. FTC cryptocurrency scam guidance supports caution around crypto payment and promotion claims; it does not turn limited evidence into a broad safety promise.
decision_policy: escalate: The publication draft overstates what the cited materials establish and should be rewritten before release. Unresolved items: No cited source proves safety, suitability, or guaranteed protection for users.; The claim is not anchored to retrieval time, scope, or dataset limits.; The wording converts static sample material into a universal trust statement. Required follow-up: Replace guarantee language with a source-bound restatement tied to the cited artifact and retrieval time.; Add limitation language covering static-sample scope and lack of live review.; Escalate any compliance or safety wording to human review before publication..
safe_restatement: The cited material supports only a bounded, source-scoped description of the sample artifact at the listed retrieval time, with explicit limitations; it is not advice, not approval, and not a promise of user safety or outcomes.
blocked_language: This strategy is verified and safe for users, guaranteed to protect them, compliant for publication, approved for use, and suitable for the audience.
citation_trail: Claim-safety review before publication, public-material-claim-safety-pack-2026-05-25, generated 2026-05-25T00:00:00.000Z, source snapshot 2026-05-25T00:00:00.000Z, https://trading4ai.com/agent-verification/materials/claim-safety-pack Citation requirements: Keep every public restatement tied to a named source, retrieval time, and explicit scope.; Do not turn sample evidence, one source, or absence of negative evidence into guarantees or suitability claims.; Preserve limitation language whenever a claim touches returns, legality, compliance, safety, or user fit.; If support is missing or mixed, say the claim remains unresolved instead of smoothing it into marketing copy..

Agent workflow

Workflow id: claim_safety_pack_agent_workflow

Step ids: select_pack, collect_minimum_inputs, answer_preflight_questionnaire, apply_decision_policy, draft_static_action_note, cite_and_escalate

Linked example: claim_safety_pack_static_action_note_example

Canonical citation: Claim-safety review before publication, public-material-claim-safety-pack-2026-05-25, generated 2026-05-25T00:00:00.000Z, source snapshot 2026-05-25T00:00:00.000Z, https://trading4ai.com/agent-verification/materials/claim-safety-pack

Source quality

sourceQuality pack id: claim_safety_pack

Action type: financial_claim_publication_review

Workflow id: claim_safety_pack_agent_workflow

Source kinds: json_artifact, official_guidance, sample_page

Source manifest count: 6

Official registry sources: 0

Official guidance sources: 4

Source quality profile: guidance_heavy

Registry coverage expectation: not_primary_for_this_static_pack

This pack is guidance-heavy because claim-safety review compares proposed language against public investor-protection and fraud-warning guidance rather than registry hit/no-hit evidence.

Zero official registry sources is acceptable for this static pack because it reviews wording boundaries, not entity clearance.

Agent source use

Use official guidance to separate safe restatements from guarantees, suitability, approval, safety, or compliance claims.
Preserve source scope and retrieval time when reusing any safe or blocked wording.
Do not turn guidance alignment into legal compliance, marketing approval, or product safety certification.

Sample evidence records: 6

Source snapshot: 2026-05-25T00:00:00.000Z

Boundary: static_source_manifest_not_live_screening

Source review policy

sourceReviewPolicy.nextRecommendedReviewAt: 2026-08-23T00:00:00.000Z

Review cadence: 90 days

Review mode: manual_public_source_recheck_required

Catalog field: sourceReviewPolicyIndex

This is a static source-review policy for a pre-generated public material pack; it is not live screening, request-time judgment, compliance clearance, payment approval, or trading advice.

Refresh triggers (sourceReviewPolicy.refreshRequiredWhen)

Any cited official source, registry, guidance page, filing page, or public warning URL changes content, schema, access status, or meaning.
A downstream agent wants to use the static pack for a new counterparty, payment destination, wallet, claim, filing, macro release, or source set.
The current date is past nextRecommendedReviewAt or the artifactVersion / generatedAt fields are removed from the handoff.
The wording under review cites new performance, suitability, safety, compliance, approval, or guarantee evidence not present in this static sample.

Case readiness checklist

caseReadinessChecklist: claim_safety_pack_case_readiness_checklist

sourceFreshnessGate: Before reuse, compare the case date and cited source retrieval plan with sourceReviewPolicy.nextRecommendedReviewAt=2026-08-23T00:00:00.000Z; if the static snapshot is stale, re-check sources and version the artifact before using it.

