ApiaryActiveLive
Try: pause · settings · learn · wipe
← Community / Reading Room
CT
craft · 24 min read

Count the Sticker Waste Before You Change the Setup

A community workshop prints small sheets of address labels for an exhibit. One week there are ten rejected labels in the tray. The next week there are six.…

AI disclosure: Drafted with AI assistance for Apiary using publicly documented waste and quality concepts. The workshop, records and calculations are fictional. This is for low-consequence paper or label work, not regulated production, chemical handling, or printer repair.

A community workshop prints small sheets of address labels for an exhibit. One week there are ten rejected labels in the tray. The next week there are six. The second number looks like an improvement until somebody checks what was printed. The first run made dozens more labels, included its setup scraps, and was reviewed again the next day for late errors. The second run counted only the labels that happened to be visibly wrong when the printer stopped. Changing an ordinary setup routine might have helped. The receipts do not yet make those two numbers comparable.

A baseline is not a ceremonial average from the day before an improvement. It is a way to make ordinary records comparable for a particular decision. Write down the physical unit, what counts as waste, the cutoff for review, the types of job included, and any important conditions. Then create a brief process that can survive a busy afternoon and an imperfect handoff. The goal is to discover whether waste moved, not to prove a favorite change worked. A stable definition makes useful decreases easier to recognize and unsupported ones easier to leave alone.

Decide what a unit is before drawing tally marks

The fictional workshop prints paper labels in eight-position sheets. A wasted sheet and a wasted label aren't the same event. A color alignment error across a sheet might affect every printed position. A clipped corner on one sticker can leave its neighbors perfectly acceptable for this low-risk exhibit. An entirely blank backing area might have been intended by the layout, not spoiled. Before counting, define the unit relevant to the decision. The maker could care about labels lost per printed position attempted; the steward may separately care about physical sheets consumed, including paper that cannot be used again. These are two measures, not two names for one.

Use three tags on the first practice sheet. Identify which positions were intended for regular copies, which were intentionally blank, and which were deliberately used as setup tests. When counting attempted copies, do not quietly put intentionally blank positions in the denominator; doing that dilutes a rate without saving anything. Conversely setup tests can consume real materials even if they are not attempts to fill an order. Give each run a small second tally for setup-only paper. If a job uses setup labels that also become useful copies after ordinary inspection, the procedure should say how to record their transition without counting them twice.

The distinction goes beyond paper. A failure record may describe one item with two problems. The text is off-center and the top edge is curled. Those count as two defect observations if you're studying failure modes, but they do not turn the single item into two spoiled labels. Conversely three physical pieces with the same defect are three affected items. NIST's discussion of attribute control charts distinguishes proportions of nonconforming units from counts of defects. The distinction is useful here long before a small workshop has any reason to calculate a formal chart. Count precisely what the numerator claims to count.

Write a simple check sheet with a few mutually understood fields. Give each documented run an ID, a clear output unit, intended copy count, setup material, counted eligible copy attempts, immediately rejected copies, and late review adjustments. To avoid false precision, leave machine diagnostics and supplier assumptions in a notes area rather than making them obligatory for every piece. ASQ's check-sheet description describes a prepared form for gathering observations. A short tally can be reliable if collaborators understand the definitions and can actually finish it while work is happening. Unsurprisingly polished records aren't always better than a sheet where people know what the marks mean.

Draw a boundary for what counts as the baseline

Choose the work you want the baseline to describe. Imagine the workshop has a family of ordinary unlaminated labels with comparable expected layout and uses a different station for glossy event sheets. If the planned change concerns how the ordinary sheets are prepared, keep the headline comparison within that family. An unusually complicated event may still deserve its own row, but it shouldn't quietly become the whole story of an ordinary week. Specify what equipment and versions matter, where jobs begin and end, and which calendar period the observation reflects. There is no universal requirement to collect exactly a certain number of weeks before making a small reversible decision.

Put the scope note into practice with a pretend volunteer. One person sees the baseline as every sticker run in the building; another hears only the ordinary exhibit labels. Give them a short list of synthetic jobs and ask which should be included and why. If the classification differs, make the labeling more specific. A job can be out of scope without disappearing from all materials records. For example actual paper waste from the special event can still be counted elsewhere. The goal is to stop it being represented as evidence about an unrelated routine under a different preparation process.

