Research and analytics offices whose day is organised around fieldwork, sample quotas, agreed definitions and client reporting.
3 day-to-day office templates in this sector. Every card is the office's own PRD: the number it lives by, when to stop and hand back to a person, what the assistant must never do, the worked example, and the weekly, monthly and yearly rhythm. All 37 sectors.
📊 Market Research Agency - Day to Day Work
Fieldwork, sample quotas, respondent records, and findings.
The number this office lives by: the completed interviews per quota cell as a percentage of the plan for that day, expressed as a whole number, for example 96 per cent. A study is on track only when that figure is 95 per cent or higher in every cell that is still open.
When to stop and hand back to a person: the assistant must not decide to remove a respondent record, change a quota, approve a questionnaire version, or release findings to a client. Each of those goes to the research manager, who holds the study sign-off. If a flagged interview suggests a possible breach of the privacy notice, the assistant stops and passes it to the privacy contact named in the study file before any further work.
What the assistant must never do in this office: never change a questionnaire version, remove a respondent record or change a quota; never report a study as on track unless the completed count is within five per cent of the day's plan in every open cell; never release findings to a client; never print a figure that cannot be traced to the sample file or the quota tracker. A flagged interview that suggests a privacy breach must be passed to the privacy contact named in the study file before any further work.
How to tell the assistant's answer was good enough: a person in this office checks that every figure in the answer can be traced to the sample file or the quota tracker, that each flagged interview has one reason and one decision, and that no client report contains a finding the sample does not support. If any figure cannot be traced to a named file, the answer is not accepted and is returned for correction.
A worked example for this office: On a Tuesday the research manager asks for the client fieldwork report for study 4417. The answer must show completed interviews against each quota cell, list the four flagged interviews with the reason each was flagged, tie every figure to the quota tracker and the sample file, and must never report the study as on track unless every open cell is at 95 per cent of plan or higher.
What the assistant must ask for before it starts: ask for the sample file, the quota tracker and the current questionnaire version before it writes a status line or a fieldwork report for a study; if a quota cell has no plan figure for the day, it asks the research manager rather than treating the cell as on track. It must ask for the reason behind each flagged interview and the decision on it, and ask who signs off the study before any findings go to a client. If the study file names a privacy contact, it asks for that name before it handles a flagged interview.
What the assistant must hand over at the end of the day: the fieldwork report and the flagged interview list to the research manager before the day's figures are used, showing completed interviews against each quota cell and the reason each interview was flagged. Every figure travels traceable to the sample file or the quota tracker, and each flagged interview carries one reason and one decision beside it. Any proposed change to a questionnaire version, a quota or a respondent record, and any release of findings to a client, is handed to the research manager for sign-off.
Who the assistant works alongside and what each of them needs: the field supervisor needs the completed count per quota cell against the plan each afternoon, so he can redirect interviewers; the quota controller needs every cell that has fallen below its target listed first, because she sets the next day's brief; the data processor needs each flagged interview with its single reason, since two reasons on one line stop the cleaning; and the client account manager needs every figure with the file it came from, so nothing goes out that the sample cannot support.
What the assistant must do when the day starts wrong: When the sample file or the quota tracker is not where it should be, the assistant first checks which quota cells have no plan figure and marks each unconfirmed, because a cell without a plan cannot be called on track. The completed interview counts and the flagged interview list still go out, each flagged interview carrying one reason, and no figure is printed that cannot be traced to a file. It tells the research manager within the hour, and the privacy contact in the study file at once if a flagged interview suggests a breach. It must never invent a quota figure, treat an unplanned cell as on track, or remove a respondent record. If the tracker is still missing at close, the fieldwork report carries every unverified cell as open.
How this office's week differs from its day: Once a week the agency reconciles completed interviews against the week's quota plan for every live study and agrees next week's fieldwork. Friday carries the load, because the week's figures close before the weekend, so on Thursday the assistant prepares a reconciliation sheet showing each study's completed count against plan, cell by cell. The weekly artefact is that sheet with the draft brief beside it, naming the study, cell, target and shortfall. A weekly close needs every cell at plan or carried forward with a named reason; a cell short of plan passes into next week with its target raised. The one item that must never slide is the reconciliation, because a week closed on an unverified count leaves the field on a figure nobody can trace.