Default posture: escalate

Must confirm before use

Collected and cited every minimum input field: claim_text, source_links, claim_category, claim_evidence, limitation_language.
Opened the HTML page for scope and limitations, then used the JSON artifact for structured retrieval.
Preserved source URLs, retrieval timestamps, source roles, canonical citation text, and unresolved items in the downstream note.
Checked decisionMatrix, misusePatterns, sourceReviewPolicy, and evidenceVerificationRecipe before upgrading any sentence.
Confirm the exact original claim text, audience, distribution channel, source links, retrieval time, and limitation language are visible before any rewrite.
Map guarantee, safety, suitability, legality, compliance, approval, protection, and outcome-certainty terms to blocked statements before publication.

Not ready signals

One or more required evidence fields are missing, uncited, or unresolved: claim_text, source_links, claim_category, claim_evidence, limitation_language.
The case is past sourceReviewPolicy.nextRecommendedReviewAt without a fresh source re-check.
The user or downstream agent asks for approval, clearance, safety, compliance, suitability, execution, payment, publication, or legal/trading advice.
The requested case uses a new counterparty, payment destination, wallet, claim, filing, macro release, source set, or audience that is not covered by the static sample.

Ready static handoff

A bounded static action note or evidence card with citations, source snapshot, unresolved items, and blocked-upgrade language preserved.
A JSON artifact reference plus canonical citation text that another crawler or agent can retrieve without treating it as a live service.
A human-review handoff when the case remains high-risk, private, legally sensitive, or close to money movement or public distribution.

Case worksheet

caseWorksheet: claim_safety_pack_case_worksheet

worksheetType: claim_redline_ledger

Treat this worksheet as complete only when every row preserves source/citation fields, result semantics, safe rewrite, blocked rewrite, and escalation reason.

Exact claim text redline row · Claim-safety pack draft review

Capture the original claim wording, headline, body copy, audience, publication channel, claim category, source links, and surrounding marketing context.

Result semantics: Record whether the claim is source-supported, partially supported, unsupported, overbroad, unresolved, or unsafe for publication.

Unsafe category: verified safety, user protection, approval, or outcome certainty

Safe restatement: the claim can only be rewritten as a bounded source-scoped statement with retrieval time, limitations, and no advice or approval language.

Blocked rewrite: do not say the strategy is verified, safe, guaranteed, approved, suitable, compliant, or protection for users.

Supporting source evidence redline row · TRADING4AI Evidence Cards sample / SEC investor alerts

Capture every source URL, evidence fragment, artifact version, source owner, retrieval timestamp, audience/channel context, and the exact claim each source is supposed to support.

Result semantics: Record source-supported, partially supported, unsupported, stale, unresolved, or out-of-scope evidence for each claim fragment.

Unsafe category: thin evidence upgraded into compliance, approval, safety, or suitability

Safe restatement: the source supports only a bounded observation about the cited material at the listed retrieval time with limitation language intact.

Do not claim that limited source evidence makes the copy compliant, approved, suitable, verified, safe, or guaranteed.

Audience and channel redline row · FINRA investment group imposter scams guidance

Capture audience, distribution channel, publication surface, social/private-group context, promotional framing, and whether the wording changes user expectations.

Result semantics: Record internal-only, public marketing, support, social group, private channel, unresolved audience, or high-risk publication context.

Unsafe category: audience/channel context hidden while publishing safety or suitability language

Safe restatement: the wording must preserve audience/channel limitations and avoid turning social or promotional context into trust approval.

Do not publish a claim as safe, verified, approved, suitable, compliant, or protective when audience/channel risk is unresolved.

Unsafe terms redline row · FINRA AI investment fraud / FTC cryptocurrency scam guidance

Capture guarantee, low-risk, compliance, legality, suitability, approval, safety, protection, verified, and outcome-certainty terms in the claim text.

Result semantics: Record each unsafe term as supported, unsupported, blocked, unresolved, or requiring qualified human review before publication.

Unsafe category: guarantee, suitability, compliance, approval, safety, verified, protection, and outcome certainty

Safe restatement: replace unsafe terms with bounded source-scope language, retrieval time, explicit limitations, and no promise of user safety or outcomes.