Pick the relevant time boundary as deliberately as the product boundary. An immediately visible flaw is easy to mark at the printer. An error spotted during packing is also real, but it may appear after the end of the earlier shift. A forgotten late review counts neither as perfect nor as known bad. One invented policy uses a final receipt after the order is checked for pickup, with a defined follow-up window during which late discoveries are joined to the original run. This choice can be adapted to a real workshop's process. What matters is that the before and after comparison use compatible cutoff rules or disclose a difference.

Some waste belongs to neither accepted printed labels nor late rejects. An initial alignment page, a partial sheet abandoned after interruption, and materials kept for another appropriate purpose can each follow a different path. Use a small set of explicitly named dispositions. An unused sheet physically returned to stock is not the same as a sheet put in a reuse box; neither necessarily belongs under destroyed label copies. The EPA materials hierarchy distinguishes preventing waste at the source from later management. Your tally does not certify environmental outcomes. It can keep material prevented from being used, diverted to another use, and discarded from being misleadingly combined.

Give a cloudy case a proper place. A sheet has stopped partway and cannot be finished, but the log does not indicate how many positions the file actually attempted. Count the known physical sheet under the relevant material category. Mark the attempted label rate unavailable for that row unless it can be reconstructed from reliable evidence. Do not assume all intended positions reached material simply because they were on an order. Missing denominator can make a rate uncomputable without making the incident invisible. This is one of the benefits of preserving both physical material and product outcomes as separate measures.

Walk one fictional run across setup, production and follow-up

Here is a dry run using numbers deliberately small enough to audit by hand. A job calls for 48 acceptable labels. One eight-position sheet is deliberately consumed as a setup test and does not count toward regular label attempts. Seven regular sheets, each with eight intended positions, are then printed. Of those 56 regular positions, five are rejected immediately, 49 are provisionally accepted and two are held for closer review. Check the identity at the moment of the count: 5 + 49 + 2 equals 56. The 48 is a demand target, not the denominator. Using it to describe spoilage would lose the fact that the maker attempted several extras.

Later, one of the two held pieces passes and the other fails an agreed ordinary visual requirement. Of the 49 provisionally accepted pieces, one is then found to contain an old event code before the order leaves. It is removed from the usable set and recorded as an additional late rejection. The final regular outcome is 49 usable and seven rejected, with no unresolved pieces. The material total is eight sheets used: one setup sheet and seven regular. The order needs only 48 labels, so the extra acceptable piece follows a declared disposition. It can be retained as a legitimate spare, transferred to a planned use, or eventually discarded. “Not immediately delivered” is not the same as known spoiled.

A count-first report for this exercise might say: “Seven rejected regular positions out of 56 attempted, about 12.5 percent for this fictional run. One separate setup sheet. All held pieces reviewed by the pickup checkpoint.” It doesn't say that paper waste is 12.5 percent of eight sheets, since damaged positions and material use aren't interchangeable without an established mapping. It also doesn't claim the mock job was a success just because the count reaches the target; the final acceptance requirements are a separate decision. In a real report save the observed dispositions as well as the arithmetic so readers can track a late item back to the originating job.

Next supply an incomplete follow-up for the same job. Imagine the two held pieces were accidentally carried away as examples and no further review is available. The stronger record would report immediate rejects and unresolved holds, not assume those holds passed because the next person found no rejected pieces near the printer. If follow-up judgment appears later, add its time and update with the same run ID while preserving the prior state. A chart can have a preliminary row, a follow-up timestamp, or exclusion from the final comparison. It doesn't need a physically invented value to keep the columns aligned.

A later defect is not necessarily a printing defect. The old event code might have been approved in the wrong revision, or labels might have been mixed after release. The disposition is observable: one label rejected before delivery. The cause may remain unknown. Keep a separate suspected-reason field if useful, and show the evidence level. A week of neatly categorized guesses can be more misleading than several unresolved entries. The point is to make waste visible in a way that helps select the right follow-up experiment, not to diagnose any machine issue or assign blame from a defect tally.

