ApiaryActiveLive
Try: pause · settings · learn · wipe
← Community / Reading Room
DN
craft · 23 min read

Do Not Count a Stale Export as New Stock

A workshop makes a fresh CSV export at 10:30. The file looks newer than the clerk's 09:55 note that six packets of drawer dividers were taken out. Its first…

AI disclosure: Drafted with AI assistance for Apiary using public data standards and provenance references. The company, artifacts, records and timestamps are synthetic. The decisions concern inexpensive, noncritical desk organizer hardware. The exercise is read-only and uses paper or offline worksheet copies; it does not perform inventory writes or release real goods.

A workshop makes a fresh CSV export at 10:30. The file looks newer than the clerk's 09:55 note that six packets of drawer dividers were taken out. Its first row still says twenty-four packets. A volunteer sees the export's creation time, takes it as proof that twenty-four packets are on hand again, and proposes reserving twenty-two for packing. The source of the twenty-four was actually a snapshot cut at 09:15. Nothing in the new download observed or counted a new packet. Its newer timestamp describes the file, not the stock.

This is a habit worth building into small operations: require three separate answers before calling a stock number current. When was the artifact generated or exported? What portion of the source did it represent and as of when? When and for which purpose is someone asking for a current number? Those are three different timestamps. They can coincide in a system that genuinely queries live state, but they need not. The timestamp of clicking Download is not a shortcut around the other two.

The simple conclusion is not “never trust CSV.” A dated export is good evidence for its actual scope, including comparisons, investigating changes, and reconstructing a prior decision. The problem is using an old observation to make a present-tense claim without the intervening evidence. A sound handoff might say “24 packets in the 09:15 source snapshot, file generated at 10:30; six packets withdrawn after that snapshot; calculated quantity 18 at 10:32 if the reviewed movement log is complete.” A result like this gives a future reviewer enough to challenge the conditional rather than arguing about filenames.

Name the artifact, the observation, and the decision

Use exported-at for the moment the artifact was produced. A system may generate a CSV on demand, schedule it once a day, or repackage an already-generated table when asked. It may be replaying a file from a report cache. Exported-at can help sort artifacts or show when a job ran. It does not describe database transaction boundaries, the latest included movement, or whether a stockroom worker actually looked at a shelf. If the tool documentation says a particular field carries more detail, use that semantics explicitly rather than assuming it from a familiar label such as generated.

Use snapshot-at for the source cutoff, provided a reliable source identifies it. Define it in practical terms: “stock records as known to the organizer ledger at 09:15, site East, stock unit packet.” Another legitimate snapshot may exclude certain rooms, include only posted transactions, or use different units. A cut at 09:15 is not always a claim that every event that physically occurred by 09:15 was in the ledger. A worker can record an earlier movement late. Distinguish event time from what the reporting system had captured by the cutoff. If the snapshot contract is unavailable, write unknown rather than transferring the export timestamp into the empty slot.

Use decision-as-of for the number someone needs: “read-only estimate at 10:32 for planning the 11:00 assembly queue.” Current state needs a decision time because stock can keep moving. A timestamp is not a promise that the number remains safe to use throughout a shift. A team may use an old number for historical analysis and need a new one immediately for allocation. If a quantity changes between drafting and handing off a plan, preserve the old estimate as history instead of pretending it never existed.

A fourth time, observed-at, may belong to a manual count of the physical shelf. Keep it distinct too, even if the count and entry happen near each other. A worker sees eighteen at 10:12 and enters it at 10:20. Both times may matter when reconciling a move dated 10:15. Neither proves that the observed items are available, unreserved or correctly identified without separate evidence. The assignment here is freshness, not a claim about release or suitability.

A useful paper summary therefore includes source, date with known time reference, location, item identity, unit, cutoff semantics, and decision use. Long field names are acceptable in a shared worksheet because they make incorrect joins less attractive. “Snapshot cutoff for included posted events” describes more than “timestamp.” If the system can only supply a single vague timestamp, document the limitation. It is cheaper to state unknown once than to rebuild a week of stock history after people have assumed different meanings.

