Counting the Torah: a pre-registered permutation study of gematria shape-share in the Hebrew Bible

# Version notes — deposit 4 (October 2026)

Supersedes the deposit of 22 September 2026 (10.5281/zenodo.22886436). The concept DOI
10.5281/zenodo.22718208 always resolves to the newest version.

## New in this deposit

The reopening declared at Cross-Manuscript Registry v2.1.0 is DISCHARGED. All
three manuscript stages have been re-run under Counting Policy v1.0, and one
further sealed protocol was written, failed its own gate, re-sealed and
executed.

- **Stages 1-3, all re-derived under the frozen policy.** MAM at 304,952
  letters; the Samaritan Pentateuch at 316,962, whose census reproduces
  exactly; the Dead Sea Scrolls from 1,001 scrolls and 1,430,241 sign nodes.
  Every Stage 1-3 differential had been computed in July against a superseded
  305,858-letter parse on both sides. They are now computed on one policy.
- **The scrolls forced a decision the counting policy never had to make.** It
  rules on marks, on ketiv and qere, and on final forms, and is SILENT on
  RECONSTRUCTED text, because the Leningrad Codex has none. The scrolls are
  46.3% reconstruction. Both arms are reported separately throughout rather
  than blended, and the reconstructed and preserved verdicts differ.
- **Skip-50, the last quantitative item outstanding across all three stages.**
  Westminster Leningrad 21 against 9.8 expected (ratio 2.14), MAM 19 (1.93),
  Samaritan 6 against 10.5 expected (0.57). The Samaritan figure is BELOW
  chance: on this count the phenomenon is absent there, not merely weakened,
  in a text 4% longer than the Masoretic one.
- **S2.4, the Samaritan chronology — pre-registered, sealed, failed, re-sealed,
  executed.** The protocol was timestamped before an extractor existed. The
  extractor then failed the gate the protocol demanded, which forbade execution
  until a v2.0 re-seal disclosed the repair; five defects are disclosed there,
  two of them errors of fact about what the text contains. Executed, the result
  is NEGATIVE with the power gate passed: the Samaritan totals score 9 against
  a permutation null of 10.41, two-tailed p = 0.745, and the Masoretic 8
  against 8.23. Neither chronology's lifespans are more structured than
  re-pairings of their own components. A sixth defect, found after execution,
  cost the Samaritan arm its blind status; the verdict is identical under both
  parses and the loss is reported rather than absorbed.
- **The metric battery recovered rather than invented.** Fitted only to data
  the archive already held, with Genesis 1:1 held out, and exact on all 5,852
  held-out verses. Every Score in the archive had been unreproducible.
- **A recursive version audit** of every file and folder against its previous
  version, and an audit of supplied calendar-gematria claims that files one
  closed-set test returning one hit in six where chance predicts one.
- **S2.3, the 37-thread across witnesses.** Seven claims, a closed inventory
  fixed before any Samaritan figure was read. The Masoretic gate reproduced
  all seven, and the Samaritan text agrees with the Masoretic on all seven.
  Perfect agreement across traditions -- which is the point: a 37-divisibility
  that holds in both is not evidence of design in either, because both inherit
  the same arithmetic.
- **S1.4, the manuscript sensitivity matrix.** The perturbation did not have to
  be simulated: 157 verses genuinely differ between the Westminster Leningrad
  Codex and MAM, and most of those differences are a single vav. Forty of the
  forty-four verse-naming findings are invariant; four name a differing verse
  and only ONE moves in substance. The belulah flagship is CLEARED for a
  checkable reason -- the word that differs at Exodus 29:40 is "a fourth", not
  belulah. Six of the 157 verses are invisible to standard gematria because
  they sum identically, and ALL SIX move under BoneH, Deuteronomy 6:9 by a
  single unit.
- **S-SDS, the Structural Data Study -- the last sealed protocol in the
  archive, and now spent.** Sealed 23 July 2026 and never run, it returns a
  NEGATIVE that arrived disguised as a positive: 626 of 1,123 derived
  quantities satisfy a primary criterion against 503.00 expected, p = 0.0005
  against both nulls and against the protocol's own declared expectation of
  nothing -- and Joshua through Kings do exactly the same thing at the same
  margin, so the excess measures Hebrew narrative number usage rather than the
  Torah. The geometry arm the protocol was written for returns nothing: under
  its own ban on half-unit rescaling the Ark of the Covenant's volume is 5.625
  and can satisfy no criterion at all. Its secondary criterion and its Null B
  are VOID BY OMISSION -- Appendix A was never filled in, and was not filled in
  retrospectively, because its own rule requires a justification that does not
  reference data this project has already read.

