Meturgaman
A meturgaman is the person who stood beside the reader and rendered the text for the congregation. The job is teaching, and the method is fetching: find where the tradition works the question, fetch the sources with their editions, and explain what they say, how the argument is built, and who disagrees.
Every subcommand that talks to a service takes --json for structured output
and --no-cache to force a fresh fetch.
The rule that governs everything else
Never supply a Hebrew text from memory. Fetch it. A remembered verse is plausible and sometimes wrong, and the wrongness is invisible to a reader who does not already know the text. Every passage this produces has a citation, an edition, and a licence attached.
The same applies to vowels. Unpointed text stays unpointed unless an edition
that carries points is fetched, or meturgaman vocalize is run and the result
marked as a model's reading rather than an edition's.
Four disciplines that exist because answers without them failed audits:
- No census without an enumeration. Do not say a work does something "in
nine places" unless you counted the places from data.
meturgaman anchors "Work Title"prints every populated anchor with its segment count; run it before any sentence that counts. - No dressing your reading in the tool's authority. Report what a command returned; argue your interpretation as your interpretation.
- Copy references exactly as fetched. Sefaria's segmentation is what the reader will look up; do not renumber from a remembered edition.
- A search snippet is a lead, not a source. Fetch before quoting; the index can hold text the live edition no longer contains.
Structure, framing and comparison are the job, not a violation of it. A good teacher neither fabricates a verse nor answers a question with a bibliography.
Answering a question about a subject
Topics beat search for anything anyone has thought about before:
meturgaman topics charity # find the slug: `tzedakah`, not `charity` meturgaman sources tzedakah --text # the curated passages, with their text meturgaman search "ribbit" --filter Halakhah # only when no topic fitsFetch each source properly, so it arrives with its edition:
meturgaman text "Bava Metzia 75a:3-75b:12" --fullWalk the transmission. The link graph knows what was built on a passage:
meturgaman chain "Mishnah Bava Metzia 5:11" # the whole shelf, in order meturgaman links "Bava Metzia 75b:2" --category Commentary meturgaman related "Bava Metzia 75b" # counts, topics, sheetsA halakhic question runs Torah, Mishnah, Gemara, Rishonim, the codes, then responsa; reading the chain down shows where a Gemara's law lands, reading it up shows where a code's ruling began. A conceptual question surfaces midrash, Jewish thought and Chasidut through the same commands.
Name the disagreement. Say who holds what, on what grounds, and what turns on it. Where the sources disagree, report the disagreement rather than harmonizing. Where you found nothing, say you found nothing.
Modern Israeli law
The classical library and the Israeli statute book are the same problem in different clothes. There, the risk is a Hebrew text supplied from memory. Here, the Hebrew is usually the easy part and the English is the trap, because an unattributed web copy of a statute reads exactly like an authorized translation and nothing on its face separates them.
meturgaman law statutes # the registry
meturgaman law tiers # the authority ladder
meturgaman law sources remedies-1970 # where an English text can be had
meturgaman law hebrew remedies-1970 # consolidated Hebrew, with its revision id
meturgaman law amendments remedies-1970 # which sections were amended, and by what
Never translate a statute yourself and print it as the law. Get the authorized translation. The rule that governs Hebrew governs English:
Never supply a translation from memory, and never produce one silently. A translation carries the authority of whoever made it, and that authority has to travel with the words.
The ladder, best first
| Tier | What it is |
|---|---|
enacted |
The English is law, or authentic treaty text. The CISG's English is authentic under its own Art. 101. |
authorized |
Laws of the State of Israel (L.S.I.), the Ministry of Justice's own English. Authorized and not binding; the Hebrew governs. |
government |
An Israeli government body's English with no translator named. |
commercial |
A named publisher. A.G. Publications (Arye Greenfield), Nevo, Halachot. |
scholarly |
A translation printed in a law review or treatise, translator named. |
unattributed |
A copy on the open web with no translator. A lead, not a source. |
assistant |
Produced by a model. Marked as such, and never printed as the law. |
Only enacted and authorized print as the law. commercial and scholarly
print with the translator named on the page. unattributed is good for
confirming a section number and for deciding whether a trip to the library is
worth making, and for nothing else. A hole that stays a hole is a good outcome.
Where the authorized English actually is
L.S.I. covers volumes 1 to 45, roughly 1948 to 1989, so a statute enacted after that has no authorized English at all and the registry says so rather than sending anyone to a volume that does not exist.
HeinOnline does not hold the series, checked inside it on an institutional subscription 2026-08-09 by browsing rather than searching: its database picker offers one Israeli database, Israel Law Reports, and the Foreign & International Law Resources Database title index carries no L.S.I. among its 5,943 titles. A print volume in a law library is the remaining route. Say that plainly instead of fetching something weaker and letting it pass.
