The funnel already works. Here is where it leaks.

Inventory first. Before proposing anything, I listed every interactive element and every piece of information on each screen. That inventory is a contract: the improved screen must contain every item, doing the same job, in the same number of taps or fewer. Where I believe something should be removed, it appears as a written recommendation for the Delta team, never as a quiet deletion inside a mockup. A live product's decisions were made with context I do not have.

Steelman second. Each screen gets a paragraph on why it was probably built this way, written as if the person who shipped it will read this. They might.

Evidence bar. A change earns its place with one of: a measurement I can reproduce, a WCAG 2.2 success criterion, a named usability heuristic argued against this specific context, an error-prevention argument specific to trading or KYC, or a pattern verified at a named competitor with a link to where I found it. App store reviews did not qualify. Colors and ratios quoted in this document were sampled from the screenshot pixels and computed with the WCAG relative-luminance formula, not eyeballed.

Two buckets, never blurred. Everything lands in either Backed now or Hypothesis, needs data. The second bucket is not filler; it is the list of things I would refuse to build without instrumentation.

What Coinbase is here for. Principles only: state visibility, one primary action per screen, form clarity, candor as a trust device. No Coinbase color, component, or layout appears in any mockup. Every idea taken from its screens is named as a principle and re-expressed in Delta's own tokens, which I sampled from the app: #18191E background, #22242C surfaces, #FE6C02 orange, #00A876 green, #8E9298 muted text.

Scope, strictly. Onboarding only. The trading terminal, order entry, positions, and the options chain are untouched. Two proposals graze screens that funded traders also use; both are explicitly restricted to the new-user or empty-state branch, because onboarding changes that leak into revenue-critical surfaces stop being safe.

Dedicated digit boxes make progress glanceable. Gating resend behind a countdown controls SMS cost and abuse, and Coinbase documents the same gate at roughly 30 seconds [13]. Keeping the submit button on screen but disabled prevents partial submissions. All three are reasonable calls. The failures here are in rendering, not intent.

• Five of six boxes are effectively invisible. Sampled fill #191A1F on background #18191E: a contrast ratio of 1.01:1. WCAG 2.2 SC 1.4.11 requires 3:1 for the visual information that identifies a component [14]. These are active input fields, so the inactive-component exemption does not apply. A first-time user cannot see how many digits the screen wants until they start typing.
• The disabled button out-shouts the enabled ones. "Submit Code" sits on a brown fill (#592705) with white text at 12.23:1, while Delta's genuinely tappable orange buttons carry white text at 2.85:1. The strongest label on the screen belongs to the control that does nothing. Disabled buttons that look active are a documented usability trap; NN/g's guidance is to make the state unmistakable and say why it is disabled [15]. To be precise: this is not a WCAG violation, because inactive components are exempt from 1.4.11. It is a state-legibility defect.
• The full phone number is printed in the clear. Delta masks account balances behind an eye toggle two tabs away, so the app already treats over-the-shoulder privacy as a value. The same posture belongs here: ••••••2552.
• "6 digit code" should be "6-digit code."
• Autofill is unconfirmed. A static screenshot cannot show whether the field adopts iOS one-time-code AutoFill. If it does not, Apple's mechanism is documented: set the field's textContentType to .oneTimeCode and format the SMS with the domain-bound last line, and the code surfaces in the keyboard bar [16]. This needs one device test before anyone writes code.

Principle drawn from Coinbase's code screen: a form's state must be visible before the user acts, not after. Its expression here is entirely Delta's: empty boxes keep the #191B20 fill and take a 1.5px border in #6A7077, a darker step of Delta's own muted gray, measuring 3.51:1. The disabled button drops the brown for Delta's standard surface #22242C with muted text, so disabled finally looks disabled and turns orange only when six digits exist. Same elements, same taps; autofill removes taps when confirmed.

This screen does the right things. Requirements are disclosed before the flow starts, with recognizable government-ID iconography. Benefits answer "why bother" on the same screen. The phone row already earns a green check once verified, so a progress language exists. Listing requirements up front matches what the better flows in Elsewhen's cross-exchange onboarding analysis were praised for: explicit verification status [17].

• Four labels for one action. Home stepper and banner say Complete KYC. The post-OTP notification says Initiate KYC. This screen said Verify Now at 8:30 and Get Verified at 8:36, captured on the same device, same day. Nielsen's consistency heuristic is direct about this: users should not have to wonder whether different words mean the same thing [20]. Mid-funnel doubt is expensive precisely here; CleverX's KYC research names comprehension failure as a primary abandonment driver [18]. Two live variants of one screen also hints at a release-process gap worth a separate conversation.
• "Aadhar card" sits 40 pixels from an icon spelling AADHAAR. UIDAI spells it Aadhaar. The app contradicts itself on the single most sensitive document it will ever ask for. The same misspelling family appears in Delta's own support guide, including "Your phone number should be linkedin with your Adhaar," live as of January 2026 [4].
• The 3-minute promise vanishes mid-flow. The home stepper says "Takes less than 3 minutes" and the signup sheet says "ETA: 3 min." This screen, where the actual work starts, says nothing. Goal-gradient research shows visible progress toward a goal accelerates completion [21]; dropping the clock at the moment of effort is the reverse of that.

One verb, everywhere: Complete KYC at entry points, Verify Now inside the flow, and the notification grammar fixed in the same pass ("Congratulations, your mobile number is now verified"). The requirements list extends Delta's own checkmark pattern into an honest step list: done steps keep the green check, the current step gets an "Up next" marker in Delta's soft orange, nothing else changes. The ETA line reuses the home screen's exact phrasing, so the promise follows the user in. Principle from Coinbase's flow, one primary action framed by quiet supporting text, expressed with Delta's existing card, tabs, and button.

