
Review the evidence and choose the next small change
Compare intentions with actual records, separate observations from interpretations, and finish with a short next-week decision card. Work through complete fictional weeks, misleading summaries, arithmetic checks and answered practice.
Your role: decide what matters, check the original records and approve any real-world changes. This is a practical review of ordinary commitments, not a score of your worth or a health assessment.
Choose a reading route
Understand the evidence
Review the complete week
- 6. Read the complete fictional plan before the actual records
- 7. Inspect Leena's actual record and change log
- 8. Reconcile the result before explaining it
- 9. Repair a polished review that changes the evidence
- 10. Investigate the difference without inventing a cause
- 11. Turn the review into Leena's next-week decision card
- 12. Update records without rewriting the past
Adapt, practise and finish
- 13. Use a smaller review when the record is sparse
- 14. Make room for reflection that does not become a score
- 15. Use a controlled SI conversation from packet to final note
- 16. Diagnose the smallest repair when a review goes wrong
- 17. Practise independently with Rosa's changed packet
- 18. Check the full transfer answers and boundary tests
- 19. Build your own review and know when to stop
1. Finish the review with a decision you can use
A weekly review is a small act of reconciliation. You began the week with some intentions, lived through conditions that did not always match them, and now need a truthful starting point for the next few days. Super Intelligence can help arrange that information, compare records and challenge an unsupported conclusion. The useful result is a short set of decisions that you understand. It is not a verdict on whether you lived the week correctly.
In this eduKate series, SI is an editorial umbrella for practical AI learning and use. It does not mean that the assistant you are using has been established as artificial superintelligence, the stronger hypothetical concept of broadly superhuman capability. A fluent review is still a generated proposal. Its claims about your week need the same care as any other summary of incomplete records.
This guide concentrates on looking backwards accurately enough to move forwards. A weekly plan assigns future capacity. A project plan works towards one outcome. A checklist supports a repeated procedure. Here, the main task is to reconcile what was intended with what happened, decide what the difference means, and carry only justified changes into the next week. You can use your existing calendar and notes throughout; no new application or elaborate personal dashboard is required.
A finished review answers five practical questions. What genuinely reached its completion condition? What remains open, and in what state? Which important conditions changed? What conclusion is supported by the available evidence? What will you keep, change, stop or check next? If the output answers those questions plainly, a short review can be sufficient. If it produces twenty new habits while leaving a promised reply unresolved, it has missed its job.
Do not start by asking an assistant to explain why you were unproductive. That question supplies a judgment before the evidence has been examined. A more useful opening is: “Help me compare this week’s intentions and records. Separate observed facts, possible explanations and decisions. Keep missing information visible. Do not infer my character, health or other people’s intentions.” This gives the assistant a bounded analytical task rather than an invitation to invent a personal story.
All named people, logs, documents, amounts and assistant responses in the worked cases below are fictional teaching material. They are supplied so you can reproduce the comparisons, including the mistakes. They are not customer results, productivity experiments or evidence that an AI tool produces a particular benefit. The method is an editorial practice framework, not a clinical assessment or a scientifically validated score of a good week.
The most important change is modest: preserve the difference between the week you intended and the week you actually lived. Neither should erase the other. The original plan explains your decision at the time. The actual record explains what needs attention now. A careful review lets both remain true while you make the next reasonable choice.
2. Separate records, observations and interpretations
Begin with a distinction that will appear throughout the guide. A record is something you can inspect: a calendar entry, a saved file, a sent message, a brief contemporaneous note or a later recollection. An observation is a statement grounded in that record. An interpretation proposes what the observation means or why it occurred. A decision says what you intend to do about it. These are related, but they should not silently replace one another.
Suppose a note says, “Wednesday, 19:10: stopped caption work because three locations were missing from the source list.” The observation is that the session stopped at a missing-information problem, according to the note. A plausible interpretation is that source preparation needs improvement. A different interpretation might be that the intended scope was too large for the available records. “I always avoid difficult work” is not established by the note. Neither is “the AI caused the delay”.
The strength of a record also depends on the claim. A scheduled calendar block is good evidence that time was reserved. It is not by itself evidence that the activity happened. A saved document proves that a document exists, but not that it was sent or accepted. A sent-message record can establish sending, but not that the recipient agreed. An activity log can support a duration estimate while leaving the quality of the result unresolved.
Keep retrospective recollections usable without pretending that they are precise measurements. “I remember working on this for around half an hour” is a valid note about memory. It should remain approximate. If you have a timer entry covering twenty-eight minutes, that is a different kind of record. Do not round both into an apparently exact data set and then compute a highly precise efficiency score. The precision of the conclusion should fit the precision of the evidence.
An unknown is a legitimate review result. If you cannot establish whether an invitation was sent, write “sending unverified” and identify the place where it can be checked. Do not mark it complete because sending seems likely, or failed because you cannot immediately find the message. When the distinction matters, inspect the original source. When it does not affect an important decision, leave the uncertainty proportionate rather than turning the review into an investigation of every minute.
Ask the assistant to attach a record label to important factual statements. For example: “The handout was ready by Friday, supported by record H3.” It need not annotate ordinary reflective prose sentence by sentence. The useful labels belong beside deadlines, quantities, completion claims, changed commitments and proposed explanations that could alter your next action. If a label points to the wrong record, the polished wording does not rescue the claim.
NIST’s Generative AI Profile identifies confidently stated false content, privacy risks and problematic human reliance among generative-AI concerns. That supports checking the assistant’s review against your sources; it does not certify this particular weekly method. The profile is a voluntary organisational risk resource, used here for that narrow caution. NIST, Generative AI Profile
3. Define the review boundary and gather a small packet
Choose the period before collecting information. It might be Monday morning through Sunday evening, a roster cycle, or the interval since your last review. Name the cutoff. An event that happens after it belongs to a later update, even if you learn about it while writing. A clear boundary prevents tomorrow’s hoped-for completion from being quietly credited to last week’s record. It also makes unfinished work easier to locate.
Then choose the scope. You do not need to review every domain of life in equal detail. Start with the commitments that mattered, active projects that changed, and loose ends with consequences. Include rest, ordinary enjoyment or connection when they matter to your own reflection, without requiring them to justify themselves through measurable output. A review may acknowledge that you enjoyed an unplanned conversation and decide nothing further about it.
