1. Historical accountability requires more than a reassuring form
A signed form, an institutional logo and a committee reference can make a research project look settled. None tells the whole story by itself. The important questions concern relationships: what the participant was told, which activity was reviewed, what actually happened, and how the institution responded when its records or conduct diverged from the plan. An oversight system becomes visible through those relationships, not through the number of official-looking documents in a folder.
The August 3, 1977 Senate hearing on MKULTRA includes CIA testimony identifying subprojects involving tests on people who were unaware they were being used as subjects. The hearing is a historical primary record of that account, not permission to invent the experiences of unidentified individuals or infer that every later study repeats the same conduct. Its relevance here is specific: the existence of a research purpose or institutional sponsor does not answer whether the people involved understood and chose their participation. Senate joint hearing, August 3, 1977.
It would also be misleading to give one scandal sole credit for the entire modern system. The Belmont Report's own introduction identifies the July 12, 1974 National Research Act as creating the National Commission. The 1979 report articulates respect for persons, beneficence and justice, connecting ethical consideration to consent, risk and benefit, and selection of participants. That chronology predates the 1977 hearing in an important respect. This article therefore does not claim that the hearing created the commission or that one historical episode fully explains subsequent regulation. HHS publication of the Belmont Report.
The practical question is what a reader can establish from a contemporary research packet without overstating either its safeguards or its failures. The examples below use a wholly fictional Harbor study and hand-authored records. No people were recruited, no experiments were conducted, and no real institution is being accused of noncompliance. The local computation checks consistency among invented dates, identifiers and counts. It does not determine whether a real study is ethically acceptable or legally authorized.
That boundary matters because a checklist can be mistaken for a substitute for judgment. A date comparison can reveal that two records conflict. It cannot tell us, without further evidence, whether a clock was wrong, a form was entered late, or an activity occurred without the required process. Conversely, a clean comparison does not establish understanding, freedom from pressure or scientific value. The point of a careful packet review is to identify exactly what the documents support and where further inquiry is necessary.
2. Identify the applicable framework before declaring compliance
Under the HHS Common Rule, consent requirements for covered research include understandable information and an opportunity to discuss participation; the text also provides specified pathways for waiver or alteration. IRB approval criteria address matters including minimized risks and equitable selection. These are scoped requirements, not a claim that every activity involving people follows an identical rulebook. Reading an exception's title is not enough to establish that its conditions are satisfied. 45 CFR 46.116, 45 CFR 46.111.
For Harbor, the fictional cover memo states that its review concerns an adult memory-task study at two named sites during January. It identifies the project as falling under the framework specified in that memo. We accept that premise only to perform the exercise; we have not independently classified an actual activity. A real reader would need the basis for such a determination, not just a self-description on a website. A project number is a useful locator, but it is not the reasoning behind a coverage decision.
Imagine finding three different cover sheets. One describes the memory task, another a clinical device investigation, and a third an anonymous public survey. All use the same institutional letterhead. Their appearance does not make their review requirements interchangeable. A policy applicable to one cannot be imported into another merely because both involve questionnaires. The exact activity, jurisdiction, funding or regulatory context, population and dates may change which documents are relevant. This article does not supply a universal classification tool or advice for starting a study.
The file should distinguish a determination that review was required from the decision made during that review. A memo could conclude that a particular activity is outside a specified policy's scope; a different memo could approve a covered activity subject to conditions. Those are different propositions. Describing both as “ethics approved” would conceal the distinction. Equally, absence of one familiar form does not establish wrongdoing if that form was not required in the applicable process. The premise must be established before the paperwork can be meaningfully assessed.
Dates are part of that scope. A rule or guidance document read in 2026 may have been revised after the conduct under examination. A historical article should not silently apply the current text to an earlier project and declare a retrospective violation. A contemporary article should not quote an obsolete provision as if it controlled today. The relevant version is evidence, just like the relevant project identifier. A careful account records both instead of assuming the latest search result solves every chronological question.
