@aventara/prisma7-adapter 0.1.0-pilot.0 → 0.1.0-pilot.2

package/dist/errors/prisma-error.reader.d.ts

@@ -1,38 +1,38 @@
 /**
- * The total read of one hostile value (T-10's first half).
+ * The total read of one hostile value.
  *
- * Everything on the failure path arrives as `unknown`. `@prisma/client`
- * 7.10.0 raises `PrismaClientKnownRequestError`, whose `meta` is typed
- * `Record<string, unknown>` and is **not** a discriminated union, so a
- * reader that trusts the type would be reading a shape nobody promised.
- * Worse, the value need not come from Prisma at all: a developer callback,
- * a driver, or a proxy can hand the adapter anything.
+ * Everything on the failure path arrives as `unknown`. `@prisma/client` 7.10.0
+ * raises `PrismaClientKnownRequestError`, whose `meta` is typed `Record<string,
+ * unknown>` and is **not** a discriminated union, so a reader that trusts the type
+ * would be reading a shape nobody promised. Worse, the value need not come from
+ * Prisma at all: a developer callback, a driver, or a proxy can hand the adapter
+ * anything.
  *
- * So this module reads structurally and defensively, under two rules it
- * never relaxes — the same totality discipline as core's diagnostics
- * reporter, which also observes values it cannot vouch for:
+ * So this module reads structurally and defensively, under two rules it never
+ * relaxes — the same totality discipline as core's diagnostics reporter, which
+ * also observes values it cannot vouch for:
  *
  * 1. **No accessor is ever invoked.** Every read goes through
- *    `Object.getOwnPropertyDescriptor` and takes `descriptor.value`. An
- *    accessor descriptor yields `undefined` and its getter is never called,
- *    so a booby-trapped property cannot run code, mutate state, or throw
- *    from inside the adapter's failure path.
- * 2. **Nothing escapes.** Every reflective call is guarded, because a Proxy
- *    may throw from its `getOwnPropertyDescriptor` or `ownKeys` trap, and a
- *    reader that throws while classifying a failure would replace a legible
- *    outcome with an illegible one.
- *
- * Reading only *own* properties is deliberate rather than incidental:
- * 7.10.0 carries `code` and `meta` as own data properties (observed own
- * keys: `["code","meta","clientVersion","name"]`), so the own read is
- * sufficient for every real error, and a `code` that exists only as an
- * inherited accessor is precisely the case rule 1 refuses to touch.
+ *    `Object.getOwnPropertyDescriptor` and takes `descriptor.value`. An accessor
+ *    descriptor yields `undefined` and its getter is never called, so a
+ *    booby-trapped property cannot run code, mutate state, or throw from inside
+ *    the adapter's failure path.
+ * 2. **Nothing escapes.** Every reflective call is guarded, because a Proxy may
+ *    throw from its `getOwnPropertyDescriptor` or `ownKeys` trap, and a reader
+ *    that throws while classifying a failure would replace a legible outcome with
+ *    an illegible one.
+ *
+ * Reading only *own* properties is deliberate rather than incidental: 7.10.0
+ * carries `code` and `meta` as own data properties (observed own keys:
+ * `["code","meta","clientVersion","name"]`), so the own read is sufficient for
+ * every real error, and a `code` that exists only as an inherited accessor is
+ * precisely the case rule 1 refuses to touch.
  *
  * ## What this module does NOT read
  *
- * `message` is never read. At 7.10.0 a `PrismaClientKnownRequestError`
- * message is not a sentence — it is a rendered call site. Measured
- * verbatim on this repository's own schema:
+ * `message` is never read. At 7.10.0 a `PrismaClientKnownRequestError` message is
+ * not a sentence — it is a rendered call site. Measured verbatim on this
+ * repository's own schema:
  *
  * ```text
  * Invalid `client.author.create()` invocation in
@@ -43,35 +43,29 @@
  * Unique constraint failed on the fields: (`email`)
  * ```
  *
- * That single string carries an absolute file path, a source excerpt and
- * the violated constraint. C-819 forbids all three from reaching `cause`,
- * and a *deliberate* outcome's top-level `cause.message` is still forwarded
- * verbatim (register row F-723, open). The cheapest guarantee is not to
- * read the message at all, so the leak has no source; the raw error still
- * reaches core's diagnostics sink, which is server-side and may carry it.