## New since 23 September — the October work

The finding count is unchanged at 58. What this deposit adds is a register of
observations, three source sweeps, one manuscript result, and the machinery
that keeps all of it honest.

- **Part C of the Register exists.** At the September deposit it did not. It
  now holds 142 rows, X.1-X.142: hand-curated connections from external
  sources and the owner's own intake, each marked ★ ◆ ○ or ✕, each with its
  figures recomputed under Counting Policy v1.0 and its base rate stated (how
  many Torah forms carry the value). A row is an observation, never a finding;
  nothing in Part C counts toward the 58.
- **Three source sweeps, filed as their own document families.** inner.org
  (S1, 3,583 connection rows from 188 articles, pulled through the owner's
  browser because the site refuses automated clients), kabbalahempowerment.com
  (S2, 120 rows, pulled by this archive) and satyori.com (S4, 469 pages, every
  Hebrew word of which first arrived as mojibake and was repaired). Each has a
  Connections Register, a Sweep Summary and a declared tier; a sweep logs
  connections and proves nothing.
- **A video source, read through summaries, with its error rate measured.**
  S6 is a YouTube channel taken in through AI summaries because the archive
  cannot read captions from its sandbox. 33 numerical claims across eight
  videos: 22 hold, 10 fail as summarised, 1 is uncomputable, and the failures
  classify as a near-neighbour word swap, a derivation error, or
  unrecoverable. Every S6 row carries TRANSCRIPT -- VERIFY AGAINST AUDIO. A
  reverse-lookup tool lists, under declared moves only, what a wrong number
  could have been meant to be, and refuses lexicon-combination search because
  its base rate was measured and found to hit anything.
- **The Samaritan Pentateuch writes ציצית plene once (X.124).** The
  Cross-Manuscript Registry had recorded the question as the next thing to
  check. Read behind the Genesis 1:1 gate, the Samaritan has the plene singular
  at the second occurrence in Numbers 15:38 -- the only plene singular in
  either Pentateuch -- so the 600 + 8 + 5 = 613 derivation holds on that one
  token of that one witness and fails on every Leningrad token by exactly one
  yud. A manuscript result, not a numerical one.
- **The names of Moses, Jacob and Joseph (X.119-X.123)**, with the three
  recension disagreements in the ten names stated rather than smoothed.
- **The checking apparatus grew and was itself audited.** cascade.py gained a
  version-aware index after it was found checking whichever file the
  filesystem listed last; a missing openpyxl that silently retired the
  spreadsheet check now fails loudly; the Sweep-Summary family entered the
  dependency graph by prefix rule; four QA gates run at every close
  (assertions recompute, settled quantities agree, cascade and transliteration
  tamper suites). One gate's own rule was found witness-blind by X.124 and
  narrowed, with the change recorded rather than made silently. Every build
  that writes a document now recomputes every figure it states from the word
  table before writing, and that guard has caught the archive's own author
  more than once.

