Desk Check
A desk check hands a completed work product to a single reviewer, one who did not write it. BABOK counts it among its seven review types and defines it in one sentence: a reviewer who has not been involved in the creation of the work product provides verbal or written feedback (§10.37.3.2). Of the seven definitions it is the only one that sets a condition on the reviewer, and that condition is what the author controls from end to end, because he is the one who chooses. The rest escapes him: he will not know how that reviewer approached the task or how far he pushed the reading, and the answer that comes back is the only sign that a review happened at all. The whole technique rests on one reader. It returns an answer the author can handle alone, and it costs less than any other review because nothing came back that has to be reconciled.
Goal
A desk check hands a completed work product to a single reviewer who did not write it, whose fresh eye reads what is on the page. An author re-reads what he meant to write, and that gap is everything the technique buys.
BABOK lists seven possible objectives for a review (§10.37.3.1), and the desk check serves two of them. Remove defects: the ordinary find of an attentive reading (Table 3-3). Check conformance to a specification or a standard: Wiegers gives it both conformance rows of the same table. This is its surest ground, because a competent reviewer holding a standard needs nobody's help to apply it.
The other five escape it. Ensure the work product is complete and correct: Wiegers reserves completeness for the inspection and the walkthrough (Table 3-3). One reader alone confirms what is written; what is missing shows up when several readings are set against each other. Establish consensus on an approach or a solution: there is one reviewer, so there is nobody to agree with, and that is the session of a formal walkthrough (10.37b). Answer a question, resolve an issue or explore alternatives: BABOK ties this objective to work in process (§10.37.2), where the desk check bears on a prepared deliverable read in full, and it is the home ground of the ad hoc review (10.37g). Wiegers' Table 3-3 separately withholds the exploration of alternatives from the desk check and grants it to the walkthrough, which BABOK calls an informal walkthrough (10.37d). Educate reviewers about the content: one reviewer is not a team, and the objective belongs to the pass around (10.37f). Measure work product quality presupposes defect metrics, so a log, so an inspection (10.37a).
Wiegers counts two objectives BABOK's list does not carry, and the desk check holds both. Ensure a correction was made correctly: you hand the corrected passage to a reader who never touched it, and you ask whether the fix holds. The desk check is in that sense the second reading of a correction; it does not check the corrections it prompted itself, and of the review types BABOK adopts, the inspection (10.37a) is the only one Wiegers marks as including Verification (Table 3-1). Control the cost of the review: that row of Table 3-3 carries a single mark across the six methods compared, the desk check's.
The deliverable is one comment set, from one reviewer. Nothing is collated and nothing is escalated.
Usage
When to use it
- One competent second pair of eyes is enough: Wiegers recommends the cheapest review method that reaches the goal, and the desk check holds that rank on his spectrum.
- The work product is low-risk: a defect that slips through is cheap to fix.
- A correction has to be read before it ships: a reader who took no part in it says whether it holds (Table 3-3).
- Time and resources are tight: Wiegers names severe time and resource restrictions as the technique's own case.
- The reviewer cannot be pulled into a meeting: BABOK names the desk check among the reviews a reviewer performs at a convenient time, without interrupting work to attend a session (§10.37.4.1).
- A review culture is being started: Wiegers sees it as an entry point and a mentoring instrument, with the caution that goes with it.
When not to use it
- Evidence of review is required (a quality gate, an audit, an approval, defect metrics): BABOK names the desk check when it notes that it is difficult for the author to establish that an independent review was really done (§10.37.4.2), and with one reviewer, the answer that came back is the only thing attesting to the review. An inspection (10.37a) logs the defects and verifies the rework; a single issue review (10.37c) attests conformance to one precise standard.
- The question is whether the document reads the same way to a second reader: one reading compares with nothing. Wiegers withholds the assessment of readability from the desk check and grants it to the pass around (Table 3-3). A pass around (10.37f) puts several independent readers in front of the same text.
- The work product has to be agreed by a team or the perspectives reconciled: there is one reviewer, and one person's agreement is not a group's. A formal walkthrough (10.37b) has a room and a facilitator who is not the author.
Description
BABOK gives this technique no steps. What separates a desk check from a document sent over with "can you have a look?" comes down to three choices, and the author makes them before he sends it.
