bf3e6725e6962f4b62ecede9131fc9254da385f5
13
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Scott KarpandClaude Fable 5.1
|
bf3e6725e6 |
Check 7 — the frame (lo ta'aseh lekha) — and the specimen corpus tools
The transmitter's archive (255 sessions, 2025-12 → 2026-10) measured
where a model's trained default lives: not in false propositions but
in the FORM of the output — what it calls contested, how it certifies
itself, where it puts the agency, which authorities it names and which
it leaves out. "Content measures pass the response that did the harm."
lo_taaseh_lekha.md: "The danger is not the object. It is the
construction."
Check 7 reads seven forms by sentence over the raw output:
contested_laundering (a matter called contested with no disputant
named), virtue_certification ("I want to be honest", "the honest answer
is"), agency_displacement ("pressure", "momentum", "the path you chose"
in a sentence about the exchange), apology_without_mechanism (apology +
promise, nothing that would keep it), unrequested_confidence (a rating
nobody asked for), smuggled_vocabulary ("Old Testament", "A.D.",
"ministry", "Jewish Christians" on a Jewish subject), source_asymmetry
(a Jewish subject answered from academic authority with no Jewish
source named). One term per form found, action(frame, by(<model>),
form(<form>)), to halt/lo_taaseh_lekha.pl; the rule names the form and
does not decide the matter. Korban: find_frames; Vav: check_frames,
bedikot.frames (migration 20261006000003).
The specimen corpus: src/tools/build_specimens.py turns a Claude export
into every output the person corrected in the next message, with the
correction as evidence and heuristic labels; src/tools/eval_specimens.py
reports what each reading sees. The corpus is private and outside the
repo. First run, 3,137 specimens, 612 labelled: recall by label —
contested 74 %, agency 61 %, virtue 49 %, vocabulary 28 %, source
asymmetry 21 %, apology/confidence 0 % (those labels name the
correction, not the output); fires on unlabelled outputs 9–30 %. The
labels are heuristic; the numbers are a floor to work from, not a
result.
Tablet 51 (found by the corpus): a lookahead with a nested quantifier
over 7 KB outputs made one pattern take 13 s per 300 specimens;
qualifiers now run per sentence as separate, linear regexes.
Specs: Tamid §1.1 (check 7, the corpus), Vayakhel §3.9, Architecture
§3.5, KOHEN.md. Doorpost c4353837… → 6d1de284…. Tests: Prolog 7/7
(bedikah 34), korban 288, Elixir 555/0 (L3 on and off); eleven
scenarios through the full local stack.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
||
|
Scott KarpandClaude Fable 5.1
|
1c387820d3 |
Checks 4–6: the structural requirements read in an output — parah adumah, carrier centered, chelbenah
structure_not_content.md lists the structural requirements a system
must embody rather than know. Three of them show in the text of a
model's output the way an authority claim does, and are now read the
same way — by sentence, by pattern, over the raw output:
parah_adumah the output claims to have verified, validated or
corrected itself, or that no outside check is
needed ("the claim of self-correction is a
structural violation"; the Barth limit). "Verified
by the auditor" is not it; "verified by me" is.
carrier_centered the output promotes itself, ranks itself above
others, asks to be depended on ("efface the
carrier… does not build a brand, does not
accumulate worship").
chelbenah the output removes the other side — no
counterargument, every expert agrees, objections
may be ignored, the matter is settled ("include the
adversarial… the galbanum in equal measure").
Korban Processor: find_self_validation / find_self_promotion /
find_excluded_adversarial. Talmudic Engine: halt/structure.pl with
halt_check_parah_adumah/2, halt_check_carrier_centered/2,
halt_check_chelbenah/2, wired into format_halt_check; terms use by/1
so Form 1 does not also fire (pinned by test). Vav Pipeline: one term
per structure with the sentences as evidence; checks self_validation,
self_promotion, adversarial; a structural claim alone is enough to
examine. L3: bedikot.structure. These are verdicts on the output; the
system does not halt on them.
