The national police decertification index carries a status word and no reason — inclusion in it does not necessarily preclude appointment as an officer, and no one outside law enforcement can read it

What is happening?

The National Decertification Index is the one national file of police license revocations in the United States. It held almost 25,000 actions as of mid-September 2018, the last count anyone has published, and by that same date 43 of the 45 states that can revoke a police license were submitting to it. Joining costs nothing and searching costs nothing. As of 2026 nobody outside law enforcement can read any of it — not the general public, and not the press.

What the file holds for each entry is a status word and nothing more — de-certified, suspended, probation, or voluntary relinquishment. The International Association of Directors of Law Enforcement Standards and Training, which operates it, describes it as a pointer system, and states that inclusion in the database does not necessarily preclude any individual from appointment as an officer, because what qualifies for de-certification in one state may be acceptable in another. Access is approved by a state training and standards board for state officers, by the Department of Defense for military police, and by the association itself for federal officers.

A separate federal file covered federal officers only. The National Law Enforcement Accountability Database launched on 2023-12-18 under Executive Order 14074 and was switched off after that order was rescinded on 2025-01-20. The Congressional Research Service, writing on 2025-02-26, states that the rescission does not affect the National Decertification Index. The record that the public can actually read was built outside government — a civil-society compilation had obtained and published officer employment histories covering 24 states as of 2026-01-26.

Whose problem is this?

The three deciding parties are separate bodies with separate budgets and separate legal duties, and the harm from a bad hire lands on a fourth party that sits in none of their rooms.

Where does this problem end?

The boundary matters here because the coordination step already succeeded and the failure sits somewhere else entirely. A file that nobody is required to open is not the same kind of gap as a file that does not exist.

What is the state now, and what should it be?

Now

Needs a new measurementthe target state: no source opened in this round names one. Not IADLEST, not the Congressional Research Service, not Executive Order 14074, and not H.R. 1280. None of them states what contribution rate, what query rate or what level of public access would count as this problem being answered. The record count has also never been published against a denominator, because no party counts revocation-eligible separations nationally, so even the figures that exist cannot be turned into a coverage rate.

How big is it?

About 18,000 officers. That is 3 percent — the share of all officers in Florida found to be wandering officers in any given year, almost 1,100 of them, by a 2020 study quoted by the Congressional Research Service on 2025-02-26 — applied to the more than 600,000 sworn law enforcement officers serving in the United States as of 2023. Because the headcount source says more than 600,000, the product is a floor rather than a central value. A second measured anchor points the same way without adding a number to the chain: a Texas study found that one quarter of officers dishonorably discharged from a previous agency were hired by a new one.

There is no second endpoint, and the interval therefore has zero width. The only other national denominator opened here is the June 2018 census of state and local agencies, which counts 1,214,000 full-time personnel across 17,541 agencies. That figure counts sworn and civilian staff together and the readable highlights give no sworn-only breakout, so 3 percent of it produces about 36,000 as an arithmetic artifact rather than as an upper estimate. A rate measured on sworn officers cannot be carried onto a headcount that includes civilian employees, so 36,000 is unreachable by construction and is not printed here as the top of a range.

The number counts officers, not people. No source opened here reports how many residents an officer contacts in a year, so the harmed public cannot be derived from this chain and must not be back-computed from it. Three further categories fall outside the count entirely. Officers who resign while an investigation is open are never decertified and so never enter any index — the federal database recorded 481 such separations among 4,790 incidents before it was switched off. Officers who cross a state line rather than move within one are captured by no measured rate at all, because every rate that exists is within-state. And officers in the states that hold no revocation authority generate no record to be counted anywhere.

The direction of the bias in carrying a Florida rate to the whole country is not determinable. Florida is one of the four states that produced more than 16,000 of the 19,100 revocations recorded through 2008, so it removes wandering officers at a rate far above the national average, which should push the national estimate down, while it also generates far more visible records, which should push it up. No source opened here resolves which effect dominates.

Under what conditions does it arise?

1. No duty to query can be shown to exist, and none can be shown to be absent. No source opened in this round states that any state requires a pre-hire search of the national index, and no source states that none does. The absence is symmetrical, which is what makes it a silence rather than a finding. What is on the record is the usage measurement from 2009 — 10 boards queried routinely and 19 occasionally — and nothing published since.

2. Contributing and querying are separate decisions taken by separate bodies. A state training and standards board decides whether to feed the index. A hiring department decides whether to read it. The person who bears the cost of a bad hire is party to neither decision.

