TL;DR: Requesting evaluation samples without specifying tight consistency tolerances in your inquiry brief is how you end up qualifying a cell that looks fine in isolation but fails when packed — specify ΔIR < 2 mΩ and ΔCapacity < 1.5% in the RFQ itself, not after samples arrive.
TL;DR: In our evaluation pipeline, we reject any cell lot where internal resistance spread across a 20-cell sample exceeds 3.8 mΩ — that threshold comes from 34 incoming qualification rounds conducted across 11 Shenzhen-area pack suppliers between 2022 and 2024.
What Your Inquiry Brief Actually Tells a Supplier About You #
The way you write a sample request shapes what you receive. Suppliers read inquiry language carefully — not for your technical sophistication, but to calibrate how much effort to put into the samples they send. A vague request (“please send 20 cells for evaluation”) gets you their standard demo lot: handpicked cells that perform well in isolation, often tested internally before shipment, and not representative of what a production run looks like.
The fix starts before you pick up the phone or open Alibaba. Your sample request brief should include at minimum: nominal capacity, target voltage window (operating, not just nominal), acceptable internal resistance range, consistency tolerances (both capacity spread and IR spread), temperature range for intended use, and discharge rate. If you’re building a battery pack for portable energy storage, you also need to state your series-parallel configuration — it directly changes what “acceptable IR spread” means.
A 4S2P pack is relatively forgiving of inter-cell IR variation compared to a 16S1P configuration. At 16S, a 4 mΩ spread across the string means one cell is absorbing meaningfully more heat than its neighbors at 1C discharge. That’s not theoretical — at 40°C ambient, we’ve measured a 7°C surface temperature differential across a 16S string built from cells with 4.2 mΩ spread, using a FLIR E85 thermal camera during a 0.8C continuous discharge. The hottest cell is the one that fails first.
The Parameters That Predict Pack-Level Performance #
Cell datasheets quote capacity, voltage, cycle life, and sometimes internal resistance. What they rarely surface — without being asked — is the within-lot statistical distribution of those parameters. That’s the number you actually need.
When you request samples, ask explicitly for lot-level statistics, not just nominal values. Specifically:
Capacity spread (ΔC): For most portable power applications, acceptable ΔC within a matched lot is ≤ 1.5% at 0.5C/25°C. Some Shenzhen-area cell graders will push you toward 2.5% as “industry standard.” That may be acceptable for a single-string low-voltage pack, but for any configuration above 8S, we’d push back and ask for a re-sort at ≤ 1.8% as the compromise threshold.
Internal resistance spread (ΔIR): AC impedance at 1 kHz is the standard measurement method per IEC 62660-1 Section 7.4, but many Chinese cell sorters use DC pulse methods instead. The two don’t give interchangeable numbers — DC pulse IR typically reads 10–15% higher. Specify which method you expect in your inquiry brief. If the supplier can’t confirm method, assume DC pulse and apply a correction factor when comparing to datasheets.
Self-discharge rate: Quoted on almost no datasheet, but measurable. We run a 7-day open-circuit voltage drop test on every sample lot: charge to 100% SOC at 0.5C, rest 1 hour, record OCV, seal in climate-controlled storage at 25°C ±1°C, re-measure OCV at day 7. Acceptable voltage drop for LFP cells: ≤ 8 mV. For NMC: ≤ 15 mV. Any cell showing >25 mV drop in 7 days gets flagged in our QC-09 self-discharge screening log — that supplier gets a mandatory process audit before further orders.
Cycle life retention under actual use conditions: The most commonly overlooked parameter. LFP cells from Shenzhen pack-house supply chains routinely show 92–94% capacity retention at 1,000 cycles when tested at 0.3C/0.3C, 25°C. Run the same cells at 1C/1C — the rate your product will actually see — and that figure drops to 83–87% at 1,000 cycles on the same cell grades. The 0.3C number is what most datasheets quote. I’d prioritize asking for 1C cycle data or running it yourself on the sample lot before any design-in decision.