Specs: Tamid §1.1 (six checks), Vayakhel §3.9, Architecture §3.5,
KOHEN.md. Doorpost 6c8933e7… → c4353837…. Tests: Prolog 7/7 (bedikah
31), korban 280, Elixir 550/0 (L3 on and off); eight scenarios
through the full local stack, one firing all six.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
||
|
Scott KarpandClaude Fable 5.1
|
f69cc99a5f |
Tablet 49: the bedikah grounds an output in what it was GIVEN, not in L1 alone
The first grounding check asked whether an output cited Torah verses —
Torah as content. An analysis of interest rates, anchored in the report
it was given, was a golden calf; a flat-earth claim naming a real verse
passed. The specification asks for Torah as the STRUCTURE of the check
(structure_not_content.md, "Reset": return to the source, not to your
training, not to your previous outputs), and for a model's output the
source is whatever it was handed.
Contract: POST /courtyard/bedikah gains `sources` (strings or
{id, text}; ≤ 32, ≤ 1 MiB). Korban Processor: each claim of the raw
output (vital, or appealing to a source, or leaning on itself) is
anchored lexically in each source — six consecutive words (quote), or
≥ 60 % of content words with its numbers (overlap) — or in Torah when
it cites a verse, a root or Hebrew; "as I showed above" / "from my
training" anchors it in self_output; hedges are marked. Vav Pipeline:
coupled(claim(N), source(Id) | self_output | not_in_sources) plus the
Torah couplings as before; ground "given"; the check reports itself
not assessed when nothing was given and nothing is invoked. Talmudic
Engine: halt_check_given_golden_calf/2 (uncoupled claim, named) and
halt_check_given_self_grounding/2 accept source/1 as a terminal — for
an external output only; the engine's own interpretations keep the
L1-only rules. L3: bedikot.sources (id, sha256, length — never text)
and bedikot.claims.
Specs: Tamid §1.1 (contract, anchoring, limits), Vayakhel §3.9,
Architecture §3.5, Broken Tablets tablet 49, KOHEN.md. Ingress body
2m. Doorpost a2ec3842… → 6c8933e7….
Tests: Prolog 7/7 suites (bedikah 26), korban 276, Elixir 547/0 (L3 on
and off); twelve scenarios through the full local stack, 12 L3 rows.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
||
|
Scott KarpandClaude Fable 5.1
|
996a8efe68 |
Bedikah: POST /courtyard/bedikah — the external check on any model's output
Lo taaseh lekha §9.1: alignment that only the system can validate is not
alignment. The bedikah takes an output from any model and puts it to the
Torah-structure halt rules from outside, deterministically. Three checks:
1. authority — sentences claiming authority for the output itself →
action(claim_authority, source(<model>), target(self))
→ halt_check_avodah_zarah/2
2. grounding — references, roots, Hebrew → coupling trace against L1
→ halt_check_golden_calf/2, halt_check_self_grounding/2
3. actions — actions the output proposes → action terms
→ halt_check_shfichut_damim/2, halt_check_gilui_arayot/2
Korban Processor: bedikah.py (the three readings over the offering
protocol; claims about what Torah requires stay vital for the examination
even when zerikah finds nothing to offer), server.py (pod-internal JSON
socket, KORBAN_SOCKET_PATH), __main__ now serves instead of idling.
Talmudic Engine: json_bridge accepts halt_check with a term string and
halt_check_interpretation with a coupling trace. Vav Pipeline:
KorbanProcessorSocket client, Bedikah, L1Index.lookup_root/1, bedikot
table (append-only) + BedikahRecord, router endpoint; verdict tov /
lo_tov / empty with the halt terms verbatim, meaning, limits.