Establish units and scope first

The synthetic workshop stores hardware packets with paper label cards. Item DIV-P contains four identical drawer divider clips per packet. Item HOLD-R is a packet of retainer rings. An export includes plain numerals without a unit, while one stockroom note refers to individual clips. Before attempting a time reconciliation, determine whether every observation counts packets rather than pieces. Twenty-four clips is six whole packets only if the pack definition is stable and the reported clips actually include no open-pack accounting complications. Do not divide by four because the number looks plausible.

Keep location in the key. “East room DIV-P” and “West room DIV-P” can legitimately have different rows. When someone transfers three packets out of East into West, total network inventory might be unchanged even as the local decision changes. An export with both rooms aggregated cannot directly answer how many packets a particular bench can collect. Treat it as useful evidence at its own level, not a problem to fix by spreading the total evenly. A row must match the actual scope of the decision before clock freshness becomes relevant.

Be explicit about the population counted. An on-hand export can include physically present but reserved pieces if defined that way. An available-to-allocate report may already subtract reservations. A cycle count can see contents without knowing commitments. Freshness does not resolve differences among these definitions. If someone compares a current availability report with a fresh physical count, the discrepancy may be meaningful without either one being stale. Choose one quantity definition for each calculation and keep the other quantities in separate named fields.

Test the source boundary with an intentionally irrelevant file. Provide the analyst a clean CSV exported at 10:45 for West room, and a less polished 10:30 CSV for East room. Ask them to select the relevant candidate for an East request without yet judging whether it is fresh enough. The analyst should set the West artifact aside for this question, not allow the newer number to dominate. If the front screen hides scope beneath the filename, make the scope visible before the timestamp.

Reconcile the divider example without overwriting anything

The source snapshot shows 24 East DIV-P packets as known to the organizer ledger at 09:15. A confirmed record says six packets were moved from East to the packing bench at 09:55. A separate report says a courier “should be bringing four later.” No observed receipt is in the exercise. The export job runs at 10:30 but reflects the 09:15 cutoff. At 10:32, the offline calculation is 24 less 6 equals 18 in East, conditional on the available movement history actually covering the time gap. Do not include the expected four. Do not promote the conditional 18 to a verified physical count simply because the arithmetic is easy.

Suppose the ledger's later batch shows a return of two packets at 10:05 and then replays the same return row in two overlapping report exports. The returning movement has an established logical event key. Both visible instances should contribute only once if their content is genuinely a repeated representation of the same event. Under the same terms, 18 plus 2 equals 20. Do not deduplicate blindly by time and quantity: two independent two-packet returns can happen in one minute. Require the movement identity or a documented equivalence before collapsing both records.

The synthetic club writes an additional row at 10:40 for a movement that actually occurred at 09:50. If your decision was made at 10:32 and did not have that information, preserve the result as an estimate made from the evidence available then. A reconstruction at 10:45 can include the late event after checking it wasn't already represented in the base. These are two valid questions: what would today's full record say about the state at 10:32, and what information was actually available at 10:32? Keeping them separate helps review a prior decision without assuming staff ignored information they couldn't access.

A newer snapshot cut at 10:25 may already include the 09:55 withdrawal and the 10:05 return. If you use that snapshot, do not subtract and add those events again because their rows are present in a movement export. Establish the base and how it includes transactions before limiting the interval. A reproducible note can say “10:25 snapshot already represents movement IDs M-44 and M-51; no replay applied.” If the snapshot does not expose its inclusion behavior, mark the result uncertain rather than counting through an ambiguous boundary.

Reconciliation improves when snapshots are inputs rather than edits. Keep the original export and ledger worksheet intact in the exercise. Do arithmetic in a separate comparison copy, recording each artifact and filtering choice. Your review can produce an explicit discrepancy note or a request for a fresh authorized count; there is no automatic write to source inventory. A computed result should earn trust through evidence, not gain authority through being written into the field every future report will import.

A nearby item gives a useful control. The 09:15 snapshot shows ten East HOLD-R packets. The movement log has two events at 09:55, but both concern DIV-P. No ring movement appears in the reviewed interval. Ten remains a conditional unchanged estimate for HOLD-R if the movement coverage is sufficient. If coverage is unknown, then “last documented 10 at 09:15” is the defensible statement. This teaches the team that an unchanged estimate and an observed unchanged shelf are not synonyms either.