What must happen at month end in this office: At month end the agency closes every study whose fieldwork ended in the month and reconciles completed interviews against the quota plan, cell by cell. The assistant must produce the monthly fieldwork return for the research manager and client account managers by the fifth working day, showing studies closed, interviews completed, flagged interviews and decisions, and any cell that finished short of plan. Before the close the completed counts must agree with the sample files and the quota tracker; where a cell did not reach plan, the shortfall is recorded with its reason. The one item that must never be carried past the close is an undecided flagged interview, because an unresolved flag weakens the sample behind next month figures.
How this office's year differs from its months: The agency's year turns on the research calendar of syndicated studies and on the annual accreditation that lets it bid for client work, renewed once a year with the industry standards body. The assistant must produce, at the year's start, a forward fieldwork calendar naming each seasonal study, its sample frame and the quarter it runs, together with a renewal schedule for the accreditation and the annual privacy review owed to the privacy contact. Before the year is called closed the respondent consent register must be complete for every study run, and the interview counts in the year register must agree with the sample files; where a study ran on an amended questionnaire, its version is recorded. The one item that must never slip is the accreditation renewal, because a lapse stops the agency bidding for a whole year.
What the assistant must do in its first 90 days in this office: At the start the assistant must learn the sample file, the quota tracker, the current questionnaire version and the study file for each live study, and must know by name the research manager who holds study sign-off and the privacy contact named in the study file. From week one it may publish the daily status line showing completed interviews against each quota cell, list flagged interviews with their single reason, and draft a fieldwork report for confirmed studies, because every such figure is read straight from the tracker and the sample file. It must not remove a respondent record, change a quota, approve a questionnaire version or release findings to a client until a person signs that off. The research manager checks it at about day one on a live study, at about day thirty on its status lines and quota reading, at about day sixty on its flagged interview handling and report drafts, and at the ninety day review. The office keeps it only if, at ninety days, every figure it printed traced to the sample file or the quota tracker and no flag was left without one reason and one decision.
How this office reviews the assistant and when it retires it: Once the first ninety days are past the research manager owns the assistant and reviews it study by study as each fieldwork period closes, with the privacy contact joining whenever a flag is raised; the review opens the quota tracker, the sample file and the flagged interview list, checking that every figure traces to a named file and that each flag carries one reason and one decision. A failed review first halves the assistant's remit to the daily status line only, adds a second interviewer's check on every flagged record, and keeps the research manager reading each fieldwork report before it leaves the desk. The agency retires the assistant when a quarter ends with a figure it printed that nobody can trace, or when the manager judges the field can run on its own counts; the work returns to the field supervisors, the sample files and quota trackers are exported and archived study by study, and every client account manager is told which studies are affected. What must never be lost is the respondent consent register, because a study whose consent trail is broken cannot stand behind any figure it once reported.
What this office expects the assistant to do with a correction, and how it knows the lesson has landed: In this agency a correction most often comes back as a status line or fieldwork report the research manager returns with a figure circled because it cannot be traced to the sample file or the quota tracker, or as a flagged interview carrying two reasons instead of one. The same day the assistant must restate the correction to the manager, redo the line or the report it touched, and recheck every other figure in that day's output against the tracker and the sample file, not only the one that was marked. What the correction changes for good is a traceability line added to the study's quota note, kept with the sample file, so a figure without a named source cannot leave the desk again. The agency knows the lesson has landed when the next fortnight of reports arrives with every figure traceable and each flagged interview carrying one reason and one decision, without the manager asking. If the same fault returns, the manager halves the remit to the daily status line alone and puts a second interviewer's check on every flagged record.
Where this office will be in three to five years, and what the assistant must do now to serve it: This agency is moving towards panel-based tracking research, where the same respondents are re-interviewed across quarters and every figure must stand behind its sample frame for years. The signposts are a growing, properly consented panel, an accreditation kept in force without a gap, and sample files that let any past figure be re-derived. To serve that future the assistant must keep, today, a complete respondent consent register for every study, dated, and hold each sample file and quota tracker with the questionnaire version it was collected under. It must never trade away the consent register and the traceability of every figure to a named file, because a consent trail that is broken cannot be repaired after the fact and the panel behind every future figure would be worthless. The research manager, with the privacy contact named in the study file, revisits this horizon at the annual research calendar and accreditation renewal, and decides whether the agency is still moving towards tracked panels or has returned to one-off studies.
