Exploit Discovery Is Commoditized. Your Asset Inventory Isn't.
Why weekly exploit risk forecasting depends on asset inventory coverage, not on free KEV and EPSS feeds: denominators, acceptance checks, and failure modes.
Introduction
Exploit discovery is now a commodity input. The binding constraint on any weekly vulnerability forecast is the asset inventory that gives those scores a denominator.
The stakes are no longer theoretical. Verizon's 2026 Data Breach Investigations Report found that exploitation of vulnerabilities is now the most common initial access vector in breaches, at 31% of its reporting dataset, up from 20% the year before. The same report found that only 26% of critical vulnerabilities — defined there as membership in the CISA Known Exploited Vulnerabilities catalog — were fully remediated by organizations in 2025, down from 38% the previous year, and that median time to full resolution rose to 43 days from 32. In the median case, organizations had 50% more critical vulnerabilities to patch than in the prior dataset.
That combination — a rising attack vector and a remediation process losing ground — is usually read as a patching capacity problem. It is at least as much a measurement problem. You cannot forecast remediation against a population you cannot count.
The argument below runs in four parts. KEV and EPSS are different instruments measuring different things, and treating them as synonyms produces bad thresholds. The unknown-asset gap is documented, but only in specific populations, and the figures do not generalize. A weekly reforecast is a defined artifact with explicit denominators, not a dashboard refresh. And an inventory has to pass an acceptance test before any weekly number derived from it means anything.
One framing note before the detail: the seven-day cadence here is a proposed operating model for readers to adopt and adapt. It is not an observed industry standard, and outside one specific federal population it is not a requirement.
The Inputs Got Cheap; the Denominator Didn't
The marginal cost of knowing that a vulnerability exists has collapsed. CVE identifiers are assigned and published through the NVD. The CISA KEV catalog publishes vulnerabilities with evidence of exploitation in the wild. EPSS scores are published daily via CSV and API with no registration required, as FIRST's EPSS FAQ documents. Public proof-of-concept repositories and exploit tooling are indexed and searchable. Each of these sources is free, continuously updated, and available to any organization with an internet connection.
What has not collapsed is the cost of knowing which of your assets carry the vulnerability.
This is worth stating as a structural point rather than a complaint about tooling. Every prioritization score is a numerator. EPSS gives you a probability for a CVE. KEV gives you a binary flag for a CVE. CVSS gives you a severity for a CVE. None of them knows whether that CVE is present on hardware you operate, in a container you deployed last Tuesday, or on a marketing team's SaaS integration that nobody registered.
A perfect EPSS score applied to an incomplete asset list produces a confident answer about the wrong population. The score is not wrong. The denominator is.
This is why the abundance of exploit intelligence has not translated into proportional risk reduction. The intelligence layer got cheap and fast. The inventory layer stayed expensive, manual, and partial — and it is the layer that determines whether the intelligence applies to you at all.
KEV and EPSS Are Not Interchangeable
KEV and EPSS are routinely cited in the same sentence as if they were two views of one thing. They are not.
KEV is retrospective and binary. A vulnerability is either in the catalog or it is not, and inclusion reflects evidence that exploitation has occurred somewhere in the wild. It is a strong signal, and FIRST's guidance on using EPSS treats a recent KEV addition as reasonable grounds for high priority regardless of EPSS score. But that signal degrades: without other evidence of ongoing exploitation, a vulnerability added to KEV years ago reverts toward the exploitation probability its current features would otherwise predict.
EPSS is prospective and probabilistic. It estimates the likelihood that exploitation activity will be observed within 30 days, and publishes a percentile alongside the score. The 30-day window is not arbitrary. FIRST's methodology page states that the window was chosen to match practitioner decision cycles, because Patch Tuesday has anchored vulnerability response cadence for decades and a 30-day prediction maps onto that cycle.
That choice is what makes EPSS a natural input to a weekly cadence — and also what limits it. The score describes the global observed population. FIRST is explicit that EPSS does not know what is in your environment, whether an attacker could reach it, or whether it is already under attack.
