@adcp/sdk 17.0.0-beta.13 → 17.0.0-beta.15
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/lib/brand/index.d.mts +2 -0
- package/dist/lib/brand/index.d.ts +2 -0
- package/dist/lib/brand/index.js +3 -0
- package/dist/lib/brand/index.mjs +2 -0
- package/dist/lib/brand/selection.d.mts +13 -0
- package/dist/lib/brand/selection.d.ts +13 -0
- package/dist/lib/brand/selection.js +83 -0
- package/dist/lib/brand/selection.mjs +59 -0
- package/dist/lib/index.d.mts +2 -2
- package/dist/lib/index.d.ts +2 -2
- package/dist/lib/index.js +2 -0
- package/dist/lib/index.mjs +2 -0
- package/dist/lib/registry/types.d.mts +5 -0
- package/dist/lib/registry/types.d.ts +5 -0
- package/dist/lib/registry/types.generated.d.ts +4 -0
- package/dist/lib/server/decisioning/runtime/from-platform.js +12 -2
- package/dist/lib/server/decisioning/runtime/from-platform.mjs +12 -2
- package/dist/lib/server/test-controller.d.mts +9 -35
- package/dist/lib/server/test-controller.d.ts +9 -35
- package/dist/lib/server/test-controller.js +4 -0
- package/dist/lib/server/test-controller.mjs +4 -0
- package/dist/lib/testing/comply-controller.d.mts +7 -10
- package/dist/lib/testing/comply-controller.d.ts +7 -10
- package/dist/lib/testing/reporting/doctor.js +2 -1
- package/dist/lib/testing/reporting/doctor.mjs +2 -1
- package/dist/lib/testing/storyboard/account-policy.d.mts +42 -0
- package/dist/lib/testing/storyboard/account-policy.d.ts +42 -0
- package/dist/lib/testing/storyboard/account-policy.js +41 -2
- package/dist/lib/testing/storyboard/account-policy.mjs +38 -1
- package/dist/lib/testing/storyboard/path.d.mts +6 -0
- package/dist/lib/testing/storyboard/path.d.ts +6 -0
- package/dist/lib/testing/storyboard/path.js +22 -0
- package/dist/lib/testing/storyboard/path.mjs +21 -0
- package/dist/lib/testing/storyboard/request-builder.js +1 -1
- package/dist/lib/testing/storyboard/request-builder.mjs +1 -1
- package/dist/lib/testing/storyboard/runner.d.mts +1 -42
- package/dist/lib/testing/storyboard/runner.d.ts +1 -42
- package/dist/lib/testing/storyboard/runner.js +8 -38
- package/dist/lib/testing/storyboard/runner.mjs +5 -35
- package/dist/lib/testing/storyboard/seeding.js +152 -2
- package/dist/lib/testing/storyboard/seeding.mjs +152 -2
- package/dist/lib/testing/storyboard/types.d.mts +4 -2
- package/dist/lib/testing/storyboard/types.d.ts +4 -2
- package/dist/lib/testing/storyboard/types.js +1 -0
- package/dist/lib/testing/storyboard/types.mjs +1 -0
- package/dist/lib/testing/storyboard/validations.d.mts +1 -1
- package/dist/lib/testing/storyboard/validations.d.ts +1 -1
- package/dist/lib/testing/storyboard/validations.js +60 -1
- package/dist/lib/testing/storyboard/validations.mjs +61 -1
- package/dist/lib/v2/projection/creative-delivery.js +62 -16
- package/dist/lib/v2/projection/creative-delivery.mjs +62 -16
- package/dist/lib/version.d.mts +3 -3
- package/dist/lib/version.d.ts +3 -3
- package/dist/lib/version.js +3 -3
- package/dist/lib/version.mjs +3 -3
- package/docs/TYPE-SUMMARY.md +1 -1
- package/docs/guides/BRAND-ENTRY-SELECTION.md +56 -0
- package/docs/guides/REPORTING-LEDGER.md +142 -12
- package/docs/guides/REPORTING-OPERATIONS.md +18 -5
- package/docs/guides/VALIDATE-YOUR-AGENT.md +44 -0
- package/docs/llms.txt +1 -1
- package/docs/migration-14.x-rc-worksheet.md +4 -4
- package/docs/migration-reporting-row-storage-alignment.md +4 -2
- package/package.json +1 -1
|
@@ -744,16 +744,71 @@ await store.readyRowStorage();
|
|
|
744
744
|
several row sets. Every delete path (period retirement, compaction and the
|
|
745
745
|
intent sweep) therefore skips an object while a live row set outside the
|
|
746
746
|
operation still references the same binding, key and native version, or an
|
|
747
|
-
open write intent created it.
