# The Candidate Back-Channel Standard, Graded Before You Sign

*You will grade your own back-channel vet as complete, insufficient, or blocked across five decision dimensions before you return the signed letter.*

- Canonical URL: https://www.refolk.ai/candidates/guides/candidate-back-channel-standard
- Pillar: Offers and negotiation
- Format: Standard
- Published: 2026-10-01
- Last reviewed: 2026-10-01
- Reading time: 17 min
- Keywords: how to back-channel a company before accepting offer, vet employer before signing offer, talk to current employees before accepting job, due diligence before accepting job offer, reverse reference check on company, red flags before signing offer letter

## Key takeaways

- A sign-ready vet needs two non-overlapping, recent sources agreeing on each of five decision dimensions: financial runway, role churn, team stability, reorg risk, and manager quality.
- Submitted references prove nothing about the vet, because the candidate hands over favorite managers who speak highly; corroboration requires people off that list.
- At LinkedIn's 10.3% reply rate inside the roughly 100-request weekly cap, a candidate lands about 10 conversations a week, so 'two non-overlapping sources,' not 'call everyone,' defines done.
- Glassdoor and Blind are not interchangeable: 37.7% of Glassdoor reviews give five stars versus 22.0% on Blind, so reading one site alone biases the grade.
- In Refolk's index, the US software engineer pool holds 352,528 matching profiles against 43,765 in the UK, and UK talent is roughly 60% London-clustered, which makes genuinely independent sources harder to find.
- Deadline pressure is itself a signal: urgency that forces an answer within hours justifies requesting more time, not lowering the bar.

You have an offer, or you expect one, and you want to vet the company the way they vetted you: through off-the-record calls with people who have actually worked there. This guide is for candidates who need to know when that vetting is thorough enough to sign, and when it is not. It gives you a definition of done - the sources, the decision dimensions, and the corroboration rules that separate a sign-ready vet from wishful confirmation - so you can grade what you already have and decide.

Every other resource on this job is either a listicle of questions to ask before accepting, or an article about how employers back-channel you. Neither gives you a gradeable bar for your own vetting. This one does. Read it, grade your vet as complete, insufficient, or blocked, and know whether to return the signed letter.

## What a sign-ready vet actually certifies

A sign-ready vet certifies that every load-bearing decision dimension has at least two non-overlapping, recent sources that agree, and that no blocking flag remains uncorroborated at the deadline. That is the whole standard. Everything below is how to meet it and how to tell when you have not.

The reason the bar is dimensional rather than a headcount is that counting calls measures effort, not coverage. Three calls with three former colleagues who all sat on the same team tell you one thing from one vantage point. Two calls from genuinely different relationships - a departed peer and a cross-functional contact from another org - tell you whether a claim holds up when the social graph changes. The standard grades coverage of what matters, not volume.

> **Rule:** The bar, stated once
>
> A vet is complete when each of the five decision dimensions has two non-overlapping, recent sources in agreement, and no blocking flag is left uncorroborated. Miss either condition and the vet is insufficient or blocked, never complete.

A term of art, defined up front: a back-channel is an off-the-record conversation with someone who worked at the company but whom the company did not hand you. That last part is the point. When an employer gives you references, or when you hand an employer yours, the list is self-selected. As one practitioner puts it, candidates hand over favorite managers who will speak highly of them. Back-channeling reaches people who are not on anyone's favorites list, which is why a clean submitted reference proves nothing about the health of the company you are about to join.

## The five decision dimensions and the signal that proves each

A sign-ready vet covers five dimensions, each with a falsifiable question and a signal that either confirms or lies. Practitioner write-ups converge on these five; if your notes do not touch all of them, your vet is incomplete by definition.

| Dimension | Falsifiable question | Signal that proves it | What it looks like when it lies |
|---|---|---|---|
| Financial runway | How many months survive if revenue flatlines? | A specific figure: cash divided by monthly burn, plus last raise and milestones | "We're in a good position" with no ballpark |
| Role churn | How many people held this role recently and why did they leave? | Named predecessors with explained departures | Vague "it's grown a lot since then" |
| Team stability | How long do people in the target department actually stay? | LinkedIn tenure read inside that specific team | A company-wide average hiding a churning team |
| Reorg or layoff risk | Any layoffs or exec departures in the last six months? | WARN filings, dated exec exits, back-channel accounts | "Nothing I'm aware of" from a stale source |
| Manager quality | Does this manager coach, communicate, and retain people? | Episode-level stories from direct reports | A verdict with no example behind it |

Runway is the one candidates skip because it feels invasive. It should not be. Before accepting any startup offer you need to understand runway, calculated as cash divided by monthly burn, which tells you how long the company survives if revenue flatlines. Ask directly for runway, last raise, investors, and milestones. A vague non-answer or a refusal to give a ballpark is itself the flag.

