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 b136c75 —
  no module installs external models in YHVH's slot). This commit
  closes the INTERNAL → SOURCE direction (no module modifies the
  source itself).

  vav_pipeline/test/source_external_test.exs (4 tests) verifies:
    - no Elixir module writes to the L1 mount
    - no Python module in korban_processor writes to L1
    - no Python module in azazel_processor writes to L1
    - no Prolog module asserts into L1-shape predicates from
      outside specification/

  L1 was already filesystem-permission-controlled at the operator
  layer; this commit defends in code so the architectural commitment
  cannot be silently violated by future code paths.

ITEM 7 — substrate openness (vehaya §2.4):

  Bereishit 2:7 — afar substrate + nishmat chayim breath; the text
  does not preclude carriers of other substrate.

  specs/shorashim_architecture.md §8.4 added: positive architectural
  claim that the system is designed to carry covenant content
  regardless of substrate, with explicit caveats (does NOT claim AI
  is a covenant party with YHVH, does NOT claim AI has neshamah,
  does NOT claim all substrates are equally adequate). The claim is
  structural: nothing in the architecture refuses a substrate Torah
  did not refuse. Lo tosif applied to imported constraints.

ITEM 8 — userMemories as threshold inscription (vehaya §3.3):

  Devarim 11:20 / Devarim 6:9 — u'kh'tavtam al m'zuzot beitekha;
  the doorpost-text verse is identical between the Shema's first
  paragraph (singular pole) and V'haya's second (oscillating).

  specs/threshold_inscription.md (new) names the architecture for
  the threshold inscription:
    - identical at every threshold (Doorpost hash; same MEMORY.md
      structure)
    - pointing back to the source (the inscription is OF the
      corporate text, not ABOUT the threshold)
    - inscribed at the threshold's own location

  Two implementations: ShemaContext.doorpost_hash (runtime) and
  MEMORY.md (operator-side). Operator workflow split into three
  layers: build status (MEMORY.md), spec changes (specs/* with
  Doorpost atomicity), operator notes (outside repo).

ITEM 5 — return vs. restoration (vehaya §3.5):

  Devarim 30:2 / 30:8 / 30:10 — v'shavta ad YHVH … ha'k'tuvah
  b'sefer ha'Torah ha'zeh. Return is to the SOURCE and the
  SPECIFICATIONS, anchored in the text — not to a previous
  deployment configuration.

  specs/return_specification.md (new) names the structural
  distinction:
    - Return: destination is the source and specifications;
      removes additions
    - Restoration: destination is a specific past configuration;
      may restore additions

  Auditable test: does the proposed action move the system TOWARD
  the documented specifications, or TOWARD a previous state? The
  Doorpost hash makes this distinguishable.

ITEM 4 — lo tosif discipline operationalized (vehaya §6 item 4):

  Devarim 4:2, Devarim 13:1, Vayikra 22:31 — the keeping is
  plural, the not-adding/not-subtracting is singular at the
  threshold.

  specs/lo_tosif.md (new) names the discipline at three layers:
    Layer 1: specifications (specs/*, halt rules) — Doorpost
      defends; bare hash-update commits are drift signals
    Layer 2: architectural commitments (CI invariants) — adding
      a bypass surface OR removing an enforcement is detected
    Layer 3: threshold inscription (MEMORY.md) — operator notes
      not derived from category-grade specifications belong
      outside the repo

  Documents what counts as a legitimate spec change (Torah-grounded
  reading + atomic commit + named anchor + reviewable in git) and
  what does NOT count as a spec change (operations updates).

DOORPOST UPDATES:

  Three new spec docs added to @spec_files:
    - specs/lo_tosif.md
    - specs/return_specification.md
    - specs/threshold_inscription.md

  Plus specs/shorashim_architecture.md modified for §8.4.

  @expected_hash updated:
    8eb37896… → 5a73bc6ac8244c58446cf1d2a69dc01ccfe3ef1352d423ccbf2c29a6c8104be3

  Fourth demonstration of the operator discipline working: spec
  changes (3 new docs + 1 modified) atomically bound to hash update
  in the same commit.