|
|
748
|
-
finds an existing object when a writer retries the same row set.
|
|
749
|
-
|
|
750
|
-
|
|
751
|
-
|
|
752
|
-
|
|
753
|
-
|
|
754
|
-
|
|
755
|
-
|
|
756
|
-
|
|
747
|
+
open write intent created it. With unique keys, a create-only write only
|
|
748
|
+
finds an existing object when a writer retries the same row set.
|
|
749
|
+
- **Object deletes and commits take no per-object lock, with one exception.**
|
|
750
|
+
Retirement, compaction and the intent sweep run the reference check as a
|
|
751
|
+
plain query, then delete the exact native version, holding no transaction or
|
|
752
|
+
pooled connection across the provider call. They are safe without a lock
|
|
753
|
+
because of five fences, none of which is the lock:
|
|
754
|
+
1. Every writable key includes `{revision_key}` (a SHA-256 of the row set ID)
|
|
755
|
+
and `{content_sha256}`, so one row set alone can ever name a key through a
|
|
756
|
+
writable binding. A committed row set is replayed: a late writer that
|
|
757
|
+
passed the replay check can adopt its just-committed object, but its
|
|
758
|
+
commit inserts nothing (the row set exists), so it cannot reference an
|
|
759
|
+
object a retirement or compaction then deletes. A `readOnly` binding's old
|
|
760
|
+
template can map several row sets to one object, but nothing adopts through
|
|
761
|
+
it; there the reference check and fence 4 are the whole argument. Bindings
|
|
762
|
+
that share one bucket and prefix must route a row set to one binding across
|
|
763
|
+
attempts, because the reference check is per binding (the orphan collector
|
|
764
|
+
checks every binding).
|
|
765
|
+
2. A header-only revision's commit holds the account lock and `FOR SHARE`-locks
|
|
766
|
+
a predecessor that must still be `live`. Retirement and compaction mark a
|
|
767
|
+
row set `pruning` under the same account lock before they delete anything,
|
|
768
|
+
and a `pruning` row set neither counts as a reference nor can be shared. So
|
|
769
|
+
a sharer commits first (its chunk rows then make the delete skip) or fails
|
|
770
|
+
with `STATE_UNAVAILABLE` and re-uploads under its own key.
|
|
771
|
+
3. The intent sweep moves an expired `open` intent to `sweeping` under the
|
|
772
|
+
account lock; a writer closes that intent, still `open`, in the commit
|
|
773
|
+
transaction that inserts its chunk rows. Exactly one wins, so a writer past
|
|
774
|
+
its lease cannot commit objects the sweep deletes, and a retry cannot take
|
|
775
|
+
over an intent being swept. A takeover renews the expiry, so the sweep
|
|
776
|
+
cannot claim it.
|
|
777
|
+
4. Concurrent retirements and compactions count only row sets that are not
|
|
778
|
+
`pruning`, each after committing its own `pruning` state, so at least one
|
|
779
|
+
of any two deletes a shared object and none can both skip it.
|
|
780
|
+
5. Deletes name an exact native version, so one cannot remove an object
|
|
781
|
+
re-created at the same key. This needs write-unique native versions, a
|
|
782
|
+
provider requirement the conformance check `write-unique-version` tests.