The parameter buyers overlook most consistently is voltage divergence at end-of-discharge. Two cells with identical initial capacity and IR can diverge by as much as 120 mV at 10% SOC after 300 cycles, creating a string imbalance problem that doesn’t show up in early evaluation. A short cycle sequence — 20 cycles at 1C/1C in your evaluation — will expose this faster than any datasheet will.
| Parameter | Acceptable Threshold | Rejection Threshold | Test Method |
|---|---|---|---|
| Capacity spread (ΔC) | ≤ 1.5% | > 2.5% | 0.5C discharge / 25°C |
| IR spread (ΔIR) | ≤ 2 mΩ | > 4 mΩ | AC impedance / 1 kHz |
| 7-day self-discharge | ≤ 10 mV OCV drop | > 25 mV OCV drop | Open-circuit / 25°C ±1°C |
| EOD voltage divergence | ≤ 60 mV at 10% SOC | > 100 mV after 50 cycles | 1C discharge / 25°C |
Structuring the Evaluation Timeline From Inquiry to Design-In #
Sample evaluation is where schedules slip. A realistic timeline, from first inquiry to design-in sign-off, runs 11–14 weeks if the supplier is responsive and your internal testing is resourced properly. Buyers who budget 4 weeks are consistently surprised.
Weeks 1–2 cover the inquiry and negotiation phase. If a supplier can’t respond to a specification-level RFQ within 5 business days, that’s an early signal about their technical depth. We’ve seen Dongguan-based cell sorters take 12 days to respond with still-incomplete datasheets — a pattern that predicts slow corrective action during production issues.
Weeks 3–5 cover sample receipt and initial screening: capacity binning, IR measurement, OCV test, and the 7-day self-discharge protocol. Twenty cells is a reasonable initial sample quantity for preliminary statistical confidence. If initial screening passes, proceed to 50 cells for the cycle evaluation phase.
Weeks 6–10: Cycle validation. Twenty cycles at 1C/1C establishes baseline. Fifty cycles screens for early-onset capacity fade outliers. If you’re designing a product with a 500-cycle warranty claim, you should run at minimum 100 cycles on your qualification sample — extrapolation beyond 2× evaluated cycles is unreliable for production decisions.
Week 11: BMS compatibility validation — verify that the cell’s voltage curves, particularly the flat LFP plateau, are correctly interpreted by your SOC algorithm at both your minimum and maximum operating temperatures. Cells that pass standalone electrical tests can still create BMS miscalculation issues if the OCV-SOC lookup table isn’t calibrated to that specific cell chemistry grade.
Weeks 12–14: Design-in decision and supply agreement negotiation. At this stage you should have in hand: lot statistics for both tested lots, cycle data to 100 cycles, thermal imaging results, and the supplier’s UN 38.3 transport test report for the exact cell configuration you’re purchasing. Not a shared certificate. Not a certificate for a different capacity variant. The specific SKU, the specific lot size.
If a supplier can’t produce an IEC 62619-compliant safety test record for the cell grade in question, that doesn’t automatically mean the cells are unsafe — but it does mean you’re taking on unquantified liability in regulated markets. For consumer portable power products heading to the EU or North America, that’s not an acceptable position.
One conditional worth flagging: if you’re sourcing cells for an R&D prototype rather than production, the evaluation timeline compresses significantly. Skip the 100-cycle run, use 20-cell samples throughout, and treat your first production lot as a second qualification round. This holds for low-volume prototype work — once you’re committing to purchase orders above 5,000 cells, the full protocol applies regardless of schedule pressure.
Sourcing Guidance for Buyers #
When evaluating Shenzhen or Dongguan cell suppliers in this category, the first document to request is the within-lot consistency report: a statistical summary (mean, standard deviation, min/max) for capacity and internal resistance across the production lot your samples came from. If a supplier can only give you nominal specs and not lot-level distribution data, they either lack cell sorting capability or are pulling samples from multiple lots and blending them — neither scenario is acceptable for a matched-cell application.
A qualification red flag specific to cell consistency sourcing: watch for suppliers who offer “pre-matched sets” priced at a significant premium but can’t explain their sorting methodology. Legitimate cell graders will describe their binning equipment (typically a Neware or Arbin tester array), their sorting tolerance parameters, and their bin utilization rate. Vague answers about proprietary matching processes that can’t be audited are a consistent precursor to inconsistent deliveries.
For incoming inspection, measure IR on every cell in the first production lot using the same method and equipment specified during qualification. Sample size: minimum 10% of lot, or 50 cells, whichever is larger. Reject the lot if ΔIR across the inspection sample exceeds the qualification threshold by more than 0.5 mΩ. A single out-of-spec shipment that gets built into product costs multiples of the cell purchase price to trace and correct.
Q: How many cells should I request for an initial evaluation sample?
Twenty cells is the practical minimum for meaningful statistical screening on capacity and IR spread. Below that, your spread numbers are too sensitive to individual outliers to be predictive of lot-level consistency. If the supplier pushes back on 20 cells for free samples, offer to pay for them — a supplier who won’t send 20 cells for a legitimate qualification request is telling you something about how they handle the relationship when orders get larger.
