TL;DR: The spec parameters you list in your initial sample request determine whether you get a real evaluation unit or a cherry-picked marketing sample — structure your inquiry accordingly.
TL;DR: In our incoming evaluation process, we reject power bank samples where measured capacity at 0.5C discharge falls below 92% of the datasheet-stated value — a threshold that filters out roughly 35% of first-time supplier submissions.
Capacity Specification Language That Actually Controls What You Receive #
The number one way buyers get misled during sample evaluation is by asking for “10,000mAh” without specifying the measurement conditions. Every factory knows this, and the ambiguity works in their favor.
When you write your sample inquiry, the capacity specification needs to include discharge rate, cutoff voltage, temperature, and whether the rating is cell-level or system-level output. A proper inquiry line reads: Rated output capacity: ≥9,000mAh measured at USB output terminals, 0.5C constant current discharge, 20°C ±2°C ambient, to 3.0V cell cutoff. That is a fundamentally different request than “10,000mAh capacity power bank,” even though most inquiry forms treat them as equivalent.
Why does this matter? Cell-level nominal capacity and actual USB-output capacity diverge by 15–22% on a typical lithium-ion design, after accounting for DC-DC conversion losses (82–88% efficiency is common), protection circuit overhead, and the fact that many factories rate cells at 0.2C while your product will cycle at 0.5C or higher in real use. A factory quoting you “10,000mAh” at 0.2C, cell-level, is not lying — they are just measuring something you don’t actually need.
The IEC 62133-2:2017 clause 7.3.3 discharge test procedure specifies a 0.2C rate as the baseline for capacity verification. For portable charger sourcing, I’d prioritize specifying your actual use-case rate in addition to the IEC baseline — because the IEC rate tells you nominal capacity, but 0.5C or 1C tells you what the product will do in a backpack.
One parameter that almost never appears in standard inquiry templates: impedance at 1kHz. Ask for it upfront. It belongs in your spec sheet and reveals cell age, grade, and pack assembly quality more directly than any other single parameter. Grade-A fresh LFP cells in a portable application typically measure 8–18mΩ at pack level for a 4S2P configuration; values above 30mΩ on a new sample indicate either aged cells or poor tab-welding quality.
The UN 38.3 test series covers transport safety and is frequently cited by factories as a quality marker. Be clear with your supplier contact that UN 38.3 compliance is a floor requirement, not a differentiator. Every shipment to most markets requires it. Treat its absence from the sample documentation as a disqualifying flag, not a negotiation point.
Supplier Qualification — What to Request and What the Response Tells You #
Send your sample inquiry with a structured technical data request attached. We use what we internally call the SRQ-3 packet (Sample Request & Qualification Checklist, version 3), which asks for six specific documents alongside the physical sample. The response pattern tells you as much as the sample itself.
Request 1: Datasheet with all parameters measured at stated conditions (not “typical” values without test conditions). If the datasheet returns with blank fields for cycle life or operating temperature range, treat those as untested, not just unspecified.
Request 2: The factory’s internal QC outgoing report for your specific sample lot. Ask for it by serial number. A factory that can produce a per-unit outgoing test report within 48 hours has real process control. A factory that sends you a generic template with the serial number filled in by hand is operating on informal QC — and that is who will ship you off-spec production units six months after you’ve designed in their product.
Request 3: UN 38.3 test report with the actual cell model and configuration matching the sample you received. Cross-check the cell model code on the report against the physical cell inside the sample. We’ve seen factories submit UN 38.3 reports for a 4S1P 18650 configuration on a sample that is actually 2S2P — two completely different pack architectures with different abuse responses. UN 38.3 section 38.3.2.3 requires testing of the specific configuration as-shipped. Shared certificates across configurations are non-compliant by definition.
Request 4: BMS specification sheet including protection thresholds. Specifically ask for: OVP (overvoltage protection threshold), UVP (undervoltage protection threshold), OCP (overcurrent protection), and short-circuit response time. If the supplier cannot provide these four numbers, they are using a black-box BMS IC with no customization access. That’s acceptable for some consumer applications, but if your product has any demanding load profile or operates in temperature extremes, you need to know these thresholds to evaluate field safety.
Request 5: Cycle life test data at your specified rate and temperature. Not the cell manufacturer’s published data — the pack factory’s own test data on their assembled configuration. Many Shenzhen-area pack houses use cell-level cycle data from CATL or EVE datasheets to represent their pack performance, which is meaningfully different once BMS cutoff behavior is included. A pack that cuts off at 3.0V per cell versus 2.8V will show different capacity retention curves over 500 cycles even with identical cells.
Request 6: Two additional samples beyond the one you intend to destructively test. If a factory hesitates to send three units at evaluation stage, that hesitation itself is a qualification input.