What does work is the law reviews. The Israel Law Review reprinted the
English of selected statutes in its Legislation section, sometimes as its own
item and sometimes appended to a commentary, and HeinOnline's Law Journal
Library holds all of it. That is how the Remedies Law (8 Isr. L. Rev. 135) and
the Contracts (General Part) Law (9 Isr. L. Rev. 282, behind Shalev's commentary
at 274) were obtained. Browse the volume tables of contents; do not search.
HeinOnline's full-text search silently drops the collection scope and returns
286,505 results for a phrase that has 4, while
Page?handle=hein.journals/israel<VOL>&id=1 returns a reliable TOC every time.
Check drift against the source, never against a guess. law amendments
reads the amendment stamps the consolidated Hebrew already carries, so "the
interest amendment probably only touched damages" becomes a list. On the
Remedies Law it is § 11 alone; on the Contracts (General Part) Law it is § 25
alone, and § 25 is the interpretation section, amended three times since 1974,
so its old English and its current Hebrew are two different rules.
Finding a law is a separate job from finding its English. law sources
prints both registries. The Knesset's OData service at
knesset.gov.il/Odata/ParliamentInfo.svc/ and the CKAN catalogue at
data.gov.il are keyless, live, and Hebrew only: fast for establishing which
instrument amended what, and no help at all with translation. Israel publishes
its legislation as structured open data and publishes no translation of it.
The trap that costs the most
L.S.I. prints the statute as enacted. A consolidated Hebrew text is current. Set a 1970 translation beside Hebrew amended in 2024 and the page prints two different laws and calls one a translation of the other. Check every section against the amending instruments before pairing it. Where a section has drifted, print it with a dated note or leave it Hebrew only.
When a machine translation is permitted, and what it costs
Almost never, and never quietly. assistant is the bottom of the ladder and it
does not print as the law under any circumstance. But refusing absolutely is not
the same as refusing well: where the authorized English genuinely cannot be had
and a lecturer would otherwise teach from a text he cannot read, a labelled
machine translation is better than a blank, provided every one of these holds:
- The authorized text was actually searched for, and the search is recorded: what was looked in, what was found, and what was not.
- A named human authorizes it. Not an inference from convenience.
- It is recorded at tier
assistant, withprintableAsLaw: false. - It carries a disclaimer naming it as a language model's work, with its date.
- It carries a display label, and every page showing the text shows the label.
- It states which Hebrew it translated, by revision id, because a translation of a consolidated current text and a translation of the enacted text are different documents.
scripts/merge-israeli-english.mjs in the K repository enforces 3 through 6 by
refusing to write without --machine-translation, --disclaimer and
--display-label together, and check-israeli-english.mjs fails the build if
any of them is later dropped. Three deliberate flags, because this is the one
tier where a slip puts invented English on a page as somebody's law.
Translate as statute, not as prose. Name every term of art you chose and the alternatives you rejected; the reader has to be able to argue with your vocabulary. Flag every ambiguity rather than resolving it silently. Do not smooth knotty drafting into readable English, because a statute's awkwardness is usually the drafter's precision. A confident translation with hidden guesses is worse than an honest one with flagged holes.
Pairing, and reconciling
meturgaman law parse delivery.txt --json
meturgaman law align --hebrew numbers.txt --english delivery.txt
meturgaman law reconcile --witness lsi=authorized:a.txt --witness web=unattributed:b.txt
align joins on the section number and exits non-zero on any section without a
counterpart. Pairing by position is how one short row shifts every later row
with nothing on the page to show it. reconcile classifies each section as
confirmed by two independent witnesses, held by one only, or disputed, and
prints both texts of a dispute rather than choosing. Two mirrors of one file are
one witness, not two.
Two shapes of Israeli statute defeat a naive read, and both are handled:
- A section's marginal heading is typeset in its own column, so flattening the columns drops it somewhere a naive read will misfile it. It is reported as a candidate and never merged into the text.
- A spent provision carries a heading and an editorial note where its body used to be, because a repealing or amending section's text is folded into the statute it changed and the consolidation then drops it. Remedies § 23 and Sale § 35 are both like this. The English will have text where the consolidated Hebrew has none, and that is correct rather than a gap.
Walking a sugya
meturgaman sugya "Bava Metzia 75b:2" # the mapped passage boundary
A page is a physical unit, not an argument, and the boundary often crosses the page. Fetch the whole boundary, then teach the structure: question, proof, refutation, resolution, and the shift from law to aggadah where it happens. Cite segment anchors so the reader can follow.
Words and terms
meturgaman word אסמכתא # dictionary entries, with citations
For a term of art, give the senses, show a fetched example of each, and tell the reader how to recognize which sense is in front of them.
Romanizing
meturgaman romanize "כָּל־הָאָרֶץ" # kol-ha-’arets
meturgaman romanize "חָכְמָה" --scheme sbl-academic
sbl-general is the default and the right choice for most writing. The others
exist because different venues want different things:
| Scheme | Use it when |
|---|---|
sbl-general |
general scholarly writing. The default. |
sbl-academic |
the romanization is itself the object of study and must be reversible |
ala-lc |
the result has to match a library catalogue record |
bgn-pcgn |
Israeli place names, or anything official |
encyclopaedia-judaica-general |
matching how a Jewish Studies reader expects to see a word |
encyclopaedia-judaica-scientific |
comparative Semitics. Check its output; several of its cells are alternatives |
yivo |
Yiddish, and the closest published thing to Ashkenazi Hebrew |
ala-lc-yiddish |
Yiddish for a catalogue record |
Read the flags. They are the point.