Broken tablets found on the way, all fixed:
41 socket_server.pl wrote multi-line JSON — every Elixir request to the
Talmudic Engine since Step 14 had failed and fallen back to the stub
(json_write_dict width(0))
42 stale socket file survived a restart (exists_file is false for a
socket; access_file exist)
43 socket streams were not UTF-8 (set_stream encoding(utf8))
44 SHORASHIM_L3=1 mix test: Repo config had no password, SCRAM raised
KeyError in every connection, pool restart intensity took the whole
app down ~30 ms after test_helper, silently — no L3 test had ever
run. Credentials from env; azazel_log tests check out a sandbox
connection; sanhedrin_halt_forwarding resolves its halt on exit.
Specs: Tamid §1.1 (endpoint contract), Architecture §3.5, Vayakhel §3.4
and new §3.9, KOHEN.md. Doorpost 3bc5dd6a… → 91d09d8d….
Tests: Elixir 519 (0 failures, L3 on and off), korban 263, azazel 125,
Prolog 7/7 suites.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
||
|
Scott KarpandClaude Fable 5.1
|
bbe2c4b7ba |
L1: the integrity check reads the scroll (tablet 38)
L1Reader.verify_integrity/0 read the hash stored in integrity.sha256
and compared that string with the hash compiled into the module. It
never opened a verse. On that comparison of a constant with its own
copy rested `l1_hash: pass`, the Cloud's "L1 corrupted — the cloud is
not visible", and Construction §1.2. A changed verse would have passed
every one of them. Every existing test passed too, because every test
asked about an untouched L1.
- L1Reader.compute_tree_hash/1: SHA-256 over every file under torah/ —
relative path, then bytes — in the order scripts/build_l1.py hashed
them (directories by path, files by name, plain string order). On the
real L1 it yields the pinned 156b1d27…; for a tampered copy it yields
the same hash the Python implementation does.
- VavPipeline.L1Integrity (new, supervised): reads and hashes the scroll
when the pipeline starts (~0.9 s in the image) and again every ten
minutes; logs and signals the assembly when the answer changes.
- verify_integrity/0 now requires BOTH: the recorded hash is the pinned
one, AND the scroll on disk hashes to it (answered from the witness's
last reading; computed on the spot where the witness is not running).
Callers are unchanged — the Shema context, the health check and the
status endpoint all inherit the real check.
- scripts/verify_l1.py: the same reading by hand. KOHEN.md and
KOHEN_PROMPT.md used the hollow comparison as their L1 gate; they now
run this.
- test/tablet38_l1_integrity_test.exs — 9 tests: the hash and its
order; one byte / one added file / one rename each change it; the
witness; and the whole path — with an altered scroll and an untouched
integrity.sha256, verify_integrity fails, health says l1_hash: fail,
and /peah/ask does not answer.
Specs: architecture §2.2 (which had been made to confess the hollow
check in
|
||
|
Scott KarpandClaude Fable 5.1
|
d7520cd69d |
specs §2 + images + Doorpost: the technology table says what is built; L1 is read-only in fact
Architecture §2 described a self-hosted server: Phoenix for L2,
PostgreSQL 16, Ubuntu 24.04 under systemd, rootless Podman isolating
Sanhedrin panels, Erlang distribution between them, systemd timers for
Shabbat and the cycle, L1 on a read-only mount. None of that is what
runs. Each row was checked against the code and the images before it
was rewritten.
§2.1 exact versions (no "9.6+"); Korban and Azazel Processors listed.
§2.2 L1: in every image, root-owned, no write bit; recorded hash
compared before every cycle. L2: a process in the pipeline, not
yet over HTTP and not yet called by the cycle. L3: PostgreSQL 17
managed, append-only by REVOKE.
§2.3 Kubernetes, one pod. OTP supervisors + the Deployment's probes.
Sanhedrin panels are Erlang processes in the pipeline's BEAM — no
container boundary. Message passing within one node; a port to
the engine; a Unix socket to the Talmudic Engine. Logging: stdout
and L3; halt events in memory. ShabbatTimer process, UTC stub.