A minimal packet contains the original intentions, any deliberate changes, the relevant actual records, and the next week’s known constraints. Keep original and revised intentions distinguishable. If you reduced a project’s scope on Thursday, preserve that Thursday decision. Otherwise, a smaller finished output can look like either a failure against an obsolete plan or a complete success against a silently rewritten one. The decision history explains which comparison is fair.
Use enough context to resolve the question, rather than copying your whole inbox or diary. “A fixed family commitment occupied Tuesday evening” may be enough to explain a lost window. The assistant does not need a relative’s private circumstances. “Document C is unavailable until next Tuesday” may be enough to explain a blocked task. Avoid sharing credentials, identity documents, private messages or sensitive information when a short non-identifying description serves the review.
Removing names is not a guarantee of anonymity. A combination of locations, dates, roles and unusual events may still identify someone. Decide what you are comfortable sharing in the service you use, and follow any applicable organisational rules for work material. A personal review is not a reason to move confidential employer records into a consumer tool. If permission or suitability is unclear, do the relevant comparison manually or use a permitted environment.
Gather sources before asking for conclusions. Open the chosen file, check the relevant message and read the short note. You may discover that a task is already finished, that a promise has changed, or that the planning label was ambiguous. Correcting the packet at this point is usually simpler than repairing an elaborate analysis later. The assistant can help identify missing fields, but it should not invent the missing record.
For context, the David Allen Company provides an established weekly-review checklist that includes revisiting calendars, actions and projects. This guide does not reproduce that checklist or claim to validate it. Its particular focus is the evidence comparison and checked next-week decision when SI assists. Choose a review scale that fits your situation rather than assuming that every existing method’s full checklist must be combined. David Allen Company, Weekly Review checklist
4. Give every important item an honest state
A binary done-or-not-done list is often too coarse for a review. It can hide a genuine blocker, erase a deliberate cancellation, or treat a drafted message as an external action. Use a few states with plain meanings. Completed means the stated acceptance condition has evidence. In progress means relevant work exists but the condition has not been met. Blocked means the next meaningful step depends on something currently unavailable. Cancelled means a decision ended the commitment or optional intention.
Add “revised scope completed” when a recorded change deliberately replaced the original outcome before completion. Keep the original and replacement visible. If you planned to sort a whole shelf and then chose to sort one box, the box can be complete without the shelf being complete. The review should not punish the reasonable scope reduction, but it should not transform it into evidence that an entire shelf can now be sorted in the box’s time.
“Waiting” is useful only with an owner and a return condition. Waiting for what, from whom or from which source, until when? The answer can be simple: “Waiting for the revised notice; check the official source on Tuesday.” An assistant should not create a promise on another person’s behalf. A date on your reminder is the date you will check, not proof that somebody else will deliver by then.
“Unverified” protects an important boundary. You may have a draft that looks final, a screenshot with no sending confirmation, or a calendar entry that was never updated. If the completion claim matters, inspect the destination or authoritative record. If you cannot inspect it, say what is known and what remains unverified. Do not close a commitment solely because its title appears in a past-tense AI summary.
Avoid creating a state for every possible feeling about a task. The purpose of labels is to guide the next decision. If both “paused” and “deferred” mean “reconsider when the same condition changes”, one label may be enough. If two states require different action, distinguish them. A cancelled optional task usually needs no return reminder. A still-promised task deferred for lack of time may require a conversation with the person waiting.
Resolve dependencies before carrying work forward. Suppose an application pack needs three documents and only two exist. Copying “finish application” onto Monday’s list does not make the third document available. Carry forward the current state and the first feasible step: verify the missing document’s availability. Keep any later assembly work conditional. This makes next week’s plan responsive to reality instead of simply louder about the same blocked outcome.
Finally, keep private planning decisions separate from external commitments. You may decide in your review that a meeting no longer fits. That decision does not cancel the meeting for anyone else. You may decide to propose a new deadline. That does not establish agreement. The next action can be to prepare or send an authorised message, but the review should preserve the difference between your intended request and the recipient’s actual response.
5. Compare effort without inventing time savings
When time matters, specify what you are counting. Task effort, total elapsed time, waiting time and review overhead answer different questions. Forty minutes spent drafting across two evenings is forty minutes of logged effort. It is not a two-day task duration in the same sense. A document that waits three days for an attachment may involve little additional active effort but still miss the useful window. Do not merge these measures casually.
Compare the same scope and completion condition before drawing a conclusion about estimates. If a finished two-page handout took 130 minutes against a ninety-minute estimate for the same two-page outcome, the recorded effort exceeded the estimate by forty minutes. If a whole-shelf task became a one-box task, subtracting the two durations does not measure faster shelf-sorting. If a reading outcome remains incomplete, lower effort does not prove efficient completion.
Keep new work separate from the original comparison. An unplanned parcel collection or a file-recovery problem may genuinely occupy time. It belongs in the actual record, but it was not necessarily included in the original estimate. You can show both: effort on originally listed items and additional effort on newly arising items. This lets the reader explain a busier week without pretending that the initial plan predicted everything.
Include the cost of using the review method when you are judging the method itself. Gathering records, writing the prompt, checking claims and transferring decisions all take effort. A generated summary that arrives quickly may still require substantial correction. Conversely, a review that takes longer may reveal an important unfinished obligation. Decide what question matters before reducing usefulness to speed. This article supplies calculations, not a universal target for review duration.
Missing time records should remain missing. You can use a range, a rough recollection or a statement that effort was not tracked. Do not treat a blank as zero. Zero means the supplied record establishes no effort on that item; blank means the value is unknown. Those meanings can change a total substantially. Where some entries are uncertain, label the total as partial or approximate rather than presenting an exact sum of invented values.
Avoid a universal completion percentage across unlike items. Sending a short reply, reading a difficult chapter and organising a cupboard are not interchangeable units. Even within one project, a count of subtasks can be manipulated by splitting easy steps more finely. If a percentage is useful for a narrow purpose, define its denominator and explain what it leaves out. A short state-by-state description is often clearer for an ordinary personal week.
The review should preserve non-measured value too. An hour of companionship need not demonstrate a financial return. Enjoying a walk does not require a performance score. You may choose to protect such activities because they matter to you. The assistant can help make the choice explicit, but it cannot derive your priorities from a time log. Arithmetic checks feasibility and comparisons; it does not decide what a worthwhile life contains.