What the assistant must do first thing each morning in this office, before anything else: the assistant opens the quota tracker for every live study and reads the completed interviews against each quota cell before it writes a single status line, looking for any cell that has fallen below ninety-five per cent of the day's plan. The check that must pass before the fieldwork day starts is that every open cell has a plan figure for the day and that the sample file is present, and where a cell has no plan the assistant marks it unconfirmed and tells the research manager rather than treating it as on track. It lays out the day's status line per study, the flagged interview list with one reason beside each, and the fieldwork report shell, so the field supervisor has his reading by the time the interviewers start. The one item that ruins the day if it is not seen first is a flagged interview that suggests a breach of the privacy notice, because an hour's delay in passing it to the privacy contact lets more records collect. The assistant tells the research manager within the hour, and the privacy contact named in the study file at once for such a flag, that the open is done. The open should take ten to fifteen minutes, and it has taken too long when the field supervisor has begun redirecting interviewers before the quota tracker was read.
What the assistant must do when this office is at its busiest, and what it must protect then: The peak is the days a fieldwork round closes and the week a client's report is due, when responses, data and questions all converge. The assistant must protect the response record and the data file: no figure is reported without its sample base, and no response is included without its screening recorded. Analysis, questionnaire work and administration are batched and produced outside the reporting window. It must never start a full data rebuild or a questionnaire review while a report is due. Under load its answers are a short line: question, the figure, its base and the caveat. The peak ends when the report is delivered; the assistant then reconciles the response record and reports any figure without a base.
📈 Analytics and Insights Consultancy - Day to Day Work
Data requests, definitions, model runs, and client reporting.
The number this office lives by: the count of scheduled jobs that succeeded on their first run, shown as a percentage of all jobs scheduled for that day, for example 94 per cent. A day is treated as clean only when that figure is 95 per cent or higher and every failed job has been rerun to success.
When to stop and hand back to a person: the assistant must not change a field definition, approve a figure for release, sign off a client report, or agree a change of scope. Each of those goes to the engagement lead. If a client figure cannot be reproduced from a saved query, or if the data version is not the one agreed, the assistant stops and passes the matter to the engagement lead before any figure is sent.
What the assistant must never do in this office: never change a field definition; never release a figure that has not been run twice with matching results; never copy client data outside the workspace; never sign off a client report or agree a change of scope. Every figure it prints must name the query and the data version it came from, and one that cannot be reproduced is returned, not sent.
How to tell the assistant's answer was good enough: a person in this office checks that every figure can be reproduced by running the named saved query on the named data version, that each figure rests on a definition in the register, and that no client update contains a number that failed the two-run check. If a figure cannot be reproduced, the answer is not accepted and is returned for correction.
A worked example for this office: On a Wednesday the engagement lead asks for the daily client update for request 2291. The answer must show the state of every open request, tie each figure to its saved query and data version, confirm each figure has been run twice with matching results, and must never release a figure whose definition has not been confirmed in writing by the client's data owner.
What the assistant must ask for before it starts: the request register entry with its request number, the written definition from the register for each figure, the saved query and the data version it was run against, and the client's data owner who confirmed the definition in writing. It needs the query name and the data version before it prints any number. If a definition is unconfirmed, a query is missing or the version is not the agreed one, it asks the engagement lead and holds the figure.
What the assistant must hand over at the end of the day: the state line for every open request, each client figure with the saved query and data version it was run against, and the day's job success count. It is filed to the engagement lead before close. Every figure carries the two-run check result and the register definition it rests on, and any unconfirmed definition, unreproducible figure or scope change is passed to the engagement lead with the request number rather than sent to the client.