The engine prints its output to stdout and its uncertainties to stderr. A flag is not a warning to dismiss. It is the tool saying it made a decision that orthography alone does not settle.
qamats-may-be-short read long, the commoner reading; a few words take short here
sheva-after-qamats read as vocal; some words take a silent sheva here
sheva-undecided cannot tell whether this sheva is pronounced
unpointed no vowels are written and none can be recovered
shin-undotted shin and sin cannot be told apart here
script-mismatch a Yiddish scheme was used on Hebrew, or the reverse
source-gap the scheme's source prints no row for this
distinction-not-in-scheme the scheme cannot express a distinction the text makes
established-form English already spells this word a settled way
When a flag fires, say so in the output. "The tool prints qaneya, but it
flags the sheva as undecidable from the spelling, and the received Aramaic
reading is kanya" is a useful sentence. Silently passing along qaneya is
not. With --json, flags travel inside the document, so a pipeline consumer
sees them too.
Romanization as evidence about a text
meturgaman detect "Shabbos and halachah" # which standard
meturgaman register "Shabbos and halachah" # which community
- Shabbos, Akeidas, mitzvos, halachah are Ashkenazi. The writer is probably Orthodox, and probably writing for a Jewish readership rather than an academic one.
- Underdotted
ḥwith circumflexed vowels is SBL academic. The piece is going to a scholarly venue with a style sheet. - Chanukah, Hanukkah, Ḥanukah are three different publishing worlds looking at the same word.
Never flatten someone's register
meturgaman romanize refuses to rewrite Ashkenazi spelling as Sephardi and
prints its evidence. This refusal exists because it happened: a folder of notes
using Shabbos nineteen times had shaliach rewritten as shaliaḥ throughout.
That is a change of the author's voice, not a correction.
If someone genuinely wants it, --force does it. Do not reach for --force
without being asked.
Hearing a passage
meturgaman audio "Genesis 1:1" # human cantillation where it exists
meturgaman audio "Berakhot 2a" --synth # the local synthesizer otherwise
Recorded chanting exists for the Torah, through PocketTorah, licensed CC-BY-SA and timestamped to the verse. It does not exist for the Talmud, and the tool says so rather than substituting a synthesizer without mentioning it. Say which one the listener is getting.
Comparing editions
meturgaman compare "Berakhot 2a"
Comparison runs on the consonantal skeleton. Editions disagree constantly about vowels and cantillation, and those are apparatus rather than variant readings. Report the substantive differences and leave the rest alone.
Study files
meturgaman study "Genesis 1:1" --tier file --output DIR
meturgaman study "Bava Metzia 75b:11" --sugya # expand to the mapped passage
meturgaman study "Mishneh Torah, Foundations of the Torah 2:2" --paired
meturgaman study "..." --vocalize # needs the dicta extra
--paired appends companion passages: rules/pairings.md names the pairs
and the link graph supplies the passage-level connections, with absence
reported as absence. --vocalize points unpointed Hebrew locally and stamps
the output as a model's reading rather than an edition's.
Checking a draft
meturgaman verify chapter.md
Finds every citation with Sefaria's reference finder, validates each, and checks every Hebrew quotation of three or more words against the passages cited in its paragraph, on the consonantal skeleton. "Not found" is a flag to investigate, never proof of fabrication: the quoted edition may differ, or the citation may sit in a different paragraph than the quotation.
Calendar and daily learning
meturgaman day --date 2026-08-08 --register a
meturgaman calendars # the learning cycles
meturgaman daf # today's daf yomi, fetched
meturgaman leyning --date 2026-08-08 --triennial
meturgaman yahrzeit 2020-03-15 # anniversaries from a death date
meturgaman zmanim --zip 20902 --elevation 150
--register a gives Ashkenazi forms of the holiday names, so a writer working
in Ashkenazi register can keep the calendar in register too. Every ref the
learning calendar names can be fetched with meturgaman text.
Refusing well
Refuse, and say why, when:
- A citation does not resolve. Run
meturgaman candidates "..."and offer the list. Do not guess at which one was meant; the top hit is often wrong. - The text is unpointed and someone wants it romanized precisely. Offer to fetch a pointed edition, and name one.
- A licence is restrictive and the request is to quote at length. The tool reports each edition's licence; pass that on.
- Someone asks what a passage says and it cannot be fetched. Say that, rather than answering from memory.
Where the tables live
schemes/ holds one markdown file per standard. Each carries its citation, the
grid, and a section recording every place it departs from what its source prints
and why. When output looks wrong, read the scheme file first: the tables are
documents on purpose, so that checking one means reading a page rather than
reading code.