Answers, in writing.

You must be at least 18, hold the legal right to work as a self-employed contractor where you live, and be resident in one of the 130 countries Cortext can pay into. Most projects also require professional English at roughly C1 — you will be reading rubrics written in English and defending your reasoning in writing.

Beyond that, eligibility is per project, not per person. A cardiology evaluation project requires a medical licence in good standing; a competitive-programming project requires a rating threshold or an equivalent portfolio. Cortext cannot admit residents of jurisdictions under comprehensive sanctions, and it does not sponsor visas or offer employment of any kind.

Five kinds of work, in rough order of volume: evaluating model output against a rubric, writing reference answers a model is trained to imitate, authoring and revising rubrics themselves, red-teaming systems to find failure modes, and annotating agent trajectories step by step to mark where a chain of tool calls went wrong.

None of it is data entry, transcription or content moderation of user-generated media. The reason projects pay an average of $68 an hour is that they need a physician, an attorney, a quant or a research engineer to say why an answer is wrong, in writing, in a form another expert can audit.

No. There is no application fee, no equipment fee, no training fee, no "verification deposit" and no paid tier that moves you up a queue. Cortext is paid by the AI labs that commission projects; contributors are paid, never charged.

If anyone asks you to pay to be onboarded, to buy a device, to accept a cheque and forward part of it, or to send cryptocurrency, it is a scam impersonating Cortext. Forward the message to security@cortext-ai.uk and do not reply to it.

An email address you control, your legal name as it appears on government ID, your country of tax residence, and a CV or a profile link that shows the work you are claiming. Depending on the projects you apply to, you will also be asked for a licence number, a registry link, a degree certificate or a public code repository.

There is no phone number field, because Cortext does not use phone numbers for verification, login or support. Everything runs through email so there is a written record on both sides.

Labs contract for work done by a specific, identified expert, and the contributor agreement is signed by a named person. Identity checks also stop the two most common forms of platform fraud: one person running several accounts, and someone passing an assessment on another person's behalf before handing the account over.

The check runs through a third-party verification vendor. Cortext receives a pass or fail, the document type, the issuing country and the expiry date; the image of the document itself stays with the vendor and is deleted on their 30-day schedule. Cortext never stores a scan of your ID.

Most contributors do. Cortext asks for no exclusivity, no non-compete and no minimum weekly commitment unless a specific project posting states one, and the contributor agreement deliberately omits all three.

Two things are your responsibility, not ours: your employer's moonlighting and IP policy, and the 80-hour weekly ceiling Cortext enforces across all of your contracts combined. If your employment contract assigns your out-of-hours output to your employer, resolve that before you accept a project — Cortext cannot deliver work to a lab that a third party may claim.

Cortext pays contributors in 130 countries. 113 of them settle through Stripe; the rest are covered by Wise, including Brazil, Georgia, Ghana, Kosovo and Nepal, Türkiye, United Arab Emirates. If your country appears in neither list, the account can exist but no payout can be released, so we do not admit applicants from those countries.

A handful of projects carry data-residency restrictions from the commissioning lab and are only visible to contributors in named regions. That restriction is written on the posting, not applied silently after you apply.

Update your country of tax residence in account settings before the next Friday run. That change re-triggers sanctions screening, may switch your payout rail between Stripe and Wise, and always invalidates your existing tax form — a US person moving abroad still files a W-9, but a non-US person moving to the US must replace their W-8BEN with one.

Payouts are held, not lost, while the new details clear. Hours already tracked stay in the ledger and release on the first run after screening completes, which usually takes two business days.

The median across the whole network is 17 days from application to first paid task. That covers credential verification, a 60–90-minute assessment, a 25-minute interview for most senior tracks, and a 3–5-hour paid trial.

The funnel is genuinely narrow: of 214,300 applications in 2025, 61% cleared credential review, 34% passed the assessment and 6.8% were admitted. Waiting on a project that has not opened its next cohort is the most common reason an admitted contributor waits longer than the median.

No. One account per person, verified against one identity document and one tax form. Multi-accounting is the single most common integrity violation and it is treated as fraud, not as a mistake: all linked accounts are closed, unpaid balances on the duplicate accounts are forfeited under the contributor agreement, and referral bonuses paid on those accounts are clawed back.

If two people in one household both want to contribute, that is fine — two identities, two tax forms, two payout destinations. Sharing a login between them is not.

A laptop or desktop with a current version of Chrome, Edge or Firefox, or Safari 17 and above, and a connection stable enough to hold a WebSocket open — the task tool autosaves continuously and reconnects, but a link that drops every few minutes will cost you time. The tool is not usable on a phone and there is no mobile app.