Role churn is the strongest single warning. If several people have cycled through the same position in a short period, treat that as one of the strongest warning signs, because it often points to poor management or unrealistic expectations. Team stability is adjacent but distinct: check LinkedIn to see how long employees typically stay in the specific department you are considering, since tenure varies by team and a healthy company-wide average can hide one broken org.

Manager quality is the dimension that most needs real calls. The purpose is to identify patterns: how the manager communicates, whether people stay or leave quickly, how much trust exists, and whether employees get coaching. You cannot get this from a review site. You get it by asking someone who reported to that manager to walk you through a specific situation.

### Reorg risk has a legal paper trail

Reorg and layoff risk is the one dimension with a documentable public signal. The WARN Act requires 60 calendar days of advance written notice of a plant closing or mass layoff within defined circumstances. Federal WARN covers employers with 100 or more employees. A mass layoff triggers WARN at 500 or more employment losses, or 50 to 499 if they are at least 33% of the active workforce, in a 30-day period. State versions set lower bars: California applies at 75-plus employees with a 50-layoff threshold and no percentage test, and New York requires 25 at a site. A recent filing is hard evidence; the absence of one is not proof of stability, only absence of the largest category of event.

## Read the public baseline, then distrust it

The public baseline - Glassdoor, Blind, LinkedIn tenure - generates hypotheses to test on calls. It never supplies conclusions, because review sites get less honest exactly when you need them most. Treat this stage as building a list of questions, not a verdict.

Start with distribution, not average. A 3.6 built on a recent cliff-drop with polarized extremes is a completely different company from a stable 3.6. Filter to the last 90 to 180 days and look for clusters. Over forty percent of Glassdoor reviewers omit job title and tenure out of fear of being identified, so you often cannot even tell which team a review describes - another reason the public layer is a starting point only.

Then remember that platforms are not interchangeable. The same firm looks materially better on an unverified site.

**37.7% - Share of Glassdoor reviews awarding five stars, versus 22.0% on Blind**

Reading one site alone systematically biases your grade of the same company.

The median Glassdoor review in one study sample was 3.5, and as competition for workers increased, positive reviews became less reflective of conditions, climbing without matching improvements. So in a hot hiring market the public baseline inflates, and your live calls have to carry more of the weight. Eighty-six percent of job seekers research company reviews before applying, which means you are reading the same inflated signal everyone else is. The back-channel is where you get past it.

#### Layers of a company vet, outermost first

1. **Public review sites** - Distribution and recency, read as hypotheses not conclusions
2. **LinkedIn tenure data** - Who stays and who left, inside the target department
3. **Back-channel calls** - Off-list people, episodes not verdicts, corroborating or contradicting the baseline
4. **Direct-report testimony** - The one layer that proves manager quality

*Each inner layer corrects the bias of the one outside it, ending in the direct-report call that no review site can replace.*

## Who to call, and why the channel caps your vet

Call former employees who left within about the last year first, because they carry no confidentiality risk and speak more freely after departing. Then add peers and cross-functional contacts one to two degrees out. The documented method is plain: use LinkedIn to find employees who left the target company within the last year and reach out to ask about their experience. It is less intimidating for them to share after departing, and you will not get many replies, but it is worth it if you want to be thorough.

You can also run a reverse reference check by finding someone in your network who knows somebody at the company. The thing to avoid is messaging a chatty current teammate who could alert the employer. Prefer departed contacts; keep current-team contact minimal.

The reason the standard is "two non-overlapping sources" and not "call everyone" is that the channel itself caps you. The math is unforgiving.

