The Target Interview Loop, Rebuilt From Conflicting Public Reports

You have picked one target company, one role, one level. The interview section of every ranking page tells you to "read the reviews" and stops there. This guide does the part they skip: it takes a pile of scattered, contradictory public interview reports and reconciles them into one level-matched, recency-weighted expected loop you can schedule prep around. It is written for a job seeker who has a specific target and needs to know the stage sequence, round count, and end-to-end timeline before the recruiter calls.

I carry one candidate's messy raw reports all the way through, including the two wrong turns a naive read produces. Follow along on your own case as you read.

Why the company average is the wrong number

A company-level average is nearly useless for planning because the spread inside one company dwarfs the spread between companies. Your job is to find your row, not the headline.

Look at what the aggregate pages actually contain once you open them:

That is one employer. Glassdoor's own company page averages 24 days across 1,301 interviews, but Junior Designer averaged 1 day and Senior Product Manager averaged 120 days. The firm "FIRST" averages 31.22 days across 23 submitted interviews, yet Project Manager runs about 7 days and Event & Venue Specialist about 77. Meanwhile national medians barely move: the HRBench 2025 benchmark puts the national median time to hire at 24 days, with the 25th percentile at 17 and the 75th at 36.

So the within-company range runs 1 to 360 days while the cross-company median moves by roughly a week. The level filter matters more than the company choice. Every step below exists to get you off the headline average and onto your own row.

The candidate case: raw reports, as they arrive

Here is the working example I will carry through. The candidate is targeting a senior software engineering role at a large tech employer. They pull the public reports and get a mess. Below is the raw pile before any filtering, exactly the kind of contradiction that makes people give up and just "wing it."

Five sources, three different round counts, one all-titles average masquerading as a process, and one dramatic 12-round rejection story. A naive read produces one of two wrong turns from this table, and I will show both before I reconcile it.

The reports carry more structure than the table shows. Glassdoor's interview page exposes per-report fields you should capture for every row: job title, experience, difficulty, outcome, length in days, the full process description, the questions, and both interview and posting dates. Grab all of them. The dates are what make recency-weighting possible, and the outcome field tells you whether you are reading a hire or a rejection.

The two wrong turns a naive read produces

There are two failure modes so common they deserve to be walked before the method. Both come from reading the pile without filtering it first.

Wrong turn one: reading the all-titles average as your number

The candidate's eye lands on "24 days across all job titles" and writes "about three and a half weeks" into their planner. That number is a blend of a 1-day Junior Designer loop and a 120-day Senior Product Manager loop. It describes no real candidate and certainly not a senior engineer. The all-titles average is a statistical artifact of mixing roles, and it is the single most repeated mistake in reading these pages.

The correction is mechanical: confirm the figure is filtered to your title before you use it. If the page only offers an all-titles number, that page has not given you your number yet.

Wrong turn two: taking the max round count as expected

The candidate then sees the 12-round Reddit story and the 7-round Glassdoor report and concludes the loop is "brutal, at least seven rounds, budget two months." This is taking the maximum as the expectation. The 12-round report is a single, level-unclear, two-year-old rejection. The dossier records exactly this kind of extreme: one Apple candidate reported going through 12 interview rounds before being told they were not a fit. That is an anecdote, not a loop.

The correction: ask whether the count is repeated by three or more recent same-level reports. If not, drop it.

Both wrong turns share a root cause. The candidate treated the pile as a single population and reached for a summary statistic - an average or a maximum - before filtering to level and recency. The rest of this guide is the disciplined version.

The procedure: raw reports to a dated expected loop

This is the seven-step reconciliation. It converts the messy pile into a single dated profile. Two published methods disagree on order - one filters by role first, one weights recency first - but they converge before reconciliation, so either sequence works as long as both filters run before you resolve the loop.

Applied to the candidate's pile, the steps do this. Scoping fixes the key to "Senior Software Engineer." The all-titles Glassdoor row falls out at the filter step because it is not level-matched. The 3-year-old 7-round Glassdoor report and the 2-year-old 12-round Reddit report fall out at the recency step. What survives is the 4-round Blind report from four months ago, the recent Levels.fyi entry confirming the level and base-versus-total-comp split, and any other sub-year same-level reports the candidate can find. From five contradictory rows the candidate is left with a small, coherent, current subset. That subset is what you reconcile.

Reconciling round counts: the counting-convention trap

When two same-level, recent reports disagree on round count, assume a counting-convention gap before you assume the process changed. This is the single most useful judgement in the whole method.

The candidate's surviving Blind report says four rounds. An older same-company report said seven. The gap is almost never a real change in the loop. It is a disagreement about what counts as a round. A same-day onsite panel is frequently four to six separate interviews in one day - the dossier describes exactly this pattern, a panel that is four to six interviews in a day. One reporter logs that panel as a single round; another logs it as six. Add or subtract the recruiter phone screen depending on whether the reporter counts it, and a "four-round" loop and a "seven-round" loop turn out to be the same loop described two ways.

So do not reconcile the numbers. Reconcile the stage sequences. Line up the recent same-level reports as ordered stages and take the mode - the sequence that appears most often - rather than the maximum or the average. The stage list is stable and plannable; the raw count is a labeling artifact.

