An Oregonian who has worked as a poll worker in this state, and who has hand-counted ballots, told Profoundd they were not permitted to watch the machine count verify the work they had just done.
That account is one person's, and readers should weigh it as such. What can be checked is whether Oregon law would permit it, and the answer is that Oregon law permits it comfortably. This article sets out the architecture that makes it possible, separating what is verifiable in statute from what remains a claim. The distinction matters more than usual here, because the central finding is about the limits of what anyone can verify.
The clerk runs the election, and the clerk is elected
Two Oregon statutes sit next to each other and are worth reading together. ORS 204.005(1) lists the county officers who are elected, and (b) is the county clerk. ORS 246.200(1) provides that "the county clerk is the only elections officer who may conduct an election in this state," and specifies that conducting an election "includes, but is not limited to, establishing precincts, preparing ballots and sample ballots, and receiving and processing votes."
So the officer who alone may conduct an election holds the office by winning one, and the election that put them there was conducted by that same office. There is a defence of this arrangement and it is not weak: an elected clerk answers to voters directly rather than to a political appointer, which is why many states prefer it to a governor's appointee, and Oregon's 36 clerks are not a single point of control. But the circularity is real, and it is the frame for everything below. Removing a clerk requires an election the clerk administers.
What the clerk decides, and under what standard
The detail of Oregon's ballot handling is not in statute. It is in the Vote By Mail Procedures Manual, which Oregon Administrative Rule 165-007-0030 designates as the governing procedure — meaning it carries rule force and can be revised administratively, without a legislative vote.
The manual states a broad principle: "Members of the public may observe all ballot processes except confidential election processes such as the challenged ballot process." Read alone, that is generous. It does not stand alone. The same passage defines an Observer as "an individual(s) who has obtained prior approval from the county elections official," provides that the clerk "will determine the number of observers allowed at any given time based on space, security, and staff availability," and adds that the clerk "may add additional restrictions as needed."
A permission that requires the approval of the party being observed, in a quantity they set, subject to restrictions they may add at will, is not a right of access. It is discretion described in the vocabulary of a right. The statute underneath it is narrower still: ORS 254.482 entitles only "authorized persons" to watch the receiving and counting of votes, where the authorization must be written and signed by a political party officer or county affiliate, a candidate, or the clerk — and the clerk admits "only so many persons as watchers as will not interfere with an orderly procedure," a standard the clerk alone applies.
Counting ballots for your county confers nothing. The poll worker and the observer are different statutory categories, and being the first does not make you the second.
The criteria are exempt from disclosure by name
This is the part that turns discretion into something a reader cannot check. The manual's security checklist directs every county to "establish guidelines for security involving observers, county employees, volunteers, vendors or other visitors to the elections office." Those guidelines go into the county elections security plan filed under ORS 254.074.
ORS 192.502(32) exempts from public records disclosure: "A county elections security plan developed and filed under ORS 254.074." The manual states the consequence plainly — "Security plans are exempt from disclosure under the public records laws."
So the written criteria determining whether you may watch your county count ballots are, by name, in a document you have no right to read. This is not an oversight or a matter of nobody having asked. It is a statutory exemption.
What you witness, you largely may not record
ORS 165.540(1)(c) makes it unlawful to "obtain or attempt to obtain the whole or any part of a conversation by means of any device... if not all participants in the conversation are specifically informed that their conversation is being obtained." The exceptions in subsection (6)(a) cover public or semipublic meetings, hearings, trials, press conferences, classes, and private meetings where everyone knew — administrative ballot processing is none of those. The Ninth Circuit, sitting en banc, upheld the statute in January 2025 in Project Veritas v. Schmidt, reversing a 2023 panel decision that had struck it down.
Be precise about the scope, because overstating it would be the easiest way to discredit everything else here: the statute restricts recording conversations. It does not on its face prohibit silent video, still photography, or taking notes. What it means in practice is that an observer cannot capture what staff say. Meanwhile the same security plan requires each county to "include the number and location of all video surveillance cameras within the elections office." The office records the room. The observer may not record the conversations in it. The document describing both arrangements is the exempt one.
The audit that could change a result is the one nobody ran
Oregon has two post-election audits and ORS 254.529(1) and ORS 254.532(2) give the choice to the county clerk.
