Summary of Findings

A value on a Certificate of Analysis (COA) is one draw from a distribution. It says where one sample from one batch fell; it says nothing about where the next batch will fall, and nothing about how wide the supplier’s spread is. Two suppliers can report identical figures on a single report while one sits mid-range with a narrow spread and the other rides the specification limit with a wide one — and only the second will produce a batch that fails.

Assessing stability therefore requires several batches, the sampling plan behind them, and attention to three properties of the set rather than to any member of it: where the distribution is centred relative to the limit, how wide it is, and whether it is moving. This article sets out how to request that data, what to look at, and what to write into a contract so that it keeps arriving.

Key Findings

• A single value cannot distinguish a stable supplier from an unstable one. The property that matters — spread — is invisible in one report by construction.
• Distance from the limit matters more than compliance with it. A distribution centred mid-range tolerates normal variation; one centred near the limit converts normal variation into failed batches.
• Three questions replace one. Where is it centred, how wide is it, and is it drifting? A drifting distribution that is currently in specification is a different situation from a stable one at the same position.
• Averages carry a caution. Reported averages should not be accepted as absolute where the underlying conditions fluctuate (Wilkinson et al., Infacon IX) — a caution made about consumption figures that applies with equal force to supplier-reported parameter averages.
• No international sampling plan covers electrode paste. The ISO 8007 series covers cathode blocks, prebaked anodes and sidewall blocks; the sampling plan behind a paste data set therefore has to be stated by the supplier rather than assumed from a standard.

Contents

1. Why can a single reported value not rank two suppliers?
2. What produces batch-to-batch variation?
3. What should be requested, and from how many batches?
4. What to look at in the data
5. What belongs in the contract
6. Frequently asked questions
7. Related reading

1. Why Can a Single Reported Value Not Rank Two Suppliers?

Because compliance and stability are different properties, and a single report only shows the first.

Consider two suppliers whose latest reports both give the same figure for a parameter, comfortably inside the specification. Behind those two identical numbers there may be two quite different production realities: one supplier centred mid-range with batches falling close together, the other centred near the limit with batches spread widely. Both reports pass. Only one of the two suppliers will keep passing.

The asymmetry is what makes this worth attention. A distribution centred mid-range absorbs normal variation without producing a failure; a distribution centred near the limit converts the same variation into rejected consignments, furnace-side surprises, and the discussions that follow. The single report cannot show which situation is in front of you, and no amount of scrutiny applied to that one report will extract the information — it is not in the document.

2. What Produces Batch-to-Batch Variation?

Four sources, in roughly the order in which they are worth checking:

Raw material lot changes. Aggregate arrives in lots, and lots differ — in mineral matter, in sulfur, and in the calcination they received. A step change in a parameter between batches, rather than a gradual movement, usually points here. Sulfur is the most sensitive indicator, because it enters with the aggregate and is not adjusted by formulation.

Homogeneity within the batch. Mixing and kneading decide whether every part of a batch carries the specified binder proportion. Where they do not, the batch average can match the formulation while its parts do not, and the variation shows up as spread across samples rather than as a failed average. This route is treated in the process control article.

Sampling. Paste is heterogeneous by construction, so where the sample is taken and how it is divided affects the reported value before any instrument is involved. Apparent variation between batches sometimes turns out to be variation in sampling practice, which is why the sampling plan has to accompany the data.

Measurement. Method, edition, basis and specimen preparation all move a result. Data assembled across a period during which any of these changed will show variation that belongs to the laboratory rather than to the material.

Only the first two are properties of the product. The value of asking for the sampling plan alongside the data is that it lets the last two be separated out before conclusions are drawn.

3. What Should Be Requested, and From How Many Batches?

Request three things together:

1. COAs for consecutive batches over a defined period — consecutive matters more than numerous, because a supplier selecting which batches to send will select the good ones.
2. The sampling plan behind them: number of increments, sampling points, division method. Without it, the spread cannot be attributed.
3. Notification of any raw-material source change within the period, which explains step changes that would otherwise look like instability.

On how many batches: there is no universal number, and any figure quoted as one should be treated with suspicion. The requirement is a set large enough that the spread is visible and a trend could be distinguished from noise — which depends on how wide the spread turns out to be and on how much of the specification range is available. In practice the workable approach is to request a defined period of consecutive production rather than a batch count, review what the spread looks like, and extend the period if the picture is ambiguous.

What can be said firmly is the lower bound: one batch shows nothing about stability, and three consecutive batches show very little. A supplier who can only produce a handful of reports has told you something about their record keeping, which is itself relevant.