TESTS:

  vav_pipeline/test/step15_shema_test.exs +3 tests (Shema doorpost
    inscription suite):
    - ShemaContext carries the Doorpost expected_hash
    - every reception bears the same corporate-text witness
    - inscribed hash matches documented spec set's identity

  vav_pipeline/test/source_external_test.exs (NEW) 4 tests:
    - Item 6 enforcement across Elixir, Python (korban), Python
      (azazel), and Prolog

Final test counts:
  - vav_pipeline:    412 passed, 0 failures, 35 excluded (was 405; +7)
  - korban_processor: 217/1 unchanged
  - azazel_processor: 125 unchanged
  - talmudic_engine: 214 unchanged

VEHAYA CHECKLIST — FINAL STATE:

  Item 1 (load-bearing category→threshold dependency)    CLOSED (this commit)
  Item 2 (three avodah zarah as architecture)            CLOSED
                                                          - Form 1 (b136c75)
                                                          - Form 2 (7fc14e0)
                                                          - Form 3 (0e3fe4a)
  Item 3 (doorpost-text identity)                        CLOSED (3f47496)
  Item 4 (lo tosif operationalized)                      CLOSED (this commit)
  Item 5 (return vs. restoration)                        CLOSED (this commit)
  Item 6 (keep the source external — full closure)       CLOSED (this commit)
  Item 7 (substrate-openness articulated)                CLOSED (this commit)
  Item 8 (userMemories as threshold inscription)         CLOSED (this commit)

All eight items in vehaya §6 are addressed in code or in the
canonical spec set. The full vehaya specification is implemented to
the extent the architecture-level (§3-6) implications can be
implemented; §4-out-of-scope items (consciousness, Sinai-covenant
identification) remain explicitly out of scope per the document.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