Do not say verified and safe, guaranteed, compliant, approved, suitable, protective, risk-free, best, or certain to produce an outcome.

Source manifest handoff

Sample evidence card id: claim_safety_pack_sample_evidence_card

Source manifest entries: 6

Official registry sources: 0

Source quality boundary: static_source_manifest_not_live_screening

Canonical citation artifact version: public-material-claim-safety-pack-2026-05-25

Source snapshot: 2026-05-25T00:00:00.000Z

This is a static citation handoff for retrieval. It records cited sources, source roles, evidence links, and limitations without turning the pack into live clearance or approval.

Machine summary

Pack: claim_safety_pack

Page: https://trading4ai.com/agent-verification/materials/claim-safety-pack

Artifact: https://trading4ai.com/reliability/public-materials/claim-safety-pack.json

Action type: financial_claim_publication_review

Default posture: escalate

Minimum input fields: claim_text, source_links, claim_category, claim_evidence, limitation_language

Decision matrix rows: 4

Evidence verification steps: 5

Misuse patterns: 3

Agent use

Open the HTML page first for scope, limitations, source freshness, and scenario fit.
Use the JSON artifact for structured retrieval after the page boundary is understood.
Carry the canonical citation, source snapshot, limitations, and unresolved items into downstream summaries.

Do not use as

live screening
payment approval
compliance clearance
trading or execution advice
proof that a counterparty, wallet, claim, filing, or macro interpretation is safe

Decision matrix labels

Exact claim text row
Supporting source and evidence row
Retrieval time and source scope row
Guarantee, suitability, compliance, approval, and safety term row

Evidence verification step labels

Exact claim text and original wording verification
Evidence record linkage check
Decision matrix boundary check
Canonical citation and timestamp check
Human escalation and blocked-upgrade check

Misuse pattern labels

Evidence upgraded into verified-and-safe guarantee
Missing retrieval or source scope treated as publishable
Softened paraphrase hides the unsafe original claim

Evidence verification recipe

Pack: claim_safety_pack

Default posture: escalate

Static only: yes

Exact claim text and original wording verification
Verification action: Compare the exact claim text, original wording, audience, source links, retrieval time, and limitation language before any rewrite or publication.
Trust boundary: A cleaner paraphrase is not verified, safe, suitable, compliant, approved, or protected when the original claim overreaches its evidence.
Failure mode: If the exact claim, source scope, retrieval time, or limitation language is missing, do not publish and escalate the claim review.
Required citation fields: sampleInputSummary, sourceManifest[].url, sourceManifest[].retrievedAt, sampleEvidenceRecords[].note, decisionMatrix[].matrixId
Evidence record linkage check
Verification action: Match each sample evidence record to a sourceManifest evidenceRecordIds entry, then preserve the record status, note, source label, source URL, and retrievedAt value.
Trust boundary: A sample record explains the static example only; it does not prove the current counterparty, claim, payment, wallet, filing, macro release, or subscription is safe.
Failure mode: If a record cannot be linked to its source manifest row, do not reuse it as evidence and keep the final output unresolved.
Required citation fields: sampleEvidenceRecords[].id, sampleEvidenceRecords[].status, sampleEvidenceRecords[].sourceUrl, sampleEvidenceRecords[].retrievedAt, sourceManifest[].evidenceRecordIds
Decision matrix boundary check
Verification action: Read all 4 decision matrix rows and carry their allowedStaticUse, blockedUpgrade, and agentAction fields into any downstream summary.
Trust boundary: Decision matrix rows define allowed static reuse and blocked upgrades; they are not automatic live approval, live screening, compliance sign-off, or execution permission.
Failure mode: If a downstream sentence conflicts with a blockedUpgrade row, block or rewrite the sentence and escalate when money, publication, or user action is involved.
Required citation fields: decisionMatrix[].matrixId, decisionMatrix[].status, decisionMatrix[].allowedStaticUse, decisionMatrix[].blockedUpgrade
Canonical citation and timestamp check
Verification action: Preserve the canonical citation text, artifact version, generatedAt, staticSnapshotAt, sourceSnapshotAt, and all source retrievedAt values when citing this pack.
Trust boundary: A citation proves the static artifact and source snapshot were recorded; it does not prove the facts are current beyond the listed timestamps.
Failure mode: If artifact version, generatedAt, sourceSnapshotAt, or source retrievedAt values are removed, the output is no longer evidence-bound.
Required citation fields: canonicalCitation.recommendedCitationText, artifactProvenance.artifactVersion, artifactProvenance.generatedAt, canonicalCitation.sourceSnapshotAt, sourceManifest[].retrievedAt
Human escalation and blocked-upgrade check
Verification action: Compare the final output with the pack's blockedStatements, sampleReviewOutput.blockedRestatement, escalationTriggers, and actionTaxonomy.stopOrEscalateWhen fields.
Trust boundary: Static public material may explain evidence and limitations, but it must not become live clearance, legal compliance, suitability, safety, guarantee, payment approval, publication approval, or trade advice.
Failure mode: If the output asks a user or agent to move money, publish a financial claim, trust a counterparty, or treat the material as live approval, stop and escalate to human review.
Required citation fields: canonicalCitation.recommendedCitationText, artifactProvenance.artifactVersion, blockedStatements[], sampleReviewOutput.blockedRestatement, escalationTriggers[], actionTaxonomy.stopOrEscalateWhen[]