|
|
783
|
+
|
|
784
|
+
The advisory lock `adcp-reporting-object:{binding_id}:{object_ref}` stays in
|
|
785
|
+
two places that pair with each other, because nothing else orders them. A
|
|
786
|
+
commit that adopts an object its own intent did not record (an earlier
|
|
787
|
+
attempt's `PUT` landed before the intent recorded it, or a writer created it
|
|
788
|
+
after a takeover) takes the lock, in key order, before it proves the object with
|
|
789
|
+
a version-pinned `stat`, and holds it until its chunk rows commit. An object its own intent recorded takes
|
|
790
|
+
no lock: until the transaction commits, the open intent names it. The orphan
|
|
791
|
+
collector deletes only objects no record names, so each of its deletes takes
|
|
792
|
+
the lock of every binding (they can share a bucket and prefix), repeats the
|
|
793
|
+
reference check under it, and then deletes. A collector that wins makes the
|
|
794
|
+
adopting commit fail its version-pinned `stat` with `STATE_UNAVAILABLE` (the
|
|
795
|
+
writer re-uploads once its intent lapses); a commit that wins is seen by the
|
|
796
|
+
collector, which skips. The collector's clone fence is a separate protection:
|
|
797
|
+
an advisory lock is per database and cannot order a clone against the
|
|
798
|
+
original. A skipped object is deleted by the retirement, compaction or sweep
|
|
799
|
+
of the last row set that referenced it, so it is not leaked.
|
|
800
|
+
- **Cross-SDK contract.** A ledger database has one writer SDK (the design record's
|
|
801
|
+
§5), so the Python and JavaScript SDKs need not agree on a lock to coexist; this
|
|
802
|
+
is the intended contract for both, with the same key text. The Python side
|
|
803
|
+
(adcp-client-python#1492) mirrors it: (1) the delete paths above take no object lock; (2) the key text
|
|
804
|
+
is `adcp-reporting-object:` + binding ID + `:` + object ref, hashed with
|
|
805
|
+
`hashtextextended(text, 0)` for `pg_advisory_xact_lock`; (3) an adopting commit
|
|
806
|
+
locks only adopted objects absent from its own intent's `created_objects`
|
|
807
|
+
(matching on object ref and native version), in object-ref order, after the
|
|
808
|
+
account lock and after closing its intent, then proves each adopted object with
|
|
809
|
+
a version-pinned `stat`; (4) the orphan collector locks every binding's key for
|
|
810
|
+
the object, sorted by binding ID, in the transaction that repeats the
|
|
811
|
+
reference check and deletes. Lock order is account lock, then object locks.
|
|
757
812
|
- An upload runs only while the obligation lease still covers
|
|
758
813
|
`writeDeadlineMilliseconds` plus five seconds. A fenced write intent records
|
|
759
814
|
every object the upload created, and the ledger commit closes the intent in
|
|
@@ -780,7 +835,7 @@ await store.readyRowStorage();
|
|
|
780
835
|
absent, without reading the body. The commit transaction uses it to prove an
|
|
781
836
|
adopted object is unchanged (native versions are write-unique, so the same
|
|
782
837
|
version at the recorded size is the same bytes) instead of downloading it
|
|
783
|
-
while holding the account
|
|
838
|
+
while holding the account lock. A provider without `stat` falls
|
|
784
839
|
back to reading the whole object there. All bundled providers implement it.
|
|
785
840
|
Also implement the optional `list({ prefix, pageToken, pageSize })`, which
|
|
786
841
|
returns `{ objects: [{ objectRef, nativeVersionRef, sizeBytes, createdAt }],
|
|
@@ -1309,6 +1364,80 @@ setInterval(() => {
|
|
|
1309
1364
|
}, 15 * 60_000);
|
|
1310
1365
|
```
|
|
1311
1366
|
|
|
1367
|
+
#### The two-table contract
|
|
1368
|
+
|
|
1369
|
+
The sink writes two tables, and the second is part of its contract. The rows
|
|
1370
|
+
table holds the report rows. The **revisions table** holds one metadata line per
|
|
1371
|
+
revision: `reporting_revision_id`, `reporting_obligation_id`, `account_id`,
|
|
1372
|
+
`revision_number`, `finality`, `period_date`, `supersedes_reporting_revision_id`,
|
|
1373
|
+
`period_start`, `period_end`, `source_timezone`, `observed_at`, `data_through`,
|
|
1374
|
+
`revision_content_sha256`, `row_count` and `control_totals`. The reference DDL
|
|
1375
|
+
(`sink.defaultTablesSql`) declares exactly these columns, and a test pins the
|
|
1376
|
+
DDL to what the sink loads. The `current_rows` view picks each obligation's
|
|
1377
|
+
current revision from the revisions table, so the rows table alone cannot tell
|
|
1378
|
+
you which revision is current.
