Digital intelligence resource centre

Core Web Vitals: interpret thresholds and measurement scope

With LCP = 3.2 s, INP = 180 ms and CLS = 0.08, LCP needs improvement and the other two values are good. This fictional set is not a measurement of CIE.

Recovery: separate restoration time from data age

Fictional objectives: RTO 120 min and RPO 30 min. Recovery takes 95 min, but recovered data is 45 min old: a 25-minute time margin and a 15-minute data overrun. The exercise therefore does not meet both objectives.

AI answers: make abstentions and denominators visible

Of 100 fictional questions where an answer is expected, the assistant answers 80 and 72 are validated. Quality among answers: 72/80 = 90%. Coverage: 80/100 = 80%. Success across the set: 72/100 = 72%. The 90% figure alone hides twenty unanswered questions.

Real photograph: separate image, caption and publishing page

Exercise: someone attributes this copy to Apollo 8 because it is titled “Earthrise”. The selected source page identifies Lunar Orbiter 1. You can flag that conflicting attribution without claiming to have verified the entire file history.

Read permissions through the actual task

A fictional extension must annotate the selected page after a click. The team compares an activeTab and scripting variant with one requesting standing access to every site. It also checks the absence of automatic scripts, transmitted data and annotation persistence. Narrower access does not remove the need for these checks.

From a received log to a verified response

In the fictional exercise below, a rule expects two failed sign-ins followed by a success for the same account within one minute. One failure arrives late. The team checks late-event handling before claiming the scenario is detected. This teaching rule alone cannot identify a real attack.

A zero rate does not establish zero risk

In a fictional sample of 20 independent requests, none was misclassified. The observed rate is 0%, yet the upper bound of the two-sided 95% Wilson interval is about 16.1%. This observation alone does not justify a claim of error-free operation.

False positives: count the cases before judging an alert

The {alerts} alerts include {tp} positive cases and {fp} false positives. Each alert still needs qualification; neither recall nor specificity directly gives its precision.

Establish the facts

Start from traceable sources, verify information and qualify dependencies before deciding.

1. OSINT guide: produce reliable decision intelligence
2. Information integrity: verify provenance, context and transformations
3. Digital due diligence: a trust checklist before commitment

Prepare response and recovery

Identify exposed assets, triage an incident signal and prove that the service can return.

1. Cyber exposure assessment: a prioritised checklist
2. Triage a cyber incident and preserve early evidence
3. Test a backup restoration from copy to working service
4. Coordinated vulnerability disclosure: receive, fix and inform

Build a usable service

Connect speed, security, accessibility and integration continuity to actual user journeys.

1. Secure, high-performance web: a journey-led audit guide
2. Evaluate web accessibility through journeys and evidence
3. Make API integrations resilient to outages

Deploy AI with evidence

Choose a use, evaluate answers and retain control of data, actions and suppliers.

1. AI agent governance: a production readiness guide
2. RAG quality: evaluate a knowledge assistant with evidence
3. AI vendor evaluation: evidence, dependencies and exit conditions
4. Accessible conversational assistants: multiple modes, recovery and human handoff

Reduce entry paths

Remediate vulnerabilities found on actual assets and strengthen sign-in without leaving weak recovery paths.

1. Prioritize vulnerabilities and verify remediation
2. Deploy phishing-resistant authentication with safe recovery
3. Identity and account recovery: govern the complete lifecycle

Measure and decide

Define events, respect privacy choices and run an interpretable experiment.

1. Build an event measurement plan and verify collection
2. Measure audiences while respecting privacy choices
3. Run an A/B experiment that supports a sound decision

Make commerce journeys reliable

Check published offers and verify that a task remains complete in every language.

1. Improve multichannel product catalog data quality
2. QA multilingual content and complete user journeys

Detect with evidence

Connect technical records to useful signals and a rehearsed response.

1. Make security event logs ready for detection
2. Triage a cyber incident and preserve early evidence

Keep pages findable and useful

Diagnose crawling and indexing, then maintain accurate content that meets a real need.

1. Diagnose crawling and indexing of useful pages
2. Govern and maintain a useful content library

Make every interaction reliable

Help people recover from form errors, evolve API contracts and verify email sending.

1. Design accessible forms and recoverable errors
2. Evolve API contracts without losing clients
3. Improve email sender trust and deliverability

Care for communities

Use clear rules and proportionate escalation for sensitive situations.

1. Moderate a community and escalate sensitive cases
2. Information integrity: verify provenance, context and transformations

Secure data graphs

Test permissions, query cost and each field’s exposure before opening a GraphQL API.

1. Secure a GraphQL API: authorization, query cost and evidence
2. Evolve API contracts without losing clients

Align sold variants

Connect catalog, URL, published offer and cart for every available option.

1. Keep product variants consistent from URL to cart
2. Improve multichannel product catalog data quality

Diagnose real experience

Start from observed visits, then reproduce responsible interactions and elements.

1. Diagnose Core Web Vitals in field data
2. Secure, high-performance web: a journey-led audit guide

Triage public signals

Keep provenance, check facts and route sensitive cases.

1. Triage social listening signals before acting
2. Moderate a community and escalate sensitive cases