Skip to content

Describe a product. Get its HTS or TARIC code, the reasons and the sources.

HTS Pilot reads a product description and proposes a declaration code for the United States, the European Union, Vietnam or Japan: 10 digits for the US and the EU, 8 for Vietnam, 9 for Japan. Each code comes with up to 10 alternatives, the reasons, the sources used and a confidence score. Cases it cannot settle go to a reviewer.

  • US HTSUS, EU TARIC, the Vietnam and Japan tariffs
  • Multilingual interface and input
  • An audit log of every decision

Results are suggestions for reference, not official classification decisions.

  1. Product description
    • Written in any of four languages
  2. HS code, 6 digits
    • International, the same in every market
  3. National line, 8 to 10 digits
    • US HTSUS, EU TARIC, the Vietnam or Japan tariff
    • One market and one tariff version
  4. Evidence
    • Reasons for every ruled-out code
    • Cited sources
    • A confidence score
  5. Review gate
    • Unclear cases go to a reviewer
    • Every decision is logged
An example, not a live result: one description, layer by layer, down to the review gate.

A code alone cannot be checked

The declarant answers for the code, so a code needs its grounds. Every proposed code comes with four things.

  • The reasonsWhy the proposed code was chosen and why each alternative was ruled out.
  • The sourcesLinked sources, each with the date it was accessed.
  • A reviewer for unclear casesWeak grounds or near-equivalent codes go to a review queue instead of a guessed code.
  • An audit trailEach decision is stored with the tariff version, its effective date and who made it.

HTS code lookup, batch classification and duty estimates

From one product to a whole catalog, with a person deciding the unclear cases.

  • Single lookup

    One description in, a declaration code out, with its grounds.

    • Up to 10 alternatives
    • Reasons and sources
    • Follow-up questions
  • Batch lookup from Excel or CSV

    Classify a whole spreadsheet in the background.

    • Column mapping
    • Per-row progress and retry
    • Excel export
    See the flow
  • Review queue and audit log

    Unclear cases go to a reviewer. Every decision is logged.

    • Accept, override, reject, reopen
    • Tariff version and date
    • Who decided
  • Code search and tree

    Browse the schedule by keyword or code.

    • Tariff line descriptions
    • Duty rates
  • SKU catalog and tariff changes

    Keep the codes you already use current.

    • Code checks
    • Version comparison
    • Alerts for affected SKUs
  • Duty and landed-cost estimates

    An estimate for every candidate code.

    • MFN, FTA or Column 2
    • Chapter 99 duties
    • MPF and HMF
    See the flow
  • API, API keys and webhooks

    Connect your ERP or PIM system.

    • REST API
    • Bulk SKU sync
    • Signed webhooks
    See the flow
  • Roles and usage credits

    Decide who reviews, and see what is used.

    • User, reviewer, admin
    • Credits per lookup
    • Your own usage

How a lookup is processed

Each stage narrows what the next one sees. The language model only compares codes that survived the earlier stages, and fixed checks run on its answer.

  1. Understand

    Facts from the description. No code is chosen yet.

    1. Product description

      Any of four languages. One market and one tariff version per lookup.

      • One market
      • One version
    2. Attribute extraction

      Rules first. A model value is kept only when it quotes the description.

      • Rules first
      • Quoted evidence only
  2. Narrow

    From the whole schedule to a short list of candidate lines.

    1. Schedule narrowing

      The search is kept to the best chapters, headings and 6-digit subheadings.

      • Chapter
      • Heading
      • Subheading
    2. Retrieval and rerank

      Candidate lines from search and precedents, ordered by a learned reranker.

      • Keywords
      • Embeddings
      • CBP rulings
      • Reviewer decisions
    3. Legal rules

      Excluded codes never reach the language model.

      • GRI
      • Section notes
      • Chapter notes
  3. Decide and check

    From the short list to one checked code.

    1. The judge compares

      A model compares what is left. It may only pick a listed code and cite what it was given.

      • Heading
      • Subheading
      • National line
    2. Validation checks

      The code must be a declarable line, and every citation must match its source.

      • Declarable line
      • Citations
      • Evidence

The result

ExampleAn example, not a live result. The confidence figure is illustrative.

Proposed

A declaration code, up to 10 alternatives, the reasons, the sources and a confidence score.