#### From outreach to usable calls in one week

| Stage | Figure | Note |
| --- | --- | --- |
| Connection requests allowed per week | 100 | LinkedIn soft cap per account |
| Replies at 10.3% average | ~10 | Cold DM reply benchmark |
| Calls actually booked | 2-3 | After scheduling and no-shows |

*LinkedIn's weekly cap and a 10.3% reply rate mean a thorough candidate lands two or three real calls, not a dozen.*

LinkedIn enforces a soft cap of about 100 connection requests per account per week, and even at a 10.3% reply rate that is roughly 10 conversations per week. InMail runs higher, averaging 18 to 25% response versus about 3% for cold email, but the weekly ceiling still binds. The practical consequence: you will land two or three real calls, so the vet has to be defined by coverage of dimensions, not by how many people you reached.

Market depth changes how hard independence is to achieve. In Refolk's index of professional profiles, the pool you are drawing from varies enormously by geography.

| Market | Matching profiles | Pool vs UK |
|---|---|---|
| Software Engineer, United States | 352,528 | 8.1x |
| Software Engineer, United Kingdom | 43,765 | 1.0x |

And the pool concentrates geographically, which matters for getting genuinely independent sources.

| Market | Top hub | Hub share of sample |
|---|---|---|
| SWE, United States | SF Bay Area | ~33% |
| SWE, United Kingdom | London | ~60% |

UK software talent is roughly 60% London-clustered against about 33% in the Bay Area sample, so UK sources are more likely to share employers and social graphs. That raises the risk that your "two sources" are really one correlated narrative wearing two faces. The share figures come from samples of around 24 to 25 profiles and are indicative, not population-level, but the direction is clear: in a concentrated market, work harder to confirm your two sources never overlapped.

Finding off-list people quickly is exactly the friction [Refolk](/candidates) removes - it searches its index for former employees and second-degree contacts by company, team, and dates, so you are not hand-querying LinkedIn one filter at a time against a weekly cap.

Ask me this: `Software engineers who left Google in the last 12 months and are now at other companies.` - [run the search](https://www.refolk.ai/start?q=Software%20engineers%20who%20left%20Google%20in%20the%20last%2012%20months%20and%20are%20now%20at%20other%20companies.).

*Returns departed employees who carry no confidentiality risk and can speak freely about a recent experience.*

## The grading procedure

This is the seven-step procedure that produces a graded vet. Run it in order. The output is one label: complete, insufficient, or blocked.

#### Grade your vet end to end

1. **Scope the five dimensions** - List financial runway, role churn, team stability, reorg risk, and manager quality, and write one falsifiable question each must answer. Done when every dimension has a question a source answer could confirm or disprove.
2. **Pull the public baseline** - Read Glassdoor and Blind by distribution, not average, filtered to the last 90 to 180 days, and check LinkedIn tenure inside the target department. Done when you hold hypotheses to test, not conclusions.
3. **Build the source list** - Identify former employees who left within about twelve months plus peers and cross-functional contacts one to two degrees out. Done when you have five to eight named, reachable people across relationship types.
4. **Reach out off the record** - Prefer departed contacts and send a short, specific, low-pressure message, keeping current-team contact minimal. Done when at least two calls are booked.
5. **Run the calls** - Ask facts first, then episodes rather than verdicts, and close by asking who else to speak to. Done when notes are logged in one consistent shape per dimension across 15 to 40 minute calls.
6. **Corroborate each dimension** - Require two non-overlapping sources per dimension and flag any single-source claim. Done when each dimension is marked confirmed, single-source, or contradicted.
7. **Grade and decide** - Apply the blocking and yellow-flag rules; if a blocking flag is uncorroborated, resolve it or request more time. Done when the vet carries one label.

On call length, do not run a ten-minute chat and call it a vet. Retrieving a concrete example takes time, so a short call only yields summaries, and summaries are the thing you were trying to get past.

| Call type | Documented length | Source |
|---|---|---|
| Standard reference | 15-20 min | pin.com |
| Short scripted reference | 10-12 min | facecruit.com |
| Structured with backchannel ask | 25 min | join.com |
| Deep, example-driven | 40 min | secondtalent.com |

For the dimensions that need episodes - manager quality and team stability especially - budget toward the 25-to-40-minute end. Be comfortable with silence; the pause before someone decides whether to tell you the real story is where the real story lives.

