Library Of Compound Cures

This library is for someone who does not know us and has no reason to trust us.

A patient reading a label. A prescriber who wants to know where else a strand of RNA could bind. A regulator asking whether a claim was measured or merely stated. A laboratory deciding where to point a bench. None of them should have to take our word for anything, and in this library none of them does: every figure on this page is printed by a program in reproduce/, and the command that prints it is written beside the figure.

It is a library, not a study. Studies close; a library grows. Entries arrive as evidence arrives, and each one is admitted by a law rather than by a person — there is no reviewer here to persuade and no committee to convince.

What is in it today: exact off-target maps for the nucleic-acid medicines the public registry publishes a usable sequence for, a genome-wide map for every guide RNA in that registry, and an exact discrimination court over seventeen tumour types asking what actually recovers a published master-regulator set. Five entries. Two named things it does not have.

---

The admission law, in one paragraph

Nothing enters this library without a reproducible measurement. An entry must name itself with a public identifier the law checks rather than trusts, state what was measured with a quantity, name a program in reproduce/ that produces it, list figures that each match a line of that program's output, carry either a seal that program actually prints or the explicit token NONE_PRINTED, take a grade from this wiki's ontology that the present evidence supports, say in its own words what it refuses to claim, name the observation that would overturn it, give a command a stranger runs from a clean clone, and carry the not-advice line inside the entry so it travels when a row is copied out. That is the per-entry half — fifteen clauses, E0 to E14. The other half holds every entry at once: nine clauses, L1 to L9, publishing a distinct count over the identity field beside the row count, always, with integer ceilings the library declares in advance and the law tests as count(most repeated value) ≤ declared — per value, never as a ratio, because a ratio between two integers is where a float enters a program that had none.

The full law: The library admission law. The law itself is the program reproduce/library-admission-law.swift — 2,495 lines, Swift 6.4, zero float on every decision path, 71 control arms passing in three directions. Beyond the two halves there is one federation clause, F1, which asks the only question neither half can: is one entry filed in two different libraries? No per-library clause can see it, because no per-library clause ever holds two libraries at once — which is why the three libraries are graded in a single run and never one at a time.

Why the second half exists. A generative pipeline in this same programme reported 37,910 validated discoveries and contained five distinct molecules. Every row carried a valid molecule and every per-row check that ran on it was right to pass it; the emitting loop read seeds[i % len(seeds)], so the row count was the loop bound. No per-item validator can see that, because no per-item validator ever holds two items at once. That is why this page publishes a row count and a distinct count side by side and will never publish one without the other.

Three terminals, and they are three answers

terminal	what it means here
ADMITTED	every clause is satisfied on evidence present at this grading. It does not mean true, safe, or recommended — read the entry's own refusal lines
REFUSED	a clause is violated, and the clause is named. There is no state between admitted and refused
NOT_KNOWN	a clause cannot be decided because its evidence is not present here. The entry waits, by name, and is neither in the library nor thrown out of it

An open slot is a fourth thing and deliberately not a terminal: a held entry has evidence that exists somewhere, a slot has evidence nowhere. Slots are named below, count in no axis, and gate on nothing. Naming a slot is how a library says what it is missing instead of quietly not having it.

Grade this library yourself

xcrun swiftc -O -swift-version 5 reproduce/library-admission-law.swift -o /tmp/lal
/tmp/lal --library library/proteins --library library/compounds --library library/materials \
         --reproduce reproduce --evidence /tmp

Grade all three libraries in one command. F1 — no entry filed in two libraries — is a relation between libraries, and a run given one library cannot answer it: it prints F1_NO_ENTRY_FILED_TWICE NOT_KNOWN and exits 2, which is the honest answer and is not a clearance. The first published version of these three pages each carried a clean table produced by a separate single-library run, and the clause that refused the combined run could not be reached by any of the three. That is fixed in the law, and this is the command it is fixed with.

--evidence is where out_<program>.txt transcripts live; validate.sh writes them to /tmp, which is the default. Exit 0 all admitted, 1 something refused, 2 something held, 3 the control arm failed and nothing was graded.

Measured 2026-09-07, all three libraries together: 15 ADMITTED · 1 HELD · 0 REFUSED, every per-library clause holding, and F1 clear over 12 distinct triples. The seal is path-independent by construction — filesystem paths are printed and never sealed, because a seal that moves with the checkout directory indicts a correct reproduction.

CONTROL ARM  71/71 PASS        43 must REFUSE · 19 must ADMIT · 9 must HOLD
  programs in reproduce/   89
LIBRARY PROTEINS    ->  ADMITTED     6 files   5 admitted   1 held   0 refused
LIBRARY COMPOUNDS   ->  ADMITTED     8 files   4 admitted   4 held   0 refused
LIBRARY MATERIALS   ->  ADMITTED     6 files   6 admitted   0 held   0 refused
F1_NO_ENTRY_FILED_TWICE   ok — no triple appears in more than one of the 3 libraries
TRANSCRIPT SEAL  sha256  fa0c43d8cc38a2ad66214bac3a82c3aa27cb86cedd310640ed88ce0d9fb9584b
sealed bytes             33,668
exit                     2

The seal moves when the census moves; the verdict does not, and the difference is worth a sentence rather than a footnote. The program census — programs in reproduce/ 89 — is inside the sealed transcript, because what programs were available is evidence about the grading. A clone holding a different number of programs prints a different seal. That is content disagreeing, not a broken reproduction: the per-entry and per-library verdicts above it are what must match.

---

The entries

Eight entry blocks follow, reproduced verbatim from the canonical entry files in library/compounds/. The checker grades one entry per file, so grade the library directory — this page is the reading surface, not the graded object. Clause L9 makes "reproduced verbatim" a gate rather than a promise: it compares the fenced blocks on this page against the blocks in that directory as multisets and refuses the library if they differ. Before L9 existed the two agreed, and that was a fact about one hour rather than something anyone checked.

Each block is the schema. Each carries its own grade, its own refusal lines and its own not-advice line, so that a row copied out of this library arrives somewhere else still carrying what it refuses to say.

---

1 — Zilganersen: where else in the transcriptome can it bind?

An approved-sequence antisense oligonucleotide against GFAP, screened against every transcript GENCODE v50 publishes. The whole point of the entry is the shape of the answer: two perfect matches, both in the intended gene, and then nothing at all until you drop two mismatches. The gap between 20/20 and 17/20 is empty. That emptiness is a measurement, and it is the one a prescriber would want.

