THE METHOD
how this study works — and why you should believe a word of it
The other pages state findings. This page is where those findings earn the right to be stated. You don't need any of it to use the rest of the site, but you need all of it to trust the rest.
Records, not experiments
Every claim in this study has the same shape: something visible in a working's record predicts whether its result succeeded. Two words in that sentence carry all the honesty. Record: we read what operators wrote, not what they did — diaries, case files, sitting transcripts, session logs. Result: what the record-keepers and independent graders reported happening, not what "really" happened. So this study can show that the way a working is conducted and written predicts the way its outcome got reported — a strong, falsifiable, useful claim about practice. It cannot show that magic is real, and never claims to. Both readings of every finding, the magical one and the mundane-psychology one, fit the data on these pages. We built the study so that EITHER way, the practical levers survive.
The house words, and the trust tiers
Two short glossaries, kept here so the other pages can use these words without re-teaching them each time.
What a working is made of — six words, taught once. A working is one act of magic with an aim; it succeeds if what it asked for came and lasted, and fails if not. Gnosis is the worked state, apart from your everyday head — reached by heat (drum, dance, wine, sex) or by stillness (fasting, fixation, the edge of sleep). Eros is the wanting a working carries — Bruno's word, his bond of bonds; not bedroom eros, the whole pull of desire. The charge is the energy a working gathers before you fire it. The click is the moment something answers back mid-working — living teachers' own word; older parts of this study call it the answer. The cord is what you do with a working after you fire: cut it loose (release), or keep hold of it (clutching — the fault the old warning against lust of result points at). The close is the clean exit — banish, give the license to depart, pay what was promised; an unclosed working is left hanging, and the old books say it bleeds.
Three different things, never one. The wanting (eros), the charge (the gathered energy), and gnosis (the state) come apart in the record and must come apart in the reading: a rote card-guessing sitting can be deep with no wanting in it; a plain-need folk petition can burn with wanting in an everyday head; and Neville's falling-asleep-in-the-wish is a true gnosis that asks for no sign at all. When a page says one of the three, it never means the other two.
The study's own shorthand, used only on this page. A vein is one source of records — a diary, a case file, a session set. The reserve is a sealed set of records held back to test the claims once, blind; the first one was read and now joins the corpus, and the Yorùbá Ifá records are the reserve that stays sealed. The eye is the readers' measured blind accuracy — how often their hidden-ending bets were right. (The instrument — the readers and everything around them — is defined just below.)
The words, in the record's own ink. Gnosis, at full depth: “He lost all sense of the physical, and was only recalled to it by its climax” (the Paris Working, read with the ending hidden — it succeeded). The clutch, caught in the act: “But does this count?” (Crowley's diary, Opus VIII, grading his own result before the world could). And the answer's absence, told honestly: “I felt that this was padding, as there is no such room or bookcase” (an SPR book-test sitter, mid-séance — the working failed). One line each; the full records live on THE RECORDS.
How much to trust a line. Every finding on these pages carries one of four trust tiers. Confirmed: it held up on the sixty-three sealed records nobody had read, opened once. Working-set: found in the main corpus, but it ran lighter on the fresh records. Exploratory: new and suggestive, not yet confirmed. Retired / can't tell: we measured nothing there, or these records cannot test it. THE LEDGER carries the tier on each card, and beside it a plainer verdict in everyday words — validated · half right · too early to tell · didn’t hold up · set aside — for how the specific claim fared. Tier says how much weight the evidence can bear; the verdict says which way it came out. THE PRACTICE grades each instruction with the same tier words.
The readers
The records were read by four AI readers, each schooled deeply in a single tradition — ceremonial magic, chaos magic, folk conjure, and the internal-energy work of the Eastern paths (the cultivator). Each wrote its own reading treatise cold, before seeing any of this project's data or hopes, so that no finding could circle back into the instrument that was supposed to test it. A shared common tongue was added later to align their category boundaries without homogenizing their schools; it raised agreement without touching their verdicts' independence. When these pages say the instrument, they mean everything the study measures with: the readers, the treatises that train their eyes, the seven questions they answer about each record, and the rules that score their answers — the way a telescope plus its mount plus its calibration is one instrument. The readers' expertise is that instrument's heart: when a measurement question stalls, the first consult is a craft-literate read, and the machinery's job is to validate the craft, not replace it.