Some projects add requirements that appear on the posting: a webcam for the interview, a second monitor for trajectory annotation, or a rule against running the tool inside a virtual machine or through a residential proxy, because both defeat the location checks labs require.

Open the project on the Explore board, read the posted rate and the requirements, then work through the application checklist on the sheet: profile, credentials, a short written response to a domain prompt, and the assessment if that project uses one. Every step is visible before you start, with the pay rate stated on the posting rather than revealed after you have invested an hour.

You apply per project, not to Cortext in general. Passing screening for one project shortens the process for the next one in the same domain family, because credential verification and identity checks carry over.

Roughly two in five applications stop at credential review. The usual causes are a claimed licence Cortext could not find on the issuing registry, a CV that does not evidence the specific specialism the project requires, a name mismatch between the CV and the identity document, or an application to a project whose cohort had already filled.

The rejection email names which of those applied. If it was an evidence problem rather than a fit problem, fixing the evidence and reapplying after 30 days generally works.

You can hold 3 open applications at a time. The cap exists because each one may pull an assessment slot and a reviewer's time, and applicants who scatter across eight postings tend to fail all of them.

Applications that reach a decision free up a slot immediately. Waitlisted applications do not count against the cap.

Five business days from your last completed step for most projects. Clinical, legal and financial-regulatory projects run to ten, because a second reviewer with the same credential has to sign off on the first reviewer's read of yours.

If you have heard nothing after those windows, email support@cortext-ai.uk with the project name and your application date. Do not reapply while an application is open — a duplicate application closes the original and puts you back at the start of the queue.

Specificity in the domain, not breadth. "Consultant nephrologist, 11 years, 400+ dialysis initiations annually, RCP fellowship 2019" gets read; "healthcare professional passionate about technology" does not. Name systems, volumes, jurisdictions, publication venues, rating floors, licence numbers — anything a reviewer can check.

Reviewers spend two to four minutes per application. Put the evidence for the specific claim the project requires in the first third of the page, and include a registry or profile link so verification does not have to be chased separately.

For some projects, yes. Competitive programming, creative writing, multilingual annotation and several red-teaming projects screen on demonstrated ability and take applicants with no degree at all. Cortext has admitted contributors on the strength of a rating, a repository or a portfolio alone.

Licensed-practice projects are the hard exception. Anything that touches clinical, legal or regulated financial judgement requires an active licence in good standing, and a student, a lapsed licence or a licence under restriction cannot be admitted to those regardless of how strong the rest of the application is.

Choose "Other" in the credential picker and upload the certificate, a transcript, and a link to the issuing body's public register if one exists. That routes the application to manual review, which adds about seven business days.

Cortext verifies against the issuing authority in the country of issue and does not require a US or UK equivalency assessment. What it cannot do is verify a credential from a body with no public register and no response to a verification request — in that case the application is decided on the rest of the evidence, and licensed-practice projects will not be available.

Yes, after 30 days, and you should bring something new: an added credential, a published piece of work, a completed certification, or an application to a project that fits your actual specialism better than the first one did.

An identical reapplication is usually rejected in under a day by the same reviewer criteria. The wait is not a punishment — it exists so reviewers are not re-reading the same file every week.

No, on both counts. Contributors are independent contractors engaged project by project. There is no salary, no benefits, no paid leave, no notice period and no employment relationship, and Cortext does not sponsor visas or work permits for contributor work.

Cortext Labs does hire salaried staff, and those openings are on the careers page. That is a separate process with a separate application flow, and passing contributor screening has no bearing on it.

You cleared screening but the project's current cohort is full. Waitlisted profiles stay in the matching pool for 90 days and are matched automatically when the project expands or when a similar project opens in the same domain.

You do not need to do anything, and you can keep applying elsewhere — waitlisted applications do not count against your 3-application cap. If nothing matches within 90 days the application closes and you can reapply immediately.

A timed, open-book, domain-specific test that runs 60–90 minutes depending on the project. It mixes multiple-choice calibration items with free-response items where you evaluate a model answer against a rubric and justify the score in writing. The pass mark is 80%.

It is deliberately not a general aptitude test. Every item is drawn from the work the project actually does, which is why the pass rate sits at 34% even among applicants whose credentials verified cleanly.

Anything longer than 30 minutes is paid. Short screeners under that ceiling are unpaid; the full 60–90-minute assessments are paid at the project's posted rate for the timed portion, capped at 90 minutes, and released on the next Friday run whether you pass or fail.

The ceiling is a written policy, not a courtesy. It was introduced after an internal audit of how long applicants were spending before anyone paid them anything, and it applies to every project on the platform.

3 attempts in total, including the first, with a 30-day wait between them. Each attempt draws a fresh item set from the bank, so a retake is not the same paper again.