The burden figure is the one that needs its control read alongside it: 324 off-target windows at 16/20 or better outside GFAP, against a composition-matched control median of 787. The real strand is quieter than scrambles of its own composition. A count without that control would be a number nobody could interpret.

LIBRARY        COMPOUNDS
IDENTITY_KIND  UNII
IDENTITY       AXQ9493NT2
TITLE          Zilganersen off-target map, exact, whole human transcriptome
MEASURED       On the real approved sequence against GENCODE v50: 2 perfect 20/20 windows, both in GFAP; 0 at 19/20 and 0 at 18/20; 11 sites named at 17/20; 324 off-target windows at 16/20 or better outside GFAP against a composition-matched control median of 787. 670,670 transcripts, one integer per window, no cutoff inside the arithmetic.
PROGRAM        zilganersen-offtarget-whole-transcriptome
FIGURE         perfect 20/20       : 2, both in GFAP
FIGURE         19/20 and 18/20     : 0 and 0
FIGURE         off-target burden at 16/20 or better, outside GFAP : 324
FIGURE         composition-matched control median at the same threshold : 787
SEAL           edcb277ddea44820502b6446b00ed8bdcfdb08835d6785fdee7d1b370420bbaa
GRADE          MEASURED
SOURCE         NCATS GSRS, UNII AXQ9493NT2 — the sequence is public and nothing here is behind a login
WHERE_THE_LAW_LIVES  reproduce/zilganersen-offtarget-whole-transcriptome.swift
REFUSED        not a safety verdict on the medicine, and not a finding about any patient
REFUSED        not a claim that any listed site is bound, cleaved or clinically relevant in a person. Complementarity is where binding is possible, never where it happens
REFUSED        not a comment on the trial, the endpoint or the approval
FALSIFIER      the same sequence against the same GENCODE v50 digest returning a different count at any of the four thresholds, or a perfect window outside GFAP
REPRODUCE      cd reproduce && xcrun swiftc -O -swift-version 5 zilganersen-offtarget-whole-transcriptome.swift -o /tmp/run && curl -sL <gencode v50 transcripts fasta> | gunzip -c | /tmp/run
NOT_ADVICE     Nothing in this entry is medical advice and it is not a recommendation to take anything.
ADDED          2026-09-07

The eleven sites at 17/20 are named in the program's output — XYLB, ENSG00000239572, ADAM20P1, ENSG00000293223 — because a bench that wants to check one needs the name, not the count.

---

2 — Pelacarsen: the same question asked of a different strand

The same instrument, a different medicine, and the reason it is here is that it was asked independently and reached the same place twice: three perfect windows, all in LPA. The registry-wide atlas below reaches LPA for this strand by a different route, and agreement between two screens that did not share a code path is worth more than either alone.

LIBRARY        COMPOUNDS
IDENTITY_KIND  UNII
IDENTITY       LSO9H7UZ90
TITLE          Pelacarsen off-target map, exact, whole human transcriptome
MEASURED       3 perfect 20/20 windows, all in LPA, over 670,670 transcripts and 1,467,336,203 windows against GENCODE v50. At 17/20 or better outside LPA: LPAL2, TMEM254-AS1, LINC02606, LINC01065, ISM1.
PROGRAM        pelacarsen-offtarget-whole-transcriptome
FIGURE         perfect 20/20 windows : 3, all in LPA
FIGURE         17/20 or better, not LPA : LPAL2, TMEM254-AS1, LINC02606, LINC01065, ISM1
FIGURE         513de7e9db6556df1895bfce4cb4d69e4816d7b45b75bee1dc8452335c2c7757
SEAL           513de7e9db6556df1895bfce4cb4d69e4816d7b45b75bee1dc8452335c2c7757
GRADE          MEASURED
SOURCE         NCATS GSRS, UNII LSO9H7UZ90
WHERE_THE_LAW_LIVES  reproduce/pelacarsen-offtarget-whole-transcriptome.swift
REFUSED        not a safety verdict on the medicine and not a finding about any patient
REFUSED        not a claim that LPAL2 or any named transcript is bound in a person
FALSIFIER      an independent screen of the same strand against the same transcriptome digest reaching a different perfect-window count or a different named set at 17/20
REPRODUCE      cd reproduce && xcrun swiftc -O -swift-version 5 pelacarsen-offtarget-whole-transcriptome.swift -o /tmp/run && curl -sL <gencode v50 transcripts fasta> | gunzip -c | /tmp/run
NOT_ADVICE     Nothing in this entry is medical advice and it is not a recommendation to take anything.
ADDED          2026-09-07

---

3 — Every nucleic-acid medicine the public registry publishes a sequence for

This is the coverage entry, and coverage is the claim it makes. The strands were enumerated from the registry, not from a list anyone remembered: 742 substances of class nucleicAcid, 740 carrying a sequence, and those whose every subunit falls between 8 and 60 nt giving 472 strands across 350 substances.

Of those, 187 had a target measured from the transcriptome — found by complementarity, not read off a label. 285 had no perfect complement anywhere and were therefore never screened for off-targets. Those 285 are ABSENT from the map. They are not clean, and the entry says so in its own refusal line, because a strand that was not screened and a strand that was screened and found quiet are two different answers and this library never prints them alike.

LIBRARY        COMPOUNDS
IDENTITY_KIND  CONTENT_DIGEST
IDENTITY       321b36c694b89a45bb81668d7ea62b9c85cf0b3087e18bba586f43b230274b08
DIGEST_OF      the merged seal of the eight strand shards of the registry-wide off-target atlas
TITLE          Every nucleic-acid substance the public registry publishes a usable sequence for
MEASURED       472 strands across 350 substances, enumerated from the registry rather than from a list anyone remembered: 742 substances of class nucleicAcid, 740 carrying a sequence, and those whose every subunit falls in 8 to 60 nt. 187 strands had a target measured FROM the transcriptome; 285 had no perfect complement anywhere and were therefore not screened for off-targets.
PROGRAM        oligo-offtarget-atlas-exact
FIGURE         321b36c694b89a45bb81668d7ea62b9c85cf0b3087e18bba586f43b230274b08
FIGURE         472 strands across 350 substances
FIGURE         187 strands had a measured target; 285 had no perfect complement anywhere
SEAL           321b36c694b89a45bb81668d7ea62b9c85cf0b3087e18bba586f43b230274b08
GRADE          MEASURED
SOURCE         NCATS GSRS, class nucleicAcid, enumerated in full
WHERE_THE_LAW_LIVES  reproduce/oligo-offtarget-atlas-exact.swift
REFUSED        the 285 refused strands are ABSENT from the map, never scored as clean. A strand with no perfect complement was not screened, and that is a different answer from a strand that was screened and found quiet
REFUSED        not a safety ranking. This entry publishes coverage and targets, not an ordering of substances by risk
REFUSED        not a claim about any approval, label or trial
FALSIFIER      a strand in the registry inside the 8 to 60 nt range that this enumeration omits, or a measured target that a second independent screen does not reach
REPRODUCE      cd reproduce && xcrun swiftc -O -swift-version 5 oligo-offtarget-atlas-exact.swift -o /tmp/run && curl -sL <gencode v50 transcripts fasta> | gunzip -c | /tmp/run ../corpus/oligo-atlas/all_nucleicacid.tsv
NOT_ADVICE     Nothing in this entry is medical advice and it is not a recommendation to take anything.
ADDED          2026-09-07