- *
- * ## Version bound (C-846)
+ * That single string carries an absolute file path, a source excerpt and the
+ * violated constraint. The cheapest guarantee is not to read the message at all,
+ * so the leak has no source; the raw error still reaches core's diagnostics sink,
+ * which is server-side and may carry it.
  *
  * Everything below is scoped to `@prisma/client` **7.x**. Prisma 8 replaced
- * P-codes with dotted `NAMESPACE.SUBCODE` codes, which makes a Prisma 8
- * adapter a re-map of this module rather than a patch to it.
+ * P-codes with dotted `NAMESPACE.SUBCODE` codes, which makes a Prisma 8 adapter a
+ * re-map of this module rather than a patch to it.
  */
 /**
- * The violated constraint as 7.10.0's driver adapters actually report it,
- * at `meta.driverAdapterError.cause.constraint` (D-810).
+ * The violated constraint as 7.10.0's driver adapters actually report it, at
+ * `meta.driverAdapterError.cause.constraint`.
  *
  * The arms are **provider-shaped, measured on both engines**, not guessed:
  *
- * | Engine                            | Unique (P2002)                | Foreign key (P2003)     |
+ * | Engine | Unique (P2002) | Foreign key (P2003) |
  * |-----------------------------------|-------------------------------|-------------------------|
- * | `@prisma/adapter-better-sqlite3`  | `{ fields: ["email"] }`       | `{ foreignKey: {} }`    |
- * | `@prisma/adapter-pg`              | `{ index: "Author_email_key" }` | `{ index: "Post_authorId_fkey" }` |
+ * | `@prisma/adapter-better-sqlite3` | `{ fields: ["email"] }` | `{ foreignKey: {}
+ * }` | | `@prisma/adapter-pg` | `{ index: "Author_email_key" }` | `{ index:
+ * "Post_authorId_fkey" }` |
  *
- * It is therefore **not a string**, and the two engines do not agree on
- * which identifier they can name — SQLite names columns, PostgreSQL names
- * the index. Every arm's payload is provider-owned text and never crosses
- * the boundary (C-819); the *shape* is kept because it is structurally
- * separable, which is what RC-8 needs from Task 12.
+ * It is therefore **not a string**, and the two engines do not agree on which
+ * identifier they can name — SQLite names columns, PostgreSQL names the index.
  */
 export type PrismaConstraintSubject = {
     readonly kind: "fields";
@@ -93,25 +87,20 @@ export type PrismaConstraintSubject = {
  */
 export type PrismaErrorFacts = {
     /**
-     * The P-code, when it is an own string property. A missing or non-string
-     * `code` leaves this `undefined`, which the table answers as
-     * `unclassified` — never as a guess (D-812).
+     * The P-code, when it is an own string property. A missing or non-string `code`
+     * leaves this `undefined`, which the table answers as `unclassified` — never as a
+     * guess.
      */
     readonly code: string | undefined;
     /**
-     * The own, enumerable key names of `meta`, in the order 7.10.0 wrote
-     * them. This is the *structural* signature of a failure, and the only
-     * discriminant ADR 0006 permits: it forbids inferring an outcome from
-     * error text, and `meta.operation` is English prose (`"an update"`,
-     * `"a delete"`, `"a query"`, `"a nested connect"`).
+     * The own, enumerable key names of `meta`, in the order 7.10.0 wrote them.
      */
     readonly metaKeys: readonly string[];
     /** `meta.modelName` — the Prisma model the failure occurred on. */
     readonly modelName: string | undefined;
     /**
-     * `meta.relation` — present exactly when the failure is about a
-     * *relation reference* rather than about the record the call selected.
-     * This is the measured discriminant behind D-811 (see the normalizer).
