# The Source-Company Raidability Read: Raid Now, Watch, or Deprioritize

*You will score any source company on vulnerability and extractability from public signals and land a raid-now, watch, or deprioritize verdict with a dated re-check.*

- Canonical URL: https://www.refolk.ai/guides/source-company-raidability-read
- Pillar: Recruiting and sourcing
- Format: Framework
- Published: 2026-10-11
- Last reviewed: 2026-10-11
- Reading time: 17 min
- Keywords: which companies to source from, best companies to poach talent from, signs a company's employees will leave, when to target a competitor for hires, talent raidability signals

## Key takeaways

- An RTO mandate is the cleanest datable distress trigger: one study of 3 million LinkedIn profiles found a 14% spike in departures immediately after a return mandate, concentrated in senior managers and the most-skilled profiles.
- Handcuffs invert the distress read on a schedule, not a vibe: 90% of private companies use the identical four-year, one-year-cliff structure, so a company that hired a wave 11 months ago is briefly un-raidable and the same company at month 13 is suddenly open.
- Scarcity should scale pursuit intensity. Refolk's index shows 40x more US Senior engineers with Kubernetes (51,293) than with Rust (1,282), so a thin skill justifies a raid even on a weak signal.
- An unexpired IPO lockup manufactures fake retention. With 180 days the most common term, a company looks sticky only because no one can sell yet, and the highest-yield window straddles lockup expiry.
- The legal gate is a hard ceiling, not a weight: a 9-out-of-10 raidability score inside an M&A non-solicit or a strict non-compete state must still resolve to do not pursue.

You have a source list of twenty companies and forty hours this week. This guide is for in-house recruiters, sourcers, and talent leaders who need to decide, for one company already on that list, whether its people are worth pursuing hard right now or whether the hours go elsewhere. It gives you a two-axis scoring model built from public signals, a verdict of raid now, watch, or deprioritize, and a re-check date attached to every call.

The published playbooks cover clearance, archetype, universe mapping, and single triggers like layoffs or return-to-office. This one does something they do not: it grades one company's current pursuit-worthiness by combining distress signals with golden-handcuff retention signals into a timed verdict. The generic listicles that rank for "signs a company's employees will leave" are written for managers defending their own teams. This is a scoring model a sourcer uses to rank where to spend finite hours.

## What raidability actually measures

Raidability is the product of two things a sourcer can read from the outside: how likely people are to leave (vulnerability) and how easy they are to extract once they do (extractability). A company can be high on one and low on the other, and the combination, not either number alone, decides your verdict.

Vulnerability is distress. It is whatever is pushing people toward the door: a layoff, a return-to-office mandate, a stalled round, a wave of leadership exits. Extractability is the inverse of the handcuffs holding people in: unvested equity, an unexpired IPO lockup, a recent tender offer that just paid people out. The distinction matters because the two can point in opposite directions. A company 60 days from a layoff announcement looks vulnerable, but if half the team is also 60 days from a lockup expiry, your window is better than the distress signal alone suggests. A company with no distress at all can still be extractable the week a cohort clears its cliff.

This guide scores them separately and then combines them, because no published source ranks distress against handcuff timing. You set the weighting, you record it, and you can defend it later.

#### The raidability verdict grid

Horizontal axis runs from Low extractability (handcuffs tight) to High extractability (handcuffs loose or expiring). Vertical axis runs from Low vulnerability (people content) to High vulnerability (people leaving).

| Quadrant | What it means |
| --- | --- |
| Watch | People are restless but locked in; date the handcuff and re-check when it loosens |
| Raid now | Distress plus loose handcuffs; queue names and move this week |
| Deprioritize | Nothing pushing, nothing loosening; spend your hours elsewhere unless the skill is scarce |
| Opportunistic | Easy to extract but no push; worth a soft touch, not a hard raid |

*Vulnerability and extractability are read separately, then combined into one of three verdicts.*

## Vulnerability: reading distress from public signals

Vulnerability is how likely people are to leave, and the strongest externally datable signal is a return-to-office mandate. A University of Pittsburgh working paper analysing LinkedIn profiles of more than 3 million people across 54 financial and technology S&P 500 companies that issued RTO mandates between April 2020 and June 2023 found a 14% spike in departure rates immediately after requiring an office return.