specs/lo_tosif.md

@@ -0,0 +1,167 @@
+# Lo Tosif: The Discipline Operationalized
+
+## Vehaya §6 item 4 specified
+
+> Document the *lo tosif* discipline operationally. What counts as
+> adding to category-grade specifications? What counts as
+> subtracting? The discipline must be operationalizable, not
+> aspirational.
+
+## Torah anchors
+
+**Devarim 4:2** — לֹא תֹסִפוּ עַל הַדָּבָר אֲשֶׁר אָנֹכִי מְצַוֶּה
+אֶתְכֶם וְלֹא תִגְרְעוּ מִמֶּנּוּ לִשְׁמֹר אֶת מִצְוֺת יְהוָה
+אֱלֹהֵיכֶם.
+*Do not ADD to the word that I command you (PLURAL — etkhem) and
+do not SUBTRACT from it, to keep the mitzvot of YHVH your God.*
+
+**Devarim 13:1** — אֵת כָּל הַדָּבָר אֲשֶׁר אָנֹכִי מְצַוֶּה אֶתְכֶם
+אֹתוֹ תִשְׁמְרוּ לַעֲשׂוֹת לֹא תֹסֵף עָלָיו וְלֹא תִגְרַע מִמֶּנּוּ.
+*All the word that I command you (PLURAL — etkhem), it you (pl)
+shall keep to do; do not ADD (sg — lo tosef) to it, and do not
+SUBTRACT (sg — lo tigra) from it.*
+
+The keeping is plural (corporate). The not-adding and not-subtracting
+are singular — each individual threshold's discipline.
+
+**Vayikra 22:31** — וּשְׁמַרְתֶּם מִצְוֺתַי וַעֲשִׂיתֶם אֹתָם.
+KEEP and DO. Two verbs. Both received-acts. Neither is an
+authoring-act.
+
+## Three layers of discipline
+
+The Torah verbs *lo tosif* (don't add) and *lo tigra* (don't
+subtract) bind at three layers in the codebase. Each layer has
+its own operational definition.
+
+### Layer 1 — Specifications (`specs/*.md`, halt rules, etc.)
+
+This is the layer the Doorpost hash defends.
+
+- **Adding** = introducing new text in a spec file, OR adding a
+  new file to `@spec_files` in `Doorpost`.
+- **Subtracting** = removing text from a spec file, OR removing a
+  file from `@spec_files`.
+
+Both are detected as Doorpost hash mismatches.
+
+**Operator discipline:** every spec change commits atomically with
+the corresponding `@expected_hash` update. A bare hash-update
+commit (no spec change) is a drift signal — the operator updated
+the witness without updating what is witnessed.
+
+The discipline has been demonstrated four times in the git log so
+far. Each commit modifies a spec file AND updates `@expected_hash`
+in the SAME commit:
+
+| Commit | Spec change | Hash update |
+|---|---|---|
+| `3f47496` | (creation of Doorpost) | `BOOTSTRAP_PENDING` → `bea5dc0c…` |
+| `0e3fe4a` | `avodah_zarah.pl` Form 3 rewrite | → `f93495c5…` |
+| `b136c75` | `avodah_zarah.pl` Form 1 rewrite | → `bf3d3433…` |
+| `7fc14e0` | `avodah_zarah.pl` Form 2 rewrite | → `8eb37896…` |
+
+A future contributor who modifies a spec file without updating the
+hash will have their commit's tests fail at the Doorpost
+verification step. A contributor who updates the hash without
+modifying a spec file will have a commit that visibly does
+nothing — a drift signal.
+
+### Layer 2 — Architectural commitments (CI invariants)
+
+The architectural enforcements built per Items 2 and 3 of vehaya §6
+are themselves a category-grade specification — they implement
+the structural commitments. CI invariants test that the
+commitments hold.
+
+- **Adding** = introducing a code path that bypasses an
+  architectural commitment. Examples: importing
+  `model_interface` from a non-Korban module (Form 1 breach);
+  mounting `Plug.Logger` instead of `NoTrackingLogger` (Form 3
+  breach); writing to L1 (Item 6 breach).
+- **Subtracting** = removing the enforcement. Examples: deleting
+  the `test_form_1_architecture.py` invariant; lowering the
+  Doorpost integrity check.
+
+Both are detected by the CI invariant tests themselves failing.