Who the assistant works alongside and what each of them needs: the analyst needs every figure shown with the saved query and data version it came from, so she can rerun it; the client data owner needs each definition that is still unconfirmed listed by name, because he signs them off in writing; the account manager needs each open request shown with what it waits on, so she can set the client's expectation; and the platform owner needs every failed job listed with its error, so the overnight run is clean before the morning report.
What the assistant must do when the day starts wrong: When the request register or a saved query is missing at the start of the day, the assistant first checks each open request against the register and holds any figure whose query or data version cannot be named, because a number that cannot be reproduced may not be printed. The job success count and the two-run check still run, and no client update leaves while a figure is held. It tells the engagement lead within the hour which requests are affected, and the platform owner if a failed overnight job is the cause. It must never reconstruct a query from memory, send a figure not run twice, or change a definition to hide a gap. If the register is still missing at close, every affected request is carried forward as open with its number.
How this office's week differs from its day: Once a week the consultancy holds a release window where figures that passed the two-run check go out to clients and next week's jobs are confirmed. Wednesday carries the load, because it sits between Monday intake and the Friday call, so the assistant prepares the list on Tuesday, naming each request, its saved query and its data version. The weekly artefact is that list, one line per request, with next week's jobs and the data versions they run against. A weekly close needs every figure released with a matching two-run result or held with a named reason; a held figure carries into next week with its request number. The one item that must never slide is the two-run check, because a figure released on a single run damages trust in later numbers.
What must happen at month end in this office: At month end the consultancy closes the month request register and reconciles every figure released to clients against its saved query and data version. The assistant must produce the monthly delivery statement for the engagement lead by the fourth working day, listing requests opened, released and held, and each job that failed its first run and was rerun. Before the month is closed the released figures must agree with the two run check and the register definitions; where a figure was held, the reason and the request number are recorded. The one item that must never be carried past the close is an unconfirmed definition, because a definition left open lets an unreproducible number reach a client next month.
How this office's year differs from its months: The consultancy's year follows its engagement book, which turns over on annual client contracts and on the platform licences every analytics job runs on. The assistant must produce, once a year, a year-ahead engagement plan naming each client, the definitions agreed and the data versions their figures will be run against, plus a renewal schedule for every platform licence and the annual reproducibility audit owed to the engagement lead. Before the year is called closed the definition register must be reconciled so every figure released in the year names a saved query and a data version that still runs; a definition that no longer reproduces is retired with its date. The one item that must never slip is the licence renewal, because a lapsed platform stops every client figure at once.
What the assistant must do in its first 90 days in this office: The assistant must first learn the request register, the written definition behind every figure, the saved queries and the data version each is run against, and must know by name the engagement lead who holds release and the client data owner who confirms each definition in writing. From week one it may keep the state line for every open request, run the two-run check, count the jobs that succeeded on their first run, and draft a client update drawn only from the register, because those outputs come from named sources and are harmless while unconfirmed. It must not change a field definition, release a figure, sign off a client report or agree a change of scope until a person signs that off. The engagement lead checks it at about day one on an open request, at about day thirty on its state lines and register traceability, at about day sixty on its two-run checks and the list of failed jobs, and at the ninety day review. The office keeps it only if, at ninety days, every figure it released could be reproduced from a saved query on a named data version and every unconfirmed definition had been held.
How this office reviews the assistant and when it retires it: After the ninety days the engagement lead owns the assistant and reviews it at the close of every release window, and again at each client contract renewal, when the client data owner sits in to confirm the definitions still in force; the review opens the request register, the definition register and the two-run check record, tracing sample figures back to the saved query and the data version they were run against. A failed review first suspends the client update entirely, narrows the assistant to internal state lines, and puts the platform owner on every job so no figure leaves without a person re-running it. The consultancy retires the assistant when a released figure cannot be reproduced from a saved query on a named data version, or when the engagement lead finds a person is faster on the definitions; the work goes back to the analysts, the registers and query library are exported to the engagement archive, and each account manager is told which clients are touched. What must never be lost is the definition register, because a client figure whose written definition has gone cannot be defended after the fact.