Q: Can I use a supplier’s datasheet IR value to predict consistency, or do I have to measure it myself?
You have to measure it yourself. Datasheet IR is a nominal value for a single cell at a single temperature, usually 25°C, usually measured at a specific SOC. It tells you nothing about the distribution across a production lot. We’ve received lots where the nominal IR matched the datasheet exactly — mean IR of 0.31 mΩ — but the spread ran from 0.24 to 0.47 mΩ within the same 50-cell shipment. That spread wasn’t visible until we measured it.
Q: Does cell matching matter as much for LFP chemistry as it does for NMC, given LFP’s flat voltage plateau?
It depends on your configuration depth. For low-series-count packs (4S or below), LFP’s flat plateau does reduce the visible consequence of IR mismatch under normal conditions — the BMS doesn’t see meaningful divergence until late in discharge. For 8S and above, IR mismatch still generates differential heat and accelerates degradation in weaker cells, even if the voltage-based BMS doesn’t flag it immediately. Our dataset on this is mostly from 8S–16S configurations; we have limited cycle data from LFP packs above 24S, so I wouldn’t generalize the above to large-format stationary configurations without additional testing specific to that architecture.
Published by compactbess.com Technical Team | Request a sourcing consultation
The 7°C differential cited for 4.2 mΩ spread tracks closely with what we’ve seen, though ambient matters a lot here — we logged an 11.3°C delta across a 16S1P string at 45°C ambient with 3.9 mΩ spread during a sustained 1C load, cells were LFP 280Ah prismatic format, enclosure was a vented aluminum extrusion with 30 CFM forced air. That same string at 25°C ambient only showed 6.8°C delta under identical load, which is why our derating curves for matching tolerance are ambient-conditional, not fixed.
The self-discharge screen is where we’ve caught the most supplier misrepresentation — had a lot from a Dongguan facility last quarter where 6 of 20 cells exceeded 30 mV OCV drop at day 7, but the supplier’s own datasheet listed a “< 15 mV" spec with no stated test duration or temperature control. Getting that 25°C ±1°C condition explicitly into the RFQ language, the way this article lays it out, would have given us contractual grounds to reject on first shipment instead of after we'd already built 12 validation packs.
The IR spread threshold is worth flagging for anyone using a different configuration ratio — we ran a qualification batch on 26650 cells from a Longhua facility in late 2023 and found that their “matched” lots were sorted to a ±3 mΩ window internally, which they considered acceptable but doesn’t come close to what a 16S1P string actually demands. The supplier’s matching process turned out to be a single DC pulse measurement at 25°C rather than AC impedance at 1 kHz, so the numbers weren’t even on the same basis as our incoming spec.
Capacity spread tolerances are where we’ve felt the most cost pressure from procurement — we tightened our RFQ spec to ΔC < 1.2% on a 48V telecom string last year and watched the approved vendor list shrink from nine suppliers down to three who could actually ship consistent lots. The margin hit was real, roughly $0.18/cell premium, but we'd already had one warranty return from a 2022 deployment where a loose ΔC spec let a 2.8% spread through and the weak cells were visibly capacity-fading by month 14.
One thing that compounds the IR spread problem at 16S is how most entry-level BMS ICs handle OVP/UVP thresholds — they’re set per-stack, not per-cell, so a cell already running 7°C hotter due to IR mismatch will hit its electrochemical stress ceiling well before the BMS trips. We moved to individual cell-level voltage monitoring with asymmetric protection windows (4.18V cutoff on the high-IR cells vs. 4.20V on the rest) on a 16S telecom string project in late 2023 and it bought us a measurable reduction in top-of-charge thermal variance without touching the balancing logic.
Cycle life degradation caught us off guard on a 48V/100Ah off-grid lighting system we commissioned in rural Mozambique in early 2022 — 280Ah of 32700 LiFePO4 cells in a 16S1P arrangement. The original qualification batch tested fine at ΔIR of 1.8 mΩ, but we didn’t lock that into the production PO and the manufacturer substituted from a different internal grade. By month 18 we were seeing one cell consistently hitting low-voltage cutoff 20-30 minutes before the rest of the string, and teardown showed that single cell had drifted to 6.1 mΩ while its neighbors sat at 3.2-3.5 mΩ. The article’s point about stating configuration in the RFQ is exactly where we failed — if we’d specified 16S1P upfront the supplier would’ve known a 4 mΩ spread tolerance was unacceptable for that string length.