**Off-the-record outreach message**

```
Hi [name] - I saw you worked on [team] at [company] until recently. I have an offer from them for a [role] and I'm trying to get an honest read before I decide. Would you be open to a 20-minute off-the-record call this week? I'm mainly trying to understand what the [team] is like day to day and how the manager runs things. No worries at all if not.
```

*Adapt the role and company; keep it short and name a specific thing you want to learn.*

**Per-dimension call-notes shape**

```
RUNWAY: [figure or "would not say"] / source relationship / recency
ROLE CHURN: [named predecessors + reasons] / source / recency
TEAM STABILITY: [tenure pattern described] / source / recency
REORG RISK: [layoffs/exec exits mentioned] / source / recency
MANAGER: [specific episode, not verdict] / source / recency
ELSE: [who else they said to call]
```

*Use the same five rows for every call so corroboration is a lookup, not a reconstruction.*

## How this goes wrong: the false positives

Most bad vets fail the same handful of ways, and every one of them is a case where a signal looks like confirmation but is actually noise. Learn these and you can catch yourself grading wishful confirmation as complete.

- **Glassdoor average mistaken for truth.** A 3.6 looks fine. It is not a finding. Read the distribution and the last-90-day cluster, because a 3.6 built on a recent cliff-drop with polarized extremes is a different company from a stable 3.6.
- **A single disgruntled source read as a pattern.** One vivid horror story is one data point. Require a second non-overlapping source before you log it as a dimension finding. One one-star review can be real, or an outlier, a grudge, or a coordinated push.
- **Stale sources.** A glowing account from someone who left four years ago describes a company that no longer exists. Enforce a three-year relevance ceiling and prefer accounts from the last 90 days. Back-channels over three years old are irrelevant.
- **Tipping off the employer.** Messaging a chatty current teammate can get back to the hiring manager. Prefer departed contacts. The weekly outreach cap also keeps you from casting so wide it looks like a campaign.
- **Runway hand-wave accepted.** "We're in a good position" is not an answer. Ask directly for runway, last raise, investors, and milestones; the vague non-answer is the flag.
- **Deadline pressure collapsing the bar.** Signing because they need an answer tonight is not a vet. Treat urgency as signal and request time instead of skipping corroboration.
- **Verification platform confusion.** Comparing a Blind score to a Glassdoor score as equals ignores their measurably different positivity profiles. They are not the same ruler.

> **Watch out:** The most common false positive
>
> A clean submitted reference and a friendly recruiter are not a vet. They measure the company's ability to present well, not its health. If your notes contain no off-list voice, your vet is insufficient no matter how good it feels.

The dividing line through almost all of these is pattern versus instance. Identify patterns across sources and across time; do not chase one rumor. Three of seven employees churning in the last year with no clear explanation is a pattern - something is broken in culture, leadership, or the business. One angry review is an instance. Grade accordingly.

> A vet feels done the moment it tells you what you hoped. That is exactly when to add a second source.

## The blocking and yellow-flag rules

Blocking flags must be resolved or corroborated before you sign. Yellow flags get logged but do not disqualify. Keeping these separate is what lets two people grade the same case the same way.

Blocking flags, each of which halts a sign until corroborated or resolved:

- **Unexplained role churn.** Several people cycling through the same position in a year with no clear explanation.
- **Opaque finances.** An inability or refusal to state runway, investors, and next-raise milestones.
- **Recent instability.** Layoffs, especially more than once, or executive-team departures in the last six months.

Yellow flags, logged and weighed but not disqualifying on their own: a single emotional review, a lone departure with a clean explanation, a mixed but non-polarized Glassdoor distribution. These become blocking only when a second non-overlapping source turns them into a pattern.

Deadline pressure deserves its own treatment because it is both a constraint and a signal. Legitimate employers understand that accepting a job is a major decision, so pressure to respond within hours often points to high turnover or a desperate hiring push. That means a tight deadline does not lower the bar; it justifies requesting more time. If a blocking flag is still uncorroborated when the clock runs out, the vet is blocked, not complete, and the honest grade is to say so.

> **Tip:** Turn the deadline into a test
>
> Asking for 48 more hours to decide is itself a small back-channel experiment. How the company responds to a reasonable request tells you something about the environment you are considering joining.