The base rate is the part most teams misread. Exploitation is rare at the population level, so most scores cluster near zero. FIRST puts the base rate at roughly 2–3% and notes that a 5% score sits in the top 10% of all published vulnerabilities — a relatively high exploitation forecast, not a low one. A team that sets a 50% threshold because "half sounds like a coin flip" has misunderstood the distribution entirely.
Two further distinctions matter for how you build a report. CVSS measures severity, not risk; the CVSS v4.0 FAQ states directly that base scores are not risk. And EPSS uses individual CVSS metrics as input features while ignoring the scores themselves, learning weights from observed exploitation rather than expert consensus. Multiplying EPSS by CVSS does not compute probability × severity — it produces a number with no interpretable meaning.
What neither instrument can tell you is whether the vulnerable software is on an asset you know about. That question belongs to the inventory.
The Unknown-Asset Gap, and Who It Was Measured In
The gap between what organizations track and what they expose has been measured, and the measurement has a specific scope that should be preserved when citing it.
Randori, an IBM company, commissioned Enterprise Strategy Group to survey 398 IT and security decision-makers in the US and Canada. The resulting report, The State of Attack Surface Management 2022, published in October 2022, reported that 69% of organizations had been compromised by an unknown, unmanaged, or poorly managed internet-facing asset in the past 12 months, and that organizations had on average 30% more exposed assets than their traditional asset management programs tracked.
What those figures cover: survey responses from a North American decision-maker sample, collected in 2021–2022, about internet-facing assets specifically. What they do not cover: a census of exposed hosts, a universal ratio applicable to any organization, or anything about internal-only assets. A 30% average is an average across respondents — it is not a target, a benchmark, or a prediction for your estate. Treat it as evidence that the gap is common and material, not as a number to plan against.
NIST's position in SP 800-40r4, §3.2 is the more useful framing for planning. The publication states that while a comprehensive inventory of all assets is ideal, it may be impossible to achieve given the highly dynamic nature of assets and software, and that a realistic goal is to maintain a close-to-comprehensive inventory by relying on automation to constantly discover new assets and collect up-to-date information on all assets. It also notes that some vendors provide machine-consumable data on software composition, such as an SBOM, which can augment organizational inventories.
That is the shape of the response: not a one-time reconciliation project, but continuous automated discovery feeding a maintained inventory. The gap is the premise. Continuous discovery is the only response that keeps the premise from re-establishing itself every quarter.
What a Rolling 7-Day Reforecast Should Contain
The weekly artifact described here is a construct proposed for adoption, not a description of an established practice. Its value is that it forces every number to carry its base.
A defensible weekly reforecast contains, at minimum:
A fixed inventory snapshot date and a discovery coverage percentage. The snapshot date tells the reader what the numbers describe. The coverage percentage tells them how much of the estate the snapshot represents.
Assets in scope, and assets newly discovered in the window. New discoveries are the direct output of continuous discovery and the leading indicator that the denominator is moving.
The vulnerability population matched against those assets, split by KEV membership and by EPSS band. Splitting by band rather than by a single threshold preserves the coverage/effort tradeoff as a visible decision rather than a buried one.
The remediation backlog, expressed against a stated denominator. Open criticals per tracked asset, or per discovered asset, or as a share of KEV-matched findings — but never as a bare count.
The denominator rule is the one that gets violated most often. "Critical vulnerabilities open" is meaningless without knowing whether the base is all CVEs, KEV members only, or EPSS above a chosen threshold, and whether the asset base is the tracked inventory or the discovered inventory. Two teams can report the same open count against different bases and reach opposite conclusions about whether posture improved.