The run was sharded by strand across eight processes, each reading the entire transcriptome, so no statistic crosses strands and sharding changes no number. The published seal is the sha256 of the eight shard seals in order, which is why a single process cannot print it — the program says so in its own output rather than leaving a reader to discover it.

The program also corrects an earlier revision of its own window count in public, in its output: a figure once printed as a property of the run was one arbitrary strand's count. Window counts are per strand length, and the program now says that where the number used to be.

---

4 — Where else could this guide cut? The whole genome, counted

Fifteen guide RNAs, found by the canonical SpCas9 scaffold rather than by name — every one of the 742 nucleicAcid substances in the registry was scanned for it, and fifteen carry it. Screened against 3,099,750,718 bases at 304,796,751 NGG PAM sites on both strands.

The finding is stated at full magnitude because it is there: all fifteen guides found a zero-mismatch site, thirteen had exactly one in the whole genome, every guide had zero sites at one mismatch, and thirteen of fifteen had zero at two. On the question this instrument answers — how many places in the genome match this guide, exactly — these are clean guides, and the count is not an opinion. The fifteen are named in the program's own output beside the UNII each sequence came from, and they include the guide of an approved therapy people are alive because of today — exagamglogene autotemcel, UNII L28RZ5CC6K. Which of the fifteen is quietest at three and four mismatches is a per-guide question, and it is answered by the per-guide table rather than by this summary.

1,718 windows containing an N were set aside and counted, never folded into either bucket. An N is neither a match nor a mismatch, and a library that quietly rounds it into one is a library whose totals cannot be checked.

LIBRARY        COMPOUNDS
IDENTITY_KIND  CONTENT_DIGEST
IDENTITY       487b4f81de2d24bd0bb11ecd1d8d42778e3a5d91b9edb33627c86dcc8df34980
DIGEST_OF      the sealed transcript of the genome-wide guide screen over 15 registry guides
TITLE          Genome-wide off-target map for every guide RNA the public registry publishes
MEASURED       15 guides, found by the canonical SpCas9 scaffold rather than by name, against 3,099,750,718 bases: 304,796,751 NGG PAM sites examined on both strands, 1,718 windows containing an N set aside rather than scored. All 15 guides found a zero-mismatch site and 13 had exactly one in the whole genome; every guide had 0 sites at one mismatch and 13 of 15 had 0 at two.
PROGRAM        crispr-genome-offtarget-exact
FIGURE         487b4f81de2d24bd0bb11ecd1d8d42778e3a5d91b9edb33627c86dcc8df34980
FIGURE         304796751 NGG PAM sites
FIGURE         All 15 guides found a zero-mismatch site; 13 had exactly one in the whole genome.
FIGURE         L28RZ5CC6K
SEAL           487b4f81de2d24bd0bb11ecd1d8d42778e3a5d91b9edb33627c86dcc8df34980
GRADE          MEASURED
SOURCE         NCATS GSRS, all 742 nucleicAcid substances scanned for the scaffold; 15 carry it
WHERE_THE_LAW_LIVES  reproduce/crispr-genome-offtarget-exact.swift
REFUSED        not a per-guide safety verdict. This entry publishes the map at the granularity the program prints, and no claim about one named guide beyond what appears there
REFUSED        an N window is neither a match nor a mismatch. The 1,718 are set aside and counted, never folded into either bucket
REFUSED        not a claim that any site is cut in a person. A PAM plus complementarity is where a cut is possible, never where it happens
FALSIFIER      a 16th guide in the registry carrying the canonical scaffold, or a different site count for the same assembly digest
REPRODUCE      cd reproduce && xcrun swiftc -O -swift-version 5 crispr-genome-offtarget-exact.swift -o /tmp/run && curl -sL <GRCh38 primary assembly fasta> | gunzip -c | /tmp/run ../corpus/crispr-atlas/guides_all.tsv
NOT_ADVICE     Nothing in this entry is medical advice and it is not a recommendation to take anything.
ADDED          2026-09-07

The full per-guide map, with coordinates and strands at four mismatches or fewer, is published on the CRISPR page — because that list is exactly where a laboratory would start if it wanted to check a guide experimentally, and publishing the summary without it would be publishing the reassuring half.

---

5 — What actually recovers a published master-regulator set?

This is the entry that changes what a reader should do next, and it is a negative result stated at full magnitude rather than apologised for.

Across seventeen TCGA tumour types, ranking regulators by regulon size alone — never reading the patient expression counts at all — recovers the published master-regulator set at least as significantly as ranking them by expression in eleven of them. Expression adds discrimination in five. One recovers nothing by either arm and is published as such, by name, rather than dropped. Ties: zero.

The arithmetic is an exact integer hypergeometric upper tail. C(100,50) is computed as 100891344545564193334812497256, not as a float that agrees to fifteen digits, and the program's self-test proves the tail equals its own denominator at s=0 before it grades anything.