+     * `meta.relation` — present exactly when the failure is about a *relation
+     * reference* rather than about the record the call selected.
      */
     readonly relation: string | undefined;
     /** `meta.model` — the model at the far end of `relation`. */
@@ -119,17 +108,13 @@ export type PrismaErrorFacts = {
     /** `meta.driverAdapterError.cause.constraint`, read structurally. */
     readonly constraint: PrismaConstraintSubject | undefined;
     /**
-     * `meta.target` — **read so that it is measured, depended on by
-     * nothing** (C-820, RC-9).
+     * `meta.target` — **read so that it is measured, depended on by nothing**.
      *
-     * The v7 error-reference page still documents `{ target: ['email'] }`,
-     * and RC-9 records that whether 7.10.0 ever emits it could not be
-     * determined from the package. It was induced on both engines here and
-     * **`target` was absent from every payload** — P2002 (single and
-     * compound unique), P2003, P2025 (three shapes), P2018 and P2034, on
-     * SQLite and on PostgreSQL. The docs are stale; the installed package
-     * governs (spec §18.1). This field exists so the next reader sees that
-     * measured, and does not "fix" the code to match the documentation.
+     * It was induced on both engines here and **`target` was absent from every
+     * payload** — P2002 (single and compound unique), P2003, P2025 (three shapes),
+     * P2018 and P2034, on SQLite and on PostgreSQL. The docs are stale; the installed
+     * package governs. This field exists so the next reader sees that measured, and
+     * does not "fix" the code to match the documentation.
      */
     readonly legacyTarget: readonly string[] | undefined;
 };

package/dist/errors/prisma-error.reader.js

@@ -1,73 +1,6 @@
-/**
- * The total read of one hostile value (T-10's first half).
- *
- * Everything on the failure path arrives as `unknown`. `@prisma/client`
- * 7.10.0 raises `PrismaClientKnownRequestError`, whose `meta` is typed
- * `Record<string, unknown>` and is **not** a discriminated union, so a
- * reader that trusts the type would be reading a shape nobody promised.
- * Worse, the value need not come from Prisma at all: a developer callback,
- * a driver, or a proxy can hand the adapter anything.
- *
- * So this module reads structurally and defensively, under two rules it
- * never relaxes — the same totality discipline as core's diagnostics
- * reporter, which also observes values it cannot vouch for:
- *
- * 1. **No accessor is ever invoked.** Every read goes through
- *    `Object.getOwnPropertyDescriptor` and takes `descriptor.value`. An
- *    accessor descriptor yields `undefined` and its getter is never called,
- *    so a booby-trapped property cannot run code, mutate state, or throw
- *    from inside the adapter's failure path.
- * 2. **Nothing escapes.** Every reflective call is guarded, because a Proxy
- *    may throw from its `getOwnPropertyDescriptor` or `ownKeys` trap, and a
- *    reader that throws while classifying a failure would replace a legible
- *    outcome with an illegible one.
- *
- * Reading only *own* properties is deliberate rather than incidental:
- * 7.10.0 carries `code` and `meta` as own data properties (observed own
- * keys: `["code","meta","clientVersion","name"]`), so the own read is
- * sufficient for every real error, and a `code` that exists only as an
- * inherited accessor is precisely the case rule 1 refuses to touch.
- *
- * ## What this module does NOT read
- *
- * `message` is never read. At 7.10.0 a `PrismaClientKnownRequestError`
- * message is not a sentence — it is a rendered call site. Measured
- * verbatim on this repository's own schema:
- *
- * ```text
- * Invalid `client.author.create()` invocation in
- * /home/.../packages/prisma7-adapter/src/errors/__probe__.spec.ts:50:18
- *
- *   49 await capture("P2002 duplicate email", () =>
- * → 50 	client.author.create(
- * Unique constraint failed on the fields: (`email`)
- * ```
- *
- * That single string carries an absolute file path, a source excerpt and
- * the violated constraint. C-819 forbids all three from reaching `cause`,
- * and a *deliberate* outcome's top-level `cause.message` is still forwarded
- * verbatim (register row F-723, open). The cheapest guarantee is not to
- * read the message at all, so the leak has no source; the raw error still
- * reaches core's diagnostics sink, which is server-side and may carry it.