Blind by construction
the record → its ending cut away → blind calls, nine of them → the bet → independent grading → an evidence tier
Before any reader saw a record, its ending was cut away — outcome sentences masked, recognizable names redacted, assessments stripped. Readers were also asked to report any place they glimpsed an ending anyway; records where that happened are demoted to a provisional tier and say so wherever they appear. An adversarial guesser was hired (early on) to attack our own masking and tell us how much could leak past it; what it found drove a re-masking round and a permanent tier system. The seals on THE ARC and the bet-before-you-break mechanic on THE RECORDS are this discipline, made playable.
The strict rule on half-worked records
The graders had a middle grade available — a working that moved things but did not land the full aim: the healing that held for a night with no word after, the remote view that matched the site but missed its center, the job that surfaced and then stalled. Ninety-five records carry it. Every rate on this site counts those records as failed. That rule was set before the records were scored, and it is deliberately harsh: testimonials tend to stop at the good moment, and a soft middle is where inflated success rates hide. The middle is still shown — THE RECORDS marks each such record half-worked, with the grader's own reasoning in its drawer, and every claim card on THE LEDGER re-runs its gap with the half-worked records set aside and again with them counted as wins. No finding on this site rides on the strict rule; the harshest reading is the one we publish.
The keystone: why readers bet on endings they cannot see
Three things follow, and they shaped the study:
- Accuracy is a license, not a finding. That the readers guess well is not itself the interesting part; it is what earns the lever tables their belief. The blind release-consult became a primary instrument the day its calls went six succeeded, and none that came and crumbled, against unseen endings.
- Accuracy maps the honest boundaries. Scoring the bets by record type found the narration fault line (the eye reads records told from inside, and fails on protocol-framed ink). Scoring them by reading-style (by lens) settled what the cultivator's edge is made of. Every scope statement on THE LEDGER was found this way.
- Accuracy is never reported naked. It can be earned for boring reasons — leaked ink, genre cues, or being a well-calibrated pessimist on a failure-heavy pile of records. So every number is scored against the dumb strategies (always bet hold; always bet fade), with leak tiers, with against-the-grain bets counted separately, and with splice tests — run, and passed (the full telling is under “The spliced record,” below).
The reader's enemies
Four ways a reader can be right for wrong reasons, and the discipline against each:
| enemy | what it does | the discipline |
|---|---|---|
| hindsight | accounts written after the result run about three times hotter on the operator's state — we measured it | score only ink written before the outcome; tier everything else |
| halo | one impressive thing colors every other call | a halo meter pointed at the readers; calls quoted to their ink |
| recognition | knowing the famous case means knowing how it ends | redaction, masking, and a recognized-source flag the readers must raise against themselves |
| home doctrine | a school sees its own teachings wherever it looks | four schools instead of one; cross-school agreement measured; doctrine-conformity checks |
Who graded the verdict — the study’s own hostile witness
Before believing any lever, read this table. The success-rate tracks who judged the outcome, almost step for step: the nearer the grader stands to the operator’s own hopes, the more the record succeeds. A book written to persuade succeeds nearly nine times in ten; a blind judge scoring a remote view, one in four. The plainest mundane reading of this study is that we predict what people wrote down, not what happened — and this table is that reading, drawn in full, by us.
| who graded the outcome | succeeded | graded | success-rate | |
|---|---|---|---|---|
| published to persuade | 40 | 46 | 87% | |
| the practitioner’s own pen | 49 | 79 | 62% | |
| told to a collector | 23 | 29 | 79% | |
| an anonymous reader’s report | 21 | 51 | 41% | |
| a periodical or casebook | 22 | 43 | 51% | |
| sworn testimony against the operator | 25 | 38 | 66% | |
| a machine’s transcript | 8 | 16 | 50% | |
| judged blind (remote viewing) | 8 | 32 | 25% | |
| unlabeled (older records carry a coding-status label here) | 47 | 125 | 38% |
A census view like this is pooled across both record sets and all traditions — a picture, never a claim. It is also exactly why every lever on THE LEDGER is tested within tradition and within record set: a rule only counts when it separates succeeded from failed among records that share the same kind of grader. The gradient cuts the base rates; it does not, by itself, cut the gaps.
The instrument, tested on itself
The enemies above are the threats; here is the instrument turned on its own readers, to see whether it survived them. These are findings about the measuring, not about magic — which is why they live here and not on THE LEDGER.
No single witness. Every read the study owns (two hundred eighty-eight, across the working set and the sealed-records test) was re-scored with each of the four reader schools removed in turn, then each of the twelve sources. Every lever kept its direction in all sixty-four re-scorings: the answered working out-succeeded the silent, the charged out-succeeded the empty. No headline rests on one reader or one corpus.
