Free tools Windows power users keep installed
One-click scans. No signup required.
Reduce production misalignment as an operational risk, not as a prompt-writing problem. Define the system’s intended use and owners, test it under deployment-like conditions, limit what it can do, watch its behavior in real time, and keep an authorized way to pause, change, contain, or retire it. No single model choice, policy, evaluation, or human-review step can guarantee aligned behavior once an AI system meets changing users, data, tools, and incentives.
The sequence below adapts the National Institute of Standards and Technology (NIST) AI Risk Management Framework (AI RMF) to production operations. It is designed to help a team decide how much autonomy to allow, what evidence to collect, and what to do when observed behavior exceeds acceptable limits.
As an Amazon Associate I earn from qualifying purchases.
What “misaligned behavior” means in production
An AI system is misaligned when its observed behavior departs from the outcomes, constraints, or values that the responsible organization has defined for its use. The departure can be obvious, such as an agent taking an unauthorized action, or subtle, such as a classifier becoming less reliable for a user group after a data or workflow change. It can arise from the model, prompts, retrieval data, tools, interfaces, downstream automation, or the way people respond to outputs.
Production conditions make the problem different from a laboratory demonstration. Inputs are incomplete and adversarial; users discover unexpected workarounds; connected services change; and a plausible answer may trigger a financial, safety, privacy, employment, or reputational consequence. A system that appears acceptable in a static test can therefore fail when its context changes.
#1 Best Overall
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
Treat alignment as a control loop: establish a purpose and risk tolerance, measure behavior against that purpose, intervene when evidence changes, and revise or retire the system when controls no longer work. NIST states the principle directly: “Risk management should be continuous, timely, and performed throughout the AI system lifecycle dimensions.” See the NIST AI RMF Core.
Use the AI RMF lifecycle as the operating model
NIST’s voluntary AI RMF organizes risk work into four functions: Govern, Map, Measure, and Manage. The NIST AI RMF Playbook, released for AI RMF 1.0 on January 26, 2023, suggests actions that help teams reach those outcomes. The functions overlap; they are not a one-time project checklist.
| Function | Production question | Evidence to keep |
|---|---|---|
| Govern | Who is accountable, what is the risk tolerance, and who can authorize or stop the system? | Named owners, escalation paths, policies, approvals, training records, and retirement criteria. |
| Map | What is the system intended to do, who is affected, and where could expected use or foreseeable misuse cause harm? | System inventory, data and component map, use-case boundaries, impact assessment, assumptions, and known limitations. |
| Measure | How will the team detect unsafe, insecure, unreliable, unfair, or confusing behavior before and after launch? | Test sets, metrics, scenarios, red-team findings, uncertainty analysis, monitoring queries, and review records. |
| Manage | What controls reduce exposure, and what happens when evidence shows a deviation? | Permission settings, policy gates, intervention procedures, incident tickets, recovery tests, change approvals, and shutdown plans. |
Governance is continuous and cross-cutting. A model update, new tool, changed user population, or altered business objective can invalidate an earlier risk assessment, so the four functions must be revisited whenever the system or its environment changes.
Recommended Free Tools
Step 1: Define intended use, scope, and impact
Write an operational use statement
Describe the task in terms that an operator can verify: the users, geography or business unit, inputs, outputs, connected systems, expected workload, and decisions that may follow. State what the system is not allowed to do. “Helps support customer service” is too broad; “drafts replies for trained agents, without sending messages or changing account records” is testable.
Record capability and uncertainty limits alongside the intended use. Note languages, data types, edge cases, latency assumptions, and conditions under which accuracy or safety has not been established. A limit is useful only when it changes a control, such as routing a case to a person or refusing an automated action.
Map the complete system, not just the model
Inventory the foundation model, prompts, retrieval indexes, data pipelines, filters, orchestration code, tools, identity systems, vendors, and downstream automations. Mark which components are controlled internally and which can change outside the team’s release process. A third-party data feed or API can alter behavior even when the model binary is unchanged.
For each component, document the information it receives, the authority it has, and the failure modes it could introduce. Include privacy, security, reliability, fairness, accessibility, and human-AI interaction concerns where they are relevant to the use case. The NIST discussion of AI risks and trustworthiness characteristics emphasizes tailoring these considerations to context rather than applying a universal checklist.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Rank #2
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
Set impact categories and risk ownership
Estimate the severity and likelihood of plausible harm if the system is wrong, manipulated, unavailable, or used outside its intended scope. Consider direct and indirect effects: an incorrect recommendation may be low impact when a person independently checks it, but high impact when it automatically changes eligibility or access.
Set an organizational risk tolerance and assign a named owner for every material risk. The owner must have authority and resources to accept residual risk, require additional controls, or stop deployment. Record affected stakeholders and a way for them to report harm or challenge an outcome; otherwise the impact assessment is not connected to correction.
