Polishing isn't free
Machine time spent on a scratch that was never going to clear buys nothing — and can cost the coating. A heavy recipe runs 480 seconds.
Display restoration systems · Designed & built in the USA
Every scratch, measured before it's polished.
XTR builds the complete restoration line for used phones: machine vision that grades the glass, a model that predicts what polishing will remove, and the hardware that restores the display and its anti-fingerprint finish.
- Of polished devices upgraded
- 73%
- Average grade gained
- +1.0
- Polish-decision accuracy
- 94%
- Units per line, per day
- 600
At least one full grade, across 1,292 paired devices
Per polished device, measured before and after
ML triage vs the actual outcome, 1,871 devices
One 7.5-hour shift with 3 operators
Production data from XTR Shopfloor, July–August 2026.
01 · The restoration line
Seven stages. One flow.
From vision triage to final vision, every device moves through the same measured sequence — scanned, prepared, polished, re-coated, cured and inspected again.
02 · The lever
One grade step is the whole margin.
The phone is the same device before and after. Cosmetic grade is what the resale channel pays for — and on a returned handset it is the one variable a line can still move.
So the question is never "can we polish it?"
It is will this specific scratch, on this specific panel, come out — and is the grade it buys worth the machine time? That is a measurement problem, then a prediction problem. XTR does both.
03 · Why XTR
Built to fix what conventional polishing gets wrong.
Most refurbishment lines still polish by feel: wet, slow to prepare, graded by eye and managed on spreadsheets. XTR replaced each weak link in that process with an engineered, measured one.
01
Preparation & water
Conventional polishing
Hours of prep to keep water out
Conventional polishing floods the glass with water and slurry. Before a device can touch a pad it has to be sealed, taped, jigged and protected against liquid ingress. Even then, water damage still turns repairable units into scrap.
The XTR line
Near-dry by design
XTR's chemical-mechanical process runs on a small, controlled volume of water and slurry, so there is no sealing, taping, protective jig or waterproofing step. After plasma etching and cleaning, the device goes straight onto the polisher. The risk of liquid damage is designed out rather than managed.
Plasma clean → straight to polish02
Proof of result
Conventional polishing
No record of what polishing achieved
Most lines polish blind. With no inspection built into the process, there is no record of what the device looked like before, what the pass removed, or whether the machine time paid off.
The XTR line
Measured before and after
Vision is part of the line. Every device is imaged on six surfaces before polishing and again after, so each unit carries a defect-by-defect before-and-after record. Yield is proven with data rather than estimated.
Before / after on every unit03
The decision to polish
Conventional polishing
Subjective calls, no cost model
Whether to polish is a judgement made at the bench. It varies by operator and by shift, and nobody can say what the unit will be worth afterwards or whether the labour will pay for itself.
The XTR line
Predicted, priced, then decided
A machine-learning model scores every scratch for its chance of clearing and predicts the post-polish grade. It then weighs the resale uplift against machine time and labour. Operators get a clear Polish, Review or Skip recommendation with the expected profit behind it.
Expected profit per unit04
Flow & inventory
Conventional polishing
Spreadsheets and whiteboards
Polishing is usually run without a dedicated shop-floor system. Units go missing between stations, bottlenecks are found late, and inventory is only as accurate as the last manual update.
The XTR line
A shop floor built for restoration
XTR Shopfloor tracks every IMEI through every gate as an immutable transaction. Queues, dwell times, bottlenecks, staging and shipments are visible in real time, and every report comes from the same ledger the floor writes as it works.
Every IMEI, every gate05
Polishing recipe
Conventional polishing
One setting for every device
Without defined recipes, light scuffs are over-processed and deep scratches under-processed. Results depend on who ran the machine that day.
The XTR line
A recipe for the condition and the target
A catalogue of defined recipes, each with its own time and pressure, is matched to the device's condition and target grade. The machine records the recipe it ran and the pressure it actually applied, so every result can be repeated and traced.