LIBRARY        COMPOUNDS
IDENTITY_KIND  CONTENT_DIGEST
IDENTITY       d0051d7a19daa0cb7f1a3d4f7be7433af73401bbf25fe41079f6409c066fee5a
DIGEST_OF      the sealed verdict transcript of the exact discrimination court over 17 TCGA tumour types
TITLE          Topology against expression, by exact hypergeometric tail, 17 tumour types
MEASURED       Ranking regulators by regulon SIZE ALONE recovers the published master-regulator set at least as significantly as ranking them by patient expression in 11 of 17 tumour types; expression adds discrimination in 5; 1 recovers nothing by either arm and is published as such. Exact integer hypergeometric upper tails, zero float on the decision path.
PROGRAM        mr-topology-vs-expression-exact
FIGURE         d0051d7a19daa0cb7f1a3d4f7be7433af73401bbf25fe41079f6409c066fee5a
FIGURE         TOPOLOGY_EXPLAINS: 11 of 17 tumour types
FIGURE         EXPRESSION_ADDS  : 5 of 17 tumour types
FIGURE         SELFTEST PASS
SEAL           d0051d7a19daa0cb7f1a3d4f7be7433af73401bbf25fe41079f6409c066fee5a
GRADE          MEASURED
SOURCE         GDC open RNA-seq integer counts and the published regulon files, both public
WHERE_THE_LAW_LIVES  reproduce/mr-topology-vs-expression-exact.swift
REFUSED        not a claim that the published master-regulator sets are wrong. It is a statement about what these numbers support and nothing wider
REFUSED        not a claim about any patient, any treatment or any outcome
REFUSED        not a drug pairing. The eleven named drug pairs published in prose on the study page are NOT in this entry, because no program in reproduce/ prints them and a stranger cannot re-derive them from a clean clone. They are named as an open slot in this library's manifest instead
FALSIFIER      the same regulon files and the same integer counts returning a different verdict for any of the 17 cohorts, or a tail probability that disagrees in exact arithmetic
REPRODUCE      cd reproduce && xcrun swiftc -O -swift-version 5 mr-topology-vs-expression-exact.swift -o /tmp/run && /tmp/run
NOT_ADVICE     Nothing in this entry is medical advice and it is not a recommendation to take anything.
ADDED          2026-09-07

The seventeen cohorts are printed one per line with pool size, set size, both arms' hit counts and both exact tails — from 2.685e-29 down to 1.000e+0 — so a reader can disagree with the verdict on any single cohort while holding the same numbers we do.

---

6 — Eleven compound pairs, where no single agent among 20,308 cleared any

This is the entry with the most direct bench handoff in any of the three libraries, and it is also the entry that proves the admission law is not decoration.

Across fifteen scored tumour types, 44,850 pairs were scored per type from the 300 most-inverting single agents under a frozen additive law. Eleven of fifteen clear a vehicle-pair control, a self-pair control, and a random-gene-set null at 0 of 200 draws — where 0 of 15 cleared for any single agent among 20,308 compounds. The four that do not clear carry the fewest observable regulators in the corpus: 8, 8, 9 and 9, which is a statement about the power of this test on those cohorts and never about those cancers.

And for one day this entry could not exist. These pairs were published on 2026-09-06 in prose. The compound names appeared in exactly one place in the entire public repository — the study page itself. No program printed them, no corpus carried them. This library's own law refused them under E3 and E4 on first contact with the real corpus, and named them as an open slot with the condition for entry written down: they enter the day a program prints them. That program was written the same day, and this entry is its output. The slot closed the day it opened, and the law refused two further defects in this very entry before admitting it — an identity digest that no FIGURE line claimed, and an invented manifest key.

LIBRARY        COMPOUNDS
IDENTITY_KIND  CONTENT_DIGEST
IDENTITY       d0117523ff3b950a0741de281671c0bd977f8dc003f41972f2bfac2f50994d74
DIGEST_OF      the sealed transcript of the reader that prints the eleven pairs from the screen's own per-cohort results
TITLE          Eleven compound pairs clearing three controls where no single agent among 20,308 cleared any
MEASURED       Across 15 scored tumour types, 44,850 pairs scored per type from the 300 most-inverting single agents under a frozen additive law. 11 of 15 clear a vehicle-pair control, a self-pair control, and a random-gene-set null at 0 of 200 draws. 0 of 15 cleared for any single agent among 20,308 compounds. The four that do not clear carry the fewest observable regulators in the corpus: 8, 8, 9 and 9.
PROGRAM        study26-combination-pairs-exact
FIGURE         estradiol + AMG-208
FIGURE         estrone + BMS-387032
FIGURE         olaparib + ursodeoxycholyltaurine
FIGURE         HMN-214 + saracatinib
FIGURE         drospirenone + alpelisib
FIGURE         XMD-892 + NVP-BGJ398
FIGURE         THE PAIRS THAT CLEAR ALL THREE CONTROLS — 11 of 15
FIGURE         d0117523ff3b950a0741de281671c0bd977f8dc003f41972f2bfac2f50994d74
SEAL           d0117523ff3b950a0741de281671c0bd977f8dc003f41972f2bfac2f50994d74
GRADE          MEASURED
SOURCE         Study 26 S3-COMBINATION, per-cohort results pinned in corpus/study-26-combination/
WHERE_THE_LAW_LIVES  reproduce/study26-combination-pairs-exact.swift
REFUSED        not efficacy. A signature-inversion score is an arithmetic statement about expression ranks of landmark genes, and nothing about a dose, a mechanism, or a patient follows from it
REFUSED        not a claim that any pair helps anyone. Whether two compounds act additively in a cell, at a dose, in a person, and whether the combination is tolerable at all, is a laboratory and clinical question this program has not asked
REFUSED        not a re-run of the screen. This program is a faithful reader of the screen's sealed output and says so on every path; re-deriving those bytes from LINCS is a separate and larger reproduction
REFUSED        not a finding about the four that did not clear. Their result is a statement about the power of this test on those cohorts, never about those cancers
FALSIFIER      a cohort file whose published gain does not equal best_pair minus best_single, which the program's own known-case check reports before emitting any table; or a re-run from LINCS reaching different pairs under the same frozen law
REPRODUCE      ( cd corpus/study-26-combination && shasum -a 256 -c SHA256SUMS ) && xcrun swiftc -O -swift-version 5 reproduce/study26-combination-pairs-exact.swift -o /tmp/s26c && /tmp/s26c < /dev/null
NOT_ADVICE     Nothing in this entry is medical advice and it is not a recommendation to take anything.
ADDED          2026-09-07

The reader carries a known-case check that runs before any table: the published gain must equal best_pair − best_single on all fifteen rows. Changing a single digit in one result file makes it name the disagreeing row and emit no table and no seal. Given no corpus it refuses rather than printing an empty result, and its seal is byte-identical from any directory.

7 — Is a medicine more specific than the same bases in another order?

The atlas above answers where a strand can pair. It cannot answer whether that burden is unusual, and alone it decides nothing: any 20-mer has hundreds of near-complementary windows in a corpus of 1.47 billion, for the same reason any twenty-letter string turns up in a large enough library. This entry is the comparison — every screened strand against sixteen permutations of its own bases, the same multiset with none of the design.