Step 2: Specify oversight and authority
Define an autonomy boundary
List actions the system may perform without approval, actions that require a policy gate, and actions that always require a qualified person. Base the boundary on potential consequence, reversibility, and the speed at which a mistake could spread. Reading public information and drafting text are generally easier to reverse than transferring funds, changing a medical record, deleting data, or communicating a binding decision.
| Action class | Typical control | Required authority |
|---|---|---|
| Informational output with no automatic downstream effect | Logging, quality checks, user disclosure, and a route to report errors. | Product owner can approve normal operation; users remain responsible for consequential use. |
| Reversible workflow action | Allow-list of tools, parameter validation, rate limits, and an undo or rollback path. | Trained operator can review exceptions and reverse the action. |
| High-impact or difficult-to-reverse action | Policy gate, dual approval where appropriate, complete audit trail, and explicit confirmation of scope. | Designated decision-maker with authority to approve, reject, or pause the workflow. |
| Action outside approved scope or showing anomalous behavior | Automatic deny or safe hold, alerting, containment, and incident response. | On-call responder can disable tools or traffic; accountable owner decides on restart. |
Make human oversight real
A “human in the loop” is not a control if the reviewer lacks time, context, training, or the ability to reject the output. Define the information the reviewer sees, the maximum workload, the response time, and the exact override mechanism. Test whether reviewers detect seeded failures rather than merely confirming normal cases. For high-consequence work, make approval an affirmative action, not a passive screen that auto-continues after a timeout.
Assign an operator who can pause traffic, revoke credentials, disable a tool, switch to a known-good version, or decommission the system. Publish contact routes and escalation thresholds so that the person observing a problem does not need to negotiate authority during an incident.
Step 3: Test before release and on a recurring basis
Evaluate under deployment-like conditions
A test set should resemble actual users, inputs, languages, workload, integrations, and operating constraints. Include long conversations, malformed requests, ambiguous instructions, missing data, tool errors, latency, retries, and partial outages when those conditions can occur in production. Document which deployment conditions are not represented and do not generalize results beyond that evidence.
Use a combination of automated metrics, expert review, adversarial testing, and end-to-end workflow tests. Examine safety, security, robustness, privacy, fairness, reliability, and human-AI interaction according to the mapped risks. A high average score can hide a severe tail failure, so report error categories, confidence or uncertainty, subgroup results where appropriate, and examples of unacceptable behavior.
Rank #3
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
Build a release gate
- Define test objectives. Tie each objective to an intended outcome, a prohibited behavior, or a known limitation.
- Assemble representative and adversarial cases. Include historical incidents, boundary conditions, abuse attempts, prompt-injection patterns, and realistic tool calls.
- Set pass, fail, and review thresholds. Specify which failures block release, which require risk-owner approval, and which are tracked as accepted residual risk.
- Run independent review. Have reviewers beyond the front-line development team examine methods, results, and unresolved limitations when the impact warrants it.
- Record the decision. Preserve the model and prompt versions, data snapshot, configuration, test results, exceptions, approvers, and expiry date for the approval.
Continue testing after launch
Re-run critical scenarios after model, prompt, retrieval, tool, policy, data, or interface changes. Schedule periodic evaluations even when nothing is intentionally changed, because user behavior, external data, and connected services evolve. Sample production cases under appropriate privacy controls, compare them with pre-release baselines, and investigate drift rather than waiting for a severe incident.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Step 4: Constrain actions and build failure paths
Use least privilege and explicit allow-lists
Give the system only the tools, data, and permissions needed for the defined task. Separate read and write credentials, restrict records by tenant and role, limit destinations, and enforce quotas for volume, spend, and execution time. Do not rely on the model to remember a permission rule; enforce it in the tool or policy layer.
Place a policy gate between generated intent and consequential execution. The gate can validate identity, target, parameters, authorization, data sensitivity, rate limits, and business rules. Require structured arguments rather than free-form commands, and reject values outside an allow-list. Log the request, decision, tool result, and actor so an investigator can reconstruct what happened.
Monitor inputs, outputs, and tool calls
Input monitoring can flag prompt injection, credential requests, sensitive data, abusive content, or patterns associated with automated exploitation. Output monitoring can detect policy violations, unsupported certainty, data leakage, or anomalous volume. Tool-call monitoring should check whether the requested action matches the user’s authority and the approved workflow, not just whether the syntax is valid.
Controls should fail safely. If a classifier, policy service, identity provider, or monitoring pipeline is unavailable, define whether the workflow blocks, downgrades to a read-only mode, queues for review, or uses a bounded fallback. A silent bypass turns a component outage into an alignment risk.