## Corrections in this deposit

- **Register finding 1.7 is DOWNGRADED.** The Jacob-Israel count is 799 =
  17 x 47 read morphologically, not 803 = 11 x 73. The 73 does not survive.
- **Genesis 1:1 scores 13, not 15.** Traced to a genuine erroneous score in a
  superseded tool build, corrected in the next, and now derived from the
  recovered battery rather than read off a document.
- **The Samaritan lacuna erratum is WITHDRAWN.** Twelve verses read as missing
  are transpositions, not omissions; the adjustment computed from them was
  wrong and is retracted in both the script and the registry.
- **The deposit file count.** The licence statement gave the published deposit
  of 17 September as 118 files in four documents. It holds 91. The 118 was the
  local package, rebuilt since the upload and never published. Both sets hold
  zero .txt and zero .xml, so the conclusion stood and only the supporting
  count changed.
- **A reader bug that produced a complete, readable and entirely fictitious
  table.** The Dead Sea Scrolls feature reader skipped empty-valued lines,
  desynchronising all implicit node numbering. Nothing crashed. It was caught
  only because the results were implausible, and a reader self-test was added.
- **Three findings are now known to rest on the superseded pre-freeze parse**,
  which counted the setumah and petuchah paragraph markers as letters. Finding
  1.9 was corrected at Register v4.15, 1.10 DOWNGRADED at v4.17 -- half the
  37-thread is an artefact of that parse -- and 1.15 DOWNGRADED at v4.18: its
  Exodus 20:1 leg states the pre-freeze 1392, and the frozen 1332 has no Pi
  prefix-sum position at any index, so the leg is withdrawn rather than
  restated. All three were found by a gate or a sweep, never by reading, and
  no claim is made that the list is exhaustive.
- **The pre-freeze sweep, and the fourth finding it found.** Three findings had
  been caught resting on the superseded parse -- the engine that counted the
  setumah and petuchah paragraph markers as letters -- and every one was found
  by accident. This is the first time the question was asked directly: both
  parses recomputed and compared, so a figure counts as contaminated only when
  it equals the verse's pre-freeze value and not its frozen value for the same
  quantity. 13,951 verse references read, 21,705 figures tested. FINDING 1.28
  IS CORRECTED: its two comparison verses both end in a setumah the old engine
  counted as a word, so Leviticus 10:20 is four words at 20.00% and Deuteronomy
  5:20 is five at 9.68% -- below the ten per cent threshold the claim itself
  uses. The sentence is also independently false: 22 verses exceed 10%, not
  three, and no length restriction yields "only two". The core survives with a
  length control, where Genesis 1:1 is first of five. The Database's calendar
  sheet carried one pre-freeze verse total, Genesis 8:14 at 2,761 against the
  frozen 2,701, now corrected; its All Verses sheet came back clean on all
  5,853 rows across three quantities. The paragraph-marker census is corrected
  from 683 to 687, making the decomposition 1,008 = 687 + 321, with one letter
  unaccounted for and reported as such.
- **A standing caution on criterion density, measured rather than asserted.**
  Triangular, star, prime and perfect between them flag 0.664 of the integers
  1-9. At the 500-5,000 scale the Register's metric battery actually works on
  they flag 0.146, so this is not a downgrade of the battery -- but any claim
  that scores a SMALL integer is scoring something these criteria flag about
  half the time.

## Licence — every external source is now cited, not redistributed

The two remaining vendored third-party corpora were removed: eleven Hebrew
works from Sefaria whose licences vary per work AND per version and had never
been verified, and a copy of the tikkunkorim project, which states no licence
of any kind. 63 files, 38.7 MB. No script in the archive read either, and no
result depends on them; what made the work reproducible was the acquisition
script plus the pinned version index, and both are kept.

Every external source this project touches is now cited rather than
redistributed, with one exception: the Westminster Leningrad Codex, whose
CC BY 4.0 licence permits it. Full per-source attribution, including the
non-commercial terms on the Samaritan and Dead Sea Scrolls datasets, is in
06_Reference/THIRD-PARTY-SOURCES.md.