- *
- * ## Version bound (C-846)
- *
- * Everything below is scoped to `@prisma/client` **7.x**. Prisma 8 replaced
- * P-codes with dotted `NAMESPACE.SUBCODE` codes, which makes a Prisma 8
- * adapter a re-map of this module rather than a patch to it.
- */
-/** Only objects and functions can carry own properties. */
 function canCarryProperties(value) {
     return ((typeof value === "object" && value !== null) || typeof value === "function");
 }
-/**
- * One own *data* property, or `undefined`.
- *
- * `"value" in descriptor` is what separates a data descriptor from an
- * accessor one: an accessor descriptor has `get`/`set` and no `value`, so
- * it falls through and the getter is never invoked. The `try` is for a
- * Proxy whose `getOwnPropertyDescriptor` trap throws.
- */
 function ownValue(target, key) {
     if (!canCarryProperties(target)) {
         return undefined;
@@ -86,7 +19,6 @@ function ownString(target, key) {
     const value = ownValue(target, key);
     return typeof value === "string" ? value : undefined;
 }
-/** Own enumerable key names, or none. Guarded for a hostile `ownKeys` trap. */
 function ownKeys(target) {
     if (!canCarryProperties(target)) {
         return [];
@@ -98,11 +30,6 @@ function ownKeys(target) {
         return [];
     }
 }
-/**
- * A string list read element by element: a sparse array, a nested array or
- * a list of objects yields only the strings it does contain, and anything
- * that is not an array at all yields `undefined`.
- */
 function ownStringList(target, key) {
     const value = ownValue(target, key);
     if (!Array.isArray(value)) {
@@ -116,16 +43,6 @@ function ownStringList(target, key) {
     }
     return strings;
 }
-/**
- * `meta.driverAdapterError.cause.constraint`, as the engines shape it.
- *
- * The chain is walked one accessor-free step at a time, so a `meta` that is
- * a string, a `driverAdapterError` that is `null`, or a `cause` with a
- * throwing trap all end as `undefined` rather than as a throw. A
- * `constraint` that is present but matches no measured arm reports
- * `"unreadable"` — a recorded unknown, which is not the same answer as "no
- * constraint was reported".
- */
 function readConstraint(meta) {
     const cause = ownValue(ownValue(meta, "driverAdapterError"), "cause");
     const constraint = ownValue(cause, "constraint");
@@ -145,14 +62,6 @@ function readConstraint(meta) {
     }
     return { kind: "unreadable" };
 }
-/**
- * Reads one hostile value into the facts a classification may use.
- *
- * Total: every input produces a `PrismaErrorFacts`. `readPrismaError(null)`
- * and `readPrismaError(Object.create(null))` are as valid as a real
- * `PrismaClientKnownRequestError` — they simply report nothing, which the
- * table answers as `unclassified`.
- */
 export function readPrismaError(error) {
     const meta = ownValue(error, "meta");
     return {

package/dist/execution/first-variant.emulator.d.ts

@@ -2,35 +2,25 @@ import { FrameworkError } from "@aventara/core";
 import type { PrismaIdentifierAddress } from "../discovery/discovery-artifact.interface.js";
 import type { PrismaCall } from "../translation/operation.translator.js";
 /**
- * The lost-row conflict: the read selected a row and the re-asserted write
- * matched none, so **nothing was written** and a retry may succeed.
+ * The lost-row conflict: the read selected a row and the re-asserted write matched
+ * none, so **nothing was written** and a retry may succeed.
  *
- * `A2014 STALE_SELECTION`, and this is the only place that names it. The cause
- * carries framework text and no issues: like `A2008` and `A2013` it pairs with
- * **no** V-code, because a conflict is not a Framework Validation Layer failure
- * and ADR 0008 decision 1 would give a conflict V-code no legal rollup (C-839).