3. An IRB number is not a project approval
OHRP's registration page explicitly cautions that IRB registration does not mean the office has determined that the board complies with the human-subject protection regulations or has the expertise to review a particular project. Registration records required information about the organization and board. It is not an endorsement of every study that mentions the registration number. That limitation is stated by the agency itself, not inferred from an allegation about any particular board. OHRP IRB registration explanation.
Harbor's packet contains a fictional registration entry for Board R and a separate decision letter for Project H. The first establishes the board identity within the example. The second identifies protocol P1 and consent document C1. A third document, a website announcement, says Harbor is launching “advanced memory research.” That broad phrase does not tell us whether the announced activity matches P1. The review must connect the public description to the specific documents rather than allowing the institutional identifier to stand in for every missing link.
Suppose the decision letter applies to Site North, but an enrollment spreadsheet includes Site South. That discrepancy could have several explanations. A later amendment may add Site South; the spreadsheet may combine two studies; or the site label may be an administrative error. The packet needs the relevant clarification. A reviewer should flag the mismatch without immediately selecting the most alarming explanation. But it would be equally wrong to remove the site column to make the approval appear comprehensive.
Version labels need similar care. Protocol P1, consent C1 and recruitment flyer F1 are different objects, even though they share a numeral. The protocol may define procedures, the consent material may describe them for participants, and the flyer may invite inquiries. A decision referring to P1 does not, from its identifier alone, tell us the contents of C1 or F1. The packet should show the relationship actually recorded. A file named “approved_final_v2” is not a substitute for the decision that made a particular version operative.
For this exercise, Harbor's fictional authorization ledger makes C1 active from January 2 until January 11, when C2 replaces it for new recorded signatures. That exact window is a premise of the invented packet, not a universal regulatory timetable. It gives the date checker something concrete to compare. If the ledger instead allowed a transition period, the expected result would change. The software must apply the stated rule rather than quietly imposing a researcher's preferred interpretation.
The resulting evidence question is modest but useful: does the available decision identify the version and site that appear in the activity record? A match supports that connection. A mismatch identifies a document or explanation still needed. Neither result alone tells us whether the underlying study is beneficial, whether participants understood it, or whether every other institutional duty was met. Those conclusions require their own evidence rather than inheriting the authority of a number.
4. The conversation cannot be inferred from the signature alone
FDA's August 2023 final informed-consent guidance describes an ongoing exchange rather than a process ending when a form is signed. It also discusses recruitment information, questions and the role of the document. This article uses that narrow explanation, not the guidance as an exhaustive statement of all later regulatory changes or exceptions. Its scope is FDA-regulated clinical investigations; Harbor remains an invented illustration, not a claim that all its details fall within FDA jurisdiction. FDA informed-consent guidance.
Consider two fictional records with the same signature date. In the first, an accompanying note records a participant asking whether a memory score would be used in an employment decision, followed by the specific answer given. In the second, the note simply says “questions answered.” The first gives a later reader more information about that particular exchange. It still does not prove that every explanation was understood or that the recorded answer was delivered exactly as written. The second is less informative, but its brevity is not proof that no conversation occurred.
A useful inquiry stays with the content of the uncertainty. If the study description says results will be used only for the stated research purpose while a recruitment email promises an individual diagnostic assessment, the contradiction can be identified directly. The problem is not solved by choosing whichever document sounds more official. The investigator's actual explanation, the reviewed materials and any correction to the email would matter. A reporter need not invent a participant's reaction to show that the two descriptions differ.
The same applies to comprehension exercises. Suppose Harbor asks twenty fictional respondents a question about whether the study promises a personal benefit. Twelve answer correctly under a prewritten scoring key. That is twelve of twenty, or 60 percent, on that question in this exercise. It is not a validated measure of informed consent, a diagnosis of capacity or evidence that the other eight should be excluded from research. The question and scoring rule cover one limited proposition, and the result should be labeled accordingly.
Now imagine an explanation is revised and a later check obtains sixteen responses, twelve correct. The reported proportion rises to 75 percent. Yet the number of correct responses is unchanged, and four people are missing from the later denominator. Without person-level matching and information about the missing responses, the result does not establish that the revision improved anyone's understanding. Among the original twenty, the later count could range from twelve to sixteen correct if the missing four were evaluated. That gives 60 to 80 percent, not a measured improvement.