**This deposit contains no third-party source text**: no .txt, no .xml, no
Hebrew source corpus of any kind.

## Document versions in this deposit

- `Ana-BKoach-Values_Siddur_v1.0_2026-09-13_JE.csv`
- `Archive-Database_Torah-Numerology_JE.html`
- `C1-C8-Results_Seven-Three-Partition_WLC-Leningrad_v1.0_2026-09-23_JE.docx`
- `Connections-Register_inner-org_v2.1_2026-09-24_JE.csv`
- `Connections-Register_kabbalah-empowerment_v1.1_2026-09-24_JE.csv`
- `Connections-Register_satyori_v1.0_2026-09-23_JE.csv`
- `Counting-Policy_WLC-Leningrad_v1.0_2026-09-07_JE.docx`
- `Cross-Manuscript-Validation-Registry_Multi-Tradition_v2.16.29_2026-10-24_JE.docx`
- `Drive-Audit-and-File-Manifest_Numerology_v3.0.39_2026-10-24_JE.docx`
- `Literature-and-Prior-Art-Review_Multi-Source_v1.0_2026-09-06_JE.md`
- `Methodology-Inventory_Torah-Numerology_v1.4.36_2026-10-24_JE.docx`
- `Novel-Findings-Register_WLC-Leningrad_v4.50_2026-10-24_JE.docx`
- `Phase10-Rerun-Results_WLC-Leningrad_v1.1_2026-09-14_JE.docx`
- `Phase11-Results_WLC-Leningrad_v1.0_2026-09-12_JE.docx`
- `Phase13-Results_WLC-Leningrad_v1.0_2026-09-14_JE.docx`
- `Pi-Calibration-Results_WLC-Leningrad_v1.0_2026-09-12_JE.docx`
- `Pre-Registration-42-Letter-Name_WLC-Leningrad_v1.0_2026-09-13_JE.docx`
- `Pre-Registration-Fibonacci-Phi_WLC-Leningrad_v1.0_2026-09-12_JE.docx`
- `Pre-Registration-HA-Replication_WLC-Leningrad_v1.3_2026-09-08_JE.docx`
- `Pre-Registration-Samaritan-Chronology_Multi-Tradition_v2.0_2026-09-19_JE.docx`
- `Pre-Registration-Seven-Three-Partition_WLC-Leningrad_v1.0_2026-09-23_JE.docx`
- `Pre-Registration-Word-Value-Search_WLC-Leningrad_v1.2_2026-09-08_JE.docx`
- `Pre-Registration_Structural-Data-Study_v1.0_2026-07-23_JE.docx`
- `S-SDS-Results_Structural-Data-Study_WLC-Leningrad_v1.0_2026-09-20_JE.docx`
- `S1.4-Results_Manuscript-Sensitivity_Multi-Tradition_v1.0_2026-09-20_JE.docx`
- `S2.4-Results_Samaritan-Chronology_Multi-Tradition_v1.0_2026-09-19_JE.docx`
- `Source-Manifest_Multi-Source_v1.7.7_2026-10-24_JE.md`
- `Source-Provenance-515-Vaetchanan_Multi-Tradition_v1.0_2026-08-18_JE.docx`
- `Source-Reference-Index_Torah-Numerology_v1.3.0_2026-09-23_JE.md`
- `Sweep-Summary_inner-org_v2.7_2026-10-24_JE.docx`
- `Sweep-Summary_kabbalah-empowerment_v1.6_2026-10-24_JE.docx`
- `Sweep-Summary_satyori_v1.5_2026-10-24_JE.docx`
- `Torah-Numerology-Book_WLC-Leningrad_v2.5.37_2026-10-24_JE.html`
- `Torah-Numerology-Database_WLC-Leningrad_v2.3.38_2026-10-24_JE.xlsx`
- `Torah-Numerology-Working-Paper_WLC-Leningrad_v2.3.36_2026-10-24_JE.docx`
- `Torah-Word-Values-Database_WLC-Leningrad_v1.0_2026-09-08_JE.xlsx`
- `Torah-Word-Values_WLC-Leningrad_v1.0_2026-09-08_JE.csv`
- `Triage_inner-org_v1.1_2026-09-23_JE.csv`

## How this deposit is arranged, and why it changed

**Zenodo caps a record at 100 files.** Its own documentation says so, and adds
that twenty or more files should be packaged into an archive. The flat package
reached 210 files on 22 September 2026 and could not have been published at
all; the September deposit fit only because it was 91.