Prepare containment, recovery, and retirement
Before launch, practice disabling a tool, isolating a tenant, revoking a token, stopping a queue, rolling back a prompt or model, and restoring a known-good configuration. Preserve evidence without retaining unnecessary personal data. Define notification, customer support, appeal, and remediation procedures for affected people.
Set change-management rules for model updates, prompt edits, policy changes, data refreshes, and vendor substitutions. Define retirement triggers such as repeated control failures, unacceptable unresolved harm, loss of required oversight, or an operating context that no longer matches evaluation evidence. Safe retirement includes data handling, credential revocation, user communication, and an archive of decisions.
Rank #4
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
Step 5: Monitor production and respond to evidence
Choose signals that reveal deviation
Monitoring should cover both system health and behavior. Track latency, errors, availability, cost, and queue depth, but also signals tied to the intended use: refusal rates, escalation rates, override frequency, tool-call patterns, policy-block events, unusual destinations, user complaints, subgroup performance, and changes in output quality. Establish a baseline and alert on meaningful changes rather than every fluctuation.
| Signal | What it may indicate | First response |
|---|---|---|
| Sudden increase in blocked or refused requests | Prompt-injection activity, a policy change, a broken dependency, or a shifted user population. | Sample cases, compare configuration versions, and route uncertain traffic to review. |
| Unexpected tool-call volume or destinations | Compromised credentials, an orchestration bug, or model behavior outside its authority. | Rate-limit or disable the tool, revoke affected credentials, and preserve logs. |
| Rise in overrides, appeals, or complaints | Quality drift, unfair impact, confusing explanations, or inadequate human review. | Investigate representative cases, pause the affected decision path if harm is plausible, and notify the risk owner. |
| Performance change for a subgroup or workflow | Data drift, integration changes, or a limitation not covered by evaluation. | Re-test in the changed context and adjust routing, data, or model configuration. |
| Monitoring or policy service unavailable | Loss of a safety dependency, even if the model appears healthy. | Apply the predefined fail-safe mode; do not silently continue high-impact actions. |
Use an intervention ladder
Respond proportionately to evidence. Start with additional sampling or a temporary rate limit when the deviation is uncertain and consequences are low. Escalate to read-only mode, mandatory human approval, tool isolation, traffic reduction, or rollback when the likelihood or severity rises. Shut down the affected capability when behavior exceeds the approved boundary, monitoring cannot detect it in time, or containment cannot be trusted. The NIST guidance calls for intervention, modification, or shutdown when behavior deviates from expectations; safeguards should match the potential harm and context.
After containment, determine the triggering condition, affected scope, duration, and whether any person or property was harmed. Correct the immediate cause, add a regression test, review whether the risk assessment and autonomy boundary were adequate, and obtain explicit approval before restoring normal operation.
How to compare control options
There is no universal NIST threshold that tells every organization when to require approval or shutdown. Compare alternatives against the following dimensions, then document why the selected control is proportionate.
| Dimension | Questions to ask | Control implication |
|---|---|---|
| Potential consequence | How severe and likely is harm if the system is wrong or manipulated? | Higher consequence supports stricter gates, stronger review, narrower permissions, and faster escalation. |
| Capability and autonomy | What tools, data, and downstream decisions can the system access? Can it act without a person? | More authority requires stronger isolation, validation, logging, and reversal mechanisms. |
| Deployment similarity | How closely do tests represent real users, inputs, workload, integrations, and operating conditions? | Lower similarity means less justified autonomy and a larger monitoring and review burden. |
| Detection and response | Can monitoring identify deviation before harm spreads? Can an authorized person contain or reverse it? | Weak detection or slow response requires preventive constraints rather than reliance on after-the-fact review. |
| Scope and components | Which models, vendors, software, data sources, and services contribute to behavior? | More dependencies require clearer contracts, version tracking, fallback modes, and change notification. |
| Operational ownership | Are communication, periodic review, incident response, and decommissioning assigned and resourced? | Unowned controls are not dependable; assign an accountable owner before increasing capability. |
For each proposed control, record the expected risk reduction, operational cost, residual risk, evidence supporting the decision, and a date for reassessment. This makes a trade-off auditable without pretending that a numeric score is a universal safety guarantee.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.A practical incident runbook
When a suspected deviation appears, use a short, rehearsed sequence rather than improvising:
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →- Declare and classify. Record the time, system version, observed behavior, affected workflow, and plausible severity. Assign an incident lead and a technical lead.
- Stabilize. Apply the narrowest effective measure: pause the tool, reduce traffic, require approval, isolate a tenant, or switch to a bounded fallback. If the boundary is unclear and harm could be serious, choose the safer mode.
- Preserve evidence. Retain relevant prompts, outputs, tool arguments, policy decisions, identities, configuration, and dependency status under approved privacy and retention rules.