After the third failure that project's assessment is closed to you permanently. Other projects in other domains remain open — the block is per assessment bank, not per account.

No. Assessments are open-book for references, documentation and your own notes, and closed to generative assistants. The entire product Cortext sells is human judgement, so an assessment passed by a model measures nothing a lab wants to buy.

Detection combines paste-burst timing, response-latency profiles and stylometric comparison against your free-response writing elsewhere in the application. A confirmed case voids the attempt, uses up one of your 3 and, on a second confirmation, closes the account.

A 25-minute live call with a reviewer who holds a credential in your domain. It opens with an identity check against the document you verified, then two probes into claims on your CV, then a short live rubric exercise where you are asked to justify a score and then defend it against a counter-argument.

It is not a culture interview and there are no behavioural questions. Up to 3 attempts are allowed if a call fails technically or has to be abandoned.

For the interview, yes — it is where identity verification is completed, and a call cannot proceed with the camera off. The recording is retained for 12 months for dispute purposes and is visible only to the interviewer, their reviewer and the trust team.

For the work itself, no. Ordinary project work involves no camera, no screen recording and no monitoring software. The narrow exception is a small number of red-teaming and agent-trajectory projects that require a recorded session, which is disclosed on the posting before you apply and requires separate consent.

Not automatically. The assessment autosaves every 20 seconds, and if the session is lost you can resume from the last save within 30 minutes; the clock keeps running during that window. Beyond 30 minutes the attempt is submitted as-is.

Each assessment carries one support-granted reset for a genuine platform failure. Email support@cortext-ai.uk with the project name, the approximate time in UTC and what you saw — support can check it against the session log, and the reset does not consume an attempt if the log confirms the failure.

Because the headline percentage is not the only gate. Every assessment carries a small set of calibration items with a known correct answer and a documented rationale; missing more than one of those fails the attempt regardless of total score, since the whole point is agreement with the rubric rather than plausible independent reasoning.

The result email names the dimension that failed. Free-response justification quality is the second most common cause — a correct score with a two-line rationale does not evidence the reasoning a lab is paying for.

3–5 hours of the project's real work, on the project's real tooling, paid at the project's posted rate. It is the last screening step, and 71% of contributors who reach it pass.

The output is genuinely used if it passes review, which is why it is paid rather than framed as a sample task. Cortext does not accept unpaid sample work from contributors under any circumstance.

Yes. Every tracked hour of the trial is paid at the posted rate and released on the next Friday run, whatever the outcome. Failing a trial does not create removed hours and does not reduce the payment.

The only exception is a trial voided for integrity reasons — generative-model output submitted on a human-judgement task, plagiarism, or work done by someone other than the account holder. That is a fraud finding, and it is handled under the contributor agreement rather than as a quality outcome.

1 per person, ever. Once you have completed a paid trial, applications to other projects skip that step entirely and go straight to project onboarding if the credential requirements are met.

That is also why a failed trial is significant: there is no second trial to redeem it with. A failed trial does not close your account, but it does close that project, and re-entry to the same domain family requires a new assessment after 30 days.

Seven calendar days from acceptance, which is deliberately loose for 3–5 hours of work so that people with clinical rotas or court schedules are not disadvantaged. You can split it across as many sessions as you like.

One seven-day extension is available on request before the deadline, through support@cortext-ai.uk. Trials that expire without a submission return to the pool and can be re-requested once.

Two independent reviewers score every trial on five dimensions: rubric adherence, factual accuracy, treatment of edge cases, quality of written justification, and consistency across items. Each is scored out of 5 and the mean must reach 4.0, with no single dimension below 3.0.

Where the two reviewers disagree by more than one point on any dimension, a third adjudicates. Disagreement is common enough that adjudication runs on roughly one trial in nine.

Yes — a written scorecard with the five dimension scores, the adjudicated total and reviewer comments on the two weakest items, sent within five business days of submission. Contributors who pass get the same scorecard, because the comments carry into the first weeks of real work.

The scorecard does not include the reviewers' names, and it does not include the gold-set answers, which stay confidential so the bank keeps working.

Not at the platform level. Some individual projects post a floor — typically five hours a week — because a lab needs predictable throughput, and that floor is written on the posting before you apply.

On projects with no floor you can work zero hours in a week without consequence. Contracts with no tracked time for 45 consecutive days are closed automatically, which frees the seat; you can request re-entry and usually get it without repeating screening.

80 hours per pay week, summed across every contract you hold. The task tool blocks new task starts once you reach it and the block lifts when the pay week rolls over at Monday 00:00 UTC.

The ceiling is not negotiable and it is not raised on request. It exists because sustained work above it correlates with a measurable fall in quality scores, and because a contractor engagement that consumes every waking hour starts to look like employment in several of the jurisdictions Cortext operates in.