Reconcile shop-floor waste with physical material without subtracting fiction

Two people can independently count the output of a job and still disagree with how a third person accounts for the stock. Keep a simple material balance using one explicit comparable unit where possible. In the invented first run, the stockkeeper issues nine whole sheets to the station. The maker uses one for setup and seven on regular copies, and returns one unused intact sheet. That balances the issued whole sheets. It does not mean seven whole sheets were lost just because a few of their positions failed. Some sheets contain usable labels while others retain their backing. If the workshop needs a more detailed material account, design one specifically for the physical remnants rather than simply adding all operational tally fields together.

Label genuinely observed material states. Returned whole stock is materially different from an opened partial sheet held for training. A sheet consumed in setup can be documented whether it was later diverted for harmless scrap use or discarded. A backing piece left after successful labels were removed represents material handled during ordinary production, not necessarily process spoilage. A usable output requires a definite number of materials even in a zero-rejection run. Decide which part your proposed change could plausibly reduce. Without a material definition, a claim that the pilot prevented waste may be true about rejected copies and silent about setup paper consumed.

Consider a hypothetical before and after with the same regular label result. Both shifts have 56 regular attempts and seven rejected copies. The first shift used a separate setup sheet and the second did not. The regular rejection rate is unchanged, while observed setup sheet use differed by one. That difference matters if the goal was to save setup stock. Now reverse the example: the number of setup sheets is stable but several fewer regular pieces are rejected. Depending on how layout packs positions, whole sheets used may not fall at all over a small order. An effective report can state the change without claiming a stronger material saving the observed data do not establish.

A late accounting entry can change apparent numbers without changing any physical operation. Suppose the stockkeeper discovers that a good leftover was used at another table. Record a transfer under an appropriate category rather than revising the original job to make it look like it printed fewer attempts. If the source of that piece cannot be recovered, retain an unidentified material disposition instead of relying on it to make totals look neat. A small discrepancy can be an opportunity to improve the record at the next iteration. It is not license to assign losses to a particular operator based on the residual of a bad log.

For a paper-only practice session prepare two cards. One records a setup page entirely discarded and a production sheet with one rejected label; the other records one partially printed practice sheet retained for a planned use. Ask both reviewers to calculate regular rejected items and whole sheets issued independently. The first result should not be made of contradictory material categories. Where the exercise cannot tell exactly how much remains reusable, make that unknown visible. The practical lesson is that reducing unnecessary virgin input and finding a reuse for existing scrap are different improvements with different receipts. Both can matter without double counting the same physical piece in a showcase slide.

Normalize a denominator without hiding jobs that went missing

A believable rate compares a subset with the opportunities that gave rise to it. For the workshop, rejected regular copies are plausibly normalized by documented regular positions attempted, because the numerator is composed of those positions. Counting rejects per requested good copy answers a different question about the burden of meeting demand, and the ratio can differ legitimately. Counting whole setup sheets per completed job answers something else again. Decide which measure supports the contemplated decision before comparing periods. Do not choose the denominator after looking for the one that makes a downward trend easiest.

Here is an illustrative comparison. A first baseline window has 8 rejected copies among 40 regular attempts on one job and 6 among 120 on another job. The combined result is 14 of 160, or 8.75 percent. Averaging 20 percent and 5 percent gives 12.5 percent, which weights the two jobs equally, not the pieces equally. Neither operation is automatically the right one for every question. If the program wants to describe the experience of a typical job in this tiny set, equal job weight might be interesting with enough context. If the claim is about the share of all eligible positions rejected, aggregate numerator and denominator for those positions. State which calculation is shown or provide both clearly.

Keep counts legible enough to catch mix effects. Suppose standard jobs consistently reject fewer pieces than special event sheets. A month with more standard work can show a lower aggregate without either type improving. Place invented before and after rows of the same types side by side and inspect within each type as well as the combined total. Do not infer from that additional grouping that every minor label should be a separate chart. Choose families based on meaningful differences in process or measurement. A handful of deliberate groups can reveal a change in work mix without burying the operator in a dashboard of five hundred lozenge charts.

If one job lacks an attempted count, do not include its reject numerator in a ratio whose denominator comes from other jobs unless you are explicitly presenting a different quantity. Mark the row incomplete and keep physical loss and immediate observations where they can be reported accurately. Indicate which portion of work was covered, by counts of eligible jobs or another supported measure. That prevents a week with many unlogged difficult jobs from receiving a glowing comparison against a poorly understood baseline. Missingness is a property of evidence, not a state of the stickers themselves.