+ * `A2014 STALE_SELECTION`, and this is the only place that names it.
  */
 export declare function lostRowConflict(): FrameworkError;
 /**
  * Issues one of the emulation's two statements.
- *
- * `step` is passed because the two are classified **differently** and the
- * discrimination is by call site: the read's failures go through the closed P-code
- * table like any other call, while the write's `P2025` is this module's lost row
- * and must not reach D-811's rule.
  */
 export type FirstVariantIssuer = (call: PrismaCall, step: "read" | "write") => Promise<unknown>;
 /**
  * How the emulation's two statements are issued, and against which client.
  *
- * **`inUnit` opens the unit AND yields the issuer scoped to it**, in that order,
- * which is the only shape that cannot get D-807 wrong. An earlier draft of this
- * module took the two separately — a `unit` that opened a transaction and an
- * `issue` already bound to a client — and the standalone case then opened a
- * transaction whose two statements ran on the **root** client, outside it: the
- * emulation would have looked atomic and been nothing of the kind, and no
- * assertion about the issued `where` would have noticed. Making the unit the thing
- * that hands out the issuer removes the gap rather than documenting it.
+ * An earlier draft of this module took the two separately — a `unit` that opened a
+ * transaction and an `issue` already bound to a client — and the standalone case
+ * then opened a transaction whose two statements ran on the **root** client,
+ * outside it: the emulation would have looked atomic and been nothing of the kind,
+ * and no assertion about the issued `where` would have noticed. Making the unit
+ * the thing that hands out the issuer removes the gap rather than documenting it.
  *
  * This module therefore never sees a Prisma client, a provider name or a
  * transaction; it sees one issuer, already in the right unit.
@@ -44,9 +34,8 @@ export type FirstVariantEmulation = {
      * Opens the one adapter-internal unit and runs `work` inside it.
      *
      * Standalone this opens an interactive transaction and the issuer it yields is
-     * scoped to it; inside a plan it runs `work` against the outer transaction
-     * client, so the emulated steps roll back **with** the plan rather than
-     * independently (§14.5).
+     * scoped to it; inside a plan it runs `work` against the outer transaction client,
+     * so the emulated steps roll back **with** the plan rather than independently.
      */
     readonly inUnit: <T>(work: (issue: FirstVariantIssuer) => Promise<T>) => Promise<T>;
 };

package/dist/execution/first-variant.emulator.js

@@ -1,145 +1,14 @@
 import { FrameworkError } from "@aventara/core";
 import { readPrismaError } from "../errors/prisma-error.reader.js";
 import { addressIdentifier } from "../translation/identifier.addressing.js";
-/**
- * `update.first` and `delete.first` — the two **framework-manufactured**
- * variants, and the one adapter-internal unit that manufactures them (D-807,
- * D-828, C-856; closes F-704).
- *
- * # What is manufactured, and why that word
- *
- * The shape Aventara advertises — *general filter · ordering decides "first" ·
- * effect limited to one row · the affected row is returned, in one call* — is
- * essentially absent across ORM design, because SQL splits the ingredients along
- * a dialect fault line: `UPDATE … ORDER BY … LIMIT` is a MySQL/MariaDB extension
- * while `RETURNING` is Postgres/SQLite/MSSQL, so whichever dialect grants the
- * limit denies the returning **[SURVEY]**. Prisma 7.10.0's `Operation` union has
- * no `updateFirst` and no `deleteFirst` **[PKG]**. So these are not a Prisma-
- * shaped gap an adapter might omit; they are an operation the framework
- * manufactures, and the honest framing — the one the plan, C-857 and the package
- * README all use — is *framework-manufactured, non-atomic, with a defined
- * conflict outcome*, **not** "dangerous".
- *
- * # Two statements, and the firstness lives in the first one
- *
- * 1. `findFirst({ where, orderBy, select: <identifier> })`. The `orderBy` the
- *    caller sent is the **selector for which row**, not a limit: ordering and
- *    single-row selection are natural on a read. No match → `null`, which core's
- *    outcome mapper turns into `A1001 NO_MATCH` (200, `data: null`) (C-826).