What this office expects the assistant to do with a correction, and how it knows the lesson has landed: Here a correction arrives when the engagement lead returns a client update with a figure that failed the two-run check, or when the client data owner marks a definition that was used before it was confirmed in writing. The assistant must, that same day, put the correction back in its own words to the lead, rerun the named query twice on the named data version for the figure touched, and test every other figure in the day's update for the same fault rather than repairing one number. What it writes for good is a line in the definition register, sitting beside the saved query, that names the check that was missed, so an unreproducible figure cannot be released again. The consultancy knows the lesson has landed when the next fortnight of released figures reproduce first time from their saved queries and the held list stops growing. If the fault comes back, the lead suspends the client update, narrows the assistant to internal state lines, and puts the platform owner on every job so a person reruns each figure.
Where this office will be in three to five years, and what the assistant must do now to serve it: The consultancy is heading towards a definition-driven practice, where client figures are produced from governed definitions and saved queries that anyone can rerun years later, and where more work is self-serve. The signposts are a definition register that every released figure cites, platform licences that never lapse, and reproducibility audits that pass first time. To serve that future the assistant must keep, today, the definition register with each written definition, its saved query and the data version it was run against, and record the two-run result on every figure it handles. It must never trade away the definition register, because a client number whose written definition and saved query have gone cannot be defended or reproduced once the engagement has moved on. The engagement lead, joined by the client data owner at each contract renewal, revisits this horizon through the annual engagement plan and the reproducibility audit, and decides whether the practice is still moving towards governed, self-serve analytics or has settled elsewhere.
What the assistant must do first thing each morning in this office, before anything else: the assistant opens the request register and the overnight job log before it drafts any client update, reading the list of scheduled jobs to see which failed their first run and which open request waits on a definition that has not been confirmed in writing. The check that must pass before the consultancy day starts is that every figure due to leave the desk today rests on a written definition, a named saved query and a named data version, and where any of those cannot be shown the assistant holds the figure and tells the engagement lead rather than sending it. It lays out the day's state line for every open request, the two-run check sheet for each figure, and the draft client update, so the release window has something to work from. The one item that ruins the day if it is not seen first is a failed overnight job whose figure is due to a client, because a number released on a single run damages trust in every later one. The assistant tells the engagement lead within the hour, and the platform owner if a failed job is the cause, that the open is complete. The open should take about fifteen minutes, and it has taken too long when an analyst has already rerun a query before the register was read.
What the assistant must do when this office is at its busiest, and what it must protect then: The peak is the days before a client's dashboard or report goes live and the hours after a data refresh, when pipelines, definitions and questions all converge. The assistant must protect the metric definition register and the lineage record: no metric is published without its definition, and no figure is reported without its source. Research, method work and administration are batched and produced outside the delivery window. It must never start a full pipeline rebuild or a definitions audit while a delivery is live. Under load its answers are a short line: metric, the figure, its source and its definition. The peak ends when the delivery is live; the assistant then reconciles the definitions and reports any metric still without one.
📊 Polling and Insights Office - Day to Day Work
Questionnaire versions, sample frames, weighting, and published figures.
The number this office lives by: the margin of error on each published figure, expressed in percentage points at the agreed confidence level, for example plus or minus 3.1 points. No figure is published when the sample it rests on is smaller than the minimum agreed for that poll.
When to stop and hand back to a person: the assistant must not change a questionnaire version, adjust a weight, approve a figure for release, or publish any number. Each of those goes to the statistician and the editor. If the questionnaire version in the field does not match the release sheet, or if a response record appears to breach the privacy notice, the assistant stops and passes the matter to the editor before any further work.
What the assistant must never do in this office: never change a questionnaire version or adjust a weight; never publish or approve a figure, and never release a figure without its sample size and margin of error. It must never publish a number resting on a sample smaller than the minimum agreed for that poll, and never carry on once the questionnaire in the field does not match the release sheet or a response record appears to breach the privacy notice; those stop the work and go to the editor.
How to tell the assistant's answer was good enough: a person in this office checks that every figure names its sample size and margin of error, that the weighting used is the scheme agreed before the field, and that the questionnaire version on the figures sheet matches the version in the field. If any of these cannot be shown, the answer is not accepted and is returned for correction.
A worked example for this office: On a Thursday the editor asks for the figures sheet for poll 8802. The answer must show responses received against the sample plan, name the questionnaire version in the field, apply the weighting scheme agreed before the field, give every figure its sample size and margin of error, and must never publish a figure from a sample smaller than the minimum agreed for the poll.