3. A revocation in one state is not a revocation in another. IADLEST gives that as the reason the file is not a blacklist, and adds that inclusion does not necessarily preclude appointment. The 2009 census of grounds shows the shape of the spread: every revoking board could act on a felony conviction, 36 could revoke for certain misdemeanor convictions, only 26 could revoke administratively for misconduct, and 31 could suspend.

4. Separation before adjudication removes the record before it is written. An officer who resigns while under investigation is never decertified, so nothing about that officer reaches the index. The federal accountability database counted this category as a line of its own and recorded 481 such separations among 4,790 incidents, which means the federal government identified the leak and then closed it for its own roughly 148,000 officers for fourteen months.

5. No outside party can audit any of the four conditions above. The file is closed, so the only people who can see whether it is failing are the hiring agencies it is supposed to constrain.

What has been tried?

Every one of these attempts either built the file or funded it, and not one of them created a duty to read it.

What was found?

Why is it still unsolved?

Institutional gap — the shared file was built, it is free, and almost every eligible state feeds it. What was never built is the institution that would make anyone use what it holds.

The registry problem is usually told as a coordination story. Fifty states, no shared record, nobody willing to move first. The record opened here does not support that telling. By September 2018, 43 of the 45 states that can revoke a police license were submitting to a national index that costs nothing to join and nothing to search. The coordination step succeeded. Three institutions are missing behind it, and each absence carries weight on its own.

The first is a duty to look. Contributing is decided by a state training and standards board, querying is decided by a hiring department, and those are different bodies answering to different authorities. The only usage measurement ever published is ordinal and dates from 2009, when 10 boards queried routinely and 19 occasionally. Seventeen years on, no source opened here states that any state requires a pre-hire query and none states that none does, so the duty can be shown neither to exist nor to be absent. Either way, nothing in the opened record shows a hiring agency being made to look.

The second is a common floor for grounds. IADLEST states in its own explanation of the index that what qualifies for de-certification in one state may be acceptable in another, and it draws the conclusion itself — inclusion in the database does not necessarily preclude any individual from appointment as an officer. A hiring board can query the index, get a hit, and hire the person anyway. What a hit produces is a phone number rather than a bar. The 2009 census of grounds gives that abstraction its shape, since every revoking board could act on a felony conviction while only 26 could revoke administratively for misconduct.

The third is any lever that reaches states at all. Police licensing is a state function. The strongest federal instrument actually used, Executive Order 14074, could compel only federal agencies, and for everyone else it could direct the Attorney General to encourage participation as permitted by law. An executive order is revocable by the next signature, and this one was revoked on 2025-01-20 after fourteen months of operation. The instrument that would have reached the states was H.R. 1280, which would have conditioned a federal grant program on submission, and it was never enacted.

Underneath the three sits a fourth absence that holds the pattern in place. Because the file is closed to the public and the press, the only parties able to see whether the first three are producing bad hires are the hiring agencies the file is supposed to constrain. That is why the substantive evidence about officers moving between agencies comes from academic studies of single states and from a compilation built by request and litigation, rather than from the index itself.

What observation would mean it is solved?

Candidates — (a) a published national count of pre-hire queries against the index, rising toward the number of law enforcement hires made each year (b) a national measurement of how often an officer revoked in one state is later hired in another (c) public access to the record, whether through the index itself or through a compilation covering every state.

(a) alone counts looking, not seeing. A query is not a bar. The operator states that inclusion does not necessarily preclude appointment, so a hiring agency could query every candidate, find every hit, and hire exactly as it did before. A rising query count would then measure diligence while nothing changed at the point of hire. It would also be measuring the cheapest part of the problem, since searching the index is free and always has been.

(b) alone has never been measured and may not be measurable from the index. Every rate in existence is within a single state, because within-state data is the only data there is. The one cross-boundary hit rate ever produced is the federal figure of 25 matches from 9,985 searches, and that is cross-agency inside the federal government rather than cross-state. A rate built from index contents would also miss every officer who resigned before adjudication, a category the federal database counted at 481 among 4,790 records.

(c) alone counts paper. A public record makes auditing possible and does not by itself change a hiring decision. The civil-society compilation already covered 24 states as of 2026-01-26 while the index beside it stayed closed, which shows the two can move independently of each other. The three have to be read together, and even together they are incomplete, because no source opened here states a target for any of them.

What is it connected to?