6109.10.00.12Men's or boys', other T-shirts, men's
Confidence90%

The product is a T-shirt, which heading 6109 names.

  • knitted
  • cotton
  • men's
Alternatives
  • 6109.10.00.14Men's or boys', other T-shirts, boys'
  • 6109.10.00.40Women's or girls', other T-shirts, women's
Sources
  • Chapter 61 notes
  • USITC HTS
  • CBP ruling

Routed to a person

Needs more info, with one question to answer, or Needs expert review.

Needs more info

Men's cotton shirt

Is the fabric knitted or woven?

  • Knitted: chapter 61
  • Woven: chapter 62

Needs expert review

Gift set: ceramic mug and a tin of tea

A set is flagged for a reviewer rather than decided automatically.

Review queue

One lookup, start to finish

The T-shirt example, played through the five stages. Pick a stage to jump to it.

Stage 1 of 5
An example, not a live result. The codes and their wording were checked against the USITC schedule and the September 2026 TARIC extraction; the confidence figure is illustrative.
  1. Rules and a model find the material, construction, user and state. A missing deciding fact becomes one question.

  2. Scoring, search and precedents build a ranked pool of candidate lines.

  3. Notes and the General Rules of Interpretation exclude codes before the model sees them.

  4. Heading, subheading, then national line, with the deciding factors stated.

  5. A checked code is proposed. Weak grounds or near-equivalent codes go to a person.

The last stage: a code is proposed, or a person decides

A code is marked Proposed only when every deciding attribute is present, no near-equivalent candidate remains and a source confirms it. Everything else is routed.

Every lookup ends in one of three statuses

  • ProposedA declaration code with its reasons, sources and confidence.
  • Needs more infoYou get the question that separates the candidate codes. Answer it and run the lookup again.
  • Needs expert reviewThe lookup goes to the review queue.

A reviewer can

  • Accept the proposed code
  • Set a different code
  • Reject
  • Reopen

The audit log keeps

  • the decision
  • the tariff version
  • its effective date
  • who decided

Stored decisions are used to retrain the ranking and the confidence calibration.

How it fits into your work

From a spreadsheet to your own systems, and what a duty estimate is made of.

Batch lookup, step by step

Excel or CSV in, an Excel file of proposed codes out.

  1. TemplateUse the template or your own file.
    • Excel
    • CSV
  2. Map columnsMatch your column names to the fields.
  3. Check rowsBlank, invalid and duplicate rows are caught.
  4. ProcessIn the background, with per-row progress and retry.
  5. ExportResults with the estimated duty of each candidate.
    • Excel

API and webhooks

Your ERP or PIM system sends products. Signed webhooks report back.

  1. Your ERP or PIMSends SKUs and product descriptions.
  2. REST APIAPI key access: lookups, bulk SKU sync, duty estimates.
    • JSON
    • OpenAPI
  3. HTS PilotClassifies, checks existing codes, tracks tariff changes.
  4. Signed webhooksHMAC-SHA256 signature, delivery attempted up to 3 times.
    • lookup.completed
    • lookup.decided
    • batch.completed
    • tariff.changed
    • sku.alert

Each event is delivered to an endpoint of your own system.

What a US duty estimate is made of

For each candidate code. An estimate, not an assessment by customs.

Based on
  • Code
  • Origin
  • Entry date
  • Value
  • DutyAt the MFN, FTA preference or Column 2 rate
  • Chapter 99Added only where it certainly applies. Section 301 and 232: “may apply”
  • MPFMerchandise Processing Fee
  • HMFHarbor Maintenance Fee
  • Estimated duty and feesShown for every candidate code
Not included
  • Anti-dumping and countervailing duties
  • Quotas
  • Domestic taxes

EU estimates use the CIF value in EUR and the most favourable applicable rate, add origin-specific additional duties and report anti-dumping ranges. Import VAT and excise are not included.

Vietnam estimates use the MFN rate on a customs value in VND. FTA rates by origin, the ordinary rate, VAT, special consumption tax and environmental protection tax are not included.

Japan estimates use the lowest rate the origin can claim, on the CIF value in JPY, before Japan Customs' rounding. Consumption tax, excise taxes and special duties are not included.