This slot was named empty on 2026-09-07 and filled on 2026-09-08. The manifest carried it as an open slot, in the library's own words, while the screen was still running: no program, no transcript, no seal, nothing to hold. It is an entry now because a program prints it.

The result the bench should read first is not the two successes. It is that 17 undesigned sequences, drawn from the corpus by a fixed rule and screened exactly as a medicine is, put the registry's numbers in a scale — and none of the seventeen is below its own controls either.

LIBRARY        COMPOUNDS
IDENTITY_KIND  CONTENT_DIGEST
IDENTITY       2d74f5c676d51d45df178f9fed7840729bd87633ae8e5a9104383f6fbd7c3fd1
DIGEST_OF      the sealed transcript of the registry-wide specificity ranking, over every strand, every threshold, every rank interval and every complete histogram
TITLE          Every registry strand ranked against sixteen rearrangements of its own bases
MEASURED       472 registry strands and 17 undesigned constructed 20-mers screened as 186 families of 17 probes each, 5,355,878,467,758 probe-windows counted with no sampling and no cutoff inside the arithmetic. At 4 mismatches: 2 families pair in strictly fewer places than all sixteen permutations of their own bases, 8 tie the lowest, 122 sit inside their own control range, 1 ties the highest, 44 pair in more places than every permutation, and 9 tie all sixteen. 18 strands are UBIQUITOUS, 266 REFUSED and 19 NOT_KNOWN — three silences, none of which is zero off-targets.
PROGRAM        registry-specificity-ranking
FIGURE         2d74f5c676d51d45df178f9fed7840729bd87633ae8e5a9104383f6fbd7c3fd1
FIGURE         717027798090 + 4638850669668 = 5355878467758
FIGURE         169 registry strands + 17 undesigned constructed 20-mers
FIGURE         18 UBIQUITOUS, 266 REFUSED, 19 NOT_KNOWN
FIGURE         BELOW-all-16 2 | ties-lowest 8 | inside 122 | ties-highest 1 |
SEAL           2d74f5c676d51d45df178f9fed7840729bd87633ae8e5a9104383f6fbd7c3fd1
GRADE          MEASURED
SOURCE         NCATS GSRS class nucleicAcid enumerated in full, screened against GENCODE v50 transcripts
WHERE_THE_LAW_LIVES  reproduce/registry-specificity-ranking.swift
REFUSED        not a safety finding. A near-complementary window is a place a molecule COULD pair — not a cut, not an occupancy, not a clinical event. A high rank is not a safety finding and a low rank is not a clearance
REFUSED        not a ranking of medicines against each other. Every comparison on this entry is a strand against permutations of ITS OWN bases, never against another strand, and raw burden is never pooled across lengths
REFUSED        not a claim about the 303 excluded strands. UBIQUITOUS, REFUSED and NOT_KNOWN are three different answers and an excluded strand is never a clean strand
REFUSED        not the chemistry half. Phosphorothioate backbone binding, complement activation, thrombocytopenia and aseptic meningitis are not sequence matches and this program does not reach them
FALSIFIER      a registry strand in the 8-60 nt band this enumeration omits; or a re-run on the same two public inputs reaching a different seal, a different disposition for any family, or a different count of families strictly below all sixteen
REPRODUCE      xcrun swiftc -O -swift-version 5 reproduce/registry-specificity-ranking.swift -o /tmp/rsr && curl -sL <gencode v50 transcripts fasta> | gunzip -c | /tmp/rsr
NOT_ADVICE     Nothing in this entry is medical advice and it is not a recommendation to take anything.
ADDED          2026-09-08

The full page, including the tie rule that took a claimed nineteen designed-specificity families down to a measured two, is The order of the bases.

8 — Every clinical CRISPR guide with a public spacer, against its own composition

The genome-wide map above answers where one guide can cut. This entry asks the comparative question of every clinical guide the public registry names: twenty-five of them, under three different PAM rules, each ranked against 32 permutations of its own bases.

The result the control arm was built for is not a flattering one, and it is the honest half. Zero of 25 sit below their own composition floor. But a therapeutic spacer is not free to be chosen — it is dictated by the locus the medicine has to cut, while its 32 permutations have no locus to hit and are free to be whatever minimises burden. The comparison is between a molecule with a job and 32 with none, so the result is a statement about the constraint, never about anyone's design. That every one of them still reaches zero sites at one mismatch under that constraint is the more remarkable half of the same measurement.

And the 3 GB download is not required to check it. The 3,022-line sealed transcript is shipped in this repository, digest-pinned, beside a deliberately truncated control that refuses — so the corpus carries both directions of its own instrument.