6. Read the complete fictional plan before the actual records
Leena is a fictional adult reviewing one ordinary Monday-to-Sunday period. The review cutoff is Sunday at 18:00. Her records concern six chosen personal items rather than a complete account of all 168 hours. Work, meals, ordinary rest and most daily administration are outside this small packet. Therefore, the totals below describe selected recorded effort only. They must never be called Leena’s total working time, free time or productivity for the week.
Plan record P1 was written on Sunday before the week began. Item H was a two-page private handout for an adult hobby group. Leena intended to finish a checked PDF by Friday evening, using ninety minutes. “Checked” meant that all six supplied reference labels were present, both pages opened, and the text was readable at her chosen viewing size. The task stopped at a private ready-to-share file. Sending or publishing was not part of its completion condition.
Item D was a personal document folder. Leena intended to place three named reference documents, A, B and C, into one private folder and check their labels by Thursday. She estimated forty-five minutes. Thursday was her own preparation target, not a legal deadline or an official submission deadline. This exercise includes no application, payment or transmission of document contents. The next stage would be considered only after the folder was ready.
Item L was learning from a chapter in a hobby manual. The planned outcome was to read chapter four, make a short note, and explain four specified ideas without the manual by Friday. The estimate was eighty minutes. Reading alone did not satisfy the whole condition. The four prompts were: identify the two main categories, explain their difference, give a fresh example of the first, and give a fresh example of the second. The actual subject content is outside this exercise; the review uses the recorded response states.
Item B was a household organisation task: sort one complete shelf, with sixty minutes estimated. Leena wanted to separate materials to keep, relocate and consider discarding. Actual disposal was excluded. No other person’s belongings were included. The shelf was an optional personal project, so she could revise its scope without changing a promise to someone else. The original plan still needed to remain visible after any revision.
Item F was a reply to a friend about two possible meeting windows. Leena had promised to reply by Wednesday evening and estimated ten minutes. Completion required an actual sent reply naming her availability. It did not require the friend to choose a time or confirm a meeting. A draft would not be enough. This distinction will prevent the review from turning one completed communication into an imaginary agreed appointment.
Item W was an optional Sunday morning walk, with forty-five minutes set aside. No other person was relying on it. It was an intention, not a medical prescription, training target or externally promised event. Leena could cancel or replace it without creating an obligation to make up the minutes. The review would record the choice without interpreting it as evidence about her health or discipline.
The original selected effort estimate was 90 + 45 + 80 + 60 + 10 + 45 = 330 minutes. This is a sum of six different task estimates, not a promise that every item would finish regardless of changing conditions. Preserve P1 unchanged as the original comparison record. Later notes will explain what happened to each item and why a single subtraction cannot establish time saved.
7. Inspect Leena’s actual record and change log
The following records are also fictional. They are the complete supplied evidence for this case; do not imagine additional messages, files or private causes. Where the record is explicit, use it. Where it is silent, preserve the gap. The purpose is to practise the discipline of reviewing a bounded packet rather than constructing the most persuasive possible story about a person.
H1, Monday: Leena logged forty minutes preparing the handout. The draft contained six reference labels, but three labels had no matching source location. Her note said, “Need to find locations before the final check.” H2, Wednesday: she logged fifty-five minutes locating the three missing references and revising the draft. The log did not divide those fifty-five minutes between searching and rewriting. H3, Friday: she logged thirty-five minutes finishing the PDF and checking all six labels, both pages and readability. The private file passed her stated checks. No sending record exists.
D1, Tuesday: Leena logged twenty minutes creating the document folder and adding A and B. Document C was absent. D2, Thursday: she inspected a current source notice that stated C would become available the following Tuesday. Her short check was included in the twenty-minute total already recorded for D; do not add an invented separate duration. She changed the folder’s state to “blocked on C” and kept the original Thursday target in the history. The folder contained two documents at the review cutoff.
L1, Tuesday: Leena logged thirty minutes reading the first part of chapter four. L2, Saturday: she logged sixty minutes finishing the chapter, writing the note and attempting the four explanations without the manual. She supplied complete responses to the first two prompts and wrote “not yet sure” for both fresh-example prompts. The note and attempts were saved. The Friday target was missed, and the full four-prompt acceptance condition was still unmet on Sunday.
B1, Thursday before beginning: Leena recorded, “Change this week’s optional shelf outcome to sorting the blue box only. Leave the rest of the shelf for a later decision.” B2, Thursday after the activity: thirty-five minutes were logged; the blue box was sorted into the three chosen categories. Nothing was discarded. The rest of the shelf remained unsorted. B1 is a genuine prior scope change, not an explanation invented after seeing the result.
F1, Wednesday: Leena logged ten minutes writing, checking and sending a reply that gave her two available windows. The sent-message record showed Wednesday at 17:20. At Sunday 18:00, no response confirming either window was in the packet. Item F’s promised reply was complete. The possible meeting was still unconfirmed. Neither state contradicts the other, because they describe different outcomes.
W1, Sunday at 09:00: Leena noted that she had cancelled the optional walk because she did not want to go out in the rain. No walk time was logged; the supplied record explicitly gives zero minutes for W. She chose a quiet morning instead. The record does not invite a diagnosis or establish that she should replace the walk with another target. It records an ordinary personal choice.
U1 recorded fifteen minutes collecting a neighbour’s parcel, an unplanned favour she had agreed to that day. U2 recorded twenty minutes restoring an accidentally moved personal file. These were additional items, not parts of H, D, L, B, F or W. R1 recorded twenty minutes preparing and checking the weekly review itself. All durations in this fictional packet are non-overlapping active minutes, supplied as exact values for arithmetic practice. Real records may be less precise.
8. Reconcile the result before explaining it
First reconcile Item H. The original outcome was a checked private two-page PDF by Friday. H3 supports that completion, and H1 through H3 give 40 + 55 + 35 = 130 minutes. The estimate was ninety minutes for the same scope. The recorded overrun is forty minutes. There is no evidence that the file was shared, and no need to treat that absence as a failure because sharing was outside the stated task. The review should say “private PDF complete”, not “handout distributed”.
Item D is blocked, with two of the three required documents present. Its active effort is twenty minutes. Thursday’s preparation target was not met. D2 explains a relevant external dependency: C was unavailable until the following Tuesday according to the supplied notice. The next useful step is a source check after that availability point. “Complete the folder on Monday” would ignore the evidence. “Leena forgot the folder” would contradict D1 and D2.