So the documents sit at the top level, where each can be opened, linked and
cited, and the rest travel in three archives:

- **`timestamps.zip`** — the proof chain: every stamping's certified manifest
  with its `.ots`, `.tsq` and `.tsr`, plus `timestamps_README.md`.
- **`scripts.zip`** — every program: the statistical engine, the
  cross-manuscript analyses, and the audit machinery that keeps the archive
  honest.
- **`results-data.zip`** — the raw results, as JSON.

The split is by kind, not by importance. An importance rule needs a judgment
call per file, and judgment calls nobody wrote down are this project's recorded
failure mode.

**The cross-manuscript programs are in this deposit for the first time.** Every
results document since 19 September cites them by path — `crossms/chronology.py`,
`crossms/s14_sensitivity.py`, `crossms/sds_collect.py` — and until now a reader
following a citation found nothing. Under a flat file budget they looked
expensive; bundled, they cost nothing.

## Integrity

`MANIFEST.md` lists every tracked file in the archive with its SHA-256, and
`DEPOSIT-MANIFEST.md` lists every file in this package with its SHA-256, its
source path in the archive, and which archive it travels in. Anchoring the
manifest anchors the lot: to prove any one document, show that its hash appears
in the manifest and verify the manifest against a proof.

**Every SHA-256 in `DEPOSIT-MANIFEST.md` is the hash of the file itself, taken
before compression.** Extract a file from one of the archives and its hash
matches the manifest unchanged — bundling moved where a file sits, never
whether it can be verified.

The stampings carried in `timestamps.zip`:

- **2026-09-11** — 214 files, `MANIFEST.md` SHA-256 `64a19a0efbdb633c0d9ec1d69631e014a1974d25a3b0a76ae261d8be8cdecd1d`
- **2026-09-17** — 282 files, `MANIFEST.md` SHA-256 `d3059d985d1c1b8134bbb8fad60ee6032c834c6b2f8b145f92443b5a331f0824`
- **2026-09-17_2** — 291 files, `MANIFEST.md` SHA-256 `55c5704a0e354393017e9b306927852e848e870d0c10c7438627db013942f58e`
- **2026-09-17_3** — 293 files, `MANIFEST.md` SHA-256 `d75b7c256ea0bce3b5d8f0b16a4289d0f020e26fe3951dcd864dee8784262fda`
- **2026-09-17_4** — 294 files, `MANIFEST.md` SHA-256 `d9fbc7480e895b8d24a8b295eedb4f561bbe852f014088ab6268917863e8649a`
- **2026-09-17_5** — 288 files, `MANIFEST.md` SHA-256 `76b7434dae9065eacdf27ffbdaeb2b221c5002fc0018846d4847a8eeb2ecaf9c`
- **2026-09-19** — 288 files, `MANIFEST.md` SHA-256 `12986fdecddf402baad999bdfc3fc1d5fa20d2f3ca0c1a2b6d827a67931fbbb2`
- **2026-09-19_2** — 292 files, `MANIFEST.md` SHA-256 `7e0dbc8b9f6d8c6b53b58b6ebeb057a34324c12f22ad31bd15ee435a9f73fcdb`
- **2026-09-19_3** — 297 files, `MANIFEST.md` SHA-256 `3c65d0269527d8c836dfe1a229b9da83139425e6386ededdbec0f101fa239709`
- **2026-09-19_4** — 349 files, `MANIFEST.md` SHA-256 `04ad51593838f9467a9edbde3eaf76a09ce10ac58e40e3aea381d2ac3f6ac69e`
- **2026-09-19_5** — 348 files, `MANIFEST.md` SHA-256 `ca36e6d1054ef83f1548f18ded90b0a6aa357f3a065141689338789d178ab52b`
- **2026-09-19_6** — 359 files, `MANIFEST.md` SHA-256 `6804eca003a6ba49514dd5e331fa44d04983184b140214db4d158ec2daceba79`
- **2026-09-19_7** — 300 files, `MANIFEST.md` SHA-256 `3b3a16d82268924e68b81b0a912c08454b605e04dbe8f1987536c2839b5a60f6`
- **2026-09-19_8** — 302 files, `MANIFEST.md` SHA-256 `f01ef02fd96b437bd58bdd1ba7a2d79e717b8d0291a631ef577a87edb782ebb1`
- **2026-09-20** — 304 files, `MANIFEST.md` SHA-256 `0c71d48a68b8c2a6c02ed6cc2d1af4add9647bf6da7331c448e9f62f8299de32`
- **2026-09-20_2** — 306 files, `MANIFEST.md` SHA-256 `55db407e33a0350c59171b07b8a97eaece1bada1c0f8233e378de80ed1eb359c`
- **2026-09-20_3** — 319 files, `MANIFEST.md` SHA-256 `1fe36bc61742deeb649970b4edf6ff48e4c23245d8dbea56c12c2eea89b002cf`
- **2026-09-20_4** — 333 files, `MANIFEST.md` SHA-256 `c8fe87993b98fcc39a2522059737a09f5f5a488c140a5698b5f92dcef5fe801e`
- **2026-09-20_5** — 352 files, `MANIFEST.md` SHA-256 `39f24056ba0f097c919d52f6607c32d2862097a314a3bdbb1842b722b6c0167b`
- **2026-09-22** — 364 files, `MANIFEST.md` SHA-256 `ed4960f3dec9aad583b61c432c561de5713cf7d6739e3c7cc72ad082a70a0cb9`
- **2026-09-22_2** — 364 files, `MANIFEST.md` SHA-256 `9f02c5348eaaf7b6c5102af79d8b754cd400ad9c136f1f3cf1074e66da88ddf5`
- **2026-09-22_3** — 364 files, `MANIFEST.md` SHA-256 `bcdbe79823dc69235b54926ac5e820bb12692a496edb9104b3fbb25e09ff75c9`
- **2026-09-22_4** — 363 files, `MANIFEST.md` SHA-256 `c5fe041c03401c2398f3fa108b1d6bbd97a74593ef58971fb33988bf3daa1f33`
- **2026-09-22_5** — 364 files, `MANIFEST.md` SHA-256 `bf78be8c4c4bd68bcf75da5565bef41d179344cabdaaa0c0273f281cacc1b8c5`
- **2026-09-23** — 376 files, `MANIFEST.md` SHA-256 `2f0f0ef8d5ccf39824d1dd635124bc306d01f9f11d0e345174a78d625a38cd24`