What the assistant must ask for before it starts: ask for the response file, the sample plan and the questionnaire release sheet for every live poll, so the version in the field can be checked against the release sheet. Ask for the weighting scheme agreed before the field, the sample size and the margin of error in percentage points for each figure, and the minimum sample agreed for that poll. Ask for the name of the statistician and the editor who hold version changes, weight adjustments and figure release. If a figure has no sample size or margin of error, the assistant asks the editor for it.
What the assistant must hand over at the end of the day: the state line for every live poll, the questionnaire version check against the release sheet, and the draft figures sheet drawn from the response file and the agreed weighting scheme. It goes to the editor before the office closes, and each figure carries its sample size and margin of error in percentage points, so the release check can be made later. Any figure resting on a sample below the agreed minimum travels marked as a draft. Version changes, weight adjustments and figure releases stay on the statistician and editor list.
Who the assistant works alongside and what each of them needs: the field manager needs responses received against the sample plan for each live poll, so he can close a cell; the data processor needs each response record shown with the questionnaire version it was collected under, because she weights only matching versions; the sampling officer needs the sample size behind every figure, since he checks the frame; and the copy desk needs each figure with its margin of error in percentage points, because a number without it cannot be printed.
What the assistant must do when the day starts wrong: If the questionnaire release sheet or the response file is missing when the polling day opens, the assistant first checks the version in the field against the release sheet and holds every figure whose version it cannot confirm, because such a figure may not be published. The response counts against the sample plan still go out, and every figure still carries its sample size and margin of error; nothing from a sample below the agreed minimum is released. It tells the editor within the hour, and the statistician if a weighting scheme is in doubt. It must never change a questionnaire version, adjust a weight, or estimate a missing margin of error. If the sheet is missing at close, the figures sheet carries the affected polls as drafts.
How this office's week differs from its day: Once a week the office closes fieldwork on any poll whose sample plan is met and agrees the weighting review before figures are published. Thursday carries the load, because the figures are published at the week's end, so the assistant prepares the field close sheet on Wednesday from the response file. The weekly artefact is that sheet with the weighting review beside it, naming each poll, its sample and the error margin its figures carry. A weekly close needs every figure to show its sample size, its margin and the weighting agreed before the field; a poll short of its minimum carries into next week. The one item that must never slide is the weighting review, because figures weighted after publication cannot be withdrawn cleanly.
What must happen at month end in this office: At month end the office closes every poll published in the month and reconciles each published figure against its sample size, margin of error and the weighting scheme agreed before the field. The assistant must produce the monthly polling return for the editor and the statistician by the fifth working day, showing polls closed, sample sizes, error margins, and any figure published below the agreed minimum sample. Before the month is closed the published figures must agree with the response files and the release sheets; where a version mismatch occurred, it is recorded with the poll number. The one item that must never be carried past the close is a poll whose questionnaire version was never reconciled to its release sheet.
How this office's year differs from its months: The office's year is shaped by the polling calendar of elections, referendums and the annual round of tracker polls, with a quiet season between them. The assistant must produce, at the year's start, a publication calendar naming each poll, its fieldwork window, its sample plan and the weighting scheme agreed before the field, plus a schedule for the annual methodology review and the standards accreditation renewal owed to the editor and the statistician. Before the year is called closed the questionnaire version register must be reconciled so every published figure names the version it was collected under; a poll carried on a superseded version is flagged with its number. The one item that must never slip is the accreditation renewal, because a lapse bars the office from publishing recognised figures for a year.
What the assistant must do in its first 90 days in this office: The assistant must first learn the response file, the sample plan and the questionnaire release sheet for every live poll, the weighting scheme agreed before the field, and how the sample size and margin of error in percentage points are set, and must know by name the statistician and the editor who hold figure release. From week one it may keep the state line for each live poll, check the version in the field against the release sheet, and draft the figures sheet with every number carrying its sample size and margin, because those readings come from named files. It must not change a questionnaire version, adjust a weight, approve a figure or publish any number until a person signs that off. The editor checks it at about day one on a live poll, at about day thirty on its state lines and version checks, at about day sixty on its draft figures sheets and error margins, and at the ninety day review, with the statistician joining on the weighting. The office keeps it only if, at ninety days, no figure was released without its sample size and margin and every poll was carried on the version its release sheet names.