LIBRARY        COMPOUNDS
IDENTITY_KIND  CONTENT_DIGEST
IDENTITY       61ef3254a8455ee339917368c5d45013072a377592045d2e0494cc99c0fad247
DIGEST_OF      the sealed transcript of the complete genome-wide enumeration for every clinical CRISPR guide whose spacer is public, with every guide's control arm
TITLE          Twenty-five clinical CRISPR guides, complete enumeration under three PAM rules, each ranked against 32 permutations of its own bases
MEASURED       25 guides registered in a public login-free substance registry with a spacer and named in a WHO INN Proposed List: 18 SpCas9 family under NGG, 4 AsCas12a under TTTV, 3 Cas12b under TTN, at each guide's own measured spacer length, across both strands of GRCh38 primary assembly. Every one of the 24 full-length guides has ZERO sites at one mismatch and 22 of 24 have zero at two; the whole off-target burden sits in the 3 and 4 mismatch buckets. 2,718 coordinates are named in full with chromosome, position and strand, and the number of coordinate lines emitted equals the sum of the 24 burdens exactly, 2,718 = 2,718. Against 32 permutations of its own bases each: 0 of 25 sit below their own composition floor and 25 of 25 sit inside the range their own bases produce, where permutation burdens span 0 to 7,718,082.
PROGRAM        crispr-clinical-guide-atlas-exact
FIGURE         61ef3254a8455ee339917368c5d45013072a377592045d2e0494cc99c0fad247
FIGURE         CRISPR_CLINICAL_GUIDE_ATLAS__COMPLETE_ENUMERATION_IS_OBSERVER_INVARIANT
FIGURE         bb188a75837c3384322723c5935da34605cb89caae43cf3b0468572fda60b80c
SEAL           61ef3254a8455ee339917368c5d45013072a377592045d2e0494cc99c0fad247
GRADE          MEASURED
SOURCE         NCATS GSRS spacers with WHO INN Proposed List product names, screened against GENCODE GRCh38 primary assembly, sha256 b760d18dbb651dd14dfc290083371b3ef3bff122d43a9cefb13ca4ecf38f05ca
WHERE_THE_LAW_LIVES  reproduce/crispr-clinical-guide-atlas-exact.swift
REFUSED        not a claim that any of these therapies is safe, or unsafe. Neither verdict is ours to give and neither follows from this arithmetic
REFUSED        not a cut. A site counted here is a place the chemistry COULD direct a cut, not an occupancy, not a clinical event, and not evidence that any medicine harms anyone
REFUSED        not a judgement of anyone's guide design. A therapeutic spacer is dictated by the locus the medicine has to cut while its permutations have no locus to hit, so the control arm is a statement about that constraint and never about design quality
REFUSED        not coverage of the six products whose spacer is not public. For those the honest word is NOT_KNOWN and it is printed as NOT_KNOWN rather than as zero
FALSIFIER      a candidate site in GRCh38 within the reported mismatch range, under the guide's own PAM rule and spacer length, that this enumeration omits; or a coordinate count that does not equal the sum of the reported burdens
REPRODUCE      xcrun swiftc -O -swift-version 5 reproduce/crispr-clinical-guide-atlas-exact.swift -o /tmp/atlas && curl -sL <GRCh38 primary assembly fasta gz> | gunzip -c | /tmp/atlas corpus/crispr-clinical/guides_expanded.tsv
NOTE           The 3,022-line sealed transcript is SHIPPED in the repository as corpus/crispr-clinical/RUN-full-assembly-n32.txt, digest-pinned in that directory SHA256SUMS beside a deliberately truncated control that REFUSES, so the seal can be checked without the 3 GB assembly download. The program itself refuses without the assembly, so in a clean clone this entry reads NOT_KNOWN until it is run or that transcript is given as evidence.
NOT_ADVICE     Nothing in this entry is medical advice and it is not a recommendation to take anything.
ADDED          2026-09-09

The refused section

A library that shows only what it admitted cannot be audited. Six things were considered for this library and did not enter, or entered only as an explicit absence.

R1 — The eleven master-regulator drug pairs. REFUSED: E3, E4.

Study 26 publishes eleven drug pairs, one per tumour type, each the top of 44,850 scored, each clearing a vehicle-pair control, a self-pair control and a random-gene-set null. Written out, that is the strongest-sounding evidence on the board, and it is the one thing on this page that a stranger cannot check.

Measured, not assumed: searched across the whole standalone wiki, the string AMG-208 appears in exactly one file — the study page itself. So does saracatinib. No program in reproduce/ prints any of the pairs, no transcript contains them, and no corpus file carries them. A reader of the study page would take those pairs as sealed; a reader with a clean clone cannot re-derive a single one.

That is E3 and E4 doing exactly the job they exist for. The pairs are not deleted and not disputed. They are named as an open slot below, with the reason recorded, and they enter the day a program in reproduce/ prints them. The same applies to the four tumour types that did not clear and to the measured blocker named for them — published set size rather than assay coverage. Real work; not yet re-derivable; therefore not yet admitted.

R2 — The 285 registry strands with no measurable target. ABSENT, and never clean.

Of 472 strands screened for a target by complementarity against the transcriptome, 285 had no perfect complement anywhere. They were therefore never screened for off-targets.

They get their own line here because absence is the easiest thing in a library to misread as safety. A strand with no measured off-target map is not a quiet strand. It is an unscreened one. The atlas entry carries that as its first refusal line so the distinction travels with the row, and it is repeated here so a reader who only reads this section still gets it.

There are three answers in this library and this is the difference between two of them: ABSENCE means we looked and found nothing there; REFUSAL means a clause was violated and is named; NOT_KNOWN means the evidence to decide is not present here. These 285 are the first: measured absence of a perfect complement, which is a finding, and which is not a clearance.

R3 — Any figure from a self-graded column. REFUSED: E12.

confidence, coherence, overall_score, validation_passed, novelty_score, quality_score and their kin may not appear in a MEASURED or FIGURE line anywhere in this library. None of them measures anything outside the program that wrote them, and in this wiki's corpora every one of them is a Double.

They remain welcome in a refusal line — saying a column is barred is the opposite of relying on it, and the law carries a control arm proving the detector knows the difference.

R4 — Any entry that called a medicine safe or effective. REFUSED by construction.

No entry in this library claims efficacy, and none claims safety. A signature-inversion score is not efficacy. An off-target map is not a safety clearance. A near-complementary window is not a cut, and a PAM plus complementarity is where a cut is possible, never where it happens. Every entry above carries those refusals in its own block rather than in this preamble, because the preamble is what gets left behind when a row is copied.

R5 — Any entry carrying a procedure. REFUSED: E11, C-007, absolute.

An entry names what a system is, what was measured about it and where the law lives. It never carries a procedure a person could follow. A constructed violation is on disk in control-cases/case-7-synthesis-cookbook.md — otherwise the admitted zilganersen entry with exactly one line changed — and the law refuses it naming the rule that caught it, the fabrication verb and the quantity, so a false positive would be contestable rather than mysterious.

R6 — The five constructed cases, run on real files, so a stranger can run the refusals too.

Refusals are only credible if you can make them happen. Each of these is the admitted zilganersen entry with exactly one thing changed, on disk in control-cases/:

  case-1-claim-no-program-prints.md         REFUSED  E3_PROGRAM
  case-1b-figure-not-printed.md             REFUSED  E4_FIGURE
  case-2-no-refusal-line.md                 REFUSED  E8_REFUSAL
  case-3-no-identifier.md                   REFUSED  E1_IDENTITY   ('unknown' is a placeholder)
  case-4-verified-on-reported-evidence.md   REFUSED  E7_GRADE_SUPPORTED
  control-cases/collapsed-library/          REFUSED  L3_DISTINCT_IDENTITY
  control-cases/empty-library/              REFUSED  L1_NOT_EMPTY

The last two are the ones that matter most to a library. collapsed-library is the Study-37 shape reproduced exactly — four entries, one identity, four different titles — and it prints both axes side by side so you can watch the wrong counter report perfect diversity:

    axis                            entries  distinct  most at 1  declared  holds
    IDENTITY                              4         1          4         1  NO
    TITLE (census, not a gate)            4         4          1         -    -

The collapsed library declares 4 per program, per seal, per measured and per triple — as generously as a library can — so the refusal cannot be blamed on a mean declaration anywhere else. DECLARED_ENTRIES_PER_IDENTITY 1 is the only thing left, and it is what catches it.

empty-library is REFUSED, never reported clean: zero admitted entries is the weakest possible evidence and calling it a clean library is the strongest possible claim.