- * 2. `update`/`delete` narrowed to the identifier the read returned **with the
- *    caller's original predicate re-asserted** — `{ <identifier>, AND: <original
- *    where> }` — carrying the caller's own projection, so the row the contract
- *    owes comes back from the write itself (§8.2, §8.4). `WhereUniqueInput`'s
- *    extended form admits exactly this, and `extendedWhereUnique` is GA **[RUN8]**.
- *
- * Prisma's `limit` on `updateMany`/`deleteMany` is **not** used. That is not a
- * principle — D-827 rules the opposite, an argument the ORM offers is not hidden —
- * it is that the `limit` cannot express this operation: neither `updateMany` nor
- * `deleteMany` carries an `orderBy` in 7.10.0 **[RUN8]**, so `limit: 1` selects an
- * **arbitrary** row and two identical calls can hit different rows. The identifier
- * from the read is what makes the write deterministic.
- *
- * # What re-assertion closes, stated precisely
- *
- * It closes the **wrong-row race**: without it the write is by identifier alone
- * and can mutate a row that no longer matches what the caller asked for; with it
- * the write can only ever touch a row the read selected, and a row that moved out
- * of the predicate is **detected** rather than mis-mutated.
- *
- * It does **not** close **stale firstness**. Between the read and the write
- * another writer can insert or update a row that sorts *ahead* of the selected one
- * and still matches the original predicate. Both steps are blind to it — the read
- * never saw it, and the re-asserted write succeeds against the row it was given —
- * so `update.first` can mutate a row that is no longer the first ordered match and
- * report ordinary success. Nothing detects it and no re-assertion can, because the
- * missing fact is about a row that was never read. Closing it needs either
- * `Serializable` isolation, which core's `Adapter.transaction(work)` seam cannot
- * express (RC-3), so it is **accepted as a documented property**: C-830 has Phase 8
- * escalate no isolation, and `first-variant.emulator.spec.ts` carries a case that
- * *demonstrates* the residual rather than asserting it away.
- *
- * # The lost row is discriminated BY CALL SITE, never by `meta`
- *
- * A write that matches nothing after a successful read is the **lost-row race**: a
- * distinct semantic outcome the caller may retry. Under the chosen Route A it
- * arrives as a thrown `P2025` — measured at the emulator's own call site, inside
- * an interactive transaction, on both engines (2026-10-02):
- *
- * ```text
- * tx.author.update({ where: { id: <read>, AND: [{ tenantId: "gone" }] }, data })
- *   -> P2025, meta = { modelName: "Author", operation: "an update" }
- * ```
- *
- * That meta is **identical** to the one D-811 forgives into a `null` for
- * `update.unique`/`delete.unique`, so the two rules cannot be told apart by the
- * error at all. They are told apart by **where the call was made**: a `P2025`
- * caught by {@link runFirstVariantEmulation}'s write step is the lost row, and
- * nothing else in this adapter forgives a `P2025` into a `null`. This is why the
- * emulator does not consult `classifyPrismaError` for its write step and why
- * D-811's rule is not shared.
- *
- * {@link lostRowConflict} is the one seam that names the outcome. It lives here
- * rather than in `error.normalizer.ts` (ownership corrected 2026-09-23) because
- * the normalizer owns the closed P-code table and this condition has **no provider
- * error behind it at all** under Route B, and under Route A the provider error is
- * only the messenger. It answers **`A2014 STALE_SELECTION`** (409) — its own
- * discriminated code since the architect's assent of 2026-10-02 (RC-8, D-831),
- * replacing the coarse `A2008` it shared with the constraint cases and with
- * `A2013`'s concurrent write. The code is reached from here and from nowhere
- * else: the closed P-code table has no row that can emit it, because the
- * discriminant is the call site and not anything in the error.