Standards and coverage

Four markets are loaded: the United States, the European Union, Vietnam and Japan. Every conclusion uses exactly one tariff dataset, one market and one version, and markets are never mixed.

  1. 6109.10Harmonized System, 6 digits, the same in every market
  2. 6109.10.00HTSUS, 8 digits: the line that carries the duty rate
  3. 6109.10.00.12HTSUS, 10 digits: the statistical line you declare
One product, four markets. The first six digits are the international HS code; the digits after them belong to the market.

International basis

Harmonized System (WCO), 6 digits
The schedule is narrowed chapter, heading, then subheading before a national line is chosen.
General Rules of Interpretation and legal notes
Loaded with each dataset and applied as rules that exclude or penalise codes.

United States

HTSUS, 10-digit statistical lines
The current USITC release (hts.usitc.gov) with its effective date, checked for new releases on a schedule.
CBP rulings (CROSS)
Rulings from rulings.cbp.gov serve as precedents. Revoked rulings are ignored.
PGA and AD/CVD hints
Flagged from a reference list keyed by code.

European Union

Combined Nomenclature, 8 digits, and TARIC, 10 digits
The monthly TARIC extraction of the European Commission (DG TAXUD), shown EU-style: 6109 10 00 10.
Measures loaded with it
Third-country duty, preferences, quotas, suspensions, additional and anti-dumping duties, and import controls such as CITES, CBAM and sanctions.

Vietnam

8-digit tariff lines
The 2026 Vietnam import and export tariff, built on the ASEAN Harmonised Tariff Nomenclature (AHTN 2022) and updated 5 April 2026, shown as 6109.10.10. Line descriptions and the section and chapter notes are in Vietnamese and English, and lines can be searched in either.
Rates and import taxes on each line
The MFN and ordinary rates, FTA rates under 19 agreements where the tariff gives one, and VAT, special consumption tax and environmental protection tax as the tariff lists them. A duty estimate in VND uses the MFN rate.
Source and status
Loaded from a compiled tariff workbook, not an official data service. A Vietnam result quotes it and is marked “No official source”, with its confidence capped, and the legal texts prevail. A new release is loaded by hand as a new workbook; there is no scheduled check.

Japan

9-digit statistical lines
Japan's Tariff Schedule as Japan Customs publishes it, edition of 8 August 2026, shown as 6109.10.020. A lookup runs in its English or its Japanese edition: “Japan (Tariff Schedule) - English” or “Japan (Tariff Schedule) - 日本語”.
Rates and the duty estimate
The General, Temporary, WTO, GSP, LDC and EPA rates of each line. A duty estimate in JPY takes the lowest rate the origin can claim, on the CIF value.
Source and status
Japan Customs publishes the schedule for reference only, not for official use, and every estimate repeats that notice. No section or chapter notes are loaded, so the legal stage applies the General Rules of Interpretation alone. A new edition is added by hand; there is no scheduled check.

Languages

Product descriptions
English, Vietnamese, Japanese or Korean. Japanese and Korean are rendered into English before the rules run.
Interface
The application and these pages, in the same four languages.

Where the EU tariff applies

TARIC is one tariff for the whole customs union. These are the 27 member states it covers, not separate national schedules.

Austria, Belgium, Bulgaria, Croatia, Cyprus, Czechia, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden

Not supported yet

No other national tariff is loaded. That includes Korea, China, the United Kingdom and Canada. An administrator can load a further national tariff from an Excel or CSV file; until one is loaded, lookups for that market are not available.

Browse the loaded tariff schedules

The US, EU and Vietnam schedules loaded in HTS Pilot are published as pages. Browse them by chapter and heading, with the rates as loaded, the version and the source of each schedule. Reading them is free and needs no account.

Plans and pricing

A free trial of 14 days with 10 credits and no card. Starter is USD 49 a month, Team USD 199 a month, Enterprise by quote. One lookup uses one credit.A free trial of 14 days with 10 credits and no card. Starter is VND 890,000 a month, Team VND 3,990,000 a month, Enterprise by quote. One lookup uses one credit.

Billing period

The free trial is open now. Outside Vietnam, Starter and Team are bought in the app under Settings, Billing, through Paddle, our reseller and Merchant of Record. In Vietnam they cannot be bought online yet: write to us with your question, and we answer by email. Enterprise is sold under a separate contract, arranged directly with us.