The risk-limiting audit under ORS 254.532 is the stronger instrument, and its requirements are worth quoting because they describe what a checkable election looks like: it must "permit members of the public to observe the procedures and verify the results," must rest on "direct visual human examination of elector-marked ballots," "may not rely on a scanned image of a ballot or a machine interpretation of marks," must keep the body counting ballots from being the sole judge of whether the audit's rules were met, and must be timely enough "to be corrected before an election contest is certified."
The hand count is the other path, and OAR 165-007-0290(7) provides that a county elections official "may only begin the hand count after certification of the official results." One audit is built to change a wrong result. The other is forbidden to start until the result is final. In fairness, the hand count is not toothless — ORS 254.529(7) escalates a persistent discrepancy above half a percent into a second count and then a full hand count that becomes the official tally. But that machinery runs after certification, on 3 percent of batches in races decided by more than two points, across three contests.
For the most recent statewide election no county chose. Secretary of State Tobias Read's Directive 2026-05, issued June 10, 2026, notes that ORS 254.529 requires each county to make that determination, then states: "This directive directs County Elections Officials to conduct a hand count of ballots cast at the May 19, 2026, Primary Election." Every county was placed on the path that begins after the result is official. A directive does produce uniformity and a single comparable statewide record, which a patchwork would not, and it is appealable to Marion County Circuit Court under ORS 246.910.
The one stage with no second party in the room
At nearly every stage of handling a ballot the manual requires two people who are not from the same party. Opening boards: "a minimum of two members, not all of the same political party." Inspection boards determining voter intent: "at least two board members not of the same political party," repeated for digital tally systems and again for re-inspecting ballots a machine could not read. Duplication and proofing: "at least two election personnel who are not members of the same political party."
Signature verification — the stage that decides whether a ballot enters the count at all — carries no party requirement. What it carries instead is an asymmetry: a signature "possessing obvious and predominantly matching characteristics" may be "determined to be a match by a single county elections official," while one "possessing one or more distinctive dissimilarities" must be "reviewed by at least two different county elections officials," and separately "an experienced, deputized elections staff member must review all rejected signatures prior to final assignment of that status." Accepting can take one person. Rejecting takes at least three.
The rationale for that tilt is serious and belongs beside it: a wrongly rejected signature disenfranchises a real voter, the error cannot be undone once the cure deadline passes, and national research on signature matching finds rejection falls hardest on young, first-time and minority voters whose signatures are least established. Oregon has weighted the process against discarding a genuine vote. That is a defensible choice. It is also, unavoidably, a process weighted toward counting a doubtful signature rather than questioning it, run by officials who need not be politically balanced, under a rule the Secretary of State may rewrite without asking the Legislature.
Testing the other explanation by the same standard
Every provision above has a stated or obvious justification, and it would be dishonest to list the architecture without running those justifications through the same test the architecture just received. Three of them survive that test unevenly.
The recording law survives it. ORS 165.540 is a conversational privacy statute of general application. It was not written for elections, it applies to every workplace and street corner in Oregon, and a federal appeals court sitting en banc upheld it against a First Amendment challenge in January 2025. Its effect on election observation is a side effect of a general rule, not a rule aimed at observers. That is the honest finding and it cuts against reading the statute as part of any design.
The security-plan exemption survives only in part. The rationale is sound where the plan describes what an adversary could exploit: camera placement, alarm systems, key control, transport routes, cyber response. Publishing those would defeat their purpose. But the exemption is written to cover the plan, and the manual requires the plan to contain the observer guidelines. Rules telling the public how to apply to watch a count are not a vulnerability in the way a camera map is. The exemption sweeps in a category that its own justification does not reach, and no published Oregon source we located explains why observer criteria must travel in the same sealed envelope as the alarm codes.
Clerk discretion survives least well. An elections office is a working room with live ballots in it and a finite number of chairs, so someone must be able to cap attendance — that much is unarguable. It does not explain why the cap has no floor, why the standard is "as will not interfere with an orderly procedure" judged solely by the person applying it, why the clerk may "add additional restrictions as needed" without stating them, or why the criteria are unreadable. A capacity limit and an unreviewable, unpublished, open-ended power are different things, and only the first is explained by the room.