Stage                    | Report A | Report B | Report C
Recruiter phone screen   |   yes    |   yes    |   yes
Technical phone screen   |   yes    |   no     |   yes
Take-home / OA           |   no     |   yes    |   no
Onsite panel (count as 1)|   4 int  |   5 int  |   4 int
Hiring manager / behavioral| yes    |   yes    |   yes
Team match / bar raiser  |   no     |   yes    |   no
--> Modal sequence: screen, tech screen, onsite panel, HM behavioral
--> Modal round count: 4 (panel = 1); range 4-6 if panel counted per-interview

The candidate's reconciled loop from this worksheet: recruiter screen, technical phone screen, an onsite panel of four to five interviews, and a hiring manager behavioral. Four rounds by the panel-as-one convention, and the seven-round report dissolves into the same shape.

Building the timeline and the seniority sanity check

Once you have the modal stage sequence, sum the stage gaps from your recent subset into an end-to-end range. Then sanity-check that range against seniority benchmarks so a single fast or slow report does not skew it.

The candidate's recent Blind report logged about five weeks end to end. That is one data point. Check it against the benchmark bands rather than trusting it alone:

A senior engineer is a professional IC, so 21 to 35 days is the benchmark window. The candidate's five weeks - about 35 days - sits at the top of that band, which is consistent and plausible. Had the surviving report claimed 90 days, the benchmark would flag it for a second look: either a reorg-slowed hiring freeze, a report that is actually a manager-level loop, or a stale entry that slipped the recency filter.

Cross-check the round count the same way against context bands:

The candidate's target is a large tech employer, so the enterprise band of four to six rounds is the expectation. The reconciled four-round loop sits at the low end, which is fine - the modal sequence beats the band when you have same-level reports. The band is the fallback when your subset is too thin to trust, and a warning that five or more rounds is where candidate drop-off spikes, so plan energy accordingly.

Where this goes wrong: failure modes and false positives

The method fails in predictable ways, and each failure has a specific check. This section is the one to reread before you commit a plan.

• Reading the all-titles average as your number. The "24 days across all titles" headline hides a 1-day role and a 120-day role in the same company. Check: confirm the figure is filtered to your title before using it.
• Taking the max round count as expected. A single 12-round report is an outlier, not the loop. Check: is it repeated by three or more recent same-level reports? If not, drop it.
• Trusting a thin sample. Below the roughly 200 to 300 review comfort zone users cite for a general read, one voice dominates. The false positive is a "brutal loop" verdict built from two angry posts. Check: count same-level reports before concluding.
• Stale reports read as current. Comp and process reports are often one to two years old after reorgs. Check the posting date on each report and discount pre-reorg ones.
• Selection bias mistaken for reality. Extreme, especially negative, experiences get published more often, so the median experience is under-represented. Check: is the sentiment bimodal? Weight repeated middle-of-the-road reports up.
• Censorship inflates the picture. Balanced or negative interview reviews are alleged to be removed when companies complain, which can make a loop look artificially smooth. Check: cross-read Blind, where removal is less alleged.
• Cross-source unit mismatch. Levels.fyi total comp includes stock; Glassdoor often shows base only, and the same mismatch hits round counts when phone screens are counted or not. Check: normalize what a round and a number mean per source.
• Geo contamination. A warehouse or wrong-city review tells you little about your team. Check: filter to your office or region.

Two of these push in opposite directions, which is exactly why single-source reads fail. Selection bias makes Glassdoor look too harsh because angry outliers over-publish. Censorship makes it look too smooth because negatives are removed. A single source cannot resolve a distortion that runs both ways at once. Cross-sourcing against Blind is what breaks the tie.

When the reports run out: fill the gap with first-hand contacts

When your same-level recent sample is too thin to reconcile, the fix is not to lower your standard for evidence. It is to get evidence directly from people who just ran the loop. Public report density is a geography artifact, and thin data is a reason to reach out, not a signal the loop is easy.

That ratio is the trap in one number. A UK target with almost no public interview reports has not got a simpler loop - it has fewer people writing reports because the population is smaller. Absence of reports is missing evidence, never a smooth process. In that situation the highest-value move is to find recent hires at that exact company and level and ask them directly what the loop looked like. They survived the current version of it, which is the one you care about.

This is where Refolk removes the friction the paragraph just described. Instead of scraping thin public pages, you describe the exact people whose first-hand account would reconcile your loop, and Refolk finds them from public professional records.

Ask them three questions: how many distinct days you were interviewed, whether the onsite was one panel or several separate rounds, and how many calendar days ran from recruiter call to offer. Those three answers reconstruct the stage sequence, resolve the counting-convention gap, and date the timeline - the three things the public reports were fighting over. Recent hires and technical recruiters at the target are the two contact types worth prioritizing, because they either ran the loop or ran candidates through it.

Keep the loop profile current

An interview loop profile has a shelf life, because both the platforms and the people who wrote the reports already discount old data. Treat your profile as dated the moment you build it and re-check it before you commit.

If nearly half of seekers and the platform itself ignore reports older than six months, an un-weighted read systematically over-represents a process that no longer exists. Re-run the recency filter each time you reopen the profile. A report that was "recent" when you first pulled it ages into noise. Anything predating a reorg or CEO change is noise on arrival, no matter how detailed.

Run this checklist before you call the loop profile done and lock your prep schedule against it.

The output you carry away is small and specific: a stage sequence, a round count under a stated counting convention, a calendar range, and the date you built it. That is a planning-ready profile. When it ages past six months, or when you hear the company reorganized, throw it out and rebuild it - the method takes under two hours, and a fresh dated profile beats a detailed stale one every time.