Recipe matched to target grade06
Grading policy
Conventional polishing
Inconsistent, one-size-fits-all grading
Cosmetic grading by eye differs between graders, sites and days. A single fixed standard can't serve buyers in different markets with different specifications.
The XTR line
Objective grading, a policy per customer
Grading is measured against named, versioned policies built on the CTIA standard. Each customer or market can have its own policy. Because the defect data is stored, any device can be re-graded to a new specification without being scanned again.
Per-market grading policies04 · The system
Measure. Decide. Restore. Verify.
Hardware, intelligence and workflow built as one closed loop — so the record is what happened, not what someone remembered to type.
01 · Vision triage
It doesn't look at the phone. It measures it.
- Six surfaces — front, back and all four edges — imaged under controlled raking light.
- A graded device in under 30 seconds. All six surfaces are captured, inspected and graded against the customer's policy in one automated pass.
- Severity, not just count. Every scratch is classified into four levels, from light marks that polishing removes to deep damage it can't, so polishability is known before the device reaches the polisher.
- Zones matter more than counts. The main display is graded far harder than the enclosure, because that's what the buyer looks at.
- One capture, every standard. CTIA 5.0 is the floor; run a named grading policy per customer and re-grade without touching the phone again.
Actual captures of one grade 3 unit. The front is graded on its own against zone AA: 22 scratches, grade 3. The back and all four sides are counted together against zone B: 2 scratches, grade 2.
02 · Decision assist
Which scratches actually come out?
A gradient-boosted model scores every scratch individually — length, width, area, angle, ten depth bands and distance to the panel edge — and returns the probability that buffing clears it. It's retrained on production data, because every polished device becomes a labelled example.
PolishModel predicts a better grade
ReviewSame grade, or uncertain — the operator decides
SkipPredicts no gain — send it on as-is
8 scratches scored4 likely to clearPolish recommended
03 · Buff, polish & AF-coat
Three heads. One recipe per device.
- Patented chemical-mechanical polishing removes scratches up to 10 µm deep using slurry, controlled pressure and specialised pads.
- Recipes are a catalogue, not a guess — from R-01 Standard (240 s @ 1.5 psi) to R-02 Heavy (480 s @ 2.8 psi), with light and deep cuts either side.
- The machine reports what it actually did. Commanded recipe and measured pressure are stored separately.
- AF-coating on a finished device. A proprietary low-temperature liquid process — plasma prep, ultrasonic spray, IR cure — forms a covalently bonded anti-fingerprint film on a fully assembled phone. No disassembly.
- No masking or sealing. Minimal water and slurry means the device goes from plasma etching and cleaning straight onto the polisher.
04 · Final vision
Then it measures the same device again.
Final vision is the second half of the experiment. Every polished unit carries a before grade and an after grade, so yield is observed, not asserted — and downgrades are reported as loudly as upgrades.
Because the same scratch is measured before and after, the line generates its own labelled data. No annotation, no manual review — every polished device trains the next decision.
- Capture6 surfaces
- Predictper scratch
- Polishrecipe run
- Re-capturefinal vision
- Labelbefore / after
- Retrainnext decision
Closed loopEvery unit is a data point
05 · Equipment
We design and build the whole line.
Polishing takes the factory's anti-fingerprint layer off along with the scratches. Putting it back on a fully assembled phone is beyond conventional vacuum coating — so XTR built the line that can. Owning the machine is why the software can read it rather than guess at it.
US patentedRestore · 01
Auto-Polishing Machine
Automated CMP with three heads, adjustable parameters and recorded pressure and process time per device.
XTR builtInspect · 02
Vision Inspection System
Six-side imaging with deep-learning defect detection, cosmetic grading and polishability scoring.
Prepare · 03
Atmospheric Plasma
Low-temperature surface cleaning and activation — adhesion without thermal damage.
Coat · 04
Ultrasonic AF Spray
Liquid-phase anti-fingerprint film applied directly to the assembled device.
Made in USACure · 05
IR Heat Chamber
Precision thermal curing for an OEM-level, covalently bonded finish.