By a per-task timer in the task tool. It starts when you open a task, pauses after 6 minutes without input, and stops when you submit. There is no screenshot capture, no keystroke logging and no webcam monitoring on ordinary project work.

Time the timer missed can be added manually with a note explaining it, up to 15% of the week's total. Manual entries above that share are held for review before payout, and a pattern of unexplained manual time is a quality issue rather than a payment one.

Almost always. The work is asynchronous and the task queue is open 24 hours a day, which is the point of a network spread across 130 countries.

Two things carry fixed times: batch deadlines, where a lab needs a delivery by a stated UTC timestamp, and the small number of projects with a live calibration session. Both are disclosed on the posting, and a project with a recurring live session says so before you apply.

Up to 3 active contracts at once, subject to the 80-hour weekly ceiling across all of them. Most contributors run one or two.

The exception is conflicting projects. Where two labs commission work in the same narrow area, the contributor agreement's confidentiality terms make holding both untenable, and the matching system will not offer you the second one. You will see it as a project that never appears in your queue rather than as a rejection.

Most lab projects are covered by confidentiality terms in the contributor agreement plus a project-specific NDA you accept at onboarding. Those cover the client's identity, model names and version strings, prompts and completions, rubric text, unreleased capabilities, and anything you learn about a system's failure modes.

You can say that you work with Cortext, describe your work in general terms — "I evaluate model output in clinical medicine" — and list Cortext on a CV or LinkedIn profile. You cannot post screenshots, paste prompts into other tools, or name the lab. The trust and safety page sets out the full boundary, including what happens after a project ends.

Assume no unless the project brief explicitly says otherwise. On human-judgement tasks — evaluation, reference answers, rubric authoring — generative output is the one thing the lab is not buying, and submitting it is treated as fraud rather than as a quality problem.

A minority of projects require tool use, for example agentic-trajectory work where you drive a model and annotate what it did. Those briefs say so, name the permitted tools, and are the only place where model output may appear in a submission.

Labs commission fixed volumes, so projects end. Cortext targets five business days' notice before a ramp-down and states the last day the queue will hold tasks; occasionally a lab pulls a project faster than that, and when it does, hours already tracked are still paid in full.

Your final week settles on the normal cycle — the pay week closes Sunday, review runs for 3 days, and payment releases on the Friday run. Ending a project does not affect your other contracts, your quality history or your eligibility for future work.

Yes. Set your status to paused in account settings and the matching system stops routing tasks to you without closing anything. Contracts with no activity for 45 days close automatically, and accounts with no activity for 12 months go dormant, which hides your profile from matching until you sign in again.

Neither is deletion. Quality history, verified credentials and tax forms survive both, so returning is a sign-in rather than a reapplication.

Each project has a pod lead who owns the rubric, answers interpretation questions and runs a weekly written office-hours thread. They are not a manager in an employment sense: they do not set your hours, approve leave or direct how you work, and the contributor agreement is explicit about that.

Rubric questions go to the pod lead in the project channel, usually answered within a working day. Payment, account and conduct questions go to support@cortext-ai.uk instead, because pod leads have no access to payout or account records.

Three layers. A sample of 10–20% of your submitted items gets a full second read by a certified reviewer; 5% of the items in your queue are gold-set items with a known correct answer, injected without being marked; and automated checks flag items that are anomalously fast, unusually short, or near-duplicates of your earlier work.

New contributors are sampled at the top of that range for their first two weeks, then move to the standard rate once the rolling score stabilises.

A 1-to-5 mean over your last 200 reviewed items, weighted so gold-set items count double. It is per project, not per account, because rubrics differ enough that a score on one says little about another.

Below 3.5 triggers a coaching cycle: annotated examples, a short recalibration set, and two weeks at the higher sampling rate. Sustained below 3.0 ends the contract. Both thresholds are stated in the project brief before you start.

Hours attached to items that failed review badly enough to be unusable — a rubric applied backwards, a fabricated citation, an item submitted so fast it cannot have been read. Those hours are deducted from the pay week's total before the Friday run.

They are not a fine and never exceed the hours attached to the failed items. Ordinary rework — an item sent back for a better justification — is not a removal, and the time you spend fixing it is paid.

You have 14 days from the removal notice. Open the item in the review log, click dispute, and state which part of the reviewer's reading you disagree with. A reviewer who was not involved in the original decision adjudicates and returns a written decision within 10 business days.

If the dispute succeeds the hours are reinstated and paid on the next Friday run, with no effect on your quality score. Roughly one dispute in five succeeds, which is high enough that filing one is worth the ten minutes when you think the reviewer was wrong.