That signal is better than it first looks, for two reasons. First, the exodus is skewed toward your best targets: the spike concentrated among middle and top-level managers, and employees who showcased the most skills on their LinkedIn profiles were also more likely to leave. Second, a related Gartner survey found 36% of senior-level job seekers facing a return mandate said it influenced their decision to leave. Both point at exactly the extractable senior talent you want. The Van Dijcke case study on SpaceX found an even larger loss of top talent as its sub-senior worker share rose by 15 percent, implying the stricter the mandate, the more likely senior staff jump ship.

Layoffs are the other discrete, trackable distress event. In 2025, layoffs.fyi counted 122,549 tech layoffs across 257 companies, down from 152,922 across 551 companies in 2024. The pace matters more than the total for timing a raid: between January 1 and September 10, 2026, layoffs.fyi counted 128,536 tech employees laid off at 299 companies, already more than in all of 2025, and the BLS Information sector averaged a 1.84% monthly layoff rate in 2026 against 1.07% for all industries.

**14% - Spike in departures immediately after an RTO mandate**

From a study of 3 million LinkedIn profiles across 54 S&P 500 firms, concentrated in senior managers and the most-skilled profiles.

### Leading versus lagging, and why shelf life is a model assumption

Not all distress is equally fresh. An RTO mandate is a leading signal: the exodus is front-loaded, so the people worth catching are deciding to leave in the weeks right after the announcement. A layoff is a discrete event you can date precisely. Leadership departures and stalled funding are softer leading indicators.

A precise shelf life per signal, meaning how many weeks each one predicts departures, is not established publicly. Treat shelf-life windows as a model assumption you set and label as such, not a cited fact. My working assumption: an RTO signal is useful for roughly a quarter, a fresh layoff for about 90 days, and a vesting-cliff estimate until the dated cohort crosses its mark. The point is not the exact number. The point is that you attach a re-check date and never let a stale leading signal masquerade as a live one.

> **Watch out:** RTO decays fast
>
> The 14% spike is measured immediately after the mandate. Raid a company that mandated return 18 months ago and the leavers are already gone, while the stayers have self-selected into people who will not move. Always date the mandate before you score on it.

## Extractability: reading the handcuffs

Extractability is the inverse of the golden handcuffs holding people in, and the dominant handcuff is the standard vesting schedule. Four years with a one-year cliff: nothing vests for the first 12 months, then 25% vests on the cliff date, and the remaining 75% vests in equal monthly increments over the next 36 months. This is widely regarded as the standard, with 90% of private companies sticking to this structure.

That near-uniformity is a gift. Because almost every startup uses the identical structure, the handcuff flips on a date you can estimate. A company that hired a wave 11 months ago is temporarily un-raidable, because leaving means walking away with zero: leave on day 364 and you walk away with nothing; leave on day 366 and you keep a year's worth. The same company at month 13 is suddenly open. One practitioner source flags month 36 as the "golden handcuffs" year, where most retention loss happens, as a cohort approaches full vesting and the marginal cost of staying another year drops. Treat that month-36 framing as opinion, not a study.

#### Vesting milestone windows as a receptivity map

| Stage | Figure | Note |
| --- | --- | --- |
| Month 6 | 0% vested | Pre-cliff, walks with nothing |
| Month 12 | 25% vested | Cliff clears, first real exit window |
| Month 24 | 50% vested | Mid-schedule |
| Month 36 | 75% vested | "Golden handcuffs" year |
| Month 48 | 100% vested | Fully vested, no equity drag |

*Extractability rises as a cohort climbs the standard four-year schedule; the cliff and the final year are the inflection points.*

The second handcuff is the IPO lockup. Most IPO lockups run 90 to 180 days from the offering, with 180 days the most common term in traditional IPOs. An unexpired lockup manufactures fake retention: a company can look sticky purely because no one can sell yet, and people leave the day it lifts. The highest-yield raid window is the weeks straddling lockup expiry, not the quiet months before it. Some companies use creative structures. Snowflake led creative lockups at its September 2020 IPO, allowing employees to sell a portion within the first 90 days rather than the full 180.

| Departure timing | Vested % | What it means for extractability |
|---|---|---|
| Month 6 | 0% | Un-raidable; leaving costs the whole grant |
| Month 12 | 25% | Cliff clears; first open window |
| Month 24 | 50% | Mid-schedule; moderate drag |
| Month 36 | 75% | Golden-handcuffs year; retention loss peaks |
| Month 48 | 100% | No equity drag left |

Whether a recent funding round or tender offer measurably lowers extractability is intuitive but not quantified in the sources I gathered. Note the tender offer, but do not score it as if there were a published effect size.