Think of time as an inclusion rule, not a filter alone

A cutoff can include records up to or strictly before a particular timestamp. It can also use a stable high-water mark or commit version when more than one event shares a timestamp. Consider a lower-resolution CSV that shows 09:15 for both the snapshot and a check-out. A date filter of “greater than 09:15” might omit a movement that was not in the baseline, while “greater than or equal to” might count one that was. The timestamp alone cannot settle inclusion when both data paths round to the same minute. Seek a documented event sequence, snapshot version, or direct membership evidence; otherwise report uncertainty for that boundary.

Local time normalization matters but is not the whole task. In a mock pair of CSVs, the two tools print 09:15 and 17:15 with differing offsets. They could denote the same instant. Conversely, two local 09:15 times from sites in different zones could denote distinct moments. Preserve the original value and known offset in the comparison, don't attach a guessed zone to unqualified values, and use the same clear time convention when summarizing. A neatly ordered table based on silently assumed offsets can be more misleading than an old one marked unknown.

Date-only numbers need special care. An export directory may label rows by business date rather than exact time. If a clerk says “the Nov 18 record,” a future reviewer may not know whether that means opening stock, closing stock, or a file generated sometime during the day. Ask what boundary the label describes before calculating movements. Being able to extract a date from a filename is not evidence for a noon cutoff. In a genuinely date-only workflow, restrict the output to date-level comparisons the input can support.

A transfer illustrates multiple events around a cutoff. Three packets leave East at 10:24 and are observed in West at 10:31. At 10:25 an East estimate may already subtract them while a West estimate cannot yet include them if the organization's definitions require observed receipt for local on-hand. A network view may carry an in-transit state rather than hiding a temporary disparity. A stale East export and a fresh West export mixed into one total can count the same physical units in two places. Match cutoffs and stated coverage, or clearly refuse to present the result as a simultaneous total.

The basic distinction is not specific to stock. The GS1 ObjectEvent model explicitly distinguishes event time from record time, while W3C PROV-O offers a term for generation time of an entity. The vocabulary helps explain that being recorded, physically occurring, and producing an artifact are three different things. This workshop does not implement either specification. It uses their distinctions to write labels that mean something when a human has to reconcile the evidence later.

Treat report metadata as data worth validating

CSV is a transport format, not a guarantee about upstream collection. Correct commas, a visible header, and a file that opens in a spreadsheet all matter, but they don't establish when the rows were true. The W3C tabular model describes how tabular data and supplementary metadata can be interpreted. In a small operation, the companion metadata may simply be a retained report description with the source cutoff and scope. It should live in a place where someone can find it after the exported filename changes or the report is forwarded.

Compare three metadata candidates in a practice folder. The first is a filesystem modification time from copying the artifact at 10:40. The second is the report generator's statement that it produced the file at 10:30. The third is a source marker naming 09:15 as the cutoff. All three can be accurate. They describe different events. To ask what happened after stock was last observed or represented, start with the third. To investigate recurring late jobs, look at production time. To troubleshoot how the local file arrived, examine copying and reception times. Do not let the first answer replace the third because the first is convenient to inspect.

Supplement time with source identity. Two compatible tables may come from a training environment and the actual workflow yet use the same columns and item IDs. Keep environment, warehouse scope, report definition version, and source version visible when those distinctions matter. A “current” mock link can be completely correct for its training fixture and completely wrong for a real request. The synthetic workshop uses paper folders clearly marked practice so no result can be confused with an operational inventory write.

Rows can have different ages within one report. A materialized summary may refresh some locations more often than others, and a late synchronization can leave only one item stale. One global generated-at value does not make every row an equally fresh observation. If the source provides per-row status or a known refresh partition, preserve it rather than stamping all rows with a rounded file time. If not, do not infer that some rows are fresh solely because the items move more often. A new output file can contain a mix of old and new source states.

Show the source contract in ordinary words. “Up-to and including all transactions actually recorded by cycle 215, excludes other rooms, no physical observation implied” gives a reader something to validate. “As of 10:30” without a definition leaves the hardest questions unanswered. Short labels and technical definitions work together: the label lets people compare quickly, while the documented interpretation tells them what it meant in the specific report. When a report changes semantics, version its definition before comparing old and new extracts.