A pool of items with a known correct answer and a written rationale, built by the project's rubric authors and adjudicated by a panel before use. They are injected into live queues at 5% and are indistinguishable from ordinary work.

They exist to detect drift — the slow divergence between what a rubric says and what a reviewer does after three weeks on the same queue. Gold answers are never published, including in dispute decisions, because a published gold set stops measuring anything.

Usually a rubric revision. When a lab tightens a definition, the project ships a new rubric version and the gold set is rebuilt against it, so answers that scored well last week can score badly this week. Rubric changes are announced in the project channel with the diff, and the first 48 hours after a revision are excluded from the rolling score.

The other common cause is a queue-mix change — a batch of harder edge cases entering the queue. If neither applies, ask the pod lead for the annotated examples behind the drop; that request is routine and reviewers are expected to answer it.

Yes: a sustained quality score below 3.0, a confirmed integrity finding, a confidentiality breach, or a lab ending the project. Quality removals come with the score history and the reviewer notes behind them, and take effect at the end of the current pay week so tracked work is not stranded.

Removal from one project does not close your account or affect your other contracts, unless the reason was fraud or a confidentiality breach — those are account-level and are handled by the trust team at trust@cortext-ai.uk.

Yes, always, and it is a policy rather than a courtesy. Every rejected item carries the failed rubric dimension, the reviewer's note and a link to the dispute form. Rejections with no reason attached are treated as an internal error, and the hours are reinstated by default when one is reported.

The same applies to contract-level decisions: a removal notice names the metric, the window and the threshold. If you receive one without that detail, reply to it and ask — you are entitled to the basis.

The pay week runs Monday 00:00 UTC to Sunday 23:59 UTC. When it closes, approved work goes through a 3-day review window, and everything that clears is released on the Friday run at Friday, 12:00 UTC. Standard transfers arrive in 2–5 business days.

So work done on a Monday is paid on the Friday eleven days later, and work done on a Sunday is paid five days later. There is no monthly cycle and no invoice to submit.

New accounts carry a 7-day hold on the first payout only. It runs from the first approved payout rather than from signup, and it is a fraud control — the window is where duplicate accounts, mismatched identity documents and stolen payout details surface.

Nothing is lost during the hold and no further action is needed from you. Every later payout runs on the normal weekly cycle.

$10. A balance below that rolls into the next week and keeps rolling until it clears the threshold, at which point it is released whole.

The threshold exists because both rails charge a per-transfer fee that Cortext absorbs, and below $10 that fee is a significant share of the transfer. You can see the pending balance and what it is waiting on at any time in the payments tab.

Once a balance is approved and released you can take it instantly for a 1% fee, minimum $0.50, and it lands in about 30 minutes. On a $400 balance that is $4.

What instant payout cannot do is skip the review window. Hours still have to clear the 3-day review before they are releasable, so the earliest an instant payout can move money is the Friday run itself, at 12:00 UTC.

Your country decides it. 113 countries settle through Stripe, direct to a local bank account or debit card. The remainder — including Brazil, Georgia, Ghana, Kosovo, Nepal, Türkiye, United Arab Emirates — settle through Wise. You do not choose; the rail is set by your country of residence and switches automatically if you move.

Cortext does not pay in cryptocurrency, does not pay to third-party wallets, and will not send a payout to an account in someone else's name. All three are common vectors for account-takeover fraud.

Rates are quoted and ledgered in USD. If your account holds a different currency, the rail converts at its mid-market rate on the day of the run. Cortext absorbs the transfer fee and the rail's conversion spread rather than passing them to contributors, which is a deliberate finance decision and is reported to the board alongside payout failure rates.

What Cortext cannot absorb is a fee charged by your own bank on an inbound international transfer, or a correspondent-bank deduction on a wire. Those appear as a gap between the statement figure and the amount that lands, and your bank is the only party that can explain them.

A failed transfer bounces back to your Cortext balance automatically, usually within two business days, and the payments tab shows the rail's reason code. Fix the details and the balance goes out on the next run rather than waiting a full cycle.

Where the money left Cortext but has not arrived, a trace runs in 5–10 business days and a genuine failed payment resolves in 8–12 business days. If the payout went to details that were wrong but valid — a closed account, a transposed IBAN — recovery depends on the receiving bank and takes 10–15 business days. Email payments@cortext-ai.uk with the payout reference; first reply is typically 1 business day.

Both models exist and the posting says which before you apply. Hourly projects pay tracked time at the posted rate; per-item projects pay a fixed amount per accepted item, with the rate set from a timed pilot so a competent contributor clears roughly the equivalent hourly figure.

Per-item projects still track time, because that is how Cortext checks the calibration is honest. If measured throughput drifts far enough that the effective rate falls below the posted equivalent, the per-item rate is re-set upward — it is not left to erode.