Response timeline is part of the evaluation. A factory that returns the complete SRQ-3 packet within five business days is organized. Ten days is marginal. Two weeks with incomplete data usually indicates either a factory without proper documentation infrastructure or a trading company reselling another factory’s product — in which case you will never have direct access to engineering support when you need it.
Cost-Performance Trade-offs in This Category #
The portable charger category is crowded at the low end, and that creates a real sourcing temptation: Grade-B cell packs at $0.038–0.044/Wh output capacity (ex-works Shenzhen, MOQ 500 units) versus Grade-A LFP or NMC packs at $0.061–0.071/Wh. The price gap is real. So is the performance gap — but it doesn’t always matter.
For a promotional giveaway product with an expected use life of 20–40 cycles, the Grade-B option is probably correct. The capacity fade doesn’t reach user-visible levels until cycle 80–120, the safety certification requirements are the same regardless of cell grade (from a market compliance perspective), and the cost savings are substantial at volume. I’d prioritize cell grade for applications where end users will rely on the product daily for 18+ months.
The counterargument to always buying Grade-A: warranty cost modeling. One European consumer electronics brand we worked with ran a field return analysis across two SKUs — identical external design, one with Grade-A NMC 21700 cells, one with Grade-B 18650 cells. At 18 months, the Grade-A SKU had a 1.4% warranty return rate versus 6.8% for the Grade-B SKU. At their volume, the Grade-A premium paid back in warranty savings within 14 months of market launch. That calculus only works above a certain volume threshold — roughly 10,000 units annually in their model.
Where the cheaper option is genuinely correct: internal-use charging stations for low-cycle corporate environments (conference room charging drawers, hotel amenity packs). Low duty cycle, controlled temperature, no brand reputation exposure. Grade-B cells with a known cycle life of 300+ cycles will outlast the product category relevance.
One cost factor buyers consistently underestimate: the BMS, not the cell, often drives the price difference between a $4.50 and a $7.20 OEM unit at the same capacity. BMS Engineering covers this in depth, but the short version is that BMS IC selection and firmware quality account for 30–40% of the performance delta between competitive quotes at identical cell specifications.
Technical Deep-Dive: Evaluating SOC Accuracy Under Real Load Conditions #
State-of-charge accuracy is the spec that most power bank buyers never test during sample evaluation — and it is the one that drives the most field complaints after launch.
A power bank that shows “2 bars remaining” when it has 6% actual charge left creates a user experience failure that generates reviews, returns, and social media complaints. It is also a safety-adjacent issue: a user who believes the pack has remaining charge and leaves it on a charging pad at near-empty will subject the BMS to a more demanding recovery scenario than intended.
The SOC display accuracy spec in most datasheets, when it appears at all, reads something like “±5% at stable conditions.” That number is usually measured at 0.1C discharge rate, 25°C, after a full charge-rest sequence. Under those conditions, even a basic coulomb-counting BMS will hit ±5%. The problem emerges at 1C discharge with a variable load (which is what USB PD 65W output creates), and after 200+ cycles when cell capacity has degraded but the BMS SOC algorithm hasn’t recalibrated.
We test SOC accuracy during sample evaluation using a three-condition protocol:
| Test Condition | Load Profile | Cycles Applied | Acceptable SOC Error |
|---|---|---|---|
| Baseline | 0.2C constant, 25°C | 0 (fresh) | ±4% |
| Moderate stress | 0.5C constant, 25°C | 50 cycles | ±6% |
| High-load variable | 1C pulsed (USB PD sim), 35°C | 100 cycles | ±8% |
SOC accuracy under the high-load condition at 100 cycles is where cheap BMS firmware degrades visibly. In our testing of 11 samples from 7 Shenzhen-area pack suppliers over a 14-month period, 4 samples showed SOC errors exceeding ±15% under the 1C pulsed condition at cycle 100 — all four from factories using a specific commodity BMS IC without adaptive calibration. The same factories quoted datasheets showing ±5% accuracy, which was technically accurate for the baseline condition they measured.
The technical mechanism: coulomb counting accumulates drift over each cycle. Without a full-charge calibration reset event happening at least every 10–15 cycles, and without a load-adaptive correction algorithm, SOC error compounds. A BMS that recalibrates only at 100% SOC (common in simple designs) will show poor accuracy in any application where the user rarely charges to full — which describes most power bank use patterns.
One practical test you can run on samples without lab equipment: charge to 100%, discharge under a 10W USB load until the display shows 10% remaining, then continue discharging into a resistive load while measuring actual output voltage. Record the capacity extracted between “10% display” and actual cutoff. On a well-calibrated 20,000mAh (rated output) unit, you should extract no more than 1,800–2,000mAh in that final 10% window. If you extract under 400mAh, the SOC algorithm is severely front-loaded. If you extract over 3,500mAh, the low-SOC warning is dramatically delayed — a safety-relevant scenario for lithium-ion under deep discharge.