> **Watch out:** Handcuffs are not loyalty
>
> Low attrition at a company 60 days from lockup expiry is not stickiness; it is a suppressed spring. "Low attrition, deprioritize" is exactly the wrong read here. Find the lockup expiration date in the prospectus before you call a quiet company loyal.

## Scarcity: the multiplier most sourcers skip

Scarcity should scale how hard you pursue, not just whether a company is on your list. A deprioritize verdict costs little when the pool is deep and costs a great deal when it is thin, so the same raidability score should trigger different behaviour depending on how rare the skill is.

In Refolk's index of professional profiles, there are 40x more US Senior engineers with Kubernetes than with Rust: 51,293 against 1,282. For the Kubernetes pool, deprioritizing one company is cheap, because the next company has plenty. For the 1,282-person Rust pool there is almost no spend-my-hours-elsewhere alternative, so even a weak distress signal can justify a hard raid.

| Skill (US, Senior) | Count in Refolk's index | Ratio to Rust |
|---|---|---|
| Kubernetes | 51,293 | 40.0x |
| Rust | 1,282 | 1.0x |

Country pool size works the same way. The US Senior Software Engineer pool is roughly 11.9x Germany's, 182,244 against 15,323. In a thin market, a watch verdict carries more opportunity cost because substitutes are scarcer, which argues for lower raid thresholds in smaller talent markets. A Berlin company you would watch becomes a company you raid, simply because there are fewer places to go next.

| Country | US Senior SWE count | Relative pool size |
|---|---|---|
| United States | 182,244 | 1.00 |
| Germany | 15,323 | 0.084 (US pool ~11.9x larger) |

> A deprioritize verdict is cheap in a deep pool and expensive in a thin one; let scarcity move the threshold, not just the search.

Once you have a verdict and a scarce pool, the bottleneck stops being the read and becomes turning the company into named people fast enough to act inside the window. That is where plain-language sourcing earns its keep.

I ran this search: `Rust engineers in the US currently at Cloudflare, OpenAI, or Figure` - [see the full result list](https://www.refolk.ai/s/434ytd9et3).

*Returns a named, ranked shortlist from a scarce pool where a raid-now verdict is worth hard pursuit per candidate.*

With [Refolk](/) you describe the pool the way you just reasoned about it, and get the people back across the public GitHub graph, LinkedIn, and the open web. The read tells you which companies are worth the hours; Refolk collapses the hours it takes to convert a company verdict into a queue of real names.

## The scoring procedure

Work the six steps in order, gating on legal status before you score anything. The whole pass is about 75 to 90 minutes per company, and the legal gate is first because a high score can never override an off-limits status.

#### Score a source company's raidability

1. **Confirm off-limits and legal status first** - Check client-relationship conflicts, any no-poach or non-solicit tied to a JV or M&A, and the target's home-state non-compete posture. About 15 minutes. If off-limits, stop; the score is irrelevant.
2. **Pull distress signals from public sources** - Check layoffs trackers, funding and round recency, RTO announcements, leadership departures, and requisition activity. 20 to 30 minutes. Done is a dated list of signals, each with a URL.
3. **Pull retention and handcuff signals** - Estimate time since last funding or liquidity event, infer vesting cohorts from hiring-wave start dates against the 12- and 36-month marks, and find any tender offer or IPO lockup date. About 15 minutes. Done is a handcuff profile.
4. **Score vulnerability and extractability separately** - Put likelihood-to-leave on one axis and ease-of-extraction on the other. No study ranks the two, so set and record your weighting. About 15 minutes.
5. **Land the verdict and attach a re-check date** - Resolve to raid now, watch, or deprioritize, tying the re-check interval to the shortest-lived signal. About 10 minutes.
6. **Build the target list and route to outreach** - Queue named people consistent with the legal gate, biasing pursuit intensity by skill scarcity. Ongoing. Done is a ranked list of people.