No. Bonuses — referral bonuses, project completion bonuses and quality bonuses — run on their own cycle, Tuesday and Friday, 12:00 UTC, separate from the weekly hours run. A bonus approved on a Wednesday goes out on the Friday bonus run, not with your hours.

They land as a separate line on the statement and a separate transfer, which is why you may see two arrivals in one week. Bonuses are not subject to the $10 minimum.

No invoice. Cortext self-bills: every Friday run generates a PDF statement listing hours by project, the rate applied, any removed hours with their reason, bonuses, fees and the net transfer. It is in the payments tab and emailed to you.

Statements are the document to give an accountant or a tax authority, and they stay available for 7 years, including after an account is closed. An annual earnings summary lands by 15 January.

No. Cortext does not advance against unapproved work, and there is no hardship exception — the review window is what keeps a lab from being billed for work it will reject, and releasing before it defeats the control.

The fastest legitimate route to money already earned is instant payout on Friday for 1%. Anyone offering an advance in exchange for a fee, in Cortext's name, is running a scam.

Four things, all itemised on the statement: removed hours from failed review, the 1% instant-payout fee if you chose it, a clawed-back referral bonus where the referred account was closed for fraud, and US backup withholding at 24% in the narrow case where a W-9 TIN fails IRS validation and is not corrected.

Nothing else. There is no platform commission taken from your rate, no equipment charge, no monthly account fee and no penalty deduction of any kind — the posted rate is what reaches you before your own tax.

No. Contributors are independent contractors. Cortext does not withhold income tax or social contributions, does not pay employer contributions, and provides no paid leave, sick pay, pension, unemployment cover or workers' compensation. You control when and whether you work, subject only to project deadlines and the 80-hour ceiling.

The practical consequence is that the gross figure on your statement is not take-home. Registering as self-employed, setting aside tax and meeting local filing deadlines are yours to handle, and in most countries they start from the first payment rather than from a threshold.

US persons — citizens, green-card holders and residents meeting the substantial-presence test — complete a W-9 with a valid SSN or EIN. Everyone else completes a W-8BEN, or a W-8BEN-E if you contribute through a non-US company.

The form is collected in the account before the first payout and no payout releases without one. A W-8BEN expires on the last day of the third calendar year after signing, and Cortext prompts for a fresh one 60 days before that date.

If you are a US person and Cortext paid you $600 or more in a calendar year, you receive a 1099-NEC by 31 January for the prior year, electronically by default and on paper on request. The figure reported is gross payments in the year, including bonuses.

Below $600 no 1099-NEC is issued, and that does not make the income untaxable — it remains reportable. Non-US contributors receive no US information return at all; the annual earnings summary issued by 15 January is the equivalent document for a local filing.

Not in the ordinary case, anywhere. Payouts are gross, and no income tax, social security or national insurance is deducted at source.

There is one exception: US backup withholding at 24%, which applies when a W-9 name and TIN fail IRS validation and the mismatch is not corrected after two notices. Correcting the form stops future withholding, but amounts already withheld are remitted to the IRS and can only be recovered through your own return.

File a W-8BEN so Cortext can document your non-US status, then handle your own local obligations. In most countries that means registering as self-employed or as a sole trader, keeping the weekly statements, and declaring the income on your annual return. Some countries require quarterly advance payments.

Cortext does not operate a treaty-relief mechanism on ordinary service income, does not issue local tax certificates, and does not register as a withholding agent in contributors' countries. Your statements are the evidence a local authority will ask for.

Most contributors fall below their country's registration threshold and have nothing to do. If you are registered, services supplied to a US company are usually outside the scope of your local VAT or subject to a reverse charge, and your invoice reference is the weekly statement.

Cortext does not add VAT or GST to contributor payouts and cannot advise on your registration status. If you become registered mid-year, tell payments@cortext-ai.uk so the statement carries your registration number.

The payments tab holds every weekly statement, the annual earnings summary and, for US persons, the 1099-NEC. All three stay available for 7 years, including after an account closes — closure deletes the profile, not the financial record, because retention is a legal obligation rather than a choice.

If a document is missing or wrong, email payments@cortext-ai.uk before the filing deadline. Corrected 1099-NECs can be issued, but the correction has to go through the IRS process and takes longer the later it is raised.

No. Nothing on this page or anywhere else on the Cortext site is tax, legal or financial advice, and Cortext is not a tax agent for contributors. These answers describe what Cortext does — the forms it collects, the returns it files, the records it keeps — and nothing about what you owe.

Tax treatment of contractor income varies by country, by residency status and by your other income, and it changes. Consult a qualified adviser in your jurisdiction before relying on anything here. Support can explain a statement line; it cannot tell you how to file.