A responsible account can still describe the revised wording and the observed results. It should not turn a higher percentage into proof that an ethical problem has been solved. The important next step in this fictional evaluation would be to establish who answered both times and what changed in their responses. That is a specific information need, not a demand for a longer form or a more impressive slogan about transparency.
5. Eight records show why a clean folder can mislead
The Harbor fixture contains eight invented participant identifiers, H01 through H08. Seven have a scanned signature record; H07's signature time is unknown. The folder holds nine signed-scan files because H01 and H02 each appear twice. Counting files would suggest more completed documentation than there are people. Deduplicating by the stated participant identifier produces seven people with a signed scan, not nine. The missing value remains unknown rather than being converted to a convenient timestamp.
The first three records use C1 on January 5, January 6 and January 10, respectively. Each records a signature at 9 a.m. and a session at 10 a.m. H05 uses C2 on January 12 with the same ordering. These four match the fictional authorization windows and recorded sequence. The checker labels them chronology-and-version matches only. It does not label the participants fully informed, the study lawful or the recorded discussion adequate. Those facts were not encoded in the input.
H04 uses C1 on January 12, after the stated replacement date. H06's C2 signature is recorded at 11 a.m. on January 13 while the session is recorded at 10 a.m. H07 lacks a signature time for the January 14 session. H08 uses a document labeled C2-ES on January 14, but that document is absent from the packet's authorization map. Each presents a different reconciliation question. Combining them into one category called “no consent” would claim more than these fields establish.
For H04, the researcher needs the actual transition decision and the document used. For H06, the timestamps and their meaning require examination: a time of scanning would not necessarily be a time of signing. H07 requires whatever documentation or applicable determination explains the missing field. H08 requires the relationship between C2-ES and the reviewed materials. A language suffix is not evidence, by itself, that a translation was reviewed or that it was defective. The fixture records absence from the supplied map, not absence from reality.
The code was actually run on these hand-authored records. Four rows matched its narrow criteria and four required reconciliation. Deliberate tests also introduced a duplicate person identifier, a scan referring to nonexistent H99, and an impossible recruitment count in which completions exceeded people reached. The validator rejected all three. These tests show that the local checker detected the intended input errors. They do not validate an actual research program or establish a general error-detection rate for every possible record system.
The distinction between a data flag and a finding is essential. If a later original record establishes that H06's 11 a.m. value was the scan time, the chronology assessment may change. That is a correction based on evidence, not permission to replace the inconvenient value silently. The original observation, clarification and revised conclusion should remain distinguishable. Otherwise the audit trail becomes a record of what the reviewer wanted to see rather than what the packet contained.
6. A change log should explain consequences, not merely list files
Suppose C2 differs from C1 in three specified ways within the fictional Harbor packet. It changes a contact number, replaces a vague statement about data retention with an exact interval, and clarifies that no individual clinical interpretation of the memory score is promised. These are invented content differences, not quotations from an actual consent form. A comparison that reports only “three edits” conceals their very different practical significance. The number of changed sentences is not a measure of their importance.
The contact-number change raises a delivery question: which materials reached people who might need to use it? The retention change raises a consistency question: do the storage plan and participant-facing explanation say the same thing? The benefit clarification raises a communication question: did earlier materials make a different promise? A useful change log connects each edit to those consequences. It does not assume that every revision requires the same response, and it does not decide a real study's obligations without the applicable review decision.
For Harbor, imagine a fictional implementation memo explicitly directing that C2 be used for new signatures starting January 11 and that the revised contact card be delivered to all eight listed participants. The date checker can test the first instruction, but it cannot test delivery of the contact card without another record. If the team sends eight emails, the send log establishes eight attempted transmissions. If two bounce, it does not establish eight successful deliveries. If six are delivered, it still does not establish that six people read them.
The distinctions lead to concrete numbers. Eight attempted messages minus two failed deliveries leaves six recorded deliveries, or 75 percent of the eight intended recipients. Suppose two replacement calls are attempted and one person answers. That does not make the record complete: one intended recipient still lacks documented contact under this fictional memo. A dashboard that marks a task finished when a message is sent would hide the unresolved case. A dashboard that tracks the actual requested outcome would not.