Prices are shown here in USDVND. The billing country decides the price list: VND in Vietnam, USD elsewhere. Both lists exclude taxes. The comparison of plans states both lists.

Terms of Use and Refund Policy

  • Free trial

    For one person who wants to test HTS Pilot on their own products.

    Free for 14 days, no card needed

    • 10 credits, granted once
    • 1 user
    • Catalog of up to 25 SKUs
    • Batch files of up to 20 rows
    Sign in to try it
  • Starter

    For one person who classifies new products regularly.

    USD 49 a month or USD 490 a year, two months free

    USD 490 a year or USD 49 a month

    VND 890,000 a month or VND 8,900,000 a year, two months free

    VND 8,900,000 a year or VND 890,000 a month

    • 100 credits a month
    • 1 user
    • Catalog of up to 500 SKUs
    • Batch files of up to 100 rows
    • 1 API key
    Choose this planAsk about a plan
  • Suggested for teams

    Team

    For a broker's desk or compliance team with a reviewer.

    USD 199 a month or USD 1,990 a year, two months free

    USD 1,990 a year or USD 199 a month

    VND 3,990,000 a month or VND 39,900,000 a year, two months free

    VND 39,900,000 a year or VND 3,990,000 a month

    • 500 credits a month, shared by the team
    • Up to 5 users
    • Catalog of up to 5,000 SKUs
    • Batch files of up to 500 rows
    • Up to 3 API keys and 3 webhooks
    Choose this planAsk about a plan
  • Enterprise

    For larger organizations that need their own volume and terms.

    By quote Under a separate contract

    • Credits, users, catalog size and API limits set in the contract
    • Batch files of up to 2,000 rows
    Talk to us

Compare plans

Not every limit is applied yet. Until a plan is bought, an account has every feature of the comparison below, within each user's role and the terms of the free trial. A paid plan applies its limits on batch rows, API keys and seats; the limits on SKUs and webhooks are not applied yet.

How credits are counted today

  • One lookup, one credit A single lookup, a batch row and a SKU classification each use one credit.

  • A failed lookup is not charged A lookup that ends in an error uses no credit.

  • A routed result counts A lookup that ends in Needs more info or Needs expert review uses a credit, like a proposed code. Running it again with added facts is a new lookup and uses another credit.

  • Looking up the schedule is not charged Code search, the code tree, duty estimates and checks of codes already in your catalog use no credits.

At zero credits lookups stop. A lookup is refused when the balance does not cover it. There is no overage charge.

Compare plans

What each plan includes, with both price lists. Until a plan is bought, an account has every feature listed here, within each user's role and the terms of the free trial.

  • A highlight of this plan
  • Not in this plan
FeatureFree trialStarterTeamEnterprise
Price, excluding taxes
Price a monthFree for 14 days, no card neededUSD 49 / VND 890,000USD 199 / VND 3,990,000By quote
Price a year, two months free Not in this planUSD 490 / VND 8,900,000USD 1,990 / VND 39,900,000By quote
In every plan
  • Single lookups with reasons, sources and confidence
  • Batch lookups from Excel or CSV
  • US HTS, EU CN/TARIC, Vietnam and Japan
  • Duty and landed-cost estimates
  • SKU catalog with code checks
  • Alerts when a tariff change affects a code in use
  • The interface in four languages
Credits and users
Lookup creditsA highlight of this plan. 10 credits, granted once100 credits a month500 credits a month, shared by the teamBy quote
Users1 user1 userUp to 5 usersBy quote
Catalog and batch limits
SKUs in the catalog255005,000By quote
Rows per batch file201005002,000
Review and team
User, reviewer and admin roles Not in this plan Not in this planIncluded. A highlight of this planIncluded
Review queue with an audit log Not in this plan Not in this planIncluded. A highlight of this planIncluded
Productivity reports Not in this plan Not in this planIncludedIncluded
Integration
REST API and API keys Not in this plan1 API keyUp to 3 API keysBy quote
Signed webhooks Not in this plan Not in this planUp to 3 webhooksBy quote
Support
Support channelContact formBy emailBy emailA named contact

Frequently asked questions

Short answers for customs brokers, compliance teams, sellers and integrators. Each one links to the section or page with the detail.