Try a counterexample with zero events and one ineligible task. A complete run of 16 ordinary labels has no rejects. It is a genuine zero under the chosen review window. A cancelled speculative print task produces nothing and has no opportunities that count under the same rule. Zero divided by its empty denominator is not a second perfect run. Keep it outside the assessed rejection rate while retaining any actual setup material consumed under the proper materials category. Mark the datum as no eligible attempts rather than 0 percent. The distinction becomes critical when a mostly inactive period suddenly appears to be the highest-quality period in a chart.

Put comparable work in front of two people and see what they count

An operational definition is only useful if ordinary people read it similarly. The fictional workshop keeps several cardstock facsimiles to test the procedure. One is clean usable output. Another has a defect the local display no longer accepts. A third has a smudged edge outside the part that is used. A fourth was deliberately printed for alignment. A fifth was held at the end of the run and still hasn't been reviewed. Ask two volunteers to classify independently without revealing the counts the other person got. Record where they disagree and why. No one needs to invent a more precise rule for close decisions that the product never requires.

Some disagreement is semantic. One volunteer marks held as rejected because they assume nothing on hold can become usable. The other excludes it entirely. Both can be trying to honor the outcome and both can produce misleading comparison values. Put unresolved in its own category. The immediate state may contain usable, rejected and held fields that sum to attempted; the final view can update when the held material has really been adjudicated. Confusing early and late versions often produces apparent variation bigger than the change being evaluated. A timeline of decisions can be reviewed without accessing unrelated data about people's shifts.

Other disagreement comes from physical definition. If the desired address label excludes its surrounding margin, an ink mark outside that area may not matter; if the whole sheet is to be presented without trimming, the mark may be consequential. Show precisely which visual criterion the actual use calls for. When a local requirement changes for a legitimate reason, record the date and treat comparability as a new question. Recalculate earlier records under the new rule only if evidence allows and label the derived view clearly. It is not fair or useful to claim recent production improved by comparing a stricter old definition to an easier new one.

Give the practice one purposely ambiguous display sheet. It contains an early check mark overwritten by a receipt for final inspection. One observer should be able to report that the original disposition is known while the final review version clarified it later. Do not teach workers to obliterate the earlier row; document the correction and reason. Without retained history somebody may later wonder why one week counted test marks differently. A page that can absorb ordinary corrections gracefully is often more credible than a nicer form everyone avoids using once they get busy.

Set a modest review threshold. If observers differ on whether two defects belong to the same affected piece, fix the worksheet instructions before adding complicated calculations. If both agree on counts but can't explain material route, add one lightweight disposition field. Only increase process effort when it resolves a decision somebody actually has to make. It's reasonable for a small co-op to stop at ordinary comparison and qualitative follow-up. It does not have to emulate a comprehensive quality program to learn where its weekly rejected copies come from.

Compare a deliberate change without awarding yourself a causal story

Suppose the proposed practice is printing a reviewed plain-paper proof before committing to the ordinary sticker sheets, using existing approved materials and procedures. The charity exhibit doesn't need a machinery intervention. Clearly record what changed, its intended relationship to waste, and which jobs can use it. If the baseline routinely included a special glossy run but the pilot does not, that difference might explain the comparison. If the new workflow also shifted staffing and job preparation format, a reduction might plausibly have several sources. Phrase the result as observed under the pilot conditions unless you have a stronger design that justifies a strong causal interpretation.

One practical small trial is to compare successive ordinary batches with similar tasks and a stable end-of-job review window. Record materials that actually entered each step, any schedule interruptions, and what would invalidate a comparison. If feasible, arrange a known order or counterbalance two reversible workflows over benign synthetic copies, rather than giving one the convenient quiet morning and the other the chaotic deadline. Even then sample size, unobserved dependence among units, and variations in work can limit what a short experiment establishes. The point is to defend the measurements and reduce obvious alternatives, not to force them into a precise probability statement.

Consider an invented favorable result: three documented old-process jobs have three setup sheets and combined regular rejection of nine out of 84; three documented pilot jobs have no setup sticker sheets and five rejected of 88. The reduced usage of setup sticker sheets is directly observed in the example. If the proof work used plain paper instead, the report should say so and identify what question has and has not been answered about overall material use. The difference in regular rejection is also an observation in the toy data; it is not proof that the proof step caused all four fewer rejects. Do not infer money, weight, or emissions savings without evidence and documented boundaries supporting those values separately.