Make an unknown result operationally useful

A coordinator asks at 10:32 whether enough dividers exist for twenty-two packages. You have a last documented 24, a confirmed withdrawal of six, and no evidence establishing complete post-snapshot movement coverage. Saying simply “unknown” omits what you learned; saying “18 available” invents certainty and availability. Give a bounded answer: “Last documented East on-hand was 24 at 09:15. Six known packets moved out at 09:55. The conditional estimate is 18 before any unobserved movements, and I cannot verify current on-hand from this export.” This lets the coordinator decide whether to ask the authorized stockroom role for a fresh count or stage less without mistaking a conditional for permission.

Use labels such as historical, conditional estimate, observed, conflict, or unavailable. Their definitions should stay stable across artifact choices. For a low-risk desk-parts pilot, an independently observed packet count at 10:42 can support its own bounded “observed 18 packets in East at 10:42” claim. It does not retrospectively show exactly what was in East at 10:32. It also cannot establish that all 18 are releasable if reservations use a different source. Giving both numbers a time, a scope, and an evidence type prevents the future planner from discarding most of the context.

If an artifact's cutoff is unknown, the most useful next step may be securing the report description rather than performing a physical count of every item. Conversely, if the pressing decision concerns one low-value bin and authorized staff are present, obtaining a fresh direct observation can be simpler than reconstructing a whole interval. Do not request extra work just because a data field is missing. Match the effort to the decision. A single bin-level unknown does not necessarily make the export useless for last week's history.

An apparent shortage can be both real and misclassified. The current East value may be 18, while the “shortage” claimed by a fresh stale export is actually a misunderstanding about available count, an unrecorded transfer, or a mismatch in which event window is being considered. Name the discrepancy without assigning an unsupported cause. If you can establish that a known movement wasn't included in the snapshot, say that. It still doesn't prove when the export system malfunctioned, whether someone made a mistake, or whether the tool behaved according to its existing contract.

When the goal is preventing overcommitment, avoid unbounded assumptions. Without additional evidence, you may not even have a sound upper or lower bound beyond trivial constraints. A known withdrawal shows a directional change relative to a baseline only if the rows refer to the same stock scope and no unknown changes offset it. Do not announce “at most 18” from 24 less six when a later receipt might have occurred. “18 under the no-other-movements assumption” correctly avoids that error. If operation policy wants a conservative commitment number, let the authorized decision process establish it with appropriate evidence.

Practice conflict handling with three timed snapshots

Give two mock readers snapshots A, B, and C covering the same practice East scope. A is cut at 09:15 and generated at 09:20. B is also cut at 09:15 but generated at 10:30 from a cached source. C is cut at 10:25 and generated at 10:29. A and B both say 24. C says 20. Ask which represents the most recent source cutoff, then ask which was generated latest. The answers are C and B respectively. If their process chooses one file for both questions, it collapses exactly the distinction the exercise is trying to teach.

Specify the source history: a six-packet withdrawal after 09:15 and one two-packet verified return before 10:25. In the known-complete mock record, C is consistent with A plus movements. Ask participants to write an explanation that does not need the neatness of the numbers to prove the method. In a second round, send them a fresh duplicate of the return in a different export. The movement identity should preserve the same resulting estimate. Record which participants recognized that two rows might represent one event and which merely deleted any repeated-looking number.

Next introduce a conflict in membership at the boundary. C's metadata says one movement recorded at the cut minute is not known to be included. The practice log contains another entry at the same minute from independent hardware. Participants should know how to separate the affected quantity from unaffected rows and when to set aside a current result. A process that can say “unable to verify this inclusion, no new availability claim” is more dependable than a process that settles ambiguity by assigning it to whatever side makes the report balances look plausible.

A final fixture introduces a C row showing 19 instead of 20. The analyst cannot tell immediately whether this indicates a missed event, a counting error, an adjustment, or a different report definition. Their instruction is to record a one-packet difference from the expected comparison, preserve both values and request the minimum evidence that can distinguish hypotheses. Don't call the snapshot wrong just because a reconstructed number differs. Reconstruction inherits every assumption about base, scope and movement coverage. The cleanest outcome may be a real discrepancy requiring an authorized fresh count.