4. What to Look At in the Data

Four readings, in order:

Where the distribution sits relative to the limit. This is the first thing and often the only thing that needs saying. A supplier centred mid-range has room; a supplier centred near the limit does not, regardless of whether every batch to date has passed.

How wide it is. Width relative to the available specification range is what determines the failure rate over time. Comparing widths between two suppliers is a fair comparison only where both were sampled and measured the same way — hence the sampling plan.

Whether it is moving. A set that is drifting in one direction is a different situation from a stable one at the same position, and it requires a different conversation: something upstream is changing. Drift in sulfur without movement in ash points to raw material source; drift in volatile matter with movement in softening point points to binder.

What the outliers are. An outlier with an explanation attached — a known raw material lot change, a documented process event — is more reassuring than a data set with no outliers at all, which sometimes means the outliers were not sent.

This article deliberately gives no numerical guidance on acceptable spread. An acceptable width depends on the specification range, the furnace’s own tolerance and the operating cadence, and a figure carried in from another plant would be misleading. The comparison that matters is between suppliers sampled the same way, and between a supplier’s current spread and its own history.

5. What Belongs in the Contract

Five clauses keep the data arriving after the first order:

1. Batch data provision — COAs for every consignment, with the sampling plan stated once and referenced thereafter.
2. Retained samples at both ends, with a retention period covering ocean freight plus the buyer’s inspection cycle. Retained samples are what make a later dispute resolvable rather than argumentative.
3. Notification of raw-material source change, before the affected consignment ships rather than after a question is raised.
4. Notification of formulation change, on the same terms. A formulation change and a form change should never arrive together.
5. A periodic review of the accumulated data against the previous period, at whatever interval matches the order pattern.

None of these is onerous for a supplier who is already doing the work. A supplier for whom they are onerous has answered a question about their process control, which is the second reason for asking.

6. Frequently Asked Questions

Q1. Our supplier’s every batch has passed. Is that sufficient evidence of stability? It establishes compliance to date. It does not establish where the distribution sits or how wide it is, and a supplier riding the limit will also show a record of passes until the batch that does not. Ask where the batches fell within the range, not only whether they passed.

Q2. The supplier sends a typical value rather than batch data. Is that usable? Not for this purpose. A typical value is a statement about a product, not about a production period, and it cannot show spread or drift. Reported averages should also be read with the caution that they should not be accepted as absolute where underlying conditions fluctuate — a caution made about consumption figures (Wilkinson et al., Infacon IX) that applies equally here.

Q3. Two suppliers’ data sets look different. How do we know it is the material and not the sampling? You do not, unless the sampling plans are stated and comparable. This is why the plan is requested alongside the data. Where the plans differ materially, the widths are not comparable and the honest conclusion is that the comparison cannot be made on the data available.

Q4. How much variation is acceptable? That is set by your specification range and your furnace’s tolerance, not by a general figure. The working method is to establish what width your process absorbs without a change in behaviour, and to treat that as the requirement. A spread narrower than that buys nothing; a spread wider than that will eventually be seen at the furnace.

Q5. Should batch data be requested during a supplier trial, or only after? During. A controlled trial establishes whether a paste works in a specific furnace; batch data establishes whether the next consignment will resemble the one that was trialled. The two answer different questions, and a trial conducted on a single consignment without any view of spread has verified one batch rather than a supplier.

7. Related Reading

• Reading a Seven-Parameter Test Report for Internal Consistency (companion article) — checking the individual reports that make up the data set
• How to Compare Electrode Paste Supplier Offers Without Relying on One Test Value — the purchasing decision this data supports
• How Calcination, Mixing and Kneading Set the Properties You Buy — where homogeneity, and therefore spread, is decided
• How Sampling and Sample Preparation Affect Electrode Paste Test Results — the sampling question behind every comparison here
• How to Plan a Controlled Electrode Paste Trial When Changing Suppliers — the trial that batch data accompanies
• The First 30 Days After a Paste Supplier or Batch Change — what a change looks like at the furnace

Author and Review

Written by Sterling Yang, Senior Vice President at Jiyang Carbon.

This article was reviewed by Mr. Yang, Technical Director at Jiyang Carbon, drawing on 40+ years of hands-on experience in carbon processing and weekly review of production data.

No numerical guidance on acceptable spread is given, because an acceptable width depends on the specification range, the furnace’s tolerance and the operating cadence, and no general figure would transfer. The curves in Figure 1 are illustrative and represent no measured data. Jiyang Carbon tests seven parameters on every batch in its own laboratory, issues a Certificate of Analysis traceable to the batch with every consignment, and operates a quality management system certified to ISO 9001 and audited annually.

Technical Consultation

For batch data covering a defined production period, or for assistance in reviewing an accumulated data set, please contact the technical team.

Get a Technical Consultation →