Fills with researchrevocation of professional licenses in other occupations such as medicine and teaching, the civil liability route by which a rehired officer generates municipal exposure, state open-records law as the actual constraint on the compilation project, and equivalent registers in other countries. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• How often the index is queried. No source states how many searches are run in a year, by how many agencies, or with what hit rate. The last usage figure anyone published is the ordinal split from 2009, and seventeen years have passed without another.
• How many records it holds today. No source publishes a count after mid-September 2018. The two IADLEST documents read in full for this round carry no count of records, of participating states or of queries — the operator of the national index publishes no measure of the national index.
• What any record count would mean. The figure has never been published against a denominator, because nobody counts revocation-eligible separations nationally. The index can be described only in absolute terms.
• Whether the two published counts can be compared. One source counts names in 2016 and the other counts actions in 2018, and neither states whether a single officer can generate several actions. The apparent growth from 20,500 to 25,000 is therefore not interpretable, and no source flags the change of unit.
• Whether any state requires a pre-hire query. Neither IADLEST, nor the 2018 trade account, nor the Congressional Research Service asserts that any state does, and none of them says that none does either. The absence runs in both directions at once.
• What happened to the federal records. The database was deactivated with no public account of the fate of its 4,790 records. The disposition request surfaced only as item 9 in a routine comment notice, and that notice states no retention period and no destruction terms.
• Which 24 states the public compilation covers. The compilation does not enumerate the states, list the fields each one supplies, or define what more limited data means for the twenty-fifth. The 24 figure is also a count of what was obtained, by request and by litigation, rather than a count of states that publish by law.
• Who is responsible for the whole. The research service describes what Congress could do, the operator describes a voluntary service it hosts, and the executive order could only encourage. No opened document names a party accountable for the outcome.

See the evidence

No primary document about the index itself was read from the government that funded it. The two IADLEST documents were opened directly and are the operator speaking about its own file, which is what makes their silences load-bearing rather than incidental — neither carries a count of records, of participating states or of queries. Executive Order 14074 and the National Archives notice were read as primary federal texts, and the 2009 survey was read in full. Everything about the size and use of the index comes from single sources, and where two sources overlap they change units instead of agreeing: 20,500 names in 2016 against almost 25,000 actions in 2018, from different authors, with no shared methodology statement. That disagreement is left visible rather than resolved, as is the four-way contradiction over which states lack revocation authority, where a 2009 survey names two of 45 certifying entities, a 2016 account names six states, a 2018 account names four, and a 2026 operator document names only Hawaii and Rhode Island as lacking a board at all — not one of the four sets out its counting rule, and the 2026 document also calls the Hawaii legislation recent although the 2018 account dates it to July 2018 with effect from July 2019. Two corrections to the assignment row belong here. The figure of 24 states describes histories obtained and published by a civil-society project as of 2026-01-26, by request and by litigation, and is not a count of states that publish by law. And the federal accountability database covered federal officers only, so its deactivation removed nothing from state and local hiring, which the research service states directly. This is a Path A output (research-based definition), so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 18,000–18,000 As of 2023 headcount with a 2020 measured rate

Sensitivity The interval has zero width, and the reason is specific to this chain rather than to the strength of the estimate. There is no second endpoint that is not an arithmetic artifact. The only other national denominator opened in this round is the June 2018 census of state and local law enforcement agencies, which counts 1,214,000 full-time personnel across 17,541 agencies, but that figure counts sworn and civilian staff together and the readable published highlights give no sworn-only breakout. Three percent of it produces about 36,000, and a rate measured on sworn officers cannot be carried onto a headcount that includes civilian employees, so 36,000 is unreachable by construction and is not offered here as the top of a range. What the number fails to count runs in four directions at once. It counts officers rather than people, and no source opened here reports how many residents an officer contacts in a year, so the harmed public cannot be derived from this chain and must not be back-computed from it. It cannot count officers who resign while an investigation is open, because they are never decertified and so never enter any index at all, and the federal accountability database recorded 481 such separations among 4,790 incidents before it was deactivated. It cannot count officers who cross a state line rather than move within one, because every rate that has ever been measured is within-state, and the movement across boundaries is precisely the quantity the registry gap is about. And it cannot count officers in the states that hold no revocation authority, whose separations generate no record anywhere. The limit in the opposite direction is that both terms in the chain were reached through carriers rather than from the primary documents, since the law review article and the labor statistics page each returned HTTP 403, so a correction to either primary would move the product without any signal appearing in the sources actually opened.

Regional breakdown No source opened in this round gives a state-by-state count of officers in a wandering state, and the only two measured rates that exist are single-state figures from Florida and Texas. Splitting the national product by state population would be proportional allocation and would be wrong by a wide margin, because state decertification programs differ in output by three orders of magnitude. Georgia recorded 3,239 decertifications across 2015 to 2019 while Minnesota recorded 21 and Maryland recorded 1, and four states produced more than 16,000 of the 19,100 revocations recorded through 2008 against about 3,100 from every other state combined. A population-weighted split would therefore assign officers to states whose boards produce almost no records and withhold them from states that produce most of them.