### Turning the two scores into a verdict

Score each axis 0 to 3 from publicly observable signals only. Vulnerability: count live, dated distress signals (fresh RTO, layoff inside 90 days, leadership exits, hiring freeze). Extractability: count loose or expiring handcuffs (cohort past cliff, lockup expiring, long time since liquidity event). Then read the grid.

**Raidability verdict rubric**

```
VULNERABILITY (likelihood to leave)
+1  RTO mandate dated within the last quarter
+1  Layoff announced within 90 days touching the target function
+1  Two or more senior leadership departures in 6 months
Cap at 3.

EXTRACTABILITY (ease of extraction)
+1  Target hiring cohort is 13-18 months in (past cliff)
+1  IPO lockup expiring within 60 days, or already lifted
+1  18+ months since last funding or liquidity event
Cap at 3.

VERDICT
Both axes >= 2  -> RAID NOW
One axis >= 2    -> WATCH
Both axes <= 1  -> DEPRIORITIZE

SCARCITY MODIFIER (apply after)
Skill pool very thin (e.g. ~1,000s nationally) or thin local market
  -> promote WATCH to RAID NOW; never demote.

RE-CHECK DATE
Set to the expiry of the shortest-lived signal:
  RTO -> ~90 days   Layoff -> ~90 days   Cliff estimate -> cohort month-13 date   Lockup -> expiry date
```

*Score each axis 0-3 on publicly observable signals only; apply the scarcity modifier last.*

> **Rule:** The legal gate is a ceiling, not a weight
>
> A raidability score never overrides off-limits status. A 9-out-of-10 read inside an M&A non-solicit or a strict non-compete state must still resolve to do not pursue. Run the gate first and let it veto the score, never the other way around.

## How this read goes wrong

The scoring model fails in predictable ways, and every failure is a false positive that wastes the hours the model is supposed to protect. Learn these before you trust a single verdict.

- **Layoff signal, wrong read.** A company cutting one division may be ring-fencing and over-paying to retain the team you want. The headline cut count looks huge, but your target function is untouched. Confirm which function was cut via WARN filings or role-level trackers, not the top-line number.
- **RTO signal decays fast.** The 14% spike is immediate; a mandate 18 months old means the leavers are gone and the stayers self-selected to stay. Date the mandate every time.
- **Cliff timing misread.** Inferring a hiring wave's 12-month mark from LinkedIn start dates is noisy, and refresh grants plus non-standard schedules break the model. You may target someone you think just cleared their cliff who actually re-vested. Treat vesting as probabilistic, never as known, and label it an estimate.
- **Handcuffs mistaken for loyalty.** An unexpired IPO lockup suppresses departures artificially. Scoring a company 60 days from lockup expiry as "low attrition, deprioritize" is backwards. Find the lockup expiration date in the prospectus.
- **Legal gate skipped.** A strong score can mask a no-poach tied to a live JV or M&A, which the DOJ treats criminally when naked. Confirm any non-solicit is ancillary to a legitimate collaboration before acting.
- **Non-compete over-caution.** Assuming non-competes are dead because of the FTC is wrong. The rule was vacated, so enforceability is state-by-state. Do not ignore an enforceable covenant in a strict state; check the target's governing-law state.
- **Pre-quitting behaviors unobservable externally.** The 13-behavior research is real: after controlling for age, tenure, education and managers' expectations, a one-unit rise on the pre-quitting scale was linked to 6.48 times higher odds of someone leaving. But it depends on a manager's baseline you cannot see. Do not import manager-side signals into an external model. Score only on publicly observable signals.

The last one deserves emphasis. Much of the "signs an employee is about to quit" literature is written for managers who can see daily behaviour. A sourcer cannot. Your edge is not reading individual restlessness; it is reading company-level distress and handcuff timing that managers often miss.

## The legal gate in detail

Nothing in this model matters if the company is off-limits, so resolve the legal status before scoring. Two regimes govern most raids: non-competes and no-poach or non-solicit agreements.

Non-competes are now a state-by-state question. The FTC's 2024 nationwide ban never took effect. On September 5, 2025, the FTC agreed to vacate the Non-Compete Clause Rule and dismiss all pending appeals, so regulation of these contracts now returns primarily to the states and the commission on a case-by-case basis. The practical read: do not assume covenants are dead, and check the target's governing-law state before pursuing anyone bound by one.