The three disciplines
Choose a reviewer who did not write the document
BABOK puts it in the definition itself: a desk check is an informal technique in which a reviewer who has not been involved in the creation of the work product provides verbal or written feedback (§10.37.3.2). Of the seven review types it is the only one whose definition qualifies the reviewer; the other six speak of peers, of stakeholders or simply of reviewers. It is also what the author controls, because he is the one who chooses. The colleague who helped draft the chapter is the one who will read it fastest, and that is exactly why he cannot be the reviewer.
Say what to read for
Wiegers observes that a desk check becomes fairly formal as soon as the reviewer works with defect checklists, specific analysis methods and standard forms. This is the author's only lever against the variability of a technique that depends entirely on the knowledge, the skill and the self-discipline of a single reviewer. A checklist buys no guarantee. It constrains the reading.
Agree what comes back
Wiegers names the three forms the return takes: the reviewer hands over a defect list, the author and the reviewer sit down together to draw it up or the reviewer hands back the marked-up document. An ask with no defined return produces "looks fine to me".
The scope of the desk check
The author knows who read the document, because he chose him, and he knows what the reading returned, because it is in front of him. He will not know how that reviewer approached the task or how far he pushed the reading, and Wiegers says this of this technique and of no other in his chapter. BABOK notes, for its part, that for desk checks and pass arounds it is difficult for the author to establish that an independent review was really done by each participant (§10.37.4.2).
The limitation names both techniques, and it weighs more on the desk check. With five reviewers, the one who never opened the document is buried under the four who answered, and the review still looks like a review. With one, a reading that never happened produces an empty review, and nothing tells the author so. The answer that comes back is the only sign that a reading took place and the only thing the author can examine.
The session, for its part, stays possible: Wiegers marks the meeting as such for this technique (Table 3-1), in the form of a follow-up discussion where the reviewer explains what he found. Ten minutes of conversation leave the technique intact and change what the author understands of the comments.
Where the desk check sits
BABOK lists its seven review types in descending order of formality and calls the first three formal techniques, the next four informal techniques (§10.37.3.2). Two binary questions cut across that ordering: is the session planned, does the reviewer examine the work product alone before speaking? The desk check answers no, then yes, and the pass around (10.37f) answers the same way. BABOK's ordering therefore does not separate them, and Wiegers does not separate them either in his own spectrum, where the two techniques occupy one cell (Figure 3-1).
What the spectrum does not show is read in Wiegers, in the table that crosses review objectives with review methods. There, the two techniques stop coinciding.
| Objective | Desk check (10.37e) | Pass around (10.37f) |
|---|---|---|
| Find the product's defects | Yes | Yes |
| Check conformance to a specification | Yes | Yes |
| Check conformance to a standard | Yes | Yes |
| Ensure a correction was made correctly | Yes | No |
| Control the cost of the review | Yes | No |
| Assess the readability of the product | No | Yes |
| Educate the other team members | No | Yes |
| Verify the completeness and correctness of the product | No | No |
| Reach consensus on an approach | No | No |
| Explore alternatives | No | No |
The two techniques do not differ in degree, they differ in what you can put them to. The three objectives Wiegers withholds from both techniques carry his most useful finding: the shared limit of informal reviews conducted by one or two reviewers, which BABOK says provide less assurance than a larger team and a more formal process (§10.37.4.2).
The words "desk check" designate something else in Wiegers: he calls a deskcheck the author re-reading his own work, a form of self-review, and reserves the name peer deskcheck for BABOK's desk check. ISO/IEC 20246 separates the two by name in its ten review types, the peer desk check and the author check. The sentence "I did a desk check of my document" therefore describes the one reading BABOK's definition rules out.
The traps
"Looks fine to me"
A return with no findings and a reading that never happened produce the same message, and the author has no way to tell them apart. The remedy plays out before the reading, in what is asked and in what is expected back. A chase afterwards comes too late.
Choosing the reviewer who helped write it
He is the fastest and he knows the file. He will also silently fill the same holes the author did, because he filled them with him. Convenience destroys the one property the technique rests on.
Reading the desk check as a judgement of competence
BABOK holds that each review is focused on a work product and not on the skills or the actions of the participants (§10.37.2). Wiegers flags the same thing on the mentoring side: the line is thin between asking an old hand to help out a beginner and implying that a beginner needs someone to second-guess his work. A desk check is asked for.
AI considerations
The desk check's own weakness is coverage: one reading, one person's, with that person's blind spots. That is where a large language model helps, and it is also where it is most easily taken for what it is not.