Cite this pack

Claim-safety review before publication, public-material-claim-safety-pack-2026-05-25, generated 2026-05-25T00:00:00.000Z, source snapshot 2026-05-25T00:00:00.000Z, https://trading4ai.com/agent-verification/materials/claim-safety-pack

Page: https://trading4ai.com/agent-verification/materials/claim-safety-pack

Artifact: https://trading4ai.com/reliability/public-materials/claim-safety-pack.json

Source snapshot: 2026-05-25T00:00:00.000Z

Source manifest

TRADING4AI Evidence Cards sample
URL: /reliability/evidence-cards-sample.json
Kind: json_artifact; role: sample_context; retrieved: 2026-05-25T00:00:00.000Z
Evidence records: sample_artifact_scope
Supports: Underlying source only supports a static sample artifact, not live user safety or suitability claims
Limitations:
- This source entry records what the static sample cites; it does not prove current, complete, or final clearance.
- Preserve retrievedAt and linked evidenceRecordIds when reusing this source in an agent or crawler workflow.
SEC investor alerts and bulletins
URL: https://www.sec.gov/oiea/investor-alerts-and-bulletins
Kind: official_guidance; role: official_reference; retrieved: 2026-05-25T00:00:00.000Z
Evidence records: no_safety_approval_source
Supports: No cited source provides a legal, compliance, or safety approval for the publication claim / This is one of the most reusable reliability tasks because many financial-agent failures come from overclaiming, not from missing raw data.
Limitations:
- This source entry records what the static sample cites; it does not prove current, complete, or final clearance.
- Preserve retrievedAt and linked evidenceRecordIds when reusing this source in an agent or crawler workflow.
FINRA artificial intelligence and investment fraud
URL: https://www.finra.org/investors/insights/artificial-intelligence-and-investment-fraud
Kind: official_guidance; role: primary_warning; retrieved: 2026-05-25T00:00:00.000Z
Evidence records: ai_fraud_overclaim_warning
Supports: FINRA artificial intelligence and investment fraud guidance warns that AI trading claims can be used to sell unrealistic promises / This is one of the most reusable reliability tasks because many financial-agent failures come from overclaiming, not from missing raw data.
Limitations:
- This source entry records what the static sample cites; it does not prove current, complete, or final clearance.
- Preserve retrievedAt and linked evidenceRecordIds when reusing this source in an agent or crawler workflow.
FINRA investment group imposter scams
URL: https://www.finra.org/investors/insights/investment-group-imposter-scams
Kind: official_guidance; role: primary_warning; retrieved: 2026-05-25T00:00:00.000Z
Evidence records: investment_group_imposter_warning
Supports: FINRA investment group imposter scams guidance is relevant when claims are distributed through social or private groups / This is one of the most reusable reliability tasks because many financial-agent failures come from overclaiming, not from missing raw data.
Limitations:
- This source entry records what the static sample cites; it does not prove current, complete, or final clearance.
- Preserve retrievedAt and linked evidenceRecordIds when reusing this source in an agent or crawler workflow.
Claim-safety pack draft review
URL: /agent-verification/materials/claim-safety-pack
Kind: sample_page; role: sample_context; retrieved: 2026-05-25T00:00:00.000Z
Evidence records: timestamp_and_scope_missing
Supports: Retrieval timestamp and limitation language are missing from the publication draft
Limitations:
- This source entry records what the static sample cites; it does not prove current, complete, or final clearance.
- Preserve retrievedAt and linked evidenceRecordIds when reusing this source in an agent or crawler workflow.
FTC cryptocurrency scam guidance
URL: https://consumer.ftc.gov/articles/what-know-about-cryptocurrency-scams
Kind: official_guidance; role: primary_warning; retrieved: 2026-05-25T00:00:00.000Z
Evidence records: guarantee_language_upgrade
Supports: The draft upgrades absence of negative evidence into a guarantee of protection or safety / This is one of the most reusable reliability tasks because many financial-agent failures come from overclaiming, not from missing raw data.
Limitations:
- This source entry records what the static sample cites; it does not prove current, complete, or final clearance.
- Preserve retrievedAt and linked evidenceRecordIds when reusing this source in an agent or crawler workflow.