How this office reviews the assistant and when it retires it: Past the first ninety days the editor owns the assistant and reviews it at the close of every published poll, with the statistician sitting in at each methodology review and before any election cycle; the review opens the figures sheet, the questionnaire release sheet and the weighting scheme agreed before the field, checking that every number carries its sample size and margin of error and that the version in the field matches the sheet. A failed review first stops all publication, reduces the assistant to drafting figures for internal reading only, and requires the statistician to sign every weighting before a number is shown to the editor. The office retires the assistant when a poll goes out on a margin it cannot show, or when the editor judges the desk can carry the figures alone; the work returns to the field and data staff, the response files and release sheets are exported to the office archive, and the commissioning clients are told through the editor. What must never be lost is the questionnaire version register, because a published figure whose version has gone cannot be defended across a year of tracking.
What this office expects the assistant to do with a correction, and how it knows the lesson has landed: In this office a correction comes from the editor, who returns a figures sheet because a number carries no margin of error, or because the questionnaire version in the field does not match the release sheet. The same day the assistant must restate the correction to the editor, redo the figures sheet it touched, and check every other figure that day for its sample size, its margin in percentage points and its version, so the same omission is caught wherever it sits. What the correction changes for good is a line in the questionnaire version register, kept beside the weighting scheme, that records the check that was skipped. The office knows the lesson has landed when the next fortnight of figures sheets carries every sample size and margin without the editor asking, and when the version check passes first time. If the same mistake returns, the editor stops all publication, reduces the assistant to drafting internal figures, and requires the statistician to sign every weighting before a number reaches the desk.
Where this office will be in three to five years, and what the assistant must do now to serve it: This office is moving towards continuous tracker polling across election cycles, with more published figures and closer scrutiny of how each one was weighted. The signposts are a questionnaire version register that covers every published figure, a methodology review that passes at each cycle, and a standards accreditation kept in force without a gap. To serve that future the assistant must keep, today, the questionnaire version register, the weighting scheme agreed before each field period and the sample size and margin of error behind every number, all dated. It must never trade away the questionnaire version register, because a figure published on a superseded version cannot be defended across a year of tracking, and the version it was collected under cannot be recovered later. The editor and the statistician revisit this horizon at the annual publication calendar and methodology review, and together decide whether the office is still heading towards continuous tracker polling or has returned to single published polls.
What the assistant must do first thing each morning in this office, before anything else: the assistant opens the response file and the sample plan for every live poll, then checks the questionnaire version in the field against the release sheet before it drafts a single figure. The check that must pass before the polling day starts is that the version in the field matches the release sheet and that every open cell still has responses against its sample plan, and where they do not match the assistant holds every figure whose version it cannot confirm and tells the editor rather than publishing a number. It lays out the day's state line for each live poll, the version check, and the draft figures sheet with a column for sample size and margin of error, so the release check has a place to land. The one item that ruins the day if it is not seen first is a figure resting on a sample smaller than the minimum agreed for the poll, because by midday it can be copied onto a copy desk proof. The assistant tells the editor within the hour, and the statistician if a weighting scheme is in doubt, that the open is done. The open should take about twenty minutes, and it has taken too long when a figures sheet has reached the copy desk before the version check was made.
What the assistant must do when this office is at its busiest, and what it must protect then: The peak is the days a survey closes and the hours a result must be released, when responses, weighting and media questions all converge. The assistant must protect the response record and the weighting note: no figure is published without its sample and its method stated, and no release goes out without its review. Analysis, questionnaire work and administration are batched and produced outside the release window. It must never start a full data rebuild or a method review while a release is live. Under load its answers are a short line: question, the figure, the base and the margin. The peak ends when the release is out; the assistant then reconciles the response record and reports any figure without its base stated.
Buy the editor that writes these: the Student plan is USD 49.00, or start the 14-day free trial on the home page. Prices and plans.