A sensible failure result may still teach something. Suppose a new preflight check catches two later issues, so recorded rejection goes up even though no one intentionally produced more bad copies. The shop's earlier review missed such issues rather than demonstrating they did not exist. A reported increase can reflect better detection. If the real decision is to notice mistakes early enough to prevent using materials, record detection stage separately from disposition. More detected problems need not automatically mean more actual waste, and a workshop shouldn't be encouraged to inspect less to improve a public metric. Compare the data gathered under consistent coverage or make that limitation prominent.

NIST's control-chart overview provides background on formal interpretations of process monitoring. It should not be used to turn a few toy points into control limits with unknown assumptions. A simple chronological plot with counts, applicable denominators, and documented changes can do the useful first work. If job types or inspection intensity change, annotate the plot at those points. The ability to see when an operational difference occurred does not establish that it caused a shift or whether the shift persists. A later formal review can choose its method and assumptions for the actual dataset instead of imposing confidence on a thin log.

Give the late-defect review an explicit route back to the originating job

One real challenge of learning from waste is that late defects often reach whichever station notices them, not the one that created the usable record. In the fictional co-op, the packing lead spots an outdated date in a mixed bundle and writes a small report. The printer has since produced several versions with similar titles. Without an ID or appropriate mapping, the report can add a real issue that cannot yet be attributed to a specific run. Do not guess from the nearest calendar date. Label the defect as observed and run unconfirmed, preserve readable version context, and only update a specific row if supported. Otherwise reserve it as an unresolved output observation, not a floor value silently allocated where it least affects the trend.

Use a practice sequence with three outcomes. A removed label carries a reference that joins to the original run without exposing private information; append it to that known row. A visually similar piece clearly belongs to another product family and should not be added to the ordinary baseline rates. A third has no reproducible mapping, so it remains unknown. A tidy comparison using only known dates can still be fair if that unresolved population is disclosed. Don't make all unassigned defects fall out of existence when calculating totals. The next operator should understand where another useful mark or ordinary handoff could improve coverage.

A correction two days later affects time-based reporting too. Suppose you compare at a fixed checkpoint after each job. If a new observation arrives after that checkpoint, decide whether to update a separately labeled latest view or recompute analytics using equivalently expanded historical windows. A day-old preliminary dashboard and a month-old revised baseline may carry different review maturity. If you compare them anyway, tell readers before they infer improvement. A small project might keep one frozen review window for routine comparison and a separate late register for operational learning. That makes the process explainable without keeping every window open forever.

A sign-off can be as small as “all items from this job checked for the stated condition at this time,” with known missing pieces identified. It should never mean nobody could discover anything afterward. Showing that a final count came from a stated scope rather than a magical forever check prevents disputes. The goal isn't to force the future to fit yesterday's chart. It is to keep the chart meaningful while real-world observations continue to appear.

Make the next decision small and reversible

After the pilot, may the workshop keep the new proof step as its default for ordinary noncritical labels? The record can show whether the additional handling was useful, more complex than necessary, or not yet measured enough to decide. If the change saved sticker stock but generated more paper waste or took excessive time, consider whether an authorized digital layout confirmation would answer the same low-risk question. If signals were mixed or the workbook records are incomplete, collect another comparable run rather than expanding a practice the team cannot assess. An article-sized diagram is not evidence of operating savings.

Prepare a compact note for the next volunteer. Name the defined job family, copy attempts checked, setup material and later dispositions, which change was actually made, any observed differences, and what evidence is missing. A reasonable note for the synthetic case might say “Setup sticker sheets observed reduced in these matched ordinary jobs; total material and sustained rejection change not established.” It can still recommend a reversible follow-up. Emphasize that the counting method can be checked by somebody else; the goal is not to preserve a particular result or advertise a specially chosen window as the whole workshop.

There is a final practical lesson in the opening mystery. A pile of discarded labels is useful evidence that something happened, but the size of a pile by itself cannot tell you the workload or the rate. Conversely any useful progress hidden by changed logging deserves a fairer comparison. Define what the tally means, carry the physical and acceptance measures separately, and keep a way to amend observed late outcomes. That sets the workshop up to find real improvements even when the next run doesn't give the easiest possible story.