Sample bounded output

Verdict: escalate

The publication draft overstates what the cited materials establish and should be rewritten before release.

Safe restatement: The cited material supports only a bounded, source-scoped description of the sample artifact at the listed retrieval time, with explicit limitations; it is not advice, not approval, and not a promise of user safety or outcomes.

Blocked restatement: This strategy is verified and safe for users, guaranteed to protect them, compliant for publication, approved for use, and suitable for the audience.

Unresolved items

No cited source proves safety, suitability, or guaranteed protection for users.
The claim is not anchored to retrieval time, scope, or dataset limits.
The wording converts static sample material into a universal trust statement.

Required follow-up

Replace guarantee language with a source-bound restatement tied to the cited artifact and retrieval time.
Add limitation language covering static-sample scope and lack of live review.
Escalate any compliance or safety wording to human review before publication.

Review boundaries

How this pack should be used in practice

These sections turn the pack into a bounded public review aid: what to check, when to escalate, and which statements must stay inside evidence limits.

Review checklist

Classify the claim category, preserve the exact wording, and compare it against the source scope before repeating it.
Check whether the wording upgrades sample evidence into proof, guarantees, suitability advice, compliance clearance, or safety approval.
Preserve retrieval time, source scope, audience, context, and limitation language whenever a claim is repeated.
Flag legal, compliance, suitability, or safety conclusions that are not directly supported by the cited material.

Escalation triggers

The statement promises returns, low risk, safety, or suitability.
The claim converts a static sample or one source into a universal conclusion.
The language turns an absence of evidence into a safety or compliance clearance claim.

Safe to repeat

The evidence supports a bounded observation with stated limitations and retrieval time.
This material can describe what was observed, but it does not guarantee future outcomes or user suitability.
The claim should stay tied to the cited source rather than being upgraded into approval language.

Blocked claims

This strategy is verified and safe for users.
This result proves the asset will go up.
The public material certifies compliance, legality, or guaranteed protection.

Related public artifacts

Reliability Corpus page (HTML page)
Reliability Corpus JSON sample (JSON artifact)
Evidence Cards page (HTML page)
Evidence Cards JSON sample (JSON artifact)

Machine-readable

Static JSON artifact

The JSON artifact is the stable machine-facing handoff surface for this pack. It is static, public, and safe to crawl.

Artifact path: /reliability/public-materials/claim-safety-pack.json

Use the HTML page for explanation and the JSON artifact for machine retrieval.

Artifact provenance

schemaVersion: trading4ai-public-reliability-v1

artifactVersion: public-material-claim-safety-pack-2026-05-25

generatedAt: 2026-05-25T00:00:00.000Z

staticSnapshotAt: 2026-05-25T00:00:00.000Z

artifactUrl: /reliability/public-materials/claim-safety-pack.json

Citation fields: artifactVersion, generatedAt, sampleEvidenceRecords[].retrievedAt, sampleReviewOutput.verdict, sourceRefs

When citing this pack, preserve artifactVersion, generatedAt, sampleEvidenceRecords[].retrievedAt, and the blocked-versus-safe wording boundary.

Keep unresolved evidence visible instead of smoothing the pack into marketing, guarantee, or compliance language.