- Assess scope and impact. Identify affected users, records, transactions, and time window. Check whether the behavior reproduced in a controlled environment and whether monitoring missed earlier signs.
- Communicate. Notify the accountable owner, security or privacy contacts, operators, and affected stakeholders according to the incident plan. Give users a route to appeal or obtain correction when an outcome affects them.
- Correct and verify. Fix the cause, add a regression case, re-run deployment-like tests, and verify that the containment and recovery controls work before re-enabling the capability.
- Learn and govern. Update the risk map, permissions, tests, monitoring, training, and escalation criteria. Document the restart decision or retire the capability if residual risk is no longer acceptable.
Keep an evidence trail that survives staff and model changes
Maintain a system card or equivalent operational record containing the intended use, prohibited uses, component versions, data sources, evaluation scope, known limitations, owners, autonomy decisions, monitoring definitions, and emergency procedures. Link each production release to the exact artifacts tested and approved.
Best Value
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
Review the record on a defined cadence and whenever a material change occurs. Include operators, security, privacy, legal or compliance specialists where relevant, domain experts, and representatives of affected users. The purpose is not paperwork for its own sake: it lets a new reviewer understand what was assumed, what was measured, and who can act when those assumptions fail.
Common approaches that fail
- “The prompt says not to do that.” Instructions are useful but cannot enforce permissions, withstand every adversarial input, or compensate for a compromised tool. Put constraints in code, identity, and policy layers.
- “A benchmark score proves safety.” Aggregate scores do not cover every user, integration, workload, or tail failure. Use deployment-like scenarios, expert review, and ongoing production evidence.
- “Human review solves it.” Reviewers can be rushed, over-trust fluent outputs, or lack authority to intervene. Measure reviewer performance and give them actionable controls.
- “We can fix it after launch.” A system with irreversible actions or weak observability may cause harm before a fix is possible. Build rollback, containment, and shutdown paths before enabling autonomy.
- “The model is unchanged, so risk is unchanged.” Retrieval data, tools, policies, user populations, and external services can change behavior. Reassess material changes across the full system.
Production readiness checklist
- Intended use, prohibited use, affected stakeholders, capabilities, and limits are documented.
- Every material risk has an accountable owner and an agreed risk tolerance.
- All models, data, prompts, tools, vendors, and downstream automations are mapped and versioned.
- Autonomous, approval-gated, and prohibited actions are explicit.
- Reviewers have relevant proficiency, context, workload limits, and authority to reject or pause.
- Pre-release tests represent actual users, inputs, integrations, workload, and likely attacks.
- Pass criteria, exceptions, limitations, and independent review are recorded.
- Permissions are least-privilege; tool calls pass through allow-lists and policy validation.
- Logs capture decisions and actions needed for investigation without unnecessary personal data.
- Behavioral and system-health monitors have baselines, alert owners, and tested fail-safe modes.
- Operators can contain, roll back, modify, or shut down the capability quickly.
- Incident, appeal, recovery, change-management, and safe-retirement procedures have been rehearsed.
- Periodic review is scheduled, and material changes automatically trigger reassessment.
Putting the sequence into practice
Start with the highest-consequence workflow rather than attempting to solve every AI risk at once. In the first operating cycle, write the use statement and system map, assign owners, set the autonomy boundary, and test the most harmful plausible failures. Before expanding capability, demonstrate that the team can detect a deviation, contain it, explain what happened, and restore a known-good state.
Then use production evidence to refine controls. Increase autonomy only when deployment similarity, detection speed, human authority, and recovery capacity justify it. Reduce or remove autonomy when the context changes faster than the evaluation and monitoring can keep up. This makes alignment a maintained operating property—supported by governance, measurement, constrained action, and intervention—rather than a promise attached to a model at launch.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteFrequently Asked Questions
Does using a more capable or newer model eliminate production misalignment risk?
No. Model capability does not remove risks introduced by tools, data, prompts, integrations, user behavior, or changing operating conditions. The same lifecycle controls remain necessary.
How often should a production AI system be re-evaluated?
Re-evaluate after model, prompt, retrieval, tool, policy, data, or interface changes, and on a periodic schedule even when no intentional change was made. The interval should reflect consequence, rate of change, and the time available to detect and contain harm.
What should a small team do if it cannot build extensive monitoring?
Narrow the use case and permissions, require approval for consequential actions, use simple high-value alerts and logs, and provide a tested pause or shutdown path. Limited observability is a reason to reduce autonomy, not to operate as if risk were unchanged.
Is the NIST AI RMF mandatory?
The AI RMF is voluntary guidance. Whether other obligations apply depends on the system, sector, contracts, and jurisdictions involved; the framework does not determine those legal requirements.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