The product

What does HTS Pilot do?

It reads a product description and prepares a classification proposal: a declaration code (10 digits for the US and the EU, 8 for Vietnam, 9 for Japan), up to 10 alternatives, the reasons each code was chosen or ruled out, the sources used and a confidence score. The declarant, broker or compliance team decides with that evidence in front of them, and cases the system cannot settle go to a reviewer.

See the capabilities
Which markets does it cover, and how current is the tariff data?

Four markets: the United States (HTSUS, 10 digits), the European Union (Combined Nomenclature, 8 digits, and TARIC, 10 digits), Vietnam (8 digits) and Japan (9 digits). The US data is the USITC release and the EU data is the monthly TARIC extraction of the European Commission; both are checked for new releases on a schedule. The Vietnam data is a compiled workbook of the 2026 Vietnam import and export tariff, updated 5 April 2026, not the legal text, with its rates and its notes in Vietnamese and English; a new release is loaded as a new workbook, not on a schedule. The Japan data is Japan's Tariff Schedule from Japan Customs, edition of 8 August 2026, in English and in Japanese; a new edition is added by hand, not on a schedule. In every market the data can be behind the official publication, which always prevails.

Standards and coverage
Which languages can a product description be written in?

English, Vietnamese, Japanese or Korean. Japanese and Korean descriptions are rendered into English before the rules run. The interface is available in the same four languages.

Languages in the coverage section

Trusting a result

Where a person takes over

HTS Pilot prepares the classification: the proposal, the reasons and the sources. The decision stays with the people who answer for it.

  • The declarant and the customs authority decide. HTS Pilot does not replace either. A qualified person should make the final call; for high-value or complex goods, request an advance ruling.

  • The description sets the quality of the result. Without the material, percentages, construction, use or target user, a product cannot be classified with certainty, so the system asks for what is missing.

  • Some goods always go to a reviewer. Multi-material goods, sets, parts and unfinished goods are flagged for a reviewer rather than decided automatically.

  • The language model can be wrong, so its reasoning is shown. Validation blocks invented codes and invented sources, but it cannot check the reasoning about a real code.

  • Duty figures are estimates. US estimates leave out anti-dumping and countervailing duties, quotas and domestic taxes. EU estimates leave out import VAT and excise. Vietnam estimates use the MFN rate and leave out FTA rates, VAT and the other import taxes. Japan estimates leave out consumption tax, excise taxes and special duties and come before Japan Customs' rounding.

Is the result an official classification?

A result is a documented proposal: the code, the reasons for it and against each alternative, the sources used and a confidence score, for the person who signs the declaration to review. It is a suggestion for reference only, not an official classification decision, a binding ruling, or legal or customs advice: the declarant and the customs authority decide. Where certainty is needed, a binding ruling from the customs authority is the route.

Terms of Use, section 2
How do I see why a code was proposed?

Each result states why the proposed code was chosen and why each alternative was ruled out, with the deciding factors. Its sources are linked, each with the date it was accessed. US and EU results link official sources: the USITC schedule, CBP rulings and the TARIC pages of the European Commission. A Japan result links the tariff pages of Japan Customs, published for reference only. A Vietnam result quotes the tariff workbook, which is not official evidence, and is marked “No official source”. The language model may only pick a code from the candidate list and cite the notes, sources and precedents it was given.

How a lookup is processed
What happens when the system is not sure?

It tells you, and it tells you what is missing. If a fact that decides the classification is missing, the status is Needs more info and you get the question that best separates the candidate codes. Weak grounds, near-equivalent codes or conflicting evidence give Needs expert review, and the lookup goes to the review queue for a person to decide.

The review workflow
Can it be wrong?

Before a result is shown, an automatic validation step confirms that the code exists in the loaded tariff schedule and that every cited source is one the system was given, so invented codes and invented sources are blocked. It cannot confirm that the reasoning about a real code is right, and the service works from the description you give, not from the goods. Unclear cases are routed to review instead of being guessed, the alternatives and why each was ruled out are shown, and a qualified person should make the final call.

Where a person takes over
When should I ask the customs authority for a binding ruling?