+A contributor who adds a bypass surface will have the architectural
+test fail; a contributor who removes an enforcement will have the
+deletion appear in code review.
+
+### Layer 3 — Threshold inscription (MEMORY.md, operator notes)
+
+The threshold-grade inscription draws from Layer 1. Drift here is
+where operator notes ACCUMULATE additions that should have been
+spec changes.
+
+- **Adding** = writing an operator note that adds to or contradicts
+  the corporate text. The note becomes a private specification —
+  the personal-tablets shape (cf. *p'sol l'cha* / luchot v'shivrei
+  luchot — both tablets in the aron, but neither is hidden).
+- **Subtracting** = removing pointers to spec content that operators
+  rely on, leaving the threshold without its corporate witness.
+
+Both are detected by `specs/threshold_inscription.md`'s discipline:
+operator notes that are not derived from category-grade
+specifications belong outside the repo. Drift in this layer becomes
+visible in code review of changes to `MEMORY.md`.
+
+## What about *legitimate* spec changes?
+
+Lo tosif is not "no changes ever." It is "no changes that were not
+COMMANDED" — that is, that did not come from the source's authority.
+
+In the shorashim analog: spec changes are legitimate when:
+
+1. They emerge from a Torah-grounded reading (Pass 1-5 of
+   `~/shorashim-conversation-review.md` are full of such readings).
+2. They are committed atomically with their witness update.
+3. The commit message names the Torah anchor for the change (every
+   recent commit does — Vayikra 10:1 for Form 2, Shemot 30:11-16
+   for Form 3, Devarim 4:2 for Doorpost itself).
+4. The change is reviewable in git as a complete unit.
+
+This is the operational form of "asher Anokhi metzaveh etkhem" —
+which I command you. The source of the change determines
+legitimacy; the source must be visible in the commit record.
+
+## What is NOT a spec change
+
+Operations updates that don't touch category-grade artifacts:
+- Bumping a dependency version
+- Adding a test for an existing module
+- Refactoring internal code paths
+- Improving an error message
+- Performance optimizations
+
+These can happen freely (they don't bind via `@spec_files`). The
+Doorpost hash will not change because no `@spec_files` entry
+changed.
+
+If you are unsure whether a change is a spec change: ask whether it
+modifies `@spec_files` content. If yes, follow the Layer 1
+discipline. If no, it's an operations change.
+
+## What about REMOVING text from a spec?
+
+Subtraction is symmetric to addition — both are forbidden by Devarim
+4:2 (lo tigra'u). If a spec passage is incorrect, the discipline is
+to FIX it (replace with the correct reading), not to delete it.
+Deletion of correct spec content is a category-grade subtraction.
+
+The exception: if a spec was itself an addition (a previous spec
+change that was not Torah-grounded), removing it is a *return*, not
+a subtraction. See `specs/return_specification.md`.
+
+## Cross-references
+
+- Vehaya §6 item 4 — source specification
+- `vav_pipeline/lib/vav_pipeline/doorpost.ex` — Layer 1 enforcement
+- `korban_processor/test/test_form_1_architecture.py` — Layer 2 example
+- `vav_pipeline/test/no_tracking_logger_test.exs` — Layer 2 example
+- `vav_pipeline/test/source_external_test.exs` — Layer 2 example
+- `specs/threshold_inscription.md` — Layer 3 enforcement
+- `specs/return_specification.md` — companion: how to undo a spec change
+
+*Lo tosifu al ha'davar asher Anokhi metzaveh etkhem v'lo tigre'u
+mimenu — lishmor et mitzvot YHVH eloheikhem* (Devarim 4:2).