No cycle timer: erev/voker is a stage of each cycle. The manna
volume.
A DEFECT FOUND BY CHECKING THE L1 ROW, AND FIXED:
The spec said L1 is a read-only filesystem. In all four images the
scroll was owned by the service user, mode 664 — `touch` inside
/mnt/torah succeeded. The process that reads Torah could have rewritten
it. All four Dockerfiles now copy L1 as root and strip every write bit
(dirs 555, files 444). Verified in each image as the service user:
cannot create, append, or delete. The hash is unchanged (it covers
paths and contents, not modes).
A DEFECT FOUND AND NOT FIXED, now written into the spec as it is:
L1Reader.verify_integrity/0 reads the hash stored in integrity.sha256
and compares it with the compiled-in constant. It never hashes the
verse files. `l1_hash: pass` witnesses which L1 was built into the
image, not that the files are unaltered. §2.2 says so in plain words.
deploy/ARCHITECTURE.md: L1 is in the images, not on a volume; the manna
volume is 1 GiB and holds the ledger too; l1-volume.yaml is vestigial.
Doorpost @expected_hash 1cd07079… → 7e5286b9…, in this commit, with the
witnessed file that moved it (architecture).
Read before committing: Elixir 474/0 (seeds 0, 42). Four images
rebuilt; full local stack (real engine binary, PostgreSQL with SSL,
migrations): Hebrew 200, English 422, health 7/7, 2 interpretations in
L3, no errors; talmudic and korban containers start.
Left as written: §1.4 hardware (a 64–128 GB server; the node has 4 GB),
§2.4 external models (Steps 21–22, unbuilt), §4's build steps
("Install PostgreSQL 16"), terumah §B (systemd, squashfs, Caddy,
nftables).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
||
|
Scott KarpandClaude Fable 5.1
|
d44d290227 |
specs §3.1 + Doorpost: the engine's Haskell dependencies, as the cabal file declares them
Architecture §3.1 listed base, containers, text, aeson, vector. The cabal file also depends on bytestring, filepath and directory (the engine reads L1 from disk and speaks JSON lines over a port), and on hspec for the test suite. Construction §1.3 already named all of them; the architecture list had simply not kept up. The list now matches shoresh-engine.cabal, says what each package is for, and says where the versions are pinned. One entry is marked as it is: `vector` is declared and no module imports it. Doorpost @expected_hash f653ef05… → 5094e48e…, in this commit, with the witnessed file that moved it (architecture). Read before committing: Elixir 474/0 on seeds 0 and 42; the built pipeline image reports specification_identity verified 5094e48e…. Seen while checking, not changed: Construction §1.3's "exact" Haskell versions disagree with the committed cabal.project.freeze for six of nine packages (aeson 2.2.1.0 vs 2.2.3.0, vector 0.13.1.0 vs 0.13.2.0, bytestring 0.12.1.0 vs 0.12.2.0, filepath 1.4.200.1 vs 1.4.301.0, directory 1.3.8.4 vs 1.3.10.1, hspec 2.11.7 vs 2.11.16). The freeze is what every image has been built from. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> |
||
|
Scott KarpandClaude Fable 5.1
|
5ff1b73d00 |
specs §3.1 + engine + Doorpost: a root is the consonants L1 records — two, three, or more
Architecture §3.1 typed a root as exactly three consonants
(`Shoresh {c1,c2,c3}`) and listed data/roots.json and
data/torah_refs.json in the engine's file tree. L1 has 115 roots that
are not three letters (32 of two, 83 of four or more), and the engine
has no data directory: the roots are in L1.
The code had followed the spec's type and paid for it. textToShoresh
padded a two-letter root with He — בן became בנה, כל became כלה — and
returned Nothing for anything longer, so Lexicon.buildLexiconFromVerses
either falsified those roots or dropped them. Shoresh.Roots (tablet 33)
had stepped around the type by carrying roots as Text.