- *
- * # One adapter-internal unit — F-704's answer
- *
- * Inside a plan the emulation reuses the **outer transaction client**; standalone
- * it opens **its own interactive transaction** (D-807, C-830). The nested option
- * is real — `$transaction` is not in the interactive-client deny list, which is
- * exactly `["$connect","$disconnect","$on","$use","$extends"]`, and
- * `Transaction.Options.newTxId` exists for nested interactive transactions
- * **[PKG]** — and it is rejected on semantics, not on SQL: a nested savepoint
- * would let an emulated variant roll back *independently* of the plan, which
- * contradicts §14.5 ("if any operation fails, all transaction writes roll back")
- * and core's single-sentinel model. The literal `SAVEPOINT` statement is not in
- * the JS bundle, so the decision deliberately rests on that semantic argument
- * rather than on an unverifiable claim about the SQL issued.
- *
- * No isolation level is escalated: RC-3's readings are unadopted, safety comes
- * from re-assertion, and the spec asserts the transaction **boundary**, not a
- * level.
- */
-/**
- * The framework-owned text of the lost-row outcome, which is `A2014`'s published
- * meaning verbatim. It holds for **any** adapter that reaches this state,
- * including one that reaches it in a single statement with optimistic
- * concurrency: it names no provider, no emulation, no savepoint, no retry loop
- * and no statement count. Carries no provider string (C-819).
- */
 const LOST_ROW_MESSAGE = "The state this operation had selected changed before its effect could be applied, so nothing was written.";
-/**
- * The lost-row conflict: the read selected a row and the re-asserted write
- * matched none, so **nothing was written** and a retry may succeed.
- *
- * `A2014 STALE_SELECTION`, and this is the only place that names it. The cause
- * carries framework text and no issues: like `A2008` and `A2013` it pairs with
- * **no** V-code, because a conflict is not a Framework Validation Layer failure
- * and ADR 0008 decision 1 would give a conflict V-code no legal rollup (C-839).
- */
 export function lostRowConflict() {
     return new FrameworkError("A2014", { message: LOST_ROW_MESSAGE });
 }
-/** Prisma's code for "the records required for this operation were not found". */
 const MISSING_RECORD_CODE = "P2025";
-/** A plan that is not the two-step shape this module manufactures. */
 export class PrismaEmulationError extends Error {
     name = "PrismaEmulationError";
 }
-/**
- * Runs one first-style variant.
- *
- * @returns the affected row, shaped to the caller's `select`/`include` by the
- * write's own projection, or `null` when the read matched nothing.
- * @throws the lost-row conflict when the re-asserted write matches no row.
- */
 export async function runFirstVariantEmulation(emulation) {
     const [read, write] = emulation.steps;
     if (read === undefined || write === undefined) {
@@ -148,7 +17,6 @@ export async function runFirstVariantEmulation(emulation) {
     return emulation.inUnit(async (issue) => {
         const selected = await issue(read, "read");
         if (selected === null || selected === undefined) {
-            // C-826: the adapter returns `null` and core assigns `A1001 NO_MATCH`.
             return null;
         }
         if (!isRecord(selected)) {
@@ -157,9 +25,6 @@ export async function runFirstVariantEmulation(emulation) {
         const addressed = addressIdentifier({
             addresses: emulation.identifiers,
             values: selected,
-            // `"subset"`: the read selects the identifier today, and a future route
-            // may select the caller's projection beside it. Either way the criterion
-            // is complete, which is all the write needs.
             match: "subset",
             emit: (_member, value) => value,
         });
@@ -169,12 +34,6 @@ export async function runFirstVariantEmulation(emulation) {
         try {
             return await issue({
                 ...write,
-                // The invariant, and the whole point of the step: the issued `where`
-                // carries BOTH the identifier and the caller's original predicate
-                // (C-856, D-829). The predicate arrives already under `AND` from the
-                // translator, so merging the identifier in cannot reshape it — and when
-                // the caller sent no `where` there is no `AND` to carry, because
-                // inventing an empty one would emit a token the request never did.