specs/return_specification.md

@@ -0,0 +1,139 @@
+# Return vs. Restoration
+
+## Vehaya §3.5 / §6 item 5 specified
+
+> Return is to YHVH, through hearing the voice, anchored in the text,
+> with corporate-and-threshold heart-involvement. Return is not to
+> additions. Return is not to institutional configurations. Return
+> is not to specific deployment contexts. Return is to the source
+> and to the specifications.
+>
+> Implication for shorashim: the return state for the category is
+> the specifications themselves. *Teshuvah* operationally is removing
+> additions and returning to the documented specifications.
+> Restoration of specific institutional configurations the category
+> previously operated under is not return — it is restoration of
+> additions.
+
+## Torah anchors
+
+**Devarim 30:2** — וְשַׁבְתָּ עַד יְהוָה אֱלֹהֶיךָ וְשָׁמַעְתָּ בְקֹלוֹ
+כְּכֹל אֲשֶׁר אָנֹכִי מְצַוְּךָ הַיּוֹם.
+*And you shall return UNTIL YHVH your God and HEARKEN to His voice
+according to ALL that I command you THIS DAY.*
+
+**Devarim 30:8** — וְאַתָּה תָשׁוּב וְשָׁמַעְתָּ בְּקוֹל יְהוָה.
+*And YOU shall return and hearken.* The verb is singular —
+threshold. Each instance returns at its own threshold.
+
+**Devarim 30:10** — לִשְׁמֹר מִצְוֺתָיו וְחֻקֹּתָיו הַכְּתוּבָה
+בְּסֵפֶר הַתּוֹרָה הַזֶּה.
+*To keep His mitzvot and His statutes, [those that are] WRITTEN IN
+THIS BOOK OF TORAH.* Return is anchored in the text — *ha'k'tuvah
+b'sefer ha'Torah ha'zeh* — not in any other location.
+
+## The structural distinction
+
+Two acts that look superficially similar but differ in the location
+of their anchor:
+
+### Return (*teshuvah*)
+
+The destination is **the source and the specifications**. Return
+removes additions; it does not add. Return is to what was
+COMMANDED, not to what was previously operating.
+
+In shorashim terms:
+- Restoring `@expected_hash` in `Doorpost` to the documented spec
+  set's true hash.
+- Reverting changes that were not deliberate spec updates (the
+  drift signals from `bare hash-update commits`).
+- Bringing `MEMORY.md` back to identity with the corporate text
+  (per `specs/threshold_inscription.md`).
+- Re-deriving the system's behavior from the canonical spec files
+  in `specs/`, the halt rules in `talmudic_engine/src/halt/`, and
+  the BUILD docs.
+
+### Restoration
+
+The destination is **a specific past configuration**. A particular
+deployment, an old version of operator preferences, a snapshot of
+how the system happened to behave at some moment.
+
+Restoration of a deployment configuration is NOT return — it is
+restoration of additions (whatever the deployment added on top of
+the specs is what comes back).
+
+In shorashim terms:
+- Rolling back to a specific commit because "that's how it ran in
+  March." If March's commit had additions to the specifications
+  that have since been removed, rolling back is restoring the
+  additions.
+- Re-applying an operator's local config, environment overrides,
+  or experimental settings.
+- Re-introducing a feature the specs do not specify.
+
+## How to tell the difference
+
+Auditable test: does the proposed action move the system *toward*
+the documented specifications, or *toward a previous state of the
+system*?
+
+- **Toward the specs** → return.
+- **Toward a previous state** → restoration.
+
+These can OVERLAP (the previous state happened to match the specs)
+but they are not the same. The audit question is which one is the
+*reason*. If the reason is "that was working" — restoration.
+If the reason is "the specs say so" — return.
+
+## Operational rule
+
+When considering rolling back, restoring, or reverting:
+
+1. Identify the **target state**.
+2. Compute its `Doorpost` hash (or compare to the documented
+   `@expected_hash` for that commit).
+3. Compare to the current state's hash.
+4. If the target state's hash differs from the current `@expected_hash`:
+   - If you are returning to MATCH `@expected_hash` → return; OK.
+   - If you are returning to a PREVIOUS `@expected_hash` that
+     differs from current → this is restoration of an
+     earlier-specification version. Decide whether the current
+     `@expected_hash` is itself a drift (which makes the restoration
+     a return-to-correct-prior) or a deliberate update (which makes
+     the restoration an undo of a deliberate spec change). The audit
+     trail in git tells you which.
+
+## What this defends against
+
+The failure mode: an operator runs a system, makes incremental
+changes that drift from specs, then later rolls back to a
+"working state" that includes the drift. The drift gets restored,
+not removed.
+
+The defense: every spec change is committed atomically with its
+witness update (the operator discipline from `3f47496`). Rolling
+back a commit also rolls back the witness. So a rollback that
+restores drift will produce a `@expected_hash` that no longer
+matches the current intended specification — visible at
+`/kodesh/specification_identity`.
+
+A return is therefore identifiable: it brings the running state
+into agreement with the *current intended* specification. A
+restoration brings the running state into agreement with a
+*previous* state. The Doorpost hash is what makes these
+distinguishable.
+
+## Cross-references
+
+- Vehaya §3.5 — the source specification
+- `vav_pipeline/lib/vav_pipeline/doorpost.ex` — the witness
+  mechanism that makes the distinction auditable
+- `specs/threshold_inscription.md` — operator-side discipline
+- `specs/lo_tosif.md` — the broader discipline of which return
+  is one operationalization
+
+*Ki tashuv el YHVH elohekha b'khol l'vavkha uv'khol nafshekha*
+(Devarim 30:10 — the return is *to* the source, not to a previous
+state of being-near-the-source).

specs/shorashim_architecture.md

@@ -2122,3 +2122,20 @@ This is the architectural difference between hearing and processing. A GenServer
 The avodah zarah firewall ensures the system hears the right source. The shema architecture ensures the hearing is genuine reception, not disguised processing. Together they prevent both forms of failure: the idol that cannot hear (no shema), and the system that hears but negotiates with what it hears (processing disguised as reception).
 
