Project MKUltra, and the caption instead of the
document.

The boundary

The question asked for a proportion. The ledger has no denominator.

The run’s own question was quantitative — to what extent did MKUltra’s documented research involve non-consensual experimentation. Answering that needs a numerator and a denominator, and the run recorded, before searching, that the denominator might not exist.

The scope note says why. Most MKUltra records were ordered destroyed in 1973, so the run wrote the ambiguity down in advance: conclusions differ depending on whether the question is scoped to the surviving record or to the presumed full extent. It also recorded that the subproject count and the subject totals vary between the Church Committee, the Rockefeller Commission and later FOIA releases — which would change any quantitative answer on its own.

The population was bounded to witting and unwitting human subjects in CIA-sponsored behavioural, psychological and pharmacological research, together with the funded researchers and institutions. The period runs 1953 to 1973, with BLUEBIRD and ARTICHOKE treated as background rather than population, and the 1975–77 investigations as aftermath.

Fifteen subquestions were written before any source was read. Five were answered by a passage. The other ten include every question about objectives, methods, subproject and institution counts, named cases of covert dosing, consent procedures, the Nuremberg Code, the 1973 destruction order, and what the two commissions concluded.

The question asked how much. The ledger came back with one meeting.

Competing explanations

Three explanations, and twenty-one bearings from one page.

Three rival accounts were written down before any source was read, each with the observation that would destroy it. Seven passages went on to declare a bearing on all three.

H1 is the familiar account: substantial, documented non-consensual experimentation, oversight deliberately bypassed, no meaningful consent process across most subprojects. H2 is that the extent is overstated — that most surviving records describe witting volunteers and contract university science, and the unwitting-subjects story generalises a few publicised cases across a hundred and fifty. H3 is a claim about the era rather than the Agency: that formal human-subjects protections did not yet exist anywhere in U.S. practice, so MKUltra’s lack of ethical review was typical rather than exceptional.

Every one of the seven passages bore for H1 or was neutral. None bore against it. H2 collected six weak negatives and one neutral; H3 four weak negatives, two strong ones and one neutral. On the counts alone this looks settled.

It is worth being precise about what produced those counts. All seven passages come from the same document’s description page, and each was scored against all three explanations, so one archival summary generated every cell in the matrix. The one-sidedness is real. Its independence has not been established, and the run does not claim it has.

Twenty-one bearings, and a single page produced every one of them.

The whole ledger

Four sources were stored. Three of them produced nothing.

Every quotation on this page, and every finding in the run, traces to the fourth: the National Security Archive’s page for a two-page memorandum written by the CIA Inspector General on 29 November 1963.

That memo is a record of one meeting, in the office of Deputy Director of Central Intelligence General Marshall Carter, held to settle the single point of disagreement over the Inspector General’s MKULTRA recommendations — whether to continue testing MKULTRA substances on unwitting U.S. citizens. Seven passages were read out of the page. Six findings were asserted on them. Nothing else in the run’s evidence section exists.

The run measured this itself and printed the result. Zero of four snapshots reached an upstream origin, and every asserted finding stands on exactly one distinct source family. Its confidence stage read that as unknown for all thirteen findings — not low, unknown, because a source whose provenance was never established cannot be shown to be independent of anything.

One item shows how thin it gets at the edges. The passage about what happened after the freeze — the 1975 cover memo’s authors finding “no record … that this freeze was ever lifted” — is scored neutral on all three explanations, and the run’s own note on it treats that same sentence as concerning a records-destruction freeze rather than the testing freeze. The ledger contradicts itself about what its last passage is even about. The run prints both readings rather than choosing, and observes that on either one, no record of the freeze being lifted is not a record of the testing stopping.

Six findings, seven passages, one page, zero traced origins.

Where the passages sit

What was quoted is the archive’s caption, not the memo.

Look at where the seven passages are located: paragraphs one, two and three of the page’s Description — the summary an archivist wrote about the document, not the document’s own bytes.

Read them and it is plain enough. “This memo records a meeting held in the office of Deputy Director of Central Intelligence Gen. Marshall Carter…” is not the memo speaking; it is a curator saying what the memo does, with the memo’s own phrases quoted inside. “Earman finds this ‘odd’” is the same construction. What the verbatim gate certified — all seven passages survived it, a clean hundred-per-cent round trip — is that the run quoted the description accurately. It certifies nothing about the memorandum.

The run’s own source assessment names the gap without flinching: the memo’s conclusions are narrow, a record of one meeting, while the archive’s framing “draws broader inferences about CIA policy and testing practices that extend beyond what this single document directly documents.” Several findings on this page then take the archive’s framing and step further again.

None of that says the description is wrong. The National Security Archive is a serious institution, the memo is real, and the names in it — Gottlieb, Helms, Earman, Carter, Kirkpatrick — are the right names. It is a claim about what has been checked. The digest printed in the source record below is the digest of a web page. It is not the digest of a CIA memorandum, and nothing in this run ever opened one.

The verbatim gate certified that the caption was quoted correctly. That is all it can certify.

The winner, audited

The explanation that leads still needs two things nobody found.

H1’s required conditions were written down in advance — all of them have to hold — and the run’s own open questions record which of them this ledger reached.

The oversight condition is the one the ledger speaks to, and it speaks clearly. The Inspector General’s review produced the disputed meeting; the meeting produced an agreement that continuation would need escalation; the Director then deferred for a year and directed that the Agency continue the freeze. That sequence is fixed in the documents, and a freeze imposed against the advocacy of the men running the programme is a control, not a gap.

The two quantitative conditions are the ones the run files as unresolved. It states plainly that there is no subject-level record here, no consent documentation, no subproject inventory and therefore no denominator, so it cannot say what share of the roughly 150 subprojects involved covert administration rather than witting volunteers, contract university science or animal work.