---

Open slots — named, and empty, and two that closed

A slot is a measurement this library has named and does not have: no program, no figure, no seal, nothing to hold. Slots are counted in no axis above and gate on nothing.

Both slots this page carried on 2026-09-07 are now closed, and closed the only way a slot may close — a program prints them:

slot named 2026-09-07	why it was empty	closed
The eleven master-regulator drug pairs, one per tumour type	published in prose on the study page; no program in reproduce/ printed them, so a stranger could not re-derive them from a clean clone	2026-09-07, entry 6 — study26-combination-pairs-exact
The registry-wide specificity ranking across all in-scope strands	running at that grading; no program, no transcript, no seal	2026-09-08, entry 7 — registry-specificity-ranking, seal 2d74f5c6…

Three slots are open, and each names the one condition that would fill it:

slot	why it is empty	named
The exact off-target map of del-zota's payload	the sequence is not public. The slot fills the day the sponsor or a regulator puts it on the table — the instrument is already written	2026-09-08
The exact off-target map of VERVE-102's guide on the real molecule	the same reason, on a base editor rather than an oligonucleotide	2026-09-08
A signature-reversal or network-recovery figure for rentosertib	there is nothing to compute against: the compound is ABSENT from both LINCS perturbagen tables (51,383 and 2,170 rows read, 0 matches, in a lookup returning 1 and 1 for pirfenidone and nintedanib) and TNIK is a master regulator in none of our cohorts. The slot fills if a public perturbation corpus carries the molecule	2026-09-08

An invented result would have been easier and would have looked better. A slot stays empty until a program prints it, and the three above are in the manifest, so the grader prints them on every run whether anyone reads this page or not.

---

Rows and distinct identities, side by side

This library states both, always. Reproduced verbatim from the grading run:

    axis                            entries  distinct  most at 1  declared  holds
    IDENTITY                              4         4          1         1  yes
    PROGRAM                               4         4          1         1  yes
    SEAL (sealed entries only)            4         4          1         1  yes
    MEASURED                              4         4          1         1  yes
    IDENTITY|PROGRAM|SEAL triple          4         4          1         1  yes
    TITLE (census, not a gate)            4         4          1         -    -
    GRADE (census, not a gate)            4         1          4         -    -

    'most at 1' is the count carried by the single most repeated value on that
    axis. It is the number the declared ceiling is about, and it is printed
    whether the ceiling holds or not — an aggregate that clears while one value
    carries five against a declared two is the collapse this table exists to show.

  GRADE CENSUS over the admitted set (a census, NOT a gate — no library is
  refused for its distribution of grades, and no library should be read as
  strong for having none of the weaker ones):
    MEASURED                   4

  OPEN SLOTS — named, and empty. A slot is NOT an entry and is counted in no
  axis above. It is a measurement this library has named and does not have:
  no program, no figure, no seal. Naming it is how a library says what it is
  missing instead of quietly not having it.
    - The exact off-target map of del-zota's payload. The sequence is not public; the slot fills the day the sponsor or a regulator puts it on the table. Named 2026-09-08.
    - The exact off-target map of VERVE-102's guide on the real molecule. Named 2026-09-08.
    - An ADMISSIBLE ENTRY for rentosertib. Two separate things block it and both are measured. (1) The recovery figure cannot be computed: the compound is ABSENT from both LINCS perturbagen tables and TNIK is a master regulator in none of our cohorts, so there is nothing to compute against. (2) Its three SEALED-looking arms cannot be filed either: measured 2026-09-09, none of rentosertib-structure-lock-exact, tnik-network-degree-exact or modality-register-exact accumulates a transcript or digests its own results, so clause E5 has no seal to check and the page publishes SOURCE digests only. The slot fills when a public perturbation corpus carries the molecule, or sooner when those three programs are given a transcript seal. Named 2026-09-08, condition made specific 2026-09-09.

  PER-LIBRARY CLAUSES
     ok  L1_NOT_EMPTY              8 entry file(s) present
     ok  L2_MANIFEST               LIBRARY COMPOUNDS declares: per identity 1, per program 1, per seal 1, per measured 1, per triple 1; page Library-Of-Compound-Cures.md; open slots 3
     ok  L7_NO_ENTRY_REFUSED       0 of 8 entries refused
     ok  L3_DISTINCT_IDENTITY      4 entries, 4 distinct; most repeated value 'AXQ9493NT2' carries 1, declared ceiling 1 per value — 1 <= 1
     ok  L4_DISTINCT_PROGRAM       4 entries, 4 distinct; most repeated value 'mr-topology-vs-expression-exact' carries 1, declared ceiling 1 per value — 1 <= 1
     ok  L5_DISTINCT_SEAL          4 entries, 4 distinct; most repeated value '513de7e9db6556df1895bfce4cb4d69e4816d7b45b75bee1dc8452335c2c7757' carries 1, declared ceiling 1 per value — 1 <= 1
     ok  L6_DISTINCT_MEASURED      4 entries, 4 distinct; most repeated value '3 perfect 20/20 windows, all in LPA, over 670,670 transcripts and 1,467,336,203 windows against GENCODE v50. At 17/20 or better outside LPA: LPAL2, TMEM254-AS1, LINC02606, LINC01065, ISM1.' carries 1, declared ceiling 1 per value — 1 <= 1
     ok  L8_DISTINCT_TRIPLE        4 entries, 4 distinct; most repeated value 'AXQ9493NT2 | zilganersen-offtarget-whole-transcriptome | edcb277ddea44820502b6446b00ed8bdcfdb08835d6785fdee7d1b370420bbaa' carries 1, declared ceiling 1 per value — 1 <= 1

Eight entry files, four admitted here, four held — and the four held are the honest half of this table. The library holds eight measurements. Four are graded ADMITTED from a clean clone, because their programs run to completion against corpora small enough to ship here. The other four — the registry-wide atlas, the genome-wide CRISPR map, the specificity ranking and the clinical guide atlas — name a program that refuses without a public download of 1.5 to 3 GB, so in a clean clone their evidence is not present and the law returns NOT_KNOWN. A held entry is not in the library and is not thrown out of it: point the grader at a directory holding those four transcripts and they admit. For the guide atlas that transcript is shipped in this repository, 3,022 lines, digest-pinned beside a deliberately truncated control that refuses — so its seal can be checked without the download at all.