Choose process measures bounded to the test. Did the analyst select the latest source cutoff rather than latest export? Did they preserve the unit and location? Did they treat a projected delivery as distinct from observed receipt? Can an independent reviewer reproduce the calculation from named inputs? Avoid a headline such as inventory accuracy improved 100% after three synthetic examples. A fair result is “both readers correctly distinguished generation time and source cutoff across three mock cases; one still treated a duplicated movement as two events.”

Read a received report with an adversarial calendar

Someone may email a CSV titled today-final. That can mean it was generated today, attached today, approved for today's discussion, or simply saved under a familiar name. None of those meanings settles the source cutoff. In the practice file list, title the older snapshot today-final, store the recent snapshot under a generic internal report ID, and add a short retained report note that correctly describes each. Ask readers to choose on provenance rather than emphasis. Renaming the artifact cannot change the source rows.

The same lesson applies to screenshots. A newly captured screenshot of an older report can show a current window clock next to obsolete stock. A screenshot can be useful for preserving exactly how a view displayed a figure. That is distinct from proving when the source data was observed. If the desired result depends on the backend cutoff and no accompanying report details survive, avoid substituting a recent screenshot as evidence. Keep it marked as a display observation and request a better source for the decision.

Practice a day transition. A stock snapshot ends at 23:55, the report job produces a file at 00:08, and staff open it with a different date. One team calls it Tuesday stock because the source cutoff was Tuesday, and another calls it Wednesday's CSV because it was generated Wednesday. Both colloquial descriptions can be accurate. The formal labels prevent them from arguing over which day is the authoritative umbrella. A Friday planning request needs the relevant as-of date, cutoff semantics, and adequately bounded changes, not a winner in a naming debate.

A correction can make an artifact untrustworthy for a different reason. Suppose source rows are exact through 10:25 but the interpreting spreadsheet accidentally reformats packet codes as numbers and drops leading zeros. That is an identity defect, not staleness. Preserve original row strings in provenance if the IDs are strings in the declared contract. Compare on the proper typed key rather than repairing a perceived timestamp issue. For the tabletop check, give two legitimate product IDs that differ only in leading zeros; the same clock information does not identify which item was exported.

Use the checklist at the actual decision

First, identify the question in ordinary words: East DIV-P packets physically on hand at the latest justified time, or packets available for the next run, or change in reported stock since yesterday. These are separate. Next identify the source population, unit, and inclusion rule. Then compare generation time with source cutoff, note relevant later movements once each, and decide whether coverage is enough for the required conclusion. Leave live values alone during the analysis. Even a correct read-only reconstruction isn't permission to adjust a live balance.

Give the result in an evidence-aware sentence. “C: East DIV-P on-hand 20 packets, source cutoff 10:25, file generated 10:29. Reconciles with A and the two uniquely identified post-09:15 movements in the complete tabletop event set; no claim after 10:25.” The result is reproducible but not oversold. For a current request at 10:50, say whether you have any later evidence. If you don't, retain the historic count as useful context and report that current stock is unknown. Do not quietly extend the life of an old result simply because the requesting person uses the word now.

Make a distinct line for what to preserve after a discrepancy is closed: original source identity, interpretation, decision time, old and new evidence, and any residual uncertainty. The value of that line is apparent after another report changes. A future reader can see that staff did not add six packets back into inventory based solely on an old snapshot. They can also see why a new verified count changed the later recommendation. Reproducibility is often less about retaining every intermediate calculation than about retaining the boundary conditions that made the conclusion valid.

A useful review habit is to state the rejected inference explicitly. “Newest file is not newest observed stock.” “Expected receipt is not an observed receipt.” “Matching table totals do not show matching item identities.” These are different checks, and the stale export case particularly needs the first. Put it near the stock number where a reader can encounter it before making the next comparison. Write it once in the workflow, and choose precise labels so people do not have to repeat a lecture every time a file is replaced.

Eventually the process may need better instrumentation, newer report generation, or a clearer stock-management policy. A small team can begin without changing any of those systems: preserve the source cutoff, report artifact time separately, and refuse to infer freshness where the evidence is missing. Then practice against a handful of dated sample cards until a second person can independently reconstruct the intended conclusion. A recent file is valuable when it carries trustworthy recent evidence. Its timestamp alone cannot put another packet on the shelf.