Then the run wrote down the reading that would break its own leading finding, which is the part worth carrying away. The only concrete venue attached to unwitting testing anywhere in this ledger is the domestic safehouse operation, and the extension to foreign nationals was rejected for want of controlled facilities. If “unwitting testing” in that 1963 meeting meant that one operational line rather than a practice spread across subprojects, then a policy fight at the Deputy Director’s level is entirely compatible with H2 — and the run says so, against itself.

A fight in the Deputy Director’s office proves the practice was managed. It does not say how big it was.

What was stored and not read

The denominator was already in the ledger.

The CIA’s own January 1976 briefing book — 149 numbered subprojects with their researchers, objectives, funding and monitors — was fetched, hashed and stored by this run. No passage was ever read out of it.

That document is the closest thing in the ledger to the denominator the run says it lacks. It sat in the source list, snapshotted and assessed, while the findings above reported that no subproject inventory was available to them. The run does not hide this; it prints the source and the zero beside it.

The Church Committee’s final report was stored too — a year-long investigation, roughly 110,000 pages of documents reviewed, some 800 witnesses, about 30,000 pages of transcript. It produced no passage. So did the Advisory Committee on Human Radiation Experiments report, which is precisely about federally sponsored human experimentation from 1944 to 1974 and the human-subjects protections of that era: the question H3 turns on.

There is a second reason to be careful with the briefing book, and the run recorded it. The stored copy is an Internet Archive OCR transcription with no named transcriber, no verification procedure and no editorial process, which is why the run assessed it at the bottom tier of the evidence hierarchy — unsupported, the same rank as an anonymous social-media post. It is the right document arriving in the wrong condition. Whether reading it would have moved anything is unknown; what the record shows is only that it was there.

A list of all 149 subprojects was in the ledger. Nothing was read out of it.

The counter-search

Eight searches went looking for the other side. All eight came back empty.

The falsification pass ran eight adversarial searches — two against H1, three against H2, three against H3 — and not one returned a snapshot-bound, verified passage.

Two hunted for what would have supported H2’s quantitative claim: documentation that most subprojects ran on consent or witting volunteers, and findings from either commission describing standard ethical oversight across the programme. Three hunted the opposite way — for a subproject inventory establishing systemic unwitting dosing, for an analysis of the roughly 20,000 recovered financial records, and for the consent status of patients in the Allan Memorial Institute subproject. Three went at H3: internal memos acknowledging the Nuremberg Code, the 1953 Wilson consent policy and whether the Agency evaded it, and whether the 1963 IG report treated unwitting testing as a legal violation rather than a public-relations hazard.

Nothing came back. The run scores that as a falsification yield of zero out of eight, and prints it beside a note that matters more than the number: no probe was pre-registered in this run, so no absence anywhere in the record counts as evidence of absence. Eight empty searches are eight empty searches, and the run refuses to read them as eight small confirmations.

The second retrieval round returned nothing at all — no new snapshots, no new source families. By the pipeline’s own rule that counts the search as saturated, and the last chapter is where the bill for that arrives.

Eight searches for the other side, and eight blanks that prove nothing.

The best-supported reading

The leading finding, printed with the attack made on it.

The run’s first finding is typed observed rather than inferred. The attack it filed against itself goes at one word in it: programme.

The finding reads the ledger as favouring H1 at the level it actually speaks to. Unwitting administration of MKULTRA substances to U.S. citizens was a standing programme element, deliberated in the Deputy Director’s office, defended by Gottlieb and Helms, opposed by Earman, Carter and Kirkpatrick, and headed for the Director if the Directorate for Plans insisted it was operationally necessary. Something requiring that much adjudication, and a formal freeze, is a managed policy rather than an accident of oversight.

The attack is filed as causal overreach, and it is the obvious one. An executive meeting and a freeze are also exactly what crisis containment looks like. If the Deputy Director convened that meeting to contain isolated, unauthorised safehouse experiments the Inspector General had just surfaced, every observation inside the finding still holds and the conclusion drawn from it does not.

Seven attacks are on the record against six asserted findings and one hypothesis. One argues that the internal objections were about security vulnerability and diplomatic blowback rather than ethics. One argues that the freeze was never checked against what field practice actually did afterwards. One argues that limiting unwitting testing to the safehouses assumes the deliberation surveyed every venue. This run does not record whether any person reviewed any stage of it, and the page says so rather than implying otherwise.

An executive meeting and a freeze are also exactly what crisis containment looks like.

The score

Fourteen out of a hundred — with full marks for a search that returned nothing.

Six findings asserted, two of them observed. Every one computed to unknown. The evidence score landed on the same 14 out of 100 as everything else in this archive, but the components got there by a different route.

Corroboration and provenance both scored zero, for the reason they always do here: not one of the four snapshots reached an upstream origin, and unknown provenance is unknown independence. Falsification scored 14 of 20 — for running eight counter-searches, not for what they found, which was nothing.

Saturation scored 15 out of 15, a full mark this archive has not printed before, and it is worth seeing exactly what earned it. Round one found three source families. Round two found zero snapshots and therefore zero new families, and no explanation’s counts moved when it was added. A measure of whether more searching would turn up anything new returns its most reassuring value when the second search returns nothing at all. The measurement is doing precisely what it was defined to do; it is the definition that should be read twice.

The components summed to 29 and the published score is 14, because one source could not be traced to an origin and that caps the unknown band. The single measurement that came back clean is the verbatim gate: seven of seven readable passages survived it, none refused, none unable to be checked. What it certifies is that the quotations match the stored bytes — and the stored bytes are an archive’s description page.