XTR AF-Coating Line — the integrated, turnkey system: plasma, spray and cure in one flow.
Discuss a line06 · Shopfloor platform
The floor routes itself.
XTR Shopfloor is the software the whole line reports into. Every device is an IMEI walking a state machine: no stage can be skipped, every move writes an immutable transaction, and the machines post their own results.
One screen the whole floor can read from the aisle.
- Pipeline at a glance. Awaiting and done-today counts for every gate, with the stage that is backing up flagged in red.
- Throughput against goal. Ten working days of shipped units, coloured by whether each day hit its target.
- Every result as it lands. A live feed of final-vision outcomes: triage grade, final grade, upgraded or not.
- Built for a wall screen. Auto-refreshing, full-screen and legible from thirty feet.
Customer name and device IMEIs masked.
See the after before you polish.
- Current and predicted, side by side. The pre-polish capture with each scratch scored, next to the surface the model expects after polishing, showing only the defects it predicts will remain.
- Scenarios, not guesses. Each recipe shows its target grade, machine time, pressure and chance of reaching it.
- The money, in order. Sell-as-is value, resale at the target grade, uplift, labour and expected profit, then one recommendation.
Device IMEI and operator name masked. Prices on this screen are sample values.
Yield you can observe, not assert.
- Paired devices. The latest before scan against the latest after scan for every improvable unit.
- Flow of outcomes. Each band is proportional to device count: green for upgrades, grey for no change, red for downgrades.
- Filter to the question. By project, equipment, model and date range, then export to CSV.
Report shown across all projects for one 30-day window.
The model grades itself against reality.
- Three answers per device. The before grade, the ML-predicted after grade and the actual after grade, on the same policy scale.
- Scored in the open. ML triage and machine triage each get a confusion matrix against the real outcome.
- Versioned models. Every prediction records the model version, its test accuracy and the matched pairs it was trained on.
Device IMEIs masked.
CTIA 5.0 is the floor.
- Named standards. Levels, depth rules and per-zone count thresholds: one complete policy per customer.
- Zones weighted by what buyers see. The main display is graded far harder than the enclosure.
- Re-grade without re-scanning. The defect set is stored, so any unit can be graded under a different policy.
Bottlenecks surface themselves.
- Per-step queues. Pending counts, median and P90 dwell for every gate.
- Measured, not argued. The slow gate is flagged from an open-visit ledger.
- Machines are locations. Each polisher is a routable buffer with its own queue.
Every question, one ledger underneath.
- Yield, reject rate, cycle time, throughput. By stage, operator, model and project.
- Grading calibration. Golden units check the policy against expert-graded devices.
- CSV and PDF out of every report. The invoice comes out of the same numbers.
Seven gates, one ledger
Out-of-sequence moves are refused; a supervisor override needs a written reason.
Reason codes, always
No rework or scrap decision is recordable without a base code.
Role-gated stations
A polishing badge cannot record a QC decision. Every action carries an operator.
Customer-facing by design
Customers log in to the same data, scoped to their organisation, with cost per device.
07 · Company
Born on a repair floor.
XTR was founded in 2012 inside an OEM repair center, developing the polishing and cover-glass processes that became its patented technology. Today it serves carriers, insurers and refurbishers running certified pre-owned programmes.
- US & Korean patentsChemical-mechanical polishing
- Made in USAMachines designed and built in-house
- CPO-readyOEM-level finish, full traceability
- 2012
Founded inside an OEM repair center; polishing and cover-glass processes developed.
- 2013
Polishing for OEM modules introduced.
- 2018
Curved OLED lamination developed.
- 2020
Korean patent awarded.
- 2022
US patent awarded.
- 2025
AI Vision and the BNP division launched.
- Now
Closed-loop restoration — vision, decision assist and shopfloor in one line.
08 · Contact
Measure it. Predict it. Then polish it.
Tell us about your volume and program. We'll walk you through the line, the data it produces and what it would change for your grade mix.
Or write to info@xtrtechnologies.com