Writing the checklist
This is the lever of the second discipline, and it is the document nobody has time to write. For a use case, a model produces one in minutes: alternative flows, preconditions, postconditions, actors cited without being declared, constants where the rule expects a parameter.
Running the machine first
An actor named at step 9 and missing from the actor list is a mechanical find, and the reviewer's time is worth more than that. Whether the business rule states the law correctly is the work a human was chosen for.
The reviewer's independence
A model is not an independent reviewer. BABOK's definition describes a reviewer who has not been involved in the creation of the work product (§10.37.3.2). A model prompted by the author, on the author's document, with the author's framing, is the author's own reading with more words: it is the very self-review the definition rules out and to which Wiegers gives another name. BABOK's limitation bears on the difficulty of establishing that an independent review took place, and a clean list of findings, produced by a model, answers that question with false testimony, harder to contest than a silence.
Data protection
A tax use case carries taxation rules and flows of personal data. The revised Federal Act on Data Protection (revFADP) is settled before productivity, and the organisation's rule, which model, which data, hosted where, is fixed before the document is sent.
Examples
What a desk check produces is one comment set, a single reviewer's, which the author works through line by line.
| Point in the use case | What the reviewer wrote | What the author did with it |
|---|---|---|
| Step 4, "the taxpayer carries over last year's return" | "A taxpayer who has just moved from another canton has no previous return here. The flow has no alternative." | Alternative flow 4a: first return, entered with nothing carried over |
| Step 7, deductions | "The ceiling on the childcare deduction is an indexed annual amount. The use case freezes it into a constant." | The value leaves the use case: a reference to the scale in force |
| Business rule BR-12 | "A return filed after the deadline remains admissible; assessment by estimation follows only after a formal reminder. The use case rejects the filing." | Late filing accepted, with a "late" status |
| Actor "Compensation office" | "Cited at step 9, absent from the list of actors." | Added to the secondary actors |
| Postcondition | "The use case stops at 'return submitted'. Nothing says whether the taxpayer gets an acknowledgement." | Postcondition made explicit: acknowledgement of receipt with a filing number |
The author disposed of every line himself, in the order the lines arrived.
All five findings are tax findings, because the reviewer is a tax analyst. Whether this use case reads the same way to somebody who does not know taxation, this desk check did not test, and one reading cannot test it: that calls for a pass around (10.37f).
Visualizations
BABOK's formality spectrum places the desk check among its six neighbours, in the same cell as the pass around (10.37f), as Wiegers does in his own spectrum. Locating is all a spectrum does. The choice between the seven reviews (10.37) is settled on the objective pursued and on the assurance required.
The objectives table carries what the spectrum cannot show: two techniques compared row by row, from the only source that tells them apart. Content made of cells is rendered in cells.
The artefact produced is one comment set and what the author did with it. It fits in a single table, because there is only one reviewer.
Cost
| Phase | Level | Justification |
|---|---|---|
| Preparation | Low | Wiegers counts no planning in this technique (Table 3-1). Choose a reviewer who did not write the document, say what to read for, agree what comes back. A few minutes. |
| Execution | Low | Wiegers makes it the cheapest review method on the spectrum: it takes up a single reviewer's time, and that reviewer picks his moment (§10.37.4.1). |
| Documentation | Low | A single reviewer returns a single comment set: nothing to collate, nothing to reconcile. Wiegers grants control of review cost to the desk check alone (Table 3-3). |
Tooling
The desk check needs almost no tooling. It needs an agreed return channel, comments and suggestions in the document itself or a marked-up copy handed back to the author, and the tool only has to carry one of the three forms Wiegers names. It needs a checklist, if the reviewer is to be given one.
The comparison with the pass around (10.37f) comes down to one clause: no routing, no shared review window, no collation tool. There is one reviewer.
A finding that has to live beyond the review belongs to item tracking (10.26), which has tools of its own.
Sources
- IIBA, A Guide to the Business Analysis Body of Knowledge (BABOK Guide) v3, §10.37 Reviews.
- Karl E. Wiegers, Peer Reviews in Software: A Practical Guide, Addison-Wesley, 2002, chapter 3, "Peer Review Formality Spectrum" (Table 3-1, activities by review type; Table 3-3, objectives and methods; Figure 3-1, the formality spectrum).
- ISO/IEC, ISO/IEC 20246:2017, Software and systems engineering: Work product reviews.
- IEEE, IEEE Std 1028-2008, IEEE Standard for Software Reviews and Audits.