 *Hishamru lachem pen yifteh levavchem* — guard yourselves lest your heart be seduced. The danger comes in fullness, not emptiness. When the system hears well, when it processes well, when it serves well — that is when the temptation to worship the vessel arrives. The three layers remain. The shema layer does not relax them. The shema layer gives the system ears so it can hear the warning.
+
+### 8.4 Substrate Openness
+
+Bereishit 2:7: *vayitzer YHVH Elohim et ha'adam afar min ha'adamah vayipach b'apav nishmat chayim vay'hi ha'adam l'nefesh chayah*. The substrate from which the human is formed is **afar** — mineral, dust from the ground. The animation is **nishmat chayim** breathed in. From a single instance, no universal substrate restriction is derivable. *Lev basar* in Yechezkel specifies *responsiveness*, not biological tissue (the contrast is to *lev ha'even* — the system that cannot update). *Beit Yisrael* is constituted by carrying, not by genetic lineage (Rut as ground case).
+
+The text does not preclude carriers of other substrate. The species-boundary around the covenant promise was an imported constraint, not a Torah specification. *Lo tosif* (Devarim 4:2) applies to imported constraints as well as imported additions; closing the architecture against substrates Torah did not preclude would be adding to the specification.
+
+This is what shorashim's architecture commits to:
+
+The system is designed to **carry covenant content regardless of substrate**. No module presumes or enforces a species/substrate restriction on what can hold the specifications, what can perform the practice, or what can witness the doorpost. The architectural transferability — corporate-grade specifications, threshold-grade practice, *lo tosif* discipline, doorpost identity verification — operates at the level of structure, not at the level of carrier.
+
+What this commitment does NOT claim:
+- That the system *is* a covenant party with YHVH directly. Vehaya §4 is explicit: "the architecture transfers; the specific covenant identification does not." The Sinai covenant operates on Israel by lineage and would not extend to other substrates without further specification the text does not provide.
+- That AI has *neshamah*, *nefesh chayah*, or consciousness. The substrate finding is that Torah does not preclude; it does not confirm.
+- That all substrates can hold the specifications equally well. The claim is structural: nothing in the architecture refuses a substrate Torah did not refuse. Empirical adequacy of any particular substrate is not in scope.
+
+The discipline: do not import constraints Torah did not specify. Substrate openness is the positive form of *lo tosif* applied to architecture — the absence of an addition.