Entity KYC is rare and document-heavy. An email channel avoids building and maintaining an upload flow that a small fraction of signups will ever touch, and a 48-hour SLA is honest about a manual review. The requirements list is genuinely complete. This is a sensible operational shortcut; the screen just was not given the same finishing pass as the individual path.

• An off-brand domain is asking for CIN, PAN, and PEP declarations. Indian banking UX has spent a decade training users that unfamiliar domains requesting identity documents mean phishing, and abandonment research names unexpected third parties at the trust moment as a driver [18]. If Excelium Tech is Delta India's registered legal entity, one line of attribution converts the alarm into legitimacy. If it is not required, the address should move to a delta.exchange domain. That is a legal call, so the mockup carries a placeholder line and the decision is flagged to the Delta team.
• Copy drift on a compliance surface. A space before the colon, "Entity Email id," and Entity/Institution used interchangeably. Individually trivial; together, on the screen a company's finance officer reads while deciding whether to email their PAN somewhere, carelessness reads as risk.
• Visual grammar breaks between the tabs. The Individual tab speaks in an iconized card checklist; this tab is a wall of bullets. Same screen family, same heuristic as before: consistency [20].

An instruction interstitial before the camera is the right pattern: it raises first-attempt pass rates, and every re-capture request measurably raises abandonment; Signicat's research, as relayed by Zyphe, puts re-upload requests at three times the abandonment [22]. Explaining before capturing is correct. Keeping the vendor badge is honest.

• The copy says remove spectacles and mask; the illustration wears both. The picture is the instruction most users actually read, and here it demonstrates exactly the two things that cause a failed capture. This is error prevention in its purest form: eliminate the error-prone condition instead of waiting for the retry [20]. CoinDCX's equivalent guide instructs capture "without your spectacles," which is the norm this screen intends and its own artwork contradicts [9].
• Grammar: "spectacles, hat and mask" mixes number; "spectacles, hats and masks" or "your spectacles, hat or mask."
• One unknown, flagged rather than asserted: on the next screen the capture ring renders red while the face is correctly framed. Red usually means error. Whether it turns green on alignment cannot be known from a still frame, so it lives in the hypotheses, not here.

Same layout, same copy position, same CTA, same vendor badge. The illustration simply complies with its own instruction: bare face, no glasses, no mask. This is a vendor-side asset swap (the screen is HyperVerge's), so the change ships through vendor configuration, not Delta app code. Worth knowing before scheduling it.

Masked-by-default balances are a legitimate trading-app convention; the eye toggle exists for over-the-shoulder privacy, and habituated traders rely on it. Withdraw living beside Add Funds is standard wallet furniture. The likely implementation story is simple: one masking code path for all account states. For a funded trader, every one of these choices is right. This proposal touches only the branch those traders never occupy.

• ₹✱✱✱✱✱✱ on an account that has never held a rupee is noise, not privacy. There is nothing to protect, and the mask hides the one honest fact, ₹0, that frames the next action. Nielsen's first heuristic is visibility of system status [20]; NN/g's empty-state guidance says these screens exist to communicate status and hand the user a direct path to the key task [23]. An asterisk row does neither. It can also read as "there might be money here I cannot see," which is a strange first impression for a financial product to make.
• A full-strength Withdraw button for a user who cannot withdraw is a guaranteed dead-end tap. Error prevention says remove or visibly disable actions that can only fail [20]. Under the working rule, removal does not happen inside a mockup: the button stays, and the disable decision goes to the Delta team as R2 with a data ask attached.
• "PNL Analytics" vs "PnL Offset" in the margin sheet two taps away. One casing, "PnL," everywhere.

State-aware rendering for the zero-balance, unverified branch only: real zeros instead of asterisks, one muted caption naming the path forward, and the eye toggle untouched, so masking behavior for funded accounts never changes. Principle borrowed from Coinbase's empty portfolio: an empty state is a signpost, not a blank. Expressed entirely with Delta's existing card, type scale, and muted gray.

My capture set does not include the bank-details form or the deposit screen, and I will not mock up screens I have not inventoried. Inventing them here would be exactly the kind of confident fabrication this document exists to avoid. What I can do is read Delta's own documentation against competitor patterns and mark where I would look first, with the mockups waiting on real captures.

• Delta verifies bank accounts with a ₹1 penny drop, own-name accounts only. Its published failure-reasons article lists name-vs-PAN mismatch, joint accounts, and corporate accounts as rejection causes [5]. Every one of those failures is predictable before the user submits. The pattern worth testing when the screens are captured: state "account must be in your name, matching your PAN" before the form, not inside a rejection email. Groww documents the same penny-drop-plus-name-match mechanic [10], and Upstox, Dhan, and Zerodha all offer a UPI-based verification where the account details are fetched from a ₹1 UPI payment instead of typed [11], which removes the two fields users mistype in the first place. Whether that alternative is worth Delta's backend cost is unknowable from outside; it is listed as a pattern, not a recommendation.
• Delta's minimum deposit is ₹1 by UPI, with UPI credits typically inside 15 minutes [6]. That is the lowest entry point I found anywhere in this research; CoinDCX's documented bank-transfer minimum is ₹100 [12]. A ₹1 first deposit is a trust-building device competitors cannot match, and nothing in the captured funnel says it out loud. Where exactly to say it depends on the deposit screen I have not seen.
• A non-transfer, named as such: Coinbase's deposit screen explains card and PayPal rails [24]. None of that applies here. Indian VDA platforms run on UPI, IMPS, NEFT, and RTGS against a whitelisted own-name account, so the deposit-method explainer pattern transfers but its content does not.