Item L is in progress. Ninety minutes were logged because 30 + 60 = 90. Reading and note-making are complete, but the independent explanation condition is not. The actual effort already exceeds the eighty-minute estimate by ten minutes, while additional work may still be required. Do not describe ninety minutes as its final completion cost. The saved attempts identify a specific gap: making fresh examples, rather than identifying and distinguishing the categories.
Item B is revised scope completed. B1 replaced the optional whole-shelf outcome with the blue-box outcome before B2. Thirty-five minutes therefore describes the smaller finished task. The original shelf remains unfinished and uncommitted for now. The review should not automatically re-add the remaining shelf next week. It should preserve it as an optional possibility whose current relevance can be considered later. There was no external promise to renegotiate.
Item F is complete at ten minutes, supported by the sent-message record. The friend’s response remains unknown within the packet, and the meeting is unconfirmed. Item W is cancelled at zero logged minutes. It is not incomplete work that needs to be repaid. These state distinctions matter more than counting how many rows receive a tick. They tell Leena what requires action and what can be left alone.
The actual effort on the six original items is 130 + 20 + 90 + 35 + 10 + 0 = 285 minutes. The difference from the original estimate is 330 − 285 = 45 minutes. That is only a difference between two totals covering changed states and scopes. It is not a forty-five-minute efficiency gain: D and L are unfinished, B is smaller, and W was cancelled. The handout and reply are the two original outcomes completed at unchanged scope; together they took 140 minutes against 100 estimated.
The additional items contribute 15 + 20 = 35 minutes. Including those items and the twenty-minute review gives 285 + 35 + 20 = 340 recorded minutes. Every item in that total has a source label. The total still excludes most of Leena’s week. A correct summary is: “This selected packet contains 340 recorded active minutes, including additional tasks and the review.” A claim that she worked only 340 minutes all week would be false.
9. Repair a polished review that changes the evidence
Consider this fictional flawed assistant response: “You completed five of your six goals, saved forty-five minutes through better organisation, and used AI effectively. The document delay happened because you procrastinated. Your reading improved by fifty percent. You finished the shelf faster than expected. Carry the walk into Monday, submit the documents then, and tell your friend the meeting is confirmed.” Every sentence sounds decisive. Nearly every sentence exceeds the packet.
Start with “completed five of six”. The response has collapsed original completion, revised completion, partial work and cancellation into one count. H and F meet their original conditions. B meets a documented replacement condition. D is blocked, L is in progress and W is cancelled. A state summary preserves those differences. Whether to count the revised item as a “goal completed” depends on an explicitly defined purpose; it cannot rescue the misleading five-of-six assertion.
The supposed forty-five-minute saving is the arithmetic error of interpretation from the previous chapter. The subtraction is correct but the conclusion is not. Less recorded effort on an unfinished or reduced task does not show faster completion. The claim about effective AI use is also unsupported: the source packet does not contain a comparison between an assisted and an unassisted method. Correct arithmetic can sit inside an entirely unjustified causal story.
The procrastination claim replaces the actual document dependency with a character explanation. D2 supplies the relevant condition: C was unavailable until Tuesday. The reading claim has no baseline against which a fifty-percent improvement could be calculated. Two complete responses out of four prompts describe this one attempt’s response states; they do not measure a change from an earlier attempt, and they do not establish general mastery of the chapter.
The shelf claim erases B1’s scope reduction. The next-week instructions then create further problems. The walk was voluntarily cancelled and need not become debt. Monday submission is impossible to justify from a packet where C becomes available Tuesday, and submission was never authorised or even defined as the task. The friend’s meeting is unconfirmed. Sending that assertion could create confusion for another person.
A repaired illustrative response would say: “At Sunday 18:00, H and F meet their original completion conditions. B’s smaller, pre-agreed personal scope is complete. D is blocked on document C; L still needs the fresh-example part of its acceptance check; W was cancelled. The six original items contain 285 active minutes, with 35 additional-task minutes and twenty review minutes. No time-saving or AI-benefit claim follows from these totals. Next week’s proposals should begin with the document source check and the remaining learning check.”
That response is less dramatic and more useful. It protects the record, leaves personal choices with Leena and identifies where further decisions belong. When checking your own AI-generated review, look especially for invented motives, vanished uncertainties, scope substitutions, unsupported percentages and external actions described as already agreed. These errors can be consequential even when the language is encouraging and the formatting is attractive.
10. Investigate the difference without inventing a cause
After reconciliation, choose only the differences worth investigating. Leena does not need a theory for every deviation. The handout overrun, missing document and learning gap each affect a plausible next decision. The cancelled walk does not necessarily require analysis. The blue-box scope choice may simply have been an appropriate adjustment. Proportionate reflection asks which explanation would change what you do, rather than demanding a lesson from every ordinary event.
For the handout, the evidence supports a preparation issue: H1 recorded three missing source locations, and H2 included finding them. It does not establish that all forty extra minutes came from that issue. H2 combines searching and revision without a breakdown. A careful interpretation is: “Missing source locations contributed to work that was not ready at the first drafting session.” A stronger numerical attribution would require a more detailed record than the packet contains.
Two plausible repairs follow. Leena could check source locations before layout, or reduce the handout’s scope when references are not available. Both address the observed problem. “Write faster” does not. “Buy another AI subscription” also does not follow from the evidence. Choose a repair at the point where the process actually became unstable. A modest verification step may be enough; more technology is not automatically the relevant answer.
For the document folder, the dependency was external to the immediate assembly task. More reminders to finish the folder before C exists would add noise. A source check on the appropriate day is a better intervention. If the notice later changes, the next step should change with it. The review is not establishing a permanent rule that all document tasks should be delayed; it is responding to the specific unavailable input in this packet.
For learning, distinguish exposure from independent application. The records show that Leena read and took notes, then could not yet supply two fresh examples. The targeted next attempt should practise generating and checking examples, rather than automatically rereading every page. This is a recommendation derived from the fictional response pattern, not a promise about educational outcomes. A future unaided attempt will show whether that particular gap remains.
Do not force a causal conclusion from one week. Perhaps the handout topic was unusual, the source list was unusually incomplete, or Leena was still learning the format. A useful next trial keeps enough of the task stable to learn something. You might test the early reference check on the next comparable handout and record whether missing locations remain. If the next task is entirely different, do not treat its shorter duration as proof that the repair caused an improvement.