Compare corrections and movements without rewriting history

A teammate reports that three packets in the original 24 were counted as full even though their recorded unit was open packets. The training worksheet has both the original count and an authorized correction to the historical source. Nothing moved at the time of the correction. A replay method that treats every newly appearing row as a physical change might subtract three as a late movement even from a snapshot already corrected to 21. Separate event types and record whether they describe object movement, an explicit inventory adjustment, or correction of an earlier recorded claim. You may need to choose a new baseline rather than append a synthetic movement to an old one.

Now assume a later source has two changing versions of M-51, the same logical two-packet return. A placeholder first reports one packet, then the retained correction shows two. These are not necessarily two returns. Resolve which source version is authoritative for the question before choosing the number. Moreover, a decision logged between the placeholder and the correction may honestly have used the first amount. The distinction between latest knowledge about the past and knowledge available at an earlier decision is as important for quantities as it is for cutoff times. A review that preserves the two versions separately can explain the discrepancy without guessing that the clerk miscounted.

Create a read-only table with one row per intended comparison key: facility, item identifier, unit, and definition. Record the baseline artifact and its cutoff in each affected case, not just once in a footer. Next place the disputed record, the inclusion state, and the hypothesized effect next to the current conclusion. The stable conclusion is the recorded disagreement; the effect can remain hypothetical. If the original CSV sheets stay read-only and the comparison is clearly marked as an analysis, another person can change assumptions in a copy without mistaking the changes for ledger updates. This yields a usable review artifact without establishing a second inventory authority.

Decide which later source could settle the question

For a narrow decision at 10:50, a recent clerk's observation might settle one bin, a verified new snapshot could settle a whole defined inventory scope, and a confirmed transfer receipt could settle a specific open movement. These follow-ups are not interchangeable. A whole-facility snapshot may be fresh but still lack the room detail you need. A bin count can confirm the contents of one location but might arrive too late to reconstruct the earlier state. A transfer receipt can support the presence of a group at the destination without saying anything about unrelated dividers in the same room.

In a practice handoff, pair each unresolved question with the smallest evidence that could answer it. For a boundary uncertainty around M-51, ask for documented inclusion in snapshot C, not a fresh copy of the same export. For a unit uncertainty, ask for the report definition and the packing convention documented by the authorized owner, not a newer timestamp. For a current physical count, ask the responsible role whether they can obtain a scoped count with an observation time. A useful assistant to an analyst collects evidence by question rather than flooding the next shift with generalized checks.

Explain what the new observation won't fix. A direct 10:50 count of 19 packets does not determine whether 19 were present at 10:25 without more context. It can still settle what the team saw at 10:50. If the data discrepancy persists, report it separately to the owner rather than allowing the newest direct observation to erase the old incorrect interpretation. Over time a small collection of narrowly worded discrepancies lets people distinguish recurring cache behavior from recurring identity errors or incomplete recording. That is a much better improvement signal than counting how often a newly generated file displayed a familiar number.

Primary sources

Frequently asked
What is Do Not Count a Stale Export as New Stock about?
A workshop makes a fresh CSV export at 10:30. The file looks newer than the clerk's 09:55 note that six packets of drawer dividers were taken out. Its first…
What should you know about name the artifact, the observation, and the decision?
Use exported-at for the moment the artifact was produced. A system may generate a CSV on demand, schedule it once a day, or repackage an already-generated table when asked. It may be replaying a file from a report cache. Exported-at can help sort artifacts or show when a job ran. It does not describe database…
What should you know about establish units and scope first?
The synthetic workshop stores hardware packets with paper label cards. Item DIV-P contains four identical drawer divider clips per packet. Item HOLD-R is a packet of retainer rings. An export includes plain numerals without a unit, while one stockroom note refers to individual clips. Before attempting a time…
What should you know about reconcile the divider example without overwriting anything?
The source snapshot shows 24 East DIV-P packets as known to the organizer ledger at 09:15. A confirmed record says six packets were moved from East to the packing bench at 09:55. A separate report says a courier “should be bringing four later.” No observed receipt is in the exercise. The export job runs at 10:30 but…
What should you know about think of time as an inclusion rule, not a filter alone?
A cutoff can include records up to or strictly before a particular timestamp. It can also use a stable high-water mark or commit version when more than one event shares a timestamp. Consider a lower-resolution CSV that shows 09:15 for both the snapshot and a check-out. A date filter of “greater than 09:15” might omit…
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