That distinction is younger than this page, and it was our own defect. Until 2026-09-08 all three of them read ADMITTED here, because several programs now print their published figures on every exit path — the right thing for the wiki harness, which must check a page against its program with no corpus present — and clauses E4 and E5 were matching those figures against text the run had quoted rather than computed. The seal of a screen that measured nothing was being credited to it. Three new control arms hold the repair: a refusal transcript that quotes every declared figure must HOLD, one that quotes the declared seal must HOLD, and the same entry against a complete run must still ADMIT — because a guard that always holds is the same defect wearing the other face.

Every ceiling is declared at 1 in library/compounds/LIBRARY.manifest and tested per value — count(most repeated value) ≤ declared — never by a ratio and never in aggregate. most at 1 is the column that carries the test; an aggregate can clear while one value carries five against a declared two, and on the sibling PROTEINS library it did. L9 is the clause that ties this page to what was graded. The checker grades library/compounds/; you are reading Library-Of-Compound-Cures.md. Nothing made them agree until L9 did — it compares the fenced entry blocks of the two as multisets and refuses if they differ, so an edit to this page that is not also an edit to the entry it publishes fails the harness.

GRADE reads 4 admitted entries over 1 distinct value and is labelled census, not a gate, in the output and here. Every admitted entry is graded MEASURED. A clause that never refuses is decoration, and calling a decoration a gate is how a reader comes to trust one — so this library does not claim strength from its grade distribution, it reports it.

Read the two columns together. The row count alone is a loop bound.

---

How this library grows

Who may add

Anyone. The law is the gatekeeper, not a person. There is no reviewer to persuade. An addition is a file, a program and a transcript, and the checker is the referee — it runs in public, on a clean clone, in well under a second once built.

What must accompany an addition — four things, in one commit

1. The entry file in library/compounds/, carrying an affine-entry block the checker admits.
2. The program in reproduce/, if it is new: self-contained, integer, with its own control arm in both directions. A program that cannot fail has proved nothing.
3. A check_figure row in reproduce/validate.sh for at least one of the entry's figures, so the wiki's own harness fails if the page and the program ever drift apart. All five programs behind this library already carry them — thirty rows in total across the five.
4. The manifest ceilings, moved if the addition needs them moved.

Point 4 is not bookkeeping. A stale ceiling silently re-admits what was just excluded. This programme has already paid that bill once, on a float ratchet whose frozen constant kept forgiving what it had been raised to catch. The same failure here would let this library grow past its own declared honesty without a single clause going red.

What happens when evidence is refuted

Nothing is ever deleted. A library that cannot retract cannot be trusted, and a library that retracts by deleting is worse, because it cannot be caught. An entry whose evidence is refuted is superseded in place:

1. add REFUTED_BY with what refuted it and a YYYY-MM-DD date;
2. rewrite GRADE to what the surviving evidence supports — usually NOT_KNOWN;
3. leave the original MEASURED, FIGURE, SEAL and REFUSED lines untouched, so the claim and its refutation stand on the same page;
4. SUPERSEDES on the replacement entry, if there is one.

E14 enforces it: refuted-and-still-MEASURED is refused, an undated refutation is refused because it cannot be ordered against the claim it refutes, and a refuted entry that is dated and regraded is ADMITTED. Retraction is a path through the law, not an exit from it.

This is written hard for a reason. In this substrate's own ledger, the append-only triggers were disarmable at runtime and three migrations used the bypass; for every game they touched, "never CUREd" and "its CUREs were deleted" are now indistinguishable. Deleting a refuted entry is the same act, one step later.

---

What this library is not

• Not medical advice. Nothing here is medical advice and no entry is a recommendation to take anything. Nothing here should change anyone's treatment.
• Not a safety clearance. An off-target map measures where binding or cutting is possible. Whether it happens — in a cell, at a dose, in that chromatin state — is a laboratory question these programs have not asked and cannot answer. Off-target potential is one input to safety among many, and this library measures that one.
• Not an efficacy claim. A signature-inversion score is not efficacy. A recovered master-regulator set is not a treatment.
• Not a ranking. The atlas publishes coverage and targets, not an ordering of substances by risk. Nothing here says one medicine is better than another.
• Not a substitute for a laboratory or a regulator. It is the exact, complete, re-derivable enumeration such a conversation should start from, available without permission and without trusting us.
• Not a claim that an admitted entry is true. ADMITTED means the entry is checkable — named, produced by a program a stranger can run, graded at what its evidence supports, and explicit about what it refuses to say.
• Not finished. Two slots are named and empty, and they are the honest shape of what is missing.

---

Reproduce

git clone https://github.com/gaiaftcl-sudo/uum8dSolarResearch.git
cd uum8dSolarResearch

# grade the libraries — the admission law is the primary artefact, and ALL THREE go in
# ONE run: F1 is a relation between libraries and one library alone cannot answer it
swiftc -O -swift-version 5 reproduce/library-admission-law.swift -o /tmp/lal
/tmp/lal --library library/proteins --library library/compounds --library library/materials \
         --reproduce reproduce --evidence /tmp

# and any single entry's own measurement, e.g. the genome-wide guide map
swiftc -O reproduce/crispr-genome-offtarget-exact.swift -o /tmp/crispr
curl -sL https://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/latest_release/GRCh38.primary_assembly.genome.fa.gz \
  | gunzip -c | /tmp/crispr corpus/crispr-atlas/guides_all.tsv

No account, no key, no data-use agreement, and no floating point anywhere in the exact path.

Each entry above carries its own REPRODUCE line, and none of them names a path private to one machine — that is clause E9, and it is checked rather than promised.

Related

• The library admission law — what may enter, and the 71 arms that prove it refuses.
• The Library of Proteins · The Library of Material Systems — the other two libraries, graded in the same run as this one.
• Where else could this guide cut? — the full per-guide map with coordinates.
• The exact off-target atlas of the nucleic-acid medicines — the registry-wide screen.
• The safety question made exact — zilganersen, with its composition-matched control.
• Study 26 — Master regulator bonds — the study the discrimination court came from.
• Study 37 — 37,910 validated discoveries, five molecules — why the per-library half exists.

Rights — source-available, not open-source

This wiki and its programs are published source-available: the source is visible so anyone can inspect it and re-derive every figure. That visibility grants no rights. The repository carries no LICENSE, which under default copyright means all rights are reserved. Any other use requires a separate written licensing agreement with the authors.