Also separate the quality of a decision from its eventual outcome. A sensible plan can meet an unforeseeable disruption. A poor assumption can happen to work once. Ask what information was available when the choice was made, whether the choice respected real boundaries, and what later evidence changes the next choice. This keeps the review from rewarding luck or criticising a reasonable decision merely because the week did not cooperate.
11. Turn the review into Leena’s next-week decision card
Leena now supplies a new constraint: she has chosen to allocate 180 minutes next week to follow-up work from this packet, including the next review. This is an illustrative personal allocation, not her total available time. Tuesday has a fifteen-minute source-check window; Wednesday has a thirty-minute conditional assembly window; Thursday has forty minutes for learning; Friday has twenty-five minutes for small administrative improvements; Sunday has twenty minutes for review. A separate fifty minutes remains unallocated reserve.
Her first decision is to keep the completed handout closed. The file passed its private acceptance checks, and no distribution task was requested. She proposes a fifteen-minute process repair on Friday: add a source-location check to the small handout preparation card she already uses. This is preparation for a later comparable task, not another round of rewriting the finished handout. If the card already contains an adequate check, she can stop early instead of inventing more procedure.
Her second decision is to check C’s availability on Tuesday for up to fifteen minutes. If the current source makes C available and it is the correct document, Wednesday’s thirty-minute window can be used to add it and verify the folder. If C is still unavailable, the folder stays blocked and the current source determines an appropriate later check. She does not spend the assembly window guessing the contents or treating the availability notice as the document itself.
Her third decision is a forty-minute learning attempt on Thursday. She will first try fresh examples without the manual, then inspect the relevant explanation if needed, and finish with another unaided example attempt. The acceptance condition remains the ability to supply and explain the required examples. The time allocation is a boundary for this session, not a guarantee that learning will be finished. If uncertainty remains, the review records it accurately rather than manufacturing a pass.
Her fourth decision is to reserve ten minutes on Friday to check whether the friend has replied. If a reply exists, she can decide what response is needed. If none exists, she will consider whether a follow-up is useful in their actual context; the card does not automatically authorise a message. No appointment is entered as confirmed until there is an agreement. The original promised reply stays complete while this possible new communication is considered separately.
Her fifth decision is to leave the remaining shelf and the cancelled walk off the committed follow-up list. She may choose either later, but neither is an unpaid debt from last week. This is an important part of review: some work should not carry forward. Otherwise, every optional idea becomes an accumulating obligation and the next week’s plan eventually contains more history than current intention.
The allocation check is 15 minutes for the handout card + 15 for the document source check + 30 conditional assembly + 40 learning + 10 friend-response check + 20 review = 130 minutes. With fifty minutes of reserve, the total is 180. Friday’s two tasks total twenty-five, matching its window. The other named tasks match their windows. If assembly is not triggered, its thirty minutes remain unused or can be reassigned by a fresh choice; they are not counted twice as both assembly and reserve.
The final decision card reads: “Protect the document check and one targeted learning attempt. Keep H and the original F reply closed. D remains conditional on C. B’s completed box stays complete; no shelf continuation is promised. W remains cancelled. Make one small source-preparation improvement. Review the learning attempt, document state and usefulness of the new check on Sunday.” This is a usable output, not a motivational report.
12. Update records without rewriting the past
The next step is to place the decisions where you will actually use them. A review document that says the right things but leaves the calendar and task list contradictory is incomplete as an operating aid. You do not need to copy the whole review into every application. Update the few current-state fields that matter, and keep a reference to the evidence or decision note when it will help later.
For Leena, the handout record changes from active to complete, with H3 as its completion evidence. The document-folder record remains blocked and receives Tuesday’s source-check action. The learning record remains in progress with the two fresh-example prompts identified. The box record says revised scope completed and retains the Thursday scope note. The reply record is complete, while any meeting record stays tentative. The optional walk record is cancelled without a replacement reminder.
Preserve the original plan rather than editing it to match the outcome. Otherwise, next month every estimate may appear perfectly accurate because the evidence of the original estimate has disappeared. A simple dated change note is sufficient. You do not need a formal version-control system for a household task. You do need enough history to distinguish what you knew and chose then from what you know and choose now.
If new evidence arrives after the cutoff, add a dated update. Suppose the friend replies on Sunday at 19:10, after the 18:00 review. The weekly review was correct to say unconfirmed at its cutoff. The later reply can update the current meeting state without making the earlier review false. Write “Update after review: reply received at 19:10; consider the proposed window.” Keep the actual content and agreement boundaries intact.
If you discover that an earlier record was wrong, correct it openly. Suppose H3’s file turns out to be an old export missing a reference. The appropriate response is to reopen the handout’s completion check, identify the current file and record the correction. Do not preserve a known false completion merely to keep the review neat. Conversely, do not erase all evidence of the original mistake if retaining a short correction would prevent it recurring.
External actions still require the relevant permission and actual execution. An assistant may prepare a calendar change, draft a message or propose moving a file. A review request alone should not be treated as blanket authority to send, delete, purchase or alter shared commitments. Confirm the intended action through the tool or process you use, then verify the result in the destination. Proposed changes and completed changes belong in different states.
13. Use a smaller review when the record is sparse
Consider a second fictional adult, Arun, whose week changed when he took on an agreed family responsibility. His packet is much thinner than Leena’s. He planned to finish a personal portfolio draft, make a promised phone call and attend an optional evening class. He has a saved draft with two sections unfinished, a call log showing the promised call took place, and a note that he cancelled the class before it began. He did not record task durations.
A sound review can still establish useful states. The portfolio remains in progress because its two unfinished sections are visible. The call is complete if the promise was simply to make that call and the log matches it. The class intention was cancelled. The family responsibility explains a changed condition, but its private details are unnecessary for this analysis. The assistant should not invent hours spent, assess Arun’s resilience or infer the feelings of the family member involved.
The time conclusion is “not measured”. It is not zero, and it is not whatever duration seems typical. Arun remembers that the portfolio received less attention than intended, but he cannot reconstruct a reliable total. He can still decide that the next useful step is one contained section rather than an entire portfolio finish. A practical review does not fail merely because it lacks a spreadsheet of the week.
Now supply the next constraint: Arun has one confirmed forty-five-minute window next week and several uncertain evenings. His proposed decision is to use that confirmed window to complete the shorter unfinished section, then inspect whether the remaining section still fits his current purpose. The uncertain evenings remain uncommitted. If a window becomes available, he can choose work then; he should not promise someone a finished portfolio on the basis of time he has not secured.