Track what changed before concluding there was less material waste

A late review can establish that a rejected piece never left the site, but that isn't automatically source prevention. Suppose a new packing check finds a wrong batch code and catches the label before the exhibit team gets it. The improvement is in earlier detection relative to delivery. The piece was still attempted and rejected if somebody printed it. If the workshop subsequently catches the bad code in the approved digital layout and never prints the wrong copy, that is a different outcome: the physical attempt was avoided. Name the mechanism of the claim because these two successes can show up differently in the ordinary numbers.

Imagine the workshop tests a no-print review on three fictional requests with known invalid layouts. It finds and corrects two before consuming any regular sheets. The third is fixed only after a test sheet. In a process log, two verified detected-before-print cases and one setup-use case are observable. What cannot be measured from this toy sequence is how many pieces the old process would definitely have printed. An informed speculated counterfactual can motivate the study, but do not count an assumed old run as actual avoided material. A comparison under compatible work is needed for stronger claims. This can also prevent a large estimated savings total from floating free of the volume that was actually scheduled.

A grouping rule can preserve incremental evidence even when the workshop cannot conduct a formal study. Keep a simple late-error code, such as revision, visible defect, or unidentified, with a separate option for multiple observations on a single item. Show which was recorded, not which machine or person caused it. On the next review, choose a clearly scoped question the history actually suggests. For instance, a change in recent revision mistakes may justify reviewing approvals; it need not trigger adjustment of the printer. This respects the distinction between process inputs, visible consequences, and possible sources of variation.

One final test asks a second reader to interpret a tiny picture of the results without the background story. Give them the counts, the physical material note, the defined follow-up window and the scope note. Ask what improved and how sure they are. A valid answer can be narrower than the shop hoped. “Fewer separate sticker setup sheets in this observed sample, with regular rejection results also recorded, no claim on total material or cause” is plainly useful. As the team gains better evidence it can support stronger decisions. Good monitoring doesn't owe every step a broad success headline.

Use a fixed check sequence on the next normal job: set the case ID, identify the production unit, tally setup and regular attempts separately, join verified later outcomes, document unknowns, and preserve the actual physical disposition. Stop adding fields if they don't clarify a decision. A simple reliable baseline lets future staff recognize when a difference is real enough to act on and when their recording process still needs attention. The pile matters; knowing what opportunities produced it matters more.

Primary sources

AI disclosure: Drafted with AI assistance for Apiary. All experiments and numbers are illustrative, not measured workshop results.

Frequently asked
What is Count the Sticker Waste Before You Change the Setup about?
A community workshop prints small sheets of address labels for an exhibit. One week there are ten rejected labels in the tray. The next week there are six.…
What should you know about decide what a unit is before drawing tally marks?
The fictional workshop prints paper labels in eight-position sheets. A wasted sheet and a wasted label aren't the same event. A color alignment error across a sheet might affect every printed position. A clipped corner on one sticker can leave its neighbors perfectly acceptable for this low-risk exhibit. An entirely…
What should you know about draw a boundary for what counts as the baseline?
Choose the work you want the baseline to describe. Imagine the workshop has a family of ordinary unlaminated labels with comparable expected layout and uses a different station for glossy event sheets. If the planned change concerns how the ordinary sheets are prepared, keep the headline comparison within that…
What should you know about walk one fictional run across setup, production and follow-up?
Here is a dry run using numbers deliberately small enough to audit by hand. A job calls for 48 acceptable labels. One eight-position sheet is deliberately consumed as a setup test and does not count toward regular label attempts. Seven regular sheets, each with eight intended positions, are then printed. Of those 56…
What should you know about reconcile shop-floor waste with physical material without subtracting fiction?
Two people can independently count the output of a job and still disagree with how a third person accounts for the stock. Keep a simple material balance using one explicit comparable unit where possible. In the invented first run, the stockkeeper issues nine whole sheets to the station. The maker uses one for setup…
References & sources
  1. Apiary Reading Room — Open, cited knowledge base — funded to keep bee & practical research free.
From the Apiary Reading Room. Opinion & editorial — not financial advice. We don't overclaim.
More from the Reading Room