Timing can change interpretation without changing content. A participant may receive the revised explanation after the next scheduled session rather than before it. Whether that sequence is acceptable depends on the relevant facts and decision, but the sequence itself can be reported accurately. The file should not use the date of uploading a new template as evidence that every participant received it on that date. Template publication and individual communication are separate events.
A well-maintained change log therefore records both the revised object and the actions actually taken in response. It can show that one correction was completed, another remains unresolved and a third was determined not to affect existing participants under the documented decision. That mixed result is more useful than a single green status applied to the entire amendment. It also permits later review without asking someone to reconstruct the history from whichever document happens to be newest.
7. Scientific usefulness and participant burden need different measurements
An oversight discussion can become vague when every consideration is compressed into “benefit.” Harbor's fictional protocol seeks to compare two ways of presenting a memory task. Its recruitment material does not promise diagnosis or treatment. A proposed revised analysis plan, however, drops half the scheduled follow-up observations without explaining how the original comparison will be made. That is a methodological question with practical implications: the reader needs to understand what the changed design can still answer before repeating a claim about its value.
Imagine that the original plan specifies forty observations, twenty per presentation method, and each observation requires thirty minutes of participant time. That totals twenty participant-hours. A revised plan retains forty visits but records the primary outcome at only twenty of them. The burden remains twenty participant-hours, while the number of primary outcome observations falls by half. This does not by itself establish an unacceptable study. It does identify a discrepancy between the planned information and the work being asked of participants.
Now compare a different proposal: twenty visits, all measuring the primary outcome, each lasting thirty minutes. The burden is ten participant-hours. Both revised proposals yield twenty primary observations under these invented premises, but they require different amounts of time. The comparison is not a complete ethical assessment; it omits uncertainty, other useful data and the quality of the design. It is a concrete way to identify what an explanation should address rather than accepting the word “efficient” without a denominator.
The measures must not be combined carelessly. An hour of waiting, a disclosed discomfort and a potential privacy exposure are not naturally interchangeable units. Giving each a number and adding them into an “ethics score” would require a justified valuation scheme that this packet does not supply. Nor does a high expected publication count automatically offset an individual burden. A public report can describe the separate considerations without inventing a numerical exchange rate between them.
Suppose Harbor's summary says the revised design is “twice as efficient.” The evidence could support a narrow statement that twenty primary observations require ten rather than twenty participant-hours in the second proposal. It would not establish twice the scientific reliability or half the total risk. The outcome, numerator and denominator belong in the explanation. If other aspects changed, those should be reported rather than hidden behind a ratio.
This kind of analysis remains useful even when the final judgment belongs to qualified reviewers. A journalist can ask how the stated purpose connects to the design and whether the described burden matches the schedule. A reader can distinguish a direct personal benefit from the hoped-for value of knowledge. Neither needs to pretend that a spreadsheet can replace the substantive scientific and ethical assessment. The point is to expose the inputs to that assessment, not to manufacture a verdict from incomplete arithmetic.
8. Recruitment numbers can conceal who never reached the conversation
Harbor's fictional recruitment log invites forty people through Channel A and forty through Channel B. Thirty from A reach the information discussion and eighteen complete the enrollment process. Ten from B reach the discussion and six complete enrollment. Among those reached, both channels show a completion proportion of 60 percent. A report focusing only on that conditional rate could conclude that the channels performed identically. The full invitation denominator tells a different, narrower story.
Eighteen of forty invited people in Channel A complete enrollment, or 45 percent. Six of forty in Channel B do so, or 15 percent. The threefold difference arises in this example from different reach, not different completion proportions after contact. We have not supplied a reason for the reach gap. It could involve delivery failures, scheduling, eligibility, interest or other factors. The arithmetic identifies where to investigate; it does not establish discrimination or tell the study to maximize enrollment regardless of people's preferences.
A better log distinguishes an undelivered invitation from a delivered invitation declined, a request for information from an enrollment decision, and a canceled appointment from a withdrawal after participation began. Those categories describe different experiences. Treating every non-enrollment as “refusal” can erase a practical barrier. Treating every refusal as a barrier can erase a person's choice. The analyst should use the category supported by the record rather than the category that makes the program look either successful or exclusionary.