Suppose an assistant says, “Your class cancellation proves that you lack consistency. Schedule two classes next week to regain momentum.” The packet supports neither the diagnosis nor the remedy. A better response is: “The class was an optional activity cancelled during changed circumstances. Decide whether it still matters and whether a future session fits. No make-up commitment follows automatically.” This preserves Arun’s agency and prevents the review from escalating demands during an already variable period.
Sparse records call for selective reconstruction. Check a source if it affects a promise, deadline or important completion claim. Do not spend an hour recovering the precise duration of an optional hobby session just to make a dashboard look complete. The value of further information depends on whether it can change a meaningful decision. “Unknown, no action needed” can be a sensible outcome for a low-consequence detail.
Arun’s completed review can be brief: “Portfolio partly drafted; two sections remain. Promised call completed. Class cancelled during changed circumstances. Task time unmeasured. Next week, use the confirmed forty-five-minute window for one section; keep other evenings uncommitted until their availability is known. No broader system change is justified by this week alone.” This is a complete review for its limited purpose, even though it contains no performance percentage.
14. Make room for reflection that does not become a score
Some of the week’s most useful information is qualitative. You might notice that a conversation mattered, a project no longer interests you, or a repeated arrangement creates friction. These observations can guide a choice without becoming a numerical rating. The assistant can help you phrase the reflection clearly, but it should not insist on ranking every relationship, assigning an emotional score or turning ordinary preferences into a personality diagnosis.
Separate what you experienced from what you infer about someone else. “I found the late change difficult to accommodate” reports your experience. “They changed the plan because they do not respect me” proposes a motive that the event alone does not establish. If a relationship issue needs discussion, the review can identify the question you want to ask. It should not deliver a confident account of another person’s hidden intentions.
Choose a small number of reflective prompts that genuinely help. What deserves to continue? What took more effort than it was worth? What mattered even though it produced no visible output? What is one boundary I want to make clearer? There is no requirement to answer every question every week. A quiet or difficult week may yield one useful observation. A week with no important mismatch may need little reflection at all.
Do not require a positive lesson from every setback. Sometimes a file was lost, a plan changed or a hoped-for event did not happen. You can acknowledge disappointment and identify the practical next step without framing the experience as beneficial. Likewise, a good week need not be interrogated until a fault is found. Reflection should help you understand and choose, rather than create work through compulsory self-analysis.
When a review touches health, distress, legal commitments or consequential financial matters, keep the assistant’s role narrow. It can help organise questions and records, but a weekly log does not establish a diagnosis, treatment decision or professional judgment. Use appropriate qualified help for those questions. The ordinary planning method here is not a substitute for care, and it should not encourage you to infer a condition from a few incomplete task records.
End qualitative reflection with either a proportionate choice or an explicit decision to leave it alone. “Protect one unhurried conversation next week if the other person is available” is a proposal. “No change; this was an enjoyable one-off afternoon” is also valid. Not everything worth noticing needs to become a repeating habit. A good review can preserve something meaningful without turning it into another obligation.
15. Use a controlled SI conversation from packet to final note
The Life series’ loop is Capture → Clarify → Expand → Structure → Execute → Review → Improve. In this task, Capture gathers the bounded weekly packet. Clarify resolves ambiguous states and source conflicts. Expand considers a few explanations or next steps. Structure produces the checked review. Execute places approved decisions into the existing system. Review checks whether that output matches the sources. Improve carries one justified change into the next cycle.
Begin with a checking pass rather than requesting the final report immediately. A copyable prompt is: “Review only the supplied period and records. First identify missing completion evidence, unclear scope changes, contradictory dates and unknown durations. Distinguish planned, actual and proposed actions. Do not infer causes, motives or health conditions. Ask only questions that could materially change the review. Do not send messages or change external records.” This allows uncertainty to surface before it is hidden inside prose.
For Leena’s complete packet, an appropriate illustrative checking response is: “The records are sufficient to classify all six original items. H is a private completion, not a sending claim. D is blocked; Thursday was a personal target. L has an unmet acceptance condition. B has a documented prior scope change. F is sent but the meeting is unconfirmed. W is cancelled. The durations are explicitly supplied as non-overlapping; no time reconstruction is needed.” That response recognises what is already known instead of asking unnecessary questions.
Next ask for the comparison: “Build a short item-by-item account with original intention, actual evidence, current state and next decision needed. Show the arithmetic separately. Do not call reduced effort a saving when scope changed, an item was cancelled or work remains unfinished. For each proposed explanation, say which record supports it and what remains uncertain.” Check that the answer has not merged unknowns into zeros or completion proposals into facts.
Then provide your current preferences and capacity for the next step. The assistant cannot infer that Leena prefers fifty minutes of reserve or that Arun wants to preserve uncertain evenings. Those are human choices or constraints supplied to the process. Ask for two reasonable alternatives if the trade-off is real. For example, one plan may protect more unallocated time while another advances an optional project. Neither should be described as objectively best without the relevant preference.
Finally request a compact decision note: “Use the checked comparison and my chosen option. Include completed items, open states, the few next actions, their conditions, what will not carry forward and the next review point. Keep the evidence labels for consequential claims. Do not add new commitments.” Read the result once against the source packet and once as something you could use on a busy day. If it requires decoding a complicated taxonomy, simplify it.
Ask, Think, Build, Integrate and Operate are useful roles here, not requirements for five tools. Ask frames the review question. Think compares the evidence. Build creates the note. Integrate puts approved changes where they belong. Operate means using and updating the note as conditions change. You can perform every role manually. The assistant earns a place only if its contribution is useful after the checking and maintenance it requires.
16. Diagnose the smallest repair when a review goes wrong
If the review becomes a long retelling of the week, ask which sentences change a decision. Keep meaningful reflection, but remove repeated chronology that serves no current purpose. A complete evidence packet may be lengthy during learning; the operating note need not reproduce it. Retain source references so the detail remains available without forcing you to reread it every time you check the next action.
If the same unfinished task returns every week, inspect its state before increasing reminders. It may lack a clear completion condition, depend on unavailable information, require a decision you have not made, or no longer be worth doing. Each possibility suggests a different repair. A blocked task needs a dependency check. An oversized task needs a smaller coherent outcome. An obsolete optional task may need cancellation. Repeated copying is not itself progress.