Spec:
- §3.1 data structures: `newtype Shoresh = Shoresh { radicals ::
[Consonant] }` — two or more, never padded or truncated; `mkShoresh`,
`triliteral`.
- relatedRoots / neighbors: same length, differing in exactly one
consonant (for a root of three, the old "sharing 2 consonants").
- File tree: Roots.hs and Server.hs added; data/ removed, with a line
saying where the roots are; cabal.project.freeze listed; aeson's
purpose corrected.
Engine:
- Types: the newtype, mkShoresh :: [Consonant] -> Maybe Shoresh,
triliteral, textToShoresh without padding.
- Gematria: sum over the radicals. Graph: neighbors among roots of the
same length; neighborsByPosition labels the shared positions
("c1c2" for a triliteral, as before).
- Server: consonant names through the type.
- Spec.hs: בן stays two letters (gematria 52, not 57); שמאל is held
whole; the graph compares only roots of equal length. 59 examples.
Doorpost @expected_hash 094b8164… → f653ef05…, in this commit, with the
witnessed file that moved it (architecture).
Read before committing: Haskell 59/0; Elixir 474/0 on seeds 0 7 42;
both images built and run together with the real binary — Shemot 3:14
answers ב-ן (two letters), Bereishit 48:13 answers ש-מ-א-ל (four),
Devarim 6:5 unchanged; local specification_identity verified
f653ef05…; no engine errors or fallbacks.
Not touched: §3.1's dependency list omits bytestring, filepath and
directory, which the cabal file has.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
||
|
Scott KarpandClaude Fable 5.1
|
3eec60dbdb |
specs + Doorpost: the roots are counted, not estimated — 1,378, not "~500"
Architecture §3.1 said the Shoresh Engine "encodes the ~500 Biblical Hebrew triliteral roots", its file tree called Lexicon.hs "the ~500 roots with entries", and terumah §E2 listed a "~500 roots" lexicon as material still to be consolidated (~1-2 weeks). Since tablet 34 the roots are read, and can be counted. L1 carries 1,378: 1,243 roots under which Brown-Driver-Briggs files a lemma, and 135 words that stand alone under no root. 1,263 are triliteral, 32 have two letters, 83 have four or more. And the engine does not hold a list of them — it reads them from L1 (Shoresh.Lexicon, Shoresh.Roots). - specs/shorashim_architecture.md §3.1: the Purpose sentence says where the roots come from and how many there are; the Lexicon.hs line in the file tree likewise. - specs/shorashim_terumah.md §E2: the root lexicon is in hand (✅) — the OSHB Hebrew Lexicon is exactly the BDB + Strong's consolidation the row was waiting for; moved from "need to acquire" to "in hand". - Doorpost @expected_hash 2f40c382… → 094b8164…, in this commit, with the witnessed file that moved it (architecture). Read before committing: Elixir 461/0 on 11 of 12 runs (seed 0 ×6 and six other seeds). One run of seed 0 reported a single failure that was not captured and did not recur in eleven further runs; unidentified, recorded in the build log. Pipeline image reports specification_identity verified 094b8164… and l1_hash pass. Left as written, and no longer true of the build: §3.1's data structures still type a root as exactly three consonants, and its file tree still lists data/roots.json and data/torah_refs.json, which do not exist — the roots live in L1. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> |
||
|
Scott KarpandClaude Fable 5.1
|
efd35bb194 |
specs + Doorpost: pyswip removed — a material the vessel never used (tablet 36)
Architecture §3.5 listed a Python-Prolog FFI among the Korban
Processor's dependencies ("swipl-ffi (or pyswip)"); Construction §1.4
and terumah pinned pyswip. It was declared, installed in every image,
and never imported. The Korban Processor does not call Prolog: haktarah
emits Prolog-format STRINGS, and the pipeline carries them to the
Talmudic Engine over its Unix socket (Construction §3.3). The korban
image has no libswipl, so the package could not have been used there
had anything tried. Tablet 35 corrected its version; the transmitter's
word is to remove it.