- **2026-09-23_10** — 499 files, `MANIFEST.md` SHA-256 `2facee72763a75c07a12b2ab70b285bb56a9ba9da48674fde1e4f07056d378e7`
- **2026-09-23_11** — 519 files, `MANIFEST.md` SHA-256 `4615b00bbfe19fb164e1d508ce2422a8799de0024f6926632fd25f297a4b6578`
- **2026-09-23_12** — 528 files, `MANIFEST.md` SHA-256 `fd0967f730f6b31f4cc6cbc84c0371773f77f89f988f487b350ac1a272f88163`
- **2026-09-23_13** — 538 files, `MANIFEST.md` SHA-256 `1a383ce8d35fc0476f7ce8053dbb0d2c9be84e6f2676d723da8e32c5f6ec1c84`
- **2026-09-23_14** — 550 files, `MANIFEST.md` SHA-256 `7f4bff028b58cf10fad3bff7330e1c9df901aed3c79a235e6408c7797db31643`
- **2026-09-23_15** — 561 files, `MANIFEST.md` SHA-256 `e3ed9cd7dcf156bd8ce7c3aea1d319fa2d8ecf399ed0129f2d3773906079eb72`
- **2026-09-23_16** — 569 files, `MANIFEST.md` SHA-256 `016510ce29e665b3191d334a2a8bdb4df65ce662df643f0da3ba6e336a3b5028`
- **2026-09-23_17** — 578 files, `MANIFEST.md` SHA-256 `fcc001c92a9fafd39069fa57c75f85233e97e86df34d22930fe4a70d7f69fb04`
- **2026-09-23_18** — 587 files, `MANIFEST.md` SHA-256 `8339ac6d6a673d38c6bc6849ecf3896d2ac48010f47d0d4e02d36ebe6058871c`
- **2026-09-23_19** — 606 files, `MANIFEST.md` SHA-256 `bf0be9064e5162f3530fdb0995821800f2d4110c8c786f79d679d2449d3ce1e1`
- **2026-09-23_2** — 388 files, `MANIFEST.md` SHA-256 `443134fba55939f3e8f7ccff70c4f6a6d5702d71afdbcbc9a466d2f3c439a40d`
- **2026-09-23_20** — 618 files, `MANIFEST.md` SHA-256 `9c1027dc579f010c98b76a8966caeb60457a359cb5c63a003e18a66d2639b3b8`
- **2026-09-23_21** — 634 files, `MANIFEST.md` SHA-256 `c8fb260293e65eba81ddc55c452cf8124e306e8de2ac945999b3bfb8ef1646f4`
- **2026-09-23_3** — 403 files, `MANIFEST.md` SHA-256 `79b57ae0b78290288588d991ec7e5d9bd97ba2e84f39737badeaa550a2d7287b`
- **2026-09-23_4** — 415 files, `MANIFEST.md` SHA-256 `72a48c40740e8ec5773222de5132f679dc5e2d25dcc913be52596bef0e3c848b`
- **2026-09-23_5** — 426 files, `MANIFEST.md` SHA-256 `b052a9edff75b80f06b1760447fc106583e0065010d1b799e16279c23ff0e7c2`
- **2026-09-23_6** — 435 files, `MANIFEST.md` SHA-256 `c8e2a761937a3f232d925034c6cc271c1ba60b46062d29fe8b96993686571f43`
- **2026-09-23_7** — 444 files, `MANIFEST.md` SHA-256 `544fcdea4b3f41ae6395a2b5bccdf0e93f5129f3d4d77d931000722ce9b5b1bf`
- **2026-09-23_8** — 456 files, `MANIFEST.md` SHA-256 `f9015946458a7cf1656e0fba8c4bce7fb4670c56e9303d4bf0f5d5f2c838241e`
- **2026-09-23_9** — 467 files, `MANIFEST.md` SHA-256 `07ffaaf22adb9295f658730ddb6817d29a9a691704cc4b60c279f4a5ea95cb5c`
- **2026-10-24** — 738 files, `MANIFEST.md` SHA-256 `8d12b05e1b1c6713f87a7110f7cb954bfbc91017a9de8bc8174a691b395f0ca0`

Each has an OpenTimestamps proof (`.ots`, anchored into the Bitcoin blockchain,
trusting no one) and an RFC-3161 signed timestamp (`.tsr`, from freeTSA.org,
immediate but trusting that authority). `timestamps_README.md` records the
signing times, the Bitcoin block where confirmed, the confirmation status of
each proof, and the commands to verify them. Those details are deliberately not
restated here: a Bitcoin anchor confirms hours after it is submitted, so a
figure copied into this file would be out of date the moment it was written.