The closest published precedent for a seven-day cycle is CISA Binding Operational Directive 23-01, which required FCEB agencies, by April 3, 2023, to perform automated asset discovery every 7 days covering at minimum the entire IPv4 space used by the agency, and to initiate automated ingestion of vulnerability enumeration results into the CDM Agency Dashboard within 72 hours of discovery completion. The directive also requires agencies to track how often they enumerate assets, what coverage they achieve, and how current their vulnerability signatures are — which is, in effect, a mandate to publish the denominator.
Cite that as precedent for the cadence, scoped to the agencies it binds. It is not a general mandate, and the 72-hour ingestion requirement is specific to the CDM reporting pipeline.
One design recommendation: publish week-over-week deltas alongside absolute counts. A backlog that shrinks because the inventory shrank looks identical to a backlog that shrank because remediation worked — unless the delta view shows assets in scope falling at the same time.
Testing Whether Your Inventory Is Good Enough to Forecast From
Before trusting a weekly number, run the inventory through a gate. This is not a maturity model with levels to climb; it is a set of checks that tell you which denominator your forecast is actually using.
Discovery coverage against an external reference. Enumerate a known external range independently — from a cloud provider's address allocation, a registrar record, or a third-party scan of your own perimeter — and compare it with what the inventory holds. The difference is your coverage gap, measured rather than assumed.
Time-to-detection for deliberately introduced assets. Stand up a test asset through a normal provisioning path and measure how long it takes to appear in the inventory. This tests the pipeline end to end, including the human steps that usually break it.
Owner and maintenance group assignment. NIST SP 800-40r4 structures patch management around assigning each asset to a maintenance group with a defined maintenance plan. The proportion of assets with a current owner and group is a direct measure of whether that structure exists in practice or only on paper.
Machine-readable software composition. The share of assets whose installed software is machine-readable — from agent telemetry or vendor-supplied SBOMs — determines whether vulnerability matching can be automated or has to be inferred. NIST notes SBOM data as an augmentation to organizational inventories; the operative word is augmentation, not replacement.
Scanner-to-inventory reconciliation. Compare the vulnerability scanner's host list against the inventory's host list. A host the scanner sees and the inventory does not is a direct, countable measure of the gap, and it is usually the fastest check to run.
Set the pass condition yourself. The thresholds depend on your exposure, your remediation capacity, and what the forecast is used to decide. Failing any single item does not invalidate the inventory — it tells you which denominator your forecast is actually using, which is exactly what the report needs to disclose.
Failure Modes That Break a Weekly Cadence
A seven-day cycle degrades in predictable ways. Each failure mode has a detection signal, and the signal is more useful than the remedy because it tells you when to apply one.
Silent or partial discovery failure. A discovery run that errors out, times out, or covers only part of the estate produces a smaller inventory without producing an alarm. Detection: track coverage percentage week over week and alert on any drop, not just on outright failure. A coverage number that moves down without a corresponding decommissioning event is a pipeline problem until proven otherwise.
Inventory growth outpacing scanning capacity. New assets appear faster than the scanner can enumerate them, so they exist in the inventory with no vulnerability record. Detection: count assets that have been in the inventory for one full cycle with zero vulnerability findings. A nonzero count is either a scanning gap or a genuinely clean asset — and the ratio tells you which.
Score drift mistaken for posture change. EPSS scores update daily and KEV additions are continuous. A week-over-week change in the count of high-EPSS findings may reflect the scores moving, not your estate changing. Detection: hold the asset list fixed and re-run the same week's scores against it. If the count moves with a static asset list, the movement is in the numerator.
Denominator switching between reports. A report that changes its base from tracked to discovered assets, or from all CVEs to KEV members, will show a posture change that did not happen. Detection: publish the base in the same row as every metric, and diff the bases between consecutive reports before diffing the values.
Cadence collapse under emergency patching. The weekly report gets skipped precisely when remediation is most intense and the report is most needed. NIST SP 800-40r4's emergency patching scenario anticipates a highly accelerated schedule, still using canary assets to confirm a patch is not corrupted before broad deployment. That acceleration is compatible with keeping the reporting cycle intact: remediation can speed up while the measurement cadence holds. Detection is trivial and therefore easy to ignore — a missed report date is the signal.
