How to Measure Hiring Trends With Repeated Job Snapshots

Collection setup

Open the Indeed Jobs Scraper Actor and the Indeed Job Market Data Task. The current preset uses five queries, US source, last seven days, max items five hundred, max pages fifteen, sort by date, and save only unique items true. Start a pilot with one query and small limits. Max items is shared. Country, query, and batch are input evidence, not promised row fields. Retain jobId, jobUrl, title, companyName, locationRaw, postedAt, postedAtRaw, and scrapedAt. Missing IDs stay unresolved outside distinct ID counts. Within-run uniqueness does not dedupe observations across days. Inspect output saved rows, duplicates, filters, and stop evidence. The current control checked source shape, not a multi-day trend benchmark. Preserve failed day unknown and capped day limited states.

The task uses saved input values `country` set to `us`, `searchTerm` containing five queries, `postedWithinDays` set to `7`, `maxItems` set to `500`, `maxPagesPerSearch` set to `15`, `sortBy` set to `date`, and `saveOnlyUniqueItems` set to `true`. A pilot replaces `searchTerm` with one query and sets small row and page limits. The `maxItems` limit is shared across keywords. The current two-row one-batch control checks source shape, not a live multi-day trend.

The Collection Panel Method

1. Define a collection panel. Use the same role queries, source country, site, locations, filters, sort policy, row or page caps, and intended check times. If you change your question, start a new comparison panel or label the scope change. Do not silently extend the old series.
2. Save a run manifest. Record the batch ID, intended time, actual time, input values, run outcome, rows saved, stop state, and warnings. Missing or failed collection days stay missing, not zero.
3. Keep observations keyed by batch, source, and job ID. Within each accepted batch, count each source listing once. Across batches, preserve each observation and update a first seen listing table. Do not count same title jobs as identical automatically.
4. Report raw returned rows, distinct listings observed in that batch, IDs first seen in the panel, IDs seen before, and source or collection limits. First seen today means first collected by your process, not necessarily first posted by the employer.
5. Compare only scopes and collection states that support your question. A capped sample can show new IDs to review, but it cannot establish overall inventory growth or decline. Failed days cannot be used as low market days.
6. Show the series with missing or limited flags next to the numbers. Keep an explanatory table that makes the denominator and unit clear. Do not join a line across an unknown day as if you measured it.
7. State the narrow result. Explain what this fixed observed panel found and which comparisons remain unknown. If you need a broader hiring claim, collect a wider sample or use an official labor source.
8. Deliver the daily panel table, first seen ledger, and coverage notes. This result supports further source checks and consistent sampling. It does not measure completed hires or employer intent.

Fictional Example and Verified Counts

All jobs, times, and outcomes in this section are invented. A fixed panel uses one analyst query, one source country, the same filter policy, and intended daily checks at 12:00 UTC.

• Day 1 accepted fixture set has four distinct listings: J1, J2, J3, and J4. First seen is four. Seen before is zero.
• Day 2 accepted set has four distinct listings: J2, J3, J4, and J5. First seen is one, which is J5. Seen before is three. An identical daily total does not mean no new source ID.
• Day 3 collection failed. There is no accepted set. The observed count and first seen count are unknown, not zero. Keep the last accepted ledger from Day 2.
• Day 4 accepted set has four distinct listings: J3, J4, J5, and J6. First seen is one, which is J6. Seen before is three. J6 is first observed on Day 4, but it could have appeared during the missing Day 3. Do not assign its creation to Day 4.
• Day 5 cap limited set has two distinct saved items: J4 and J7, because a cap of two was reached. First seen is one, which is J7. Seen before is one. Its lower count is caused by a changed active cap, so exclude it from an unqualified decline claim. Capture the scope change or use limited beside it.

The union across accepted observations contains seven distinct source IDs, which are J1 through J7. Total observed listing appearances equal four plus four plus four plus two, which is fourteen. Repeat appearances beyond each key's first observation equal fourteen minus seven, which is seven. Those are observations, not seven reposted vacancies. New to ledger counts equal four plus one plus one plus one, which is seven. Day 3 is excluded from arithmetic because it is unknown, not a zero measurement. Do not sum daily distinct counts of four, four, four, and two and call it fourteen unique opportunities.

The current runtime check is one batch, not a live multi day trend benchmark. The toy accepted Day 1, Day 2, and Day 4 sets are only complete for this fictional panel, not a real platform guarantee.