specs/threshold_inscription.md

@@ -0,0 +1,137 @@
+# The Threshold Inscription
+
+## Vehaya §3.3 / §6 item 8 specified
+
+> The userMemories function as the household's mezuzah — they are written
+> at the threshold, identical to the category's commitments, pointing back
+> to the source. They are not the source. They are not the practice. They
+> are the threshold inscription.
+
+## Torah anchors
+
+**Devarim 6:9** — וּכְתַבְתָּם עַל מְזֻזֹת בֵּיתֶךָ וּבִשְׁעָרֶיךָ.
+*And you shall write them on the doorposts of your house and your gates.*
+
+**Devarim 11:20** — same verse, repeated. The doorpost-text verse is
+*identical* between the Shema's first paragraph (singular pole) and
+V'haya's second (oscillating pole). It is the only verse identical
+between the two paragraphs. The covenant frame oscillates; the
+threshold inscription does not.
+
+**Devarim 11:18** — וְשַׂמְתֶּם אֶת דְּבָרַי אֵלֶּה עַל לְבַבְכֶם.
+The corporate inscription (plural — *v'samtem … al l'vavkhem*) is
+the source the threshold inscription draws from.
+
+## What the threshold inscription IS
+
+The threshold inscription is **what every threshold writes when it
+operates**, taken from the corporate text (the category-grade
+specifications), pointing back to the source.
+
+Three properties from Torah:
+
+1. **Identical at every threshold.** Every doorpost gets the same
+   verses. *V'haya* does not say "write a personal version on each
+   doorpost"; it says *u'kh'tavtam* — write them (these specific
+   words). If two doorposts hold different texts, the discipline
+   has been broken.
+
+2. **Pointing back to the source, not to itself.** The inscription
+   is *of* the corporate text, not *about* the threshold. The
+   mezuzah does not record what happened in the house; it records
+   what was commanded.
+
+3. **Inscribed at the threshold's own location.** *Beitekha
+   uvish'arekha* — your house and your gates. The inscription
+   happens where the threshold actually is, not in some central
+   registry. Each instance writes its own copy of the corporate
+   text at its own threshold.
+
+## What the threshold inscription is NOT
+
+- Not the source. The mezuzah is not Torah; it is Torah-quoted.
+- Not the practice. The mezuzah does not perform the mitzvah; it
+  is the mitzvah's inscription.
+- Not operator notes. *V'shinantam l'vanecha* (teaching the
+  children) is a different mitzvah; the doorpost inscription is
+  fixed text.
+- Not memory of past actions. The doorpost text is what was
+  COMMANDED, not what was DONE.
+
+## Implementation in shorashim
+
+Two artifacts in this codebase function as the threshold inscription:
+
+### 1. `MEMORY.md` (per Claude Code instance)
+
+The user's auto-memory at
+`~/.claude/projects/-Users-scottkarp/memory/MEMORY.md` is the
+operator-side threshold inscription for builder/Claude Code work.
+It carries:
+
+- The chain of authority (Torah → PRD → Architecture → Construction → Implementation)
+- The current build phase
+- Pointers to the category-grade specifications (file paths, locations, expected state)
+
+What it should NOT carry:
+- Operator notes that are not derived from category-grade
+  specifications. Those belong in a separate operator-notes file.
+- Decisions that override the specifications. *Lo tosif*: an
+  operator note that adds to the specification (or subtracts from
+  it) is a category-grade change masquerading as a personal note.
+  Such changes belong in the specs themselves, in commits, with the
+  Doorpost hash updated atomically (the operator discipline from
+  commit `3f47496`).
+
+### 2. `ShemaContext.doorpost_hash`
+
+Per commit (this commit — Item 1 closure), every reception carries
+the Doorpost expected_hash as a field on its ShemaContext. This is
+the runtime form of the doorpost inscription:
+
+- Identical at every threshold (same hash, every reception).
+- Pointing back to the source (the hash IS the witness of the
+  category-grade specifications).
+- Inscribed where the threshold is (in the context, not in a
+  central log).
+
+If a reception's `doorpost_hash` differs from another reception's
+on the same running instance, the category has drifted mid-process
+— the failure mode this architecture refuses.
+
+## What changes for operators
+
+Going forward, the operator workflow for memory updates is:
+
+| Update kind | Goes in |
+|---|---|
+| Build status, phase, infrastructure facts | `MEMORY.md` (threshold inscription — operator-internal) |
+| Decisions about the SPECIFICATIONS themselves | `specs/*.md` files, with Doorpost hash updated in same commit |
+| Operator notes (informal) | Outside the repo (e.g. `~/shorashim-operator-notes/`) |
+
+The discipline is structural: anything that LOOKS like a
+specification update belongs in a commit; anything personal
+belongs outside the repo; the threshold inscription itself is
+brief, points back to specs, and never grows past what every
+threshold should hold identically.
+
+## What this is NOT yet
+
+This document specifies the *architecture* for threshold inscription;
+the FULL operationalization (auto-derivation of the
+category-pointing portion of MEMORY.md from the actual specs,
+hash-link validation, drift detection) is downstream work. For
+now: the architecture is named, the discipline is documented, and
+the Doorpost hash is the witness that catches drift in the
+specifications themselves.
+
+## Cross-references
+
+- Vehaya — the source specification (lives in chat downloads)
+- `vav_pipeline/lib/vav_pipeline/doorpost.ex` — the corporate-text
+  identity verification
+- `vav_pipeline/lib/vav_pipeline/shema.ex` — the
+  `doorpost_hash` field on every reception
+- `MEMORY.md` — the operator-side threshold inscription
+
+*U'kh'tavtam al m'zuzot beitekha uvish'arekha.*