Does Compliance Require This?
The compliance answer depends entirely on which population you are in, and it is worth being precise rather than reassuring.
BOD 23-01 binds FCEB agencies. The seven-day automated discovery requirement and the 72-hour ingestion of enumeration results into the CDM Agency Dashboard apply to Federal Civilian Executive Branch agencies, with the April 3, 2023 deadlines for the initial requirements. CISA's implementation guidance notes that while the primary audience is FCEB agencies, other entities may find the content useful — useful, not binding.
NIST SP 800-40r4 is guidance, not a mandate. Its inventory recommendation sits alongside mappings to NIST SP 800-53 controls CM-8 (System Component Inventory) and SI-2 (Flaw Remediation), and to Cybersecurity Framework subcategories ID.AM-1 (physical devices and systems inventoried) and ID.AM-2 (software platforms and applications inventoried). Those mappings matter to organizations that have adopted SP 800-53 or the CSF, or that are subject to them contractually through customers, regulators, or insurers. They do not create a standalone obligation for everyone else.
For private-sector organizations outside those obligations, the case for a maintained inventory rests on forecast validity rather than on a specific rule. A forecast with an unstated denominator is not a compliance failure; it is a decision-support failure, and it fails whether or not anyone is auditing it.
Confirm which obligations actually apply to you rather than assuming the federal timeline transfers. The seven-day cadence is defensible on its own merits — it matches EPSS's 30-day window and the Patch Tuesday rhythm most programs already run on — without borrowing authority from a directive that does not reach you.
Conclusion
The exploit signal is now free and the inventory is the constraint. Everything that follows from that sentence is an operational choice about measurement.
Two shifts make the constraint visible. First, publish the denominator with every weekly number — the asset base, the vulnerability base, the coverage percentage, and the snapshot date, in the same row as the value. Second, treat inventory coverage as the first metric in the report rather than an assumed precondition. A coverage figure that leads the report sets the reader's confidence in everything below it, and a coverage figure that moves explains movements elsewhere before anyone misreads them as posture change.
The practical next step is small. Run the acceptance checklist once before the next reforecast cycle. You do not need to fix what it finds before publishing — you need to know which denominator you are publishing against, so that the first weekly number is honest about what it measures.
Frequently Asked Questions
But doesn't a high EPSS score tell us what to patch first?
No. It tells you the estimated probability of exploitation across the vulnerability population, not the exposure of your estate. FIRST's own documentation directs users to layer local context on top of the score, because EPSS does not know what is in your environment or whether an attacker could reach it. The same score applied to an asset you have and an asset you do not produces the same number but a very different risk.
Isn't KEV enough on its own, since those are the ones being exploited?
KEV membership is retrospective evidence that exploitation occurred somewhere — a strong signal, but not a statement about your exposure or your patch state. The 2026 DBIR finding that only 26% of KEV vulnerabilities were fully remediated in 2025, with a median 43 days to full resolution, points to a constraint in remediation capacity and asset identification rather than in knowing which vulnerabilities matter. The catalog already told you which ones matter. The open question is which of them are on your estate.
Our inventory will never be complete — is weekly forecasting pointless?
No, provided the forecast publishes its coverage. NIST SP 800-40r4 states that a close-to-comprehensive inventory maintained by continuous automated discovery is the realistic goal, given how dynamic assets and software are. A forecast with a stated and tracked coverage percentage is usable and improvable. One with an unstated denominator is not usable at all, because the reader cannot tell whether a change in the number reflects the estate or the measurement.
Doesn't the seven-day cycle come from a federal mandate that doesn't apply to us?
BOD 23-01's seven-day discovery requirement binds FCEB agencies specifically. For other organizations the weekly cadence is a design choice, justified by EPSS's 30-day prediction window and the Patch Tuesday rhythm most vulnerability programs already follow, rather than by regulation. Confirm your own applicable obligations before treating any federal timeline as your own.