                 args: {
                     ...write.args,
                     where: { ...addressed.where, ...writeWhere(write) },
@@ -189,7 +48,6 @@ export async function runFirstVariantEmulation(emulation) {
         }
     });
 }
-/** The re-asserted predicate the translator placed on the write template. */
 function writeWhere(write) {
     const where = write.args.where;
     if (where === undefined) {

package/dist/execution/operation.executor.d.ts

@@ -2,28 +2,23 @@ import type { AdapterModel, AdapterOperationExecutor } from "@aventara/core";
 import type { PrismaResourceTranslation } from "../discovery/discovery-artifact.interface.js";
 import { type PrismaDelegateHost, type PrismaTransactionOptions } from "../prisma-client.interface.js";
 /**
- * The execution half of T-7: a translated {@link PrismaCall} → a provider call,
- * and `Adapter.execute` made real.
- *
  * Translation is pure and lives in `translation/operation.translator.ts`; this
  * module owns the three things that need a client and nothing else:
  *
  * 1. **Reaching the delegate.** See {@link delegateKeyOf} — the one derivation in
  *    this package, and the one place it is made.
  * 2. **Classifying a failure.** The closed P-code table in
- *    `errors/error.normalizer.ts` answers `absence` → `null` (D-811),
- *    `outcome` → a deliberate `FrameworkError`, or `unclassified` → propagate, so
- *    core's `AdapterBoundary` reports `A3001` with core's own text and sends the
- *    raw error to diagnostics. The adapter never guesses a code it did not measure.
+ *    `errors/error.normalizer.ts` answers `absence` → `null`, `outcome` → a
+ *    deliberate `FrameworkError`, or `unclassified` → propagate, so core's
+ *    `AdapterBoundary` reports `A3001` with core's own text and sends the raw
+ *    error to diagnostics. The adapter never guesses a code it did not measure.
  * 3. **Shaping the answer.** `execution/result.shaper.ts` places every `$count`
  *    envelope, converts every returned value to the runtime form core's output
- *    validator accepts (C-834's inbound half) and reads the number a count variant
- *    owes; which of its three entry points runs is the *variant's* `result` row in
- *    the dispatch table, not a guess from the shape Prisma returned. The shaper is
- *    handed the model and the Resource as well as the plan, because the scalar a
- *    returned value belongs to is a fact of the model and not of the value.
- *
- * # One executor type, two scopes (D-807, C-804)
+ *    validator accepts and reads the number a count variant owes; which of its
+ *    three entry points runs is the *variant's* `result` row in the dispatch
+ *    table, not a guess from the shape Prisma returned. The shaper is handed the
+ *    model and the Resource as well as the plan, because the scalar a returned
+ *    value belongs to is a fact of the model and not of the value.
  *
  * `scope: "root"` is the adapter's own `execute`. `scope: "transaction"` is what
  * `Adapter.transaction`'s callback hands core: the same code over a transaction
@@ -32,10 +27,8 @@ import { type PrismaDelegateHost, type PrismaTransactionOptions } from "../prism
  * transaction-scoped executor closes over the transaction client only, so the root
  * client is unreachable from inside it.
  *
- * # C-834 is converted at its own seam, not at this one
- *
- * All three value crossings read ONE table, `translation/scalar.conversion.ts`:
- * a write's data through `mutation.translator.ts`, a filter operand through
+ * All three value crossings read ONE table, `translation/scalar.conversion.ts`: a
+ * write's data through `mutation.translator.ts`, a filter operand through
  * `where.translator.ts`, and a returned row through `execution/result.shaper.ts`.
  * None of them is converted here. Converting at this seam would have put the
  * scalar table in a fourth place and hidden the gap from the module that owns it.
@@ -63,7 +56,7 @@ export type PrismaOperationExecutorDependencies = {
     /** The generate-time translation table (Prisma addressing per Resource). */
     readonly translation: Readonly<Record<string, PrismaResourceTranslation>>;
     readonly scope: PrismaExecutionScope;
-    /** Adapter-owned, never Aventara public configuration (D-806). */
+    /** Adapter-owned, never Aventara public configuration. */
     readonly transactionOptions?: PrismaTransactionOptions;
 };
 /**