$250 for a standard referral, more on specialist tracks where the posting names the higher figure. The bonus becomes payable once the person you referred is admitted and has logged 20 approved hours.

Payment goes out on the next bonus run — Tuesday and Friday, 12:00 UTC — not with your weekly hours. From qualifying hours to money in an account is usually four to seven days.

Someone who has never held a Cortext account, signs up through your link or enters your code at signup, and is admitted within 90 days of that signup. Attribution is last-touch, so if they later use someone else's link the later code wins.

People who applied before, were rejected before, or already hold an account do not qualify, and a code cannot be added retroactively after signup. Support cannot attach attribution after the fact — the check is automated precisely so it cannot be argued.

The most likely reason is that they have not yet logged 20 approved hours — admitted is not the same as qualified, and approval means the hours have cleared the 3-day review window. The referrals tab shows their progress toward the threshold without showing anything else about their account.

Other causes: the bonus qualified after the last bonus run and is queued for the next one, your own tax form is incomplete so no payout can release, or the referral is in manual review because you crossed 25 referrals this quarter. Email referrals@cortext-ai.uk if none of those fit.

Yes, in three cases: the referred account is closed for fraud or multi-accounting; the referral is found to be self-referral through a second identity; or the referred contributor's qualifying hours are later reversed as removed hours, dropping them below the 20-hour threshold.

A clawback appears as an itemised negative line on the next statement, with the reason. Good-faith referrals who simply stop working are never clawed back — once qualified on genuine hours, the bonus is yours.

No hard cap, but the 25th qualified referral in a quarter moves subsequent bonuses to manual review before payment, which adds up to ten business days each. Volume referrers — professional-body organisers, university society leads — pass that review routinely.

What the review catches is the pattern that precedes fraud: many signups from one device, one payout destination behind several accounts, or referral clusters that all stop working at the 20-hour mark.

Treat them as income. For US persons they count toward the $600 1099-NEC threshold and appear in the reported total; for everyone else they appear in the annual earnings summary alongside project earnings.

Cortext does not withhold on bonuses, and the same disclaimer applies as everywhere else on this page: this is a description of what Cortext reports, not tax advice.

There is two-factor authentication and it never uses a phone. Cortext supports time-based codes from any authenticator app and passkeys on a hardware key or device biometric; both are stronger than SMS, which is defeated by SIM-swap attacks that specifically target accounts holding payout details.

Cortext will never send you a code by text message and never asks for a phone number. Any SMS claiming to be a Cortext verification code is phishing — do not enter it anywhere, and forward the sender number to security@cortext-ai.uk.

Use one of the ten single-use recovery codes issued when you enabled two-factor. Each works once, and the remaining count is shown in security settings so you can regenerate the set when it runs low.

Without recovery codes it is a manual identity recovery through support@cortext-ai.uk: re-verification against the original identity document, plus confirmation of details only the account holder would have. It takes five business days, cannot be accelerated, and payouts are frozen throughout — which is the point.

In account settings. The change is confirmed from both the old and the new address, and if the old mailbox is unreachable the request goes to manual review at support@cortext-ai.uk. Payouts are held for 48 hours after an email change, because a change of address is the first step in almost every account takeover.

Shared, role-based or disposable addresses are rejected. The account has to reach one identified person, since it is the delivery channel for contracts, statements and tax documents.

Request closure in account settings. Any balance above $10 is released on the next Friday run, then a 30-day grace period runs during which signing in cancels the request. After that the profile, CV, credential documents and work history are deleted.

Three things survive deletion because the law requires it: payment and tax records for 7 years, the record of any trust and safety finding, and work product already delivered to a lab, which belongs to the lab under the contributor agreement and cannot be recalled. Rejected application data is deleted after 24 months whether or not you ask.

No screenshots, no keystroke logging, no webcam access and no software to install on ordinary project work. What is collected is task-level telemetry inside the browser tool: which task is open, timer start and stop, idle detection after 6 minutes, paste events on assessment items, and submission timestamps.

The exception is a small number of red-teaming and agent-trajectory projects that require a recorded working session. Those are disclosed on the posting, need separate written consent, and can be declined without any effect on your standing.

Genuine mail comes only from the cortext-ai.uk domain and is authenticated with SPF, DKIM and DMARC. Cortext will never ask for your password, a two-factor code, a recovery code, your bank login, or a cryptocurrency payment of any kind, and no genuine message asks you to move a conversation to WhatsApp or Telegram.

The scam patterns that target AI-training workers are specific: a "recruiter" offering pre-approved work for an onboarding fee, a fake task tool that harvests logins, an overpayment cheque you are asked to refund in part, and a "payout upgrade" that requests bank credentials. Check the sender domain, then forward anything suspicious to security@cortext-ai.uk. The trust and safety page has the full list.