Suppose the Channel B log contains four undelivered messages, five scheduling conflicts and twenty-one people with no recorded response. Those thirty entries explain why only ten were reached in the fictional count, but the twenty-one unknown responses remain unknown. Assigning all twenty-one to lack of interest would manufacture information. Assigning them all to exclusion would do the same. A bounded report can state that the contact process has unresolved records without pretending to know the motives or circumstances of people who did not respond.
These figures also do not establish that the enrolled group represents a broader population. The example has no random sampling design or population benchmark. An enrollment proportion can describe the pathway through this invitation list while saying little about whose experiences the resulting study will capture. If a public claim extends beyond the people studied, the scientific justification for that extension needs to be examined separately.
The concrete editorial gain is a more accurate sentence: the two channels had equal completion proportions among people reached, but different proportions among everyone invited. That sentence invites a focused question about contact and access without assuming its answer. It is more informative than saying participation was “balanced” because one selected percentage matched, or calling the entire process unfair because another did not.
9. An incident can generate several reports without becoming several events
OHRP's September 2022 incident-reporting guidance addresses specified categories including unanticipated problems, serious or continuing noncompliance, and suspension or termination of approval in its applicable research context. It allows an initial report to be followed by further information and describes review of corrective actions. The guidance does not provide one fixed reporting deadline for every circumstance. This is a scoped description, not a determination that a particular fictional event below meets a regulatory reporting category. OHRP reporting guidance.
Harbor's incident folder contains four notices. N1 is an initial account of Event E1. N2 adds information about E1. N3 records a response concerning E1. N4 concerns a separate Event E2. The folder therefore contains four notices about two events. If a dashboard counts notices as incidents, it doubles the event total in this fixture. If it keeps only the last notice for E1, it may erase what was known at the start and why the response changed.
Suppose the initial E1 notice says a data file may have been sent to an unintended address. The follow-up says the transmission failed before delivery. The response records a correction to the address list. These are hypothetical facts. A final account should distinguish the initial concern, the later evidence and the corrective step. Describing E1 simply as a confirmed disclosure would exceed the final packet; describing it as “nothing happened” would conceal the documented process failure and investigation.
E2 might have a different evidentiary trajectory. The notice could remain unresolved because the delivery log is unavailable. It should not inherit E1's resolution merely because both involved email. Grouping cases by topic can help organize them, but it does not supply missing facts. Each event needs its own identity, chronology and status so that a correction in one does not silently clear all related entries.
The same distinction applies to responses. Drafting a corrective procedure, approving it, distributing it and confirming that it is in use are separate actions. A promise to retrain staff is not evidence that training occurred. A training attendance list is not evidence that the problem cannot recur. The reporting packet can document progress without calling every planned response completed or treating every remaining uncertainty as proof of deliberate misconduct.
10. Oversight effectiveness is a question to measure, not a badge
GAO's January 2023 report examined federal oversight of IRBs and the challenge of evaluating their effectiveness. As checked on October 30, 2026, its recommendation to convene stakeholders on effectiveness approaches is marked closed and implemented. The accompanying explanation credits consultations while noting that new approaches had not been implemented at the time described. That status does not establish that all boards are effective or that every oversight concern in the report is closed. GAO report and recommendation tracker.
A fictional evaluation makes the distinction tangible. Board R reviews one hundred submissions in a year and requests revisions to sixty. Board S reviews twenty and requests revisions to fifteen. Their revision proportions are 60 percent and 75 percent. It would be premature to rank S as more protective or R as more efficient. The mix of submissions, the significance of requested changes and the later outcomes are unspecified. A count of interventions is not a measure of their value without additional context.
Now suppose R's sixty revisions include forty corrections to contact information and twenty changes to study procedures. S's fifteen include fourteen contact corrections and one procedural change. That extra detail improves description but still does not determine which board produced better participant outcomes. A contact error can matter greatly in one situation and be minor in another. A procedural change can be beneficial, unnecessary or inadequately implemented. The relevant question is what the change did, not whether it fits a category with an impressive name.