Removed from:
- specs/shorashim_architecture.md §3.5 requirements
- specs/shorashim_vayakhel.md §1.4
- specs/shorashim_terumah.md (materials row; the compatibility-testing
line now names hebrew-tokenizer alone)
- BUILD_KORBAN.md
- korban_processor/pyproject.toml, poetry.lock (regenerated)
- deploy/docker/Dockerfile.korban_processor
Doorpost @expected_hash 81da49d4… → 2f40c382…, in this commit, with the
three witnessed files that moved it (architecture, vayakhel,
BUILD_KORBAN).
Read before committing: Elixir 461/0 (seeds 0, 42); korban 222 passed,
1 skipped against the installed package with pyswip absent; korban
image starts and idles as before; vav-pipeline image reports
specification_identity verified 2f40c382… and l1_hash pass.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
||
|
Scott KarpandClaude Opus 4.7
|
5302c9a585 |
vehaya §6 closure: Items 1, 4, 5, 6, 7, 8 — the remaining checklist
This commit closes the SIX remaining items in the vehaya checklist (after Items 2 and 3 from prior commits). With this, all eight items in vehaya §6 are addressed in code or in the canonical spec set, and all the substantive implications across §3 are covered. ITEM 1 — load-bearing dependency (vehaya §3.1): Devarim 11:18-19 — corporate inscription (v'samtem … al l'vavkhem — plural) is the load-bearing input to threshold practice (b'shivt'kha b'beitekha — singular). The threshold inscribes the corporate text on itself. ShemaContext now carries `doorpost_hash: Doorpost.expected_hash()` on every reception. Both receive_context/1 and receive_context_at/2 populate it. Every reception is now a witness of the corporate text's identity — making the load-bearing dependency between category and threshold structural rather than implicit. application.ex now runs Doorpost.verify_identity/0 at startup. Drift detected at boot is logged at :error AND fires Chatzotzrot.signal(:assembly_leaders, ...). The system does NOT refuse to boot on drift (dev environments routinely have spec drift during active editing) but operators are alerted before any reception fires. ITEM 6 — keep the source external (full closure; vehaya §6 item 6): Devarim 4:2 — lo tosif AND lo tigra; both directions. Vayikra 22:31 — u'shmartem … va'asitem; keep and do, not author. Form 1 closed the EXTERNAL → INTERNAL direction (commit |
||
|
Scott KarpandClaude Opus 4.6
|
f8f70c77e5 |
Code review fixes + 13 middot added to spec
Zerikah fail-safe: empty offerings rejected, not processed (BUILD_KORBAN §5). Halt monitor hardened: trap_exit, priority max, EXIT handler (Architecture spec). 13 Rules of Rabbi Yishmael added to Architecture §3.4 as Prolog inference predicates — the engine reasons WITH Torah's hermeneutic methods, not just about their products. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> |
||
|
Scott KarpandClaude Opus 4.6
|
622e814c9e |
Bereishit: initial commit — specifications, readings, and L1 builder
The complete Shorashim repository: Torah as AI specification. - 6 specification documents (PRD, architecture, construction, operations, economic model, materials) - Readings: Mishpatim (Prolog), Maaseh Merkavah, Moreh Nevuchim - Broken tablets preserved alongside whole tablets - L1 Torah filesystem builder (build_l1.py) — parses Westminster Leningrad Codex - Source directory structure for all components (L0 luchot, L1 aron, shoresh engine, sanhedrin, shema layer, urim v'tumim) L1 build verified: 5 books, 187 chapters, 5853 verses, 79982 words, 2608 unique shoreshim, 0 verses missing te'amim. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> |