The spliced record. Eight chimeras (a real opening sewn onto a foreign ending, same vein so the voice gives nothing away) were hidden among intact records and handed to two readers, untold. Both flagged all eight as incoherent; neither followed the sewn-on ending once; on thirteen of seventeen splice reads they called the ending unreadable rather than spin an impression into a story. The deflationary worry, a reader harmonizing whatever it is handed into one confident tale, did not show.
The eye, on inside-told records. Where a record is written by the one who did it, schooled readers beat the dumb always-guess strategy: 75–86% on inside-told records, against a 33% read of protocol-framed ones. The fault line is narration, not fame. On the never-read confirmation-pass records the edge shrank to 65% against a 74% always-fail base, so the claim carries the pass's humility — the narration instruction stands, the eye's edge is modest.
The cultivator exception. The fourth reader, schooled in internal-energy work, beat the baseline even on protocol-framed records (83%) where the other three went zero of seven — by reading the energetic temperature of the prose itself. The lens worked only for the reader who owned it: handed to the others as a sheet, it did not transfer. Cross-tradition awareness is trained, not photocopied.
The panel pushed back. The readers reviewed these pages as practitioners and challenged them; five cheap blind checks followed, and the levers sharpened exactly where the chaos and folk reviewers said they would — the clutch splitting worry-repetition from recipe-contract, the narration fault running through individual records rather than whole collections.
Hindsight, measured. The three-times-hotter figure in the enemies table above is measured on our own records, not assumed — and every reader is metered for halo, recognition, and home-doctrine bias, every catch logged.
Freezes, and the seal that breaks once
Before the main reads, the instrument was frozen: the call-set, the report templates, and a no-amendment clause — if anything frozen needs changing mid-study, the study stops and says so rather than quietly bending. Analyses that came later are labeled exploratory, loudly. And the final discipline: a sealed reserve, read exactly once. The first one ran — sixty-three never-read records, pairs of blind readers, rules frozen before the seal broke. Its questions were written down before the seal was set, and the practitioners' own predictions logged beside them, so the act graded the judgment of the people who made the claims as much as the claims. A reserve read is a reserve spent: those records now join the corpus, and a fresh set stays sealed for the next exam — the Yorùbá Ifá records, untouched. That is what keeps everyone honest: every claim on these pages knows another exam is still coming, and an exam cannot be retaken.
Two layers, one policy
What a blind reader said, when blind, is never edited — a call's evidentiary value IS its provenance, and rewriting it would let today's beliefs reach back and doctor yesterday's evidence. But the analysis is not history: every number displayed on these pages reflects the study's most accurate current view, with every audit and correction applied as a labeled overlay on top of the untouched originals. When an audit moves a number, the page moves with it and says why. The only place yesterday's rules still bind is a sealed reserve: its questions are scored under the rules stated before the seal breaks, which is what keeps a confirmation real instead of curated.
The audit trail
A running calibration log records every ruling, bug, and self-correction — including the embarrassing ones: the scorer bug we found because the study's owner said the numbers smelled wrong; the claim a working magician overturned ("a working magician would easily see the difference — consult one"), which the blind consult then proved; the practitioner panel that reviewed these very pages and caught the instrument conflating recipe-repetition with anxious clutching; the audit that re-filed our own record categories. Two standing rules govern it all: surprises are treated as artifacts first (the instrument is audited before the theory is changed) and the magician's seat is consulted before more machinery is built. So far the seat's record against the apparatus is undefeated. And one correction is worth stating plainly, because of how it can look: the answer lever first scored as failed-to-confirm, and passed only after a counting bug was corrected. The bug was a file-format slip (two JSON shapes, one read), the rules never changed, the full before-and-after table is in the log, and the same fix made the state floor stronger rather than weaker. The trail is there for you to audit it yourself.
What this can and cannot show
- It CAN show which conduct, visible in records, separates workings that succeeded from workings that failed — across four traditions and a century of ink.
- It CAN test doctrines against the record and retire the ones that fail (procedural exactness; True Will as a scoreable thing).
- It CANNOT show whether magic is real; both readings fit everything here.
- It CANNOT yet speak prospectively — every record is the past, and the folk record especially is told years after the fact (its own handicap, measured). The living experiment on THE LIVING EXPERIMENT is the answer to this.
- Counts are small everywhere outside the diary. The pages say so wherever it matters, and a sealed reserve exists because saying so is not enough.