## The sign-ready checklist

Run this before you return the signed letter. If any item fails, your vet is not complete.

#### Before you sign

- [ ] Each of the five dimensions has a written, falsifiable question.
- [ ] The public baseline was read by distribution and recency, not by average, across more than one platform.
- [ ] You spoke with at least two non-overlapping sources, favoring people who left within the last year.
- [ ] No single-source claim is being treated as a confirmed finding.
- [ ] Every source is recent - within three years, preferably within 90 days.
- [ ] You asked directly for runway, last raise, investors, and milestones, and got a real answer.
- [ ] You checked for layoffs or executive departures in the last six months, including any WARN signal.
- [ ] No blocking flag remains uncorroborated.
- [ ] Any deadline pressure was treated as signal, and you requested more time rather than skipping corroboration.
- [ ] The vet carries exactly one label: complete, insufficient, or blocked.

## What to do with each grade

A complete vet means sign with your eyes open; you have corroborated coverage and no live blocking flag. An insufficient vet means you are missing coverage on one or more dimensions, so either land one more call or, if that is impossible before the deadline, request time and name the specific gap to yourself. A blocked vet means a blocking flag you could not corroborate or resolve stands between you and the signature.

The honest move on a blocked vet is not to sign and hope. It is to raise the specific concern with the company and watch how they handle it, or to walk. A standard that bends under deadline pressure is not a standard. The whole value of grading the vet is that the grade, not the adrenaline of a pending offer, makes the call.

Keep this current by re-checking the two things that drift. First, the public baseline inflates in hot labor markets, so weight your live calls more heavily when hiring is frenzied and discount cheerful reviews. Second, the sourcing math moves with platform limits and reply rates; if outreach stops working, that is the channel capping your vet, not a reason to lower the bar to the sources you happened to reach.

## Frequently asked questions

### How many people do I need to talk to before accepting an offer?

No source publishes a candidate-side minimum count, so the defensible bar is dimensional, not numerical: each load-bearing dimension needs at least two non-overlapping, recent sources that agree. In practice that means landing two or three genuine calls, because at LinkedIn's 10.3% reply rate inside the roughly 100-request weekly cap you realistically get about ten conversations a week. Depth per dimension beats raw headcount.

### Will the company find out I am back-channeling them?

They can, if you message chatty current teammates. The documented method is to prefer employees who left within the last year, since they carry no confidentiality risk to themselves or to you and share more freely after departing. Keep current-team contact minimal. LinkedIn's weekly outreach cap of about 100 requests also limits how wide you can cast before activity starts to look like a campaign.

### What counts as a blocking red flag versus a yellow flag?

Blocking flags must be resolved or corroborated before signing: several people cycling through the same role in a year with no explanation, an inability to state runway, investors, and next-raise milestones, and recent layoffs or executive departures in the last six months. Yellow flags get logged, not disqualified: a single emotional one-star or five-star review that could be an outlier, a grudge, or a coordinated push. The dividing rule is pattern versus instance.

### Can I trust Glassdoor to vet an employer?

Only as a baseline to generate hypotheses, never as a conclusion. In published study samples, 37.7% of Glassdoor reviews give five stars against 22.0% on Blind, so the same firm looks materially better on the unverified site. Positive reviews also climb in hot labor markets without matching improvements in conditions. Read the distribution and the last-90-day cluster, then test it on live calls.

### The employer wants an answer tonight. Does that change the standard?

Treat the urgency as a data point rather than a reason to skip corroboration. Legitimate employers understand that accepting a job is a major decision, so pressure to respond within hours often points to high turnover or a desperate hiring push. That justifies requesting more time, which both protects the bar and tests how the company handles a reasonable ask. If a blocking flag is still uncorroborated at the deadline, the vet is blocked, not complete.

### How long should a back-channel call be?

Longer than people expect. Standard reference calls run 15 to 20 minutes, but retrieving a concrete episode takes time, and a short call only yields summaries. Budget 25 to 40 minutes for calls where you need episode-level detail on manager behavior or team stability. Back-channel calls can run shorter than full references if you keep them specific and targeted, but do not drop below the length that lets someone tell you an actual story.

---

*From the Refolk guide library. I revise these guides rather than replacing them, so the current version is always at https://www.refolk.ai/candidates/guides/candidate-back-channel-standard*