What this does and does not establish
Taken together these provisions establish something specific and, we think, important: no party outside an Oregon elections office is positioned to independently verify ballot processing. Observation is permission rather than right; the criteria governing that permission are statutorily sealed; what an observer hears may not be recorded; the audit built to catch an error before it becomes official is optional and was not used statewide in the most recent election; and the only remedy is a circuit court appeal that resolves long after certification.
What follows from that is narrower than either side usually says, and the narrow version is the one that survives argument. It does not establish that fraud occurs in Oregon. No investigation has produced evidence of fraud that changed an Oregon result, and the state's fraud prosecutions are vanishingly rare — 38 convictions across 60.9 million ballots between 2000 and 2019, according to the Legislative Fiscal Office. Critics answer, fairly, that a conviction count measures what was detected and charged rather than what occurred. Both statements are true, and neither settles the question.
What it establishes is that the question cannot be settled from outside. If ballots were mishandled at these stages, no member of the public would be positioned to see it, no recording could preserve it, and no records request would reach the rules under which it happened. An assurance that nothing is wrong, offered by the only party in a position to know, is not a check. It is a claim — and it is the same category of claim as its opposite.
The claim we are publishing as a claim
Critics of Oregon's system, including the poll worker whose account opens this article, argue something stronger: that this combination of laws is not coincidence but design, an accountability structure built so that no level of Oregon government can be held to account for how elections are run. Related arguments are made about public-sector employment protections — Oregon's Public Employee Collective Bargaining Act, in force since 1973 and administered by the Employment Relations Board, and the just-cause and progressive-discipline provisions standard in the collective bargaining agreements covering county staff — as a further layer insulating the people who handle ballots.
We are publishing that as an attributed claim, not as a finding, and readers should treat it that way. We did not examine the legislative history of any of these provisions, and reasoning backwards from an effect to an intent is how a great deal of bad election writing is produced. We also did not establish which Oregon counties' elections staff are unionized or what their agreements provide. What we can say without guessing is that the provisions are not administrative drift: each has an author, a date and a recorded vote. ORS 254.482's watcher-credential requirement is 2001 Oregon Laws chapter 805, amended in 2005. The unpostmarked-ballot presumption arrived with House Bill 3291 in 2021. And Senate Bill 589 of 2023, which authorised a live public video feed of the rooms where ballots are tallied, carried its own repeal in section 2 of the same act — the Legislature scheduled the end of that transparency in the moment it created it, effective July 1, 2026. Who wanted each provision, and why, is answerable from the legislative record by anyone willing to read it. We have not, and until someone does, intent stays a claim.
What we could not substantiate
The poll worker's account is one person's, unnamed here, describing one county on one occasion. It is evidence of what happened to them and it is consistent with what the statutes permit. It is not evidence of a pattern, and we did not obtain the county's observer records or the accounts of other Oregon observers.
We could not establish whether any Oregonian has ever used the ORS 246.910 appeal — which names county clerks explicitly — to challenge a denial of observation. If the answer is nobody, that is itself the story, and we could not confirm it either way.
We could not determine whether the Senate Bill 589 video pilot was ever funded or operated before its repeal took effect this July. Those are very different stories and we have neither.
We could not locate an equivalent statewide hand-count directive for the November 2024 general election. The finding that every county was directed onto the post-certification audit path is documented for the May 2026 primary and only for it, which materially affects how much that finding carries.
Separately, claims circulating in Oregon that we examined and could not support: a widely repeated assertion that the state admitted roughly 1,500 non-citizens voted. Oregon's admitted figure is approximately 1,600 people wrongly registered through a Driver and Motor Vehicle Services error discovered in September 2024, of whom 38 had any voting history — and the DMV states many of those 38 were in fact citizens whose records were stale. The state's own conduct after that discovery is a fair target: three cases were referred for prosecution in February 2025, a senior assistant attorney general declined them in September 2025 partly because the three had been singled out after hiring lawyers or declining to answer self-incriminating letters, he found the Elections Division was "sorting cases without well-defined protocols or guidelines," neither agency announced the outcome, and the Secretary of State's office says it has no plans to refer further cases. That sequence is documented. The 1,500-voting figure is not.