We haven’t fully characterized how NMC versus LFP cell chemistry affects this test result at lower cycle counts. LFP’s flat discharge voltage curve creates more fundamental challenges for voltage-based SOC estimation, and factories using LFP cells without adaptive algorithms should be asked specifically how they handle this. Our dataset covers mostly NMC 18650 and 21700 configurations — we’ll extend it to LFP cylindrical cells after our Q3 supplier audit cycle.
Compact BESS Products buyers sourcing larger stationary units face the same SOC accuracy challenge at much higher consequence — the underlying BMS evaluation methodology transfers directly.
Sourcing Guidance for Buyers #
When evaluating Chinese suppliers in this category, the first document to request is the outgoing QC report for your specific sample serial number. Not the generic inspection checklist, not the factory certification wall — the per-unit test record showing measured capacity, impedance, and protection threshold verification for the exact unit sent to you. A supplier that cannot produce this within three business days either doesn’t test individual units before shipment or doesn’t maintain traceable records. Either situation creates production risk that will compound at volume.
The qualification red flag specific to power banks and portable chargers: any supplier who quotes cycle life data without specifying discharge rate, temperature, and depth of discharge is presenting unanchored data. “500 cycles” means nothing without conditions. Suppliers with real engineering depth will reference their test conditions without being asked; those without it will get defensive or vague when you request the methodology. That defensiveness is itself a qualification input.
For incoming inspection of production samples, test a minimum of 5 units per 500-unit lot using the IEC 62368-1 clause 5.4 energy source classification procedure to verify protection threshold activation. Measure USB output capacity at 0.5C rate and reject the lot if the average falls below 91% of the contracted specification or if any single unit falls below 87%. Safety & Certification teams evaluating these units for market compliance should also review UL 2056 for North American market requirements alongside IEC 62133-2 for international coverage.
Progress to a production supply agreement only after evaluating a second sample lot of at least 10 units produced under the same production line conditions as your contracted order — not the R&D samples that first passed evaluation.
FAQ
What sample quantity should I request for a power bank evaluation?
Request a minimum of 3 units: one for capacity and cycle testing, one for BMS threshold and SOC accuracy testing, and one retained as a reference sample sealed against the datasheet. If your evaluation includes drop testing or ingress protection verification, request 5–6 units minimum.
How long should the full evaluation-to-design-in process take?
From initial sample inquiry to production supply agreement, expect 10–16 weeks if the supplier is organized. Roughly 2 weeks for sample preparation and shipping, 4–6 weeks for your evaluation protocol, 2 weeks for supplier response and negotiation on any out-of-spec findings, and 3–4 weeks for a second-lot verification run. Compress any of these stages and you’re accepting unquantified risk.
Can I rely on a factory’s published datasheet instead of testing samples myself?
No. Datasheets from Shenzhen pack houses reflect best-case measurement conditions unless explicitly stated otherwise, and many are copied from cell manufacturer data rather than pack-level test results. The datasheet is a starting document for your inquiry, not a substitute for incoming test data.
What’s the right impedance threshold for rejecting a power bank sample?
It depends on cell chemistry and configuration. For NMC 18650 4S2P packs, reject samples where pack-level impedance at 1kHz exceeds 28mΩ at a fresh, fully-charged state. For LFP configurations the threshold is different given the chemistry’s naturally higher impedance — I’d set the rejection threshold at 45mΩ for a comparable configuration and weight the SOC accuracy test results more heavily than impedance alone.
If a supplier passes sample evaluation but fails incoming inspection at production, what typically went wrong?
Usually one of two things: the sample was hand-selected or assembled separately from the production line, or the production cell grade changed between sample and volume order. Both are preventable. Require that your production supply agreement specifies cell model, grade, and manufacturer by name — not just “equivalent cells” — and reserve the right to inspect cell traceability documentation on each production lot.
Published by compactbess.com Technical Team | Request a sourcing consultation
Ran into this exact gap when qualifying cells for a tethered UAV ground power unit — supplier was rating at 0.2C cell-level and our flight ops profile was pulling closer to 0.8C sustained, which put us about 18% below the quoted capacity in actual field conditions. We ended up writing a dual-rate spec into every subsequent RFQ: IEC 62133-2 baseline for compliance documentation, plus a secondary 0.8C system-output figure as the contractual acceptance threshold, and that second number alone dropped our qualified supplier pool from 11 to 4.
The cell-to-system output gap caught us on a defibrillator charging module — we’d spec’d 9,800mAh at cell-level with our CM in Shenzhen, didn’t realize their protection PCB was pulling ~180mA quiescent during the high-voltage boost phase, which ate nearly 8% of usable capacity before we even saw it at the output terminals. Took three board revisions and a firmware patch to the BQ40Z80 fuel gauge before our system-level SOC estimate stopped drifting past our ±6% threshold under clinical cycling conditions.