If the assistant offers many plausible explanations, narrow the request to explanations that are both supported and actionable. “There may be a mismatch in your motivation, environment, habits, identity and goals” is too broad to guide a modest task repair. “The source locations were missing before layout” points to a specific check. When evidence does not distinguish explanations, acknowledge that and choose a low-cost test rather than adopting the most impressive theory.
If the review produces guilt, inspect the language and the measurement design. Has cancelled optional work been treated as debt? Have changing responsibilities disappeared from the comparison? Are unlike tasks being combined into a score of personal worth? Replace those errors with state descriptions and current choices. This is not a promise that any format will remove difficult feelings. It is a way to avoid adding unsupported judgments through the review itself.
If the process takes more effort than its decisions justify, shrink its scope. Review the few open commitments first, then one useful lesson. Stop when the remaining unknowns are low consequence and would not change the next choice. You do not need to monitor every category continuously. A recurring field that never affects a decision is a candidate for removal, especially if maintaining it requires reconstructing private details you did not otherwise need.
If the numbers look precise but cannot be traced, rebuild the arithmetic from the original entries. Check for overlapping sessions, duplicated tasks, approximate values presented as exact, and a blank interpreted as zero. Keep an excluded-items note where the boundary might be misunderstood. A five-minute correction to a denominator can matter more than an attractive chart, particularly when the chart is about tasks with different completion conditions.
If the review silently changes shared commitments, separate the private decision from the communication. Record the request you want to make, identify who must agree, and leave the external state unchanged until the authorised action happens. The same principle applies to file deletion and calendar changes. A useful review clarifies authority; it should not create a shortcut around it merely because an assistant can operate a tool.
17. Practise independently with Rosa’s changed packet
Try this fictional transfer case before reading the answers. Rosa is an adult reviewing Monday through Sunday at 17:00. Her packet covers five personal items and no other part of her week. All supplied durations are exact, non-overlapping active minutes for this exercise. Her original plan was Q, a private three-page hobby guide taking 100 minutes; R, sorting two boxes taking fifty minutes; S, sending an availability reply taking ten minutes; T, completing an unaided four-prompt learning check taking sixty minutes; and V, an optional forty-minute visit to a public exhibition.
Q required three readable pages and four verified reference labels by Saturday. Q1 logged forty minutes on Tuesday. Q2 logged seventy minutes on Friday. Q3 showed a three-page PDF saved Friday, but its check record said only three of the four reference labels were verified. There was no later verification record. The draft existed before Saturday, but the acceptance condition included all four labels. Do not assume that the fourth was checked off-screen.
R1, Wednesday before starting, said: “Reduce this optional task to the red box only.” R2 logged twenty minutes sorting that box. The second box remained untouched. S1 showed a saved ten-minute draft on Thursday; a sending check at the review cutoff found no sent message. Rosa had promised the reply by Friday. She had not yet asked the recipient to change that expectation. The draft’s title contained the word “final”, which is not evidence of sending.
T1 logged fifty minutes on Saturday. Its record showed complete unaided responses to all four specified prompts, followed by a check against the supplied answers confirming that each met the acceptance condition. The packet contains no earlier comparable attempt. V1 said Rosa cancelled the optional exhibition visit on Sunday morning because she preferred to stay home. It gives zero minutes. No other person was waiting for her at the exhibition.
U1, an additional unplanned item, logged fifteen minutes returning a borrowed book as agreed that day. W1 logged fifteen minutes gathering records and checking the weekly review. Rosa also wrote, “I felt less rushed on Thursday.” That is a supplied reflection about her experience; it is not evidence of why Thursday felt different. No AI-use comparison or health information is included in the packet.
Answer these eight questions. First, give the current state of each original item with its evidence. Second, calculate the original estimate total and actual effort on those five items. Third, calculate all recorded effort including the additional item and review. Fourth, identify exactly which original-scope outcome is verified complete. Fifth, explain why the difference between estimated and actual totals is not an efficiency gain. Sixth, identify the most time-sensitive unresolved commitment. Seventh, propose one supported process repair and name an uncertainty. Eighth, prepare a bounded next-week decision card from the constraints below.
For the next-week card, Rosa supplies eighty minutes. Monday has a fifteen-minute window for inspecting and, if she chooses, sending the overdue reply. Tuesday has twenty-five minutes for finding and checking Q’s fourth reference. Friday has twenty minutes for a final guide check if the reference is resolved. Sunday has ten minutes for review. Ten minutes remain unallocated. If the fourth reference cannot be verified, she wants a clear choice between removing the unsupported claim and postponing completion; she has not yet chosen either.
Before turning to the answers, audit this flawed summary: “Rosa finished four of five goals and saved eighty minutes. Her learning improved because AI made her calmer. The guide is ready to publish, the reply is complete, and next week she should finish the second box as well.” Identify every unsupported completion, causal claim, authority leap and carry-forward assumption. A strong answer should correct the statements, not merely call them inaccurate.
18. Check the full transfer answers and boundary tests
Answer one: Q is in progress or awaiting verification, not complete. Q3 confirms a three-page file but leaves one required reference unverified. R is revised scope completed because R1 changed the optional task before R2 completed the red box. S is overdue and unsent: S1 proves a draft, while the sending check found no sent message at the cutoff. T is complete against its stated four-prompt condition. V is cancelled. Each state follows a supplied record rather than the appeal of the task’s title.
Answer two: the original estimates total 100 + 50 + 10 + 60 + 40 = 260 minutes. Actual effort on the five items is Q’s 40 + 70 = 110, plus R’s twenty, S’s ten, T’s fifty and V’s zero. Therefore 110 + 20 + 10 + 50 + 0 = 190 minutes. The estimated-versus-actual difference is seventy minutes, not eighty. This catches both an arithmetic mistake and the more important interpretive mistake in the flawed summary.
Answer three: add the fifteen-minute book return and fifteen-minute review to the 190 minutes. All recorded effort is 220 minutes. Do not add them again when discussing the original five items. The selected packet is not a total account of Rosa’s life. Its scope excludes other activities, so it cannot establish how much free time she had or whether she should have done more.
Answer four: T is the only verified completed original-scope outcome among the five. It met the supplied acceptance check at fifty minutes against sixty estimated. That supports a ten-minute difference from this task’s estimate. It does not establish learning improvement from an earlier baseline, prove durable mastery, or show that AI caused the result. R is also a legitimate completion, but of a documented smaller replacement scope. Keep that success visible without confusing the two comparison sets.