We recommend it for high-value, high-volume, complex or novel goods, or when the candidate codes carry very different duty rates or controls. The instrument depends on the market: an advance ruling from U.S. Customs and Border Protection (CBP) in the United States, a Binding Tariff Information (BTI) decision from the customs authority of a Member State in the European Union, or, in Vietnam, an advance ruling on the code issued by the Customs Department. Each has legal effect before the customs authority only for the goods it describes. A result from HTS Pilot is not a ruling.

Terms of Use, section 2

Using it

Can I classify a whole file instead of one product?

A batch lookup takes an Excel or CSV file, so a whole product list is classified in one run. There is a template, column mapping and checks for blank, invalid and duplicate rows. The file is processed in the background with per-row progress and retry, and the results are exported to Excel.

The batch flow, step by step
Can I check the codes I already use?

The SKU catalog checks each code you already use against the loaded schedule: valid, not specific enough, expired, nonexistent or changed. When a new tariff version is loaded it is compared with the previous one, and alerts are raised on the SKUs it affects.

SKU catalog and tariff changes
What do the duty and landed-cost estimates leave out?

Each candidate code comes with a duty and landed-cost estimate, so the cost of the alternatives can be compared. These are estimates, not an assessment by customs. US estimates leave out anti-dumping and countervailing duties, quotas and domestic taxes, and show Section 301 and 232 duties as “may apply”. EU estimates leave out import VAT and excise. Vietnam estimates use the MFN rate on a customs value in VND and leave out FTA rates, VAT and the other import taxes; the FTA rates and those taxes are shown with each line. Japan estimates take the lowest rate the origin can claim, on the CIF value in JPY, before Japan Customs' rounding, and leave out consumption tax, excise taxes and special duties. Trade preferences depend on rules of origin that the service does not verify.

What a duty estimate is made of
How do roles and the audit log work?

There are three roles: user, reviewer and admin. A reviewer can accept the proposed code, set a different code, reject or reopen. Each decision is stored with the tariff version, its effective date and who made it.

The review workflow
Can I connect my own systems?

A REST API with API keys connects your own systems. There is a bulk endpoint that syncs SKUs and classifies those without a code, and webhooks are signed with HMAC-SHA256, with delivery attempted up to 3 times. An administrator creates the keys, and a key never has more permissions than its creator.

API and webhooks

Account, credits and data

How do I get an account?

New here? Sign in with Google to start the 14-day free trial: the account is created at the first sign-in, and no card is asked for. Accounts with an email address and a password are created by an administrator.

Start free trial
How do credits work?

One lookup uses one credit, and a new account receives free credits. A failed lookup is not charged. A result of Needs more info or Needs expert review is charged, and so is a re-run with added facts. Code search, the code tree, duty estimates and checks of codes in the catalog use no credits.

How credits are counted
How much does HTS Pilot cost?

The free trial costs nothing: it lasts 14 days, gives 10 credits once and asks for no card. On the USD list, Starter is USD 49 a month or USD 490 a year, and Team is USD 199 a month or USD 1,990 a year. A customer billed in Vietnam pays the VND list: Starter is VND 890,000 a month or VND 8,900,000 a year, and Team is VND 3,990,000 a month or VND 39,900,000 a year. Both lists exclude taxes. Enterprise is by quote, under a separate contract arranged directly with us. The free trial is open now. Outside Vietnam, Starter and Team are bought in the app under Settings, Billing, through Paddle, our reseller and Merchant of Record; taxes are added at checkout. In Vietnam they cannot be bought online yet: write to us through the contact form, and we answer by email.

Plans and pricing
What happens to my product descriptions?

The description and the optional fields are sent to third-party AI inference services that we use under contract, only to produce the result; your name, email address, account identifier and IP address are not sent. The application server and the database are hosted in Singapore. To gather evidence, a few product terms taken from the description are also sent to the public rulings search of U.S. Customs and Border Protection. When a reviewer decides a lookup, the description, the extracted attributes and the decision are stored as an example used inside this service only: as a precedent for later lookups of similar products in the same workspace, and to improve the ranking and the confidence figure as the Privacy Policy sets out. Another workspace does not see it. We do not use your content to train general-purpose language models. Entries added to the SKU catalog are visible to the members of your workspace, not to other workspaces.

Privacy Policy, section 5

Did not find your question? Contact us

Try it with one of your own products

Sign in, describe the product, and read the reasoning before you rely on the code.