No-poach and non-solicit agreements carry criminal exposure. Since 2016 the DOJ has intended to proceed criminally against naked wage-fixing or no-poaching agreements, treating them like agreements to fix prices or allocate customers. They can be legal if reasonably necessary to a larger legitimate collaboration, such as the sale of a business or a staffing arrangement. At least seven criminal indictments have been filed, and while the DOJ has struggled to get juries to convict, it remains committed. The distinction you need is whether a non-solicit is naked or ancillary to a legitimate collaboration. If it is tied to a live JV or M&A, treat it as a hard stop.

## Before you call the job done

Run this checklist before you commit a verdict to your source list. A verdict without a dated re-check is not finished; it is a guess with a timestamp missing.

#### Raidability verdict sign-off

- [ ] Legal gate cleared: no client conflict, no naked no-poach, governing-law state checked for enforceable covenants.
- [ ] Every distress signal has a date and a URL, and no signal older than its assumed shelf life is being scored as live.
- [ ] The RTO mandate, if any, is dated and falls within the current quarter.
- [ ] Vesting-cohort timing is labelled an estimate, not a known fact.
- [ ] Any IPO lockup has its expiration date recorded, and a quiet company near expiry is not misread as loyal.
- [ ] The scarcity modifier has been applied using the actual pool size for the skill and market.
- [ ] A re-check date is attached, tied to the shortest-lived signal in the profile.
- [ ] The verdict routes to named people, not just a company name.

## Keeping the read current

A raidability verdict is a perishable good, so the real discipline is re-checking, not scoring once. Tie every verdict's re-check date to its shortest-lived signal: an RTO or layoff signal pulls the whole verdict forward to roughly a 90-day horizon, while a vesting-cohort estimate holds until the cohort crosses its month-13 mark and a lockup holds until its expiry date.

Build a dated queue rather than a static ranking. A company you deprioritized today because a cohort is only 8 months in becomes a raid-now candidate on a date you can already calculate. A company you are watching because of an unexpired lockup flips the week the lockup lifts. The verdicts move on schedules you can estimate, which means your source list should be a calendar as much as a list.

When you re-check, re-run the legal gate too. Covenant enforceability shifts with state law, and a non-solicit that was ancillary to a collaboration can outlive the collaboration. The gate that was clear last quarter may not be clear this one. Score the signals, respect the ceiling, attach the date, and spend the hours where the grid tells you they will pay.

## Frequently asked questions

### How many signals does it take to call a raid?

No published source gives a numeric threshold, so treat any single-number rule as invented. The consistent research guidance is to read signals as a cluster, not one isolated red flag. In this framework, raid now means a live distress trigger plus weak or expiring handcuffs scoring high on both axes at once; one signal alone, especially a lagging one, earns a watch, not a raid.

### Is an RTO mandate really a reliable sign a company's employees will leave?

It is the cleanest datable trigger available externally. A University of Pittsburgh study of 3 million LinkedIn profiles across 54 S&P 500 firms found a 14% spike in departures immediately after a return-to-office mandate, concentrated among senior managers and the most-skilled profiles. The catch is decay: the spike is front-loaded, so date the mandate and raid within weeks, not 18 months later.

### Can I tell when someone's equity cliff clears from public data?

You can estimate it, never know it. Because 90% of private companies use the same four-year, one-year-cliff schedule, a hiring wave's LinkedIn start dates let you infer when a cohort crosses month 12 or month 36. But refresh grants, non-standard schedules, and noisy start dates break the model, so treat vesting as probabilistic and label it an estimate in your notes.

### Does a strong raidability score override an off-limits company?

Never. The legal gate is a hard ceiling, not a weight. Non-compete enforceability now varies state by state after the FTC vacated its rule in September 2025, and naked no-poach agreements remain criminally exposed under DOJ policy. A 9-out-of-10 score inside an M&A non-solicit or a strict-covenant state must still resolve to do not pursue.

### Why should skill scarcity change my verdict?

Because a deprioritize verdict costs you nothing when the pool is deep and costs you a lot when it is thin. In Refolk's index there are 40x more US Senior engineers with Kubernetes than with Rust, 51,293 against 1,282. For the Rust pool there is almost no spend-my-hours-elsewhere alternative, so even a weak distress signal can justify a hard raid.

---

*From the Refolk guide library. I revise these guides rather than replacing them, so the current version is always at https://www.refolk.ai/guides/source-company-raidability-read*