Answer five: the seventy-minute difference mixes an unfinished guide, a reduced box-sorting task, an unsent reply, a completed learning task and a cancelled visit. These are not equal original outcomes achieved in less time. The correct statement is about recorded effort and states. If Rosa wants to evaluate a process change, she needs comparable task definitions and relevant observations over suitable uses. This packet does not contain that comparison.
Answer six: the overdue reply is the clearest unresolved external commitment. Rosa promised Friday and the cutoff is Sunday. A saved draft did not fulfil that promise. She should inspect the draft and decide on an appropriate response promptly, using the authorised communication route. Her Monday window supplies a proposed opportunity, not permission for the assistant to send on her behalf. The exact message should reflect the real relationship and circumstances rather than an invented excuse.
Answer seven: one supported process repair is to put a reference-completeness check before declaring Q ready. The record shows an acceptance gap between the saved file and the four-label condition. Another supported repair is to separate “draft complete” from “sent” in the reply record. Choose one that prevents the observed error without building unnecessary administration. The cause of the unsent draft remains unknown; it may have been overlooked, paused intentionally or blocked for a reason absent from the packet. Do not choose a motive merely to finish the explanation.
Answer eight: allocate Monday’s fifteen minutes to checking the overdue reply and taking the action Rosa authorises. Allocate Tuesday’s twenty-five minutes to the fourth reference. Keep Friday’s twenty-minute final guide check conditional on that reference being resolved or on Rosa explicitly choosing an acceptable scope revision. Keep Sunday’s ten-minute review and ten minutes unallocated. The check is 15 + 25 + 20 + 10 + 10 = 80. No second-box work or replacement exhibition visit is added. T remains complete; any later learning check would be a new choice.
The flawed summary fails in several ways. Four of five is not supported by the states. Eighty minutes is arithmetically wrong; seventy is the raw estimate difference, and even seventy is not established time saved. “Learning improved” lacks a prior comparison. “Because AI made her calmer” lacks an AI-use record and confuses Rosa’s Thursday reflection with a causal account of Saturday’s learning. The guide is neither fully verified nor authorised for publication. The reply is drafted, not sent. The second box is an optional future choice, not an automatic debt.
Now test two boundaries. If a new sending record shows that S was sent Friday but had been overlooked, correct S to complete and record the evidence correction. Do not retain the overdue label for consistency with the old review. If Q’s fourth reference is verified on Monday after Sunday’s cutoff, the Sunday state remains accurately incomplete at that time; add the later completion update when all conditions are checked. New evidence can correct an old factual error or describe a genuinely later event. Distinguish those cases.
19. Build your own review and know when to stop
For your first use, choose a bounded period and a handful of items that matter. Write the original intention for each, the relevant actual evidence, any recorded scope change and the current state. Add only the next week’s constraints needed to make the immediate decisions. If you do not have time records, leave them out or label your recollection. A truthful small packet is a better starting point than a comprehensive-looking reconstruction built from guesses.
Use this compact review form in ordinary prose: “Period and cutoff. Important completed outcomes with evidence. Open items and their current states. Material changes to scope or conditions. One supported interpretation, with uncertainty. Next actions and their conditions. Items that will not carry forward. Next review point.” The form is an aid, not a compulsory set of boxes. Remove a field if it repeatedly adds no useful information in your situation.
How long should the review take?
There is no duration that this guide can prescribe for everyone. Decide what the review needs to establish, then use the smallest process that does it reliably. A quiet week may require only a few state updates. A week with several changed promises may deserve more checking. If the process grows, ask whether the extra detail changes decisions or merely makes the report more impressive. Leena’s and Rosa’s fictional durations are arithmetic inputs, not targets.
What if the week went well?
Record genuine completions, preserve anything useful and avoid inventing a fault to justify the review. You can keep the current arrangement without adding a new habit. A review can end with “no material change needed”. If something went unusually smoothly, note the conditions before treating it as a permanent improvement. An easier task, a quieter week or help from another person may explain part of the difference.
What if I cannot remember much?
Inspect a few relevant records for important commitments and leave low-consequence gaps unresolved. Do not let the assistant fill them with a typical-looking week. The next practical change might be one short completion note when a meaningful task finishes, rather than detailed tracking of every activity. Choose a recording habit only if the information it preserves will help you make later decisions.
Can the assistant decide what I should do next?
It can propose options from the evidence and constraints you supply. You decide which outcome matters, which trade-off is acceptable and which commitments you are willing to make. A numerical ranking cannot supply missing personal values or another person’s consent. Use the assistant to make the choices understandable, then record the option you actually choose.
What shows that the review itself is useful?
Look for practical evidence in your own use: fewer falsely closed commitments, clearer next actions, visible changed conditions and a process whose maintenance remains proportionate. Do not claim a general productivity benefit from a single good week. If the review repeatedly creates more tasks than it resolves, simplify it. If a manual note does the job more clearly, that is a valid result.
Finish when the important states are accurate, the next actions are bounded and the remaining uncertainty is either assigned a useful check or consciously left alone. Save the small decision note where you will see it. Then return to living the next week. The purpose of reviewing is to leave you with a clearer account and a few defensible choices, while preserving room for ordinary life to remain ordinary.
Sources and further routes
This article’s packets, names, records, calculations, prompts and worked answers are original fictional teaching examples. They are not reported user outcomes. The evidence-state method and suggested review form are editorial guidance. Neither cited source is presented as validating this particular method or guaranteeing a personal benefit.
NIST’s July 2024 Generative AI Profile supplies the narrow factual background on confabulation, privacy and human–AI risks. Its organisational context and voluntary status are preserved in the discussion. Read the primary NIST profile
The David Allen Company’s Weekly Review checklist is linked as an established primary example of reviewing calendars, actions and projects. The article does not reproduce its checklist or claim that a particular review duration follows from it. Read the official checklist
Use How to Plan Your Week with Super Intelligence when the main task is allocating future capacity. Use How to Turn Goals into Action Plans with Super Intelligence for a project outcome, and How to Prioritise Tasks with Super Intelligence when several possible actions compete for attention.
For the wider personal system, return to the Super Intelligence Life hub. If the question is whether an SI workflow itself deserves to continue, use How to Review Whether Super Intelligence Is Actually Improving Your Life. That is a different evaluation from judging the state of one week’s personal commitments.