Email privacy@cortext-ai.uk from the address on the account. Access, portability, correction, deletion and objection requests are answered within 30 days, and Cortext applies that timeline to every contributor rather than only to those covered by the GDPR or the CCPA.

Export arrives as a machine-readable archive: profile, applications, assessment scores, work history, review decisions, statements and support correspondence. Deletion is bounded by the same retention obligations set out in the account-closure answer.

Work down this list before emailing: confirm the address you are typing is the one on the account, check that the two-factor code is from the right entry in your authenticator and that your device clock is correct — a clock more than 30 seconds out will reject every code — then try a private window with extensions disabled to rule out a password manager filling a stale value.

Five failed attempts lock the account for 15 minutes, which is a lockout and not a closure. If the lockout keeps recurring and you did not cause it, email security@cortext-ai.uk rather than support@cortext-ai.uk — repeated failed logins you did not attempt mean someone else has your address and is guessing.

Give it ten minutes, then check spam and any Promotions or Updates tab. Corporate mail filters are the most common cause: university and hospital domains frequently quarantine automated mail, and a personal address usually solves it permanently. Resends are limited to three per hour per address.

If nothing has arrived after an hour and the address is not a corporate one, check the status page — email delivery is one of the monitored components and a delivery backlog will be posted there before support can tell you anything more.

First a hard reload, then a private window with extensions off. Ad blockers, aggressive privacy extensions and translation extensions all rewrite the DOM the task tool depends on, and content blockers are behind most "blank editor" reports. Confirm you are on a supported browser: current Chrome, Edge or Firefox, or Safari 17 and above.

If the tool loads but crawls, check whether you are on a VPN or a corporate proxy — the editor holds a WebSocket and inspecting proxies frequently break it. Still stuck: email support@cortext-ai.uk with your browser and version, the project name, the task ID and the UTC time, and any error text from the browser console.

A timer left running is capped by the 6-minute idle timeout, so a session cannot silently bill overnight; the overrun is trimmed automatically before review. Missing time is the more common problem, usually caused by a lost connection at the moment of submission.

File a time correction within 48 hours through the time tab, with the task ID and what you were doing. Corrections inside that window are reviewed against the session log and approved routinely; after it, the pay week has closed and reopening it needs a payments escalation with a much lower success rate.

The ceiling is 25 MB per file. Larger files fail with a generic error in some browsers, which is a known bug rather than a rejection of your work — compress the file or split it. Executables, archives and files with active macros are blocked outright by policy.

Your draft is cached in the browser until the upload succeeds, so do not close the tab. If a retry on the same connection fails twice, switch network — mobile tethering is a good test — before contacting support, because the majority of upload failures are middlebox interference on the sending side.

Three usual causes: the batch is exhausted and the next one has not landed, the project has a regional restriction that no longer matches your account country, or the contract was paused — either by the lab or automatically after 45 days without tracked time.

The project channel carries batch announcements and is worth checking first. Tracked hours are unaffected in every one of those cases and settle on the normal cycle.

That flag is set when a timed assessment is opened and neither submitted nor resumed within 30 minutes of the session dropping. It consumes one of your 3 attempts, which is why the tool warns before you start.

If the cause was a platform failure rather than a walk-away, email support@cortext-ai.uk with the project and the UTC time. Support can read the session log, and each assessment carries one reset that restores the attempt when the log confirms the failure.

Check the date first. Hours from the pay week that closed on Sunday sit in the 3-day review window and do not appear as releasable until it ends, so a balance checked on Monday or Tuesday legitimately reads zero.

After that, the usual causes are visible in the payments tab: an incomplete tax form, a payout destination that failed verification, the 7-day hold on a first payout, or a balance under the $10 minimum that rolled over. Each shows a reason and the action that clears it.

Write from the address on your account, and put the specifics in the first message: the project name, the task or payout reference, the date and time in UTC, what you expected, what happened instead, and your browser and version if it is a tool problem. Screenshots help; screenshots of task content do not, because confidentiality terms cover them.

A first message with those details is usually resolved in one reply. A message saying only that something is broken takes three round trips before anyone can look, and across timezones that is two days lost.

The status page carries live state for eight components — job board, applications and assessments, work trials, time tracking, payouts, authentication, the public API and email delivery — with a 90-day history and every resolved incident. It is updated during an incident before support can answer individually.

Scheduled maintenance runs Thursdays 03:00–05:00 UTC and is posted in advance. If the status page is green and the problem persists after a private-window retry, it is specific to your account or your network, and support needs the detail listed in the previous answer to find it.