Review speed presents another denominator problem. One office reports a median of ten days from receiving a complete submission to its first decision. Another reports twenty days from initial submission to final authorization. The second measure includes stages the first may exclude. Comparing them as though both measured the same interval could reward a reporting convention rather than a genuinely faster process. The event definitions belong alongside the numbers, particularly when a claim about institutional quality depends on the comparison.
Harbor's packet can contribute evidence to such an evaluation without pretending to settle it. It identifies four chronology or version questions and records how each was investigated. If the same error repeatedly recurs after a claimed correction, that is a specific observation worth examining. If a different data-import method merely produces more flags, an increase in the flag count may reflect detection rather than deterioration. A meaningful assessment separates the underlying process from the instrument used to observe it.
11. What should a public evidence packet reveal?
A public account of Harbor does not need to expose hypothetical participants' signatures, personal contact details or complete private conversations to make its central points. It can publish the invented identifier scheme, the stated authorization windows, the date comparisons and the aggregation rules. A real publication would need a careful assessment of privacy and relevance. Transparency about a process does not require broadcasting sensitive information that is unnecessary to understand the finding.
The report should distinguish documents inspected from documents merely referenced. If a decision letter mentions an attachment that was not supplied, say so. If a participant-facing document was available only as a later revision, do not present it as proof of what earlier participants received. If a summary says a discussion occurred, attribute that statement to the summary unless the underlying record was checked. These distinctions keep a thin evidentiary chain from becoming thicker merely through retelling.
For the fictional H08 record, the public statement could be that the supplied approval map does not identify C2-ES and further documentation is needed to reconcile that version. It should not say that a real Spanish-speaking participant lacked consent or that translation itself is suspicious. For H06, it can identify the recorded one-hour reversal and the need to determine what the timestamps represent. Precise wording preserves the concern while avoiding a personal allegation the data cannot support.
The packet should also document its calculation rules. Nine scan files map to seven people, eight participant records produce four narrow matches and four reconciliation flags, and four notices map to two events. These results are reproducible because the invented inputs and the checker were retained. Reproducibility here means that another reader can obtain the same outputs from the same exercise. It does not mean that the outputs certify real-world conduct or exhaust every question a qualified reviewer might ask.
If a correction arrives, identify which conclusion changes. A missing authorization for C2-ES might resolve H08's version question without resolving H07's missing time. Clarifying H06's scan timestamp might resolve chronology without establishing the content of the earlier discussion. A corrected count should not erase the distinction between resolved and unresolved issues. The public record is stronger when updates are tied to evidence rather than issued as a general assurance that everything is now fine.
12. The lesson is a chain of accountable decisions
The useful response to a troubling research history is neither automatic distrust of every contemporary study nor automatic confidence in every approved-looking packet. Both approaches skip the work of connecting authority, information and action. A serious account asks what was proposed, what was reviewed, what was communicated, what happened and what changed after a problem was identified. Those questions can be answered at different levels of confidence within the same investigation.
Harbor's local review illustrates that unevenness. It verifies a small set of chronology and version matches, identifies distinct unresolved records, and shows how duplicated files or repeated notices can distort counts. It does not establish the entire ethical quality of the imagined project. That limitation is a feature of honest measurement, not a reason to discard it. A narrow result can be useful when its scope is clear and when it points to the next piece of evidence needed.
A reader evaluating a real study should not use this article as individualized legal or medical advice, or as instructions to enroll, withdraw or conduct research. Those decisions require the relevant study team, qualified advisers and appropriate oversight channels. The public-interest task here is different: read the records carefully enough that a registration number does not become a universal endorsement, a data flag does not become an unsupported accusation, and a corrected spreadsheet does not masquerade as proof of participant understanding.
The final standard is whether the account lets someone check its consequential claims. Can they see why a version was treated as applicable, how a count was constructed, which contradiction remains unresolved and what evidence would resolve it? If so, the report has moved beyond ceremonial paperwork. It has made particular decisions and limitations inspectable. That is a more durable form of accountability than either a slogan about progress or an invented continuation of the worst episodes in the historical record.