|
|
1379
|
+
|
|
1380
|
+
- **The revisions table is required.** `revisionsTableId` has no default and no
|
|
1381
|
+
way to turn it off, and the sink refuses to start without it. A custom table
|
|
1382
|
+
must keep every column above: the load sets no schema and no
|
|
1383
|
+
`ignoreUnknownValues`, so a missing column fails the revisions job (and only
|
|
1384
|
+
that job: the rows job ran first and has already landed).
|
|
1385
|
+
- **Load order.** Each batch loads the rows table first, then the revisions table,
|
|
1386
|
+
as two separate load jobs. BigQuery has no transaction across tables. The
|
|
1387
|
+
cursor advances only after both jobs have finished cleanly.
|
|
1388
|
+
- **Visibility.** A revision's rows can be visible in the rows table before its
|
|
1389
|
+
metadata is in the revisions table: between the two jobs of a batch, and for as
|
|
1390
|
+
long as the revisions load keeps failing. `current_rows` shows only revisions
|
|
1391
|
+
whose metadata has loaded. While a newer revision's metadata is pending, each
|
|
1392
|
+
affected obligation keeps showing its previous revision, and an obligation
|
|
1393
|
+
whose first revision is pending shows nothing. The reverse does not happen
|
|
1394
|
+
through the sink: a revisions row implies its rows landed, because the rows job
|
|
1395
|
+
finished before the revisions job was submitted (so long as nothing outside the
|
|
1396
|
+
sink deletes or truncates the tables, which adopter-managed table retention
|
|
1397
|
+
can). Read the two tables in one query (the view does), or read the revisions
|
|
1398
|
+
table first.
|
|
1399
|
+
- **What to query.** For current figures, query `current_rows`. For revision
|
|
1400
|
+
history and audit, query the revisions table and join it to the rows table on
|
|
1401
|
+
`reporting_revision_id`. Do not sum the rows table alone: it holds every loaded
|
|
1402
|
+
revision, and cumulative snapshots double-count. To alert on metadata that is
|
|
1403
|
+
lagging, look for rows with no revisions row:
|
|
1404
|
+
|
|
1405
|
+
```sql
|
|
1406
|
+
SELECT DISTINCT r.reporting_revision_id
|
|
1407
|
+
FROM `analytics-prod.reporting.media_buy_delivery_rows` AS r
|
|
1408
|
+
LEFT JOIN `analytics-prod.reporting.media_buy_delivery_revisions` AS v USING (reporting_revision_id)
|
|
1409
|
+
WHERE v.reporting_revision_id IS NULL
|
|
1410
|
+
```
|
|
1411
|
+
|
|
1412
|
+
A non-empty result for a few minutes is the window between the two jobs. One
|
|
1413
|
+
that persists means the revisions load is failing (add a `period_date` filter
|
|
1414
|
+
to bound the scan on large tables).
|
|
1415
|
+
- **Recovery when the revisions load fails.** The batch plan stays in PostgreSQL
|
|
1416
|
+
and the cursor does not move, so nothing is lost and nothing needs editing by
|
|
1417
|
+
hand. The next `runOnce()` finds the rows job by its ID, accepts it as loaded
|
|
1418
|
+
and does not upload it again, then retries only the revisions job. A network
|
|
1419
|
+
error or timeout retries under the same job ID (BigQuery may have created the
|
|
1420
|
+
job, and the sink looks it up before it uploads). A job BigQuery ran and
|
|
1421
|
+
rejected is retried under a new attempt suffix: the run that first sees the
|
|
1422
|
+
failed job advances the attempt and rejects, and the run after it submits. A
|
|
1423
|
+
load job is atomic, so a failed one lands nothing and the rows are never
|
|
1424
|
+
duplicated. Fix the cause (usually a column the revisions table lacks), keep
|
|
1425
|
+
running, and `current_rows` catches up on its own. Do not insert or delete
|
|
1426
|
+
warehouse rows to repair it, and do not `reset()`: a reset reloads the history
|
|
1427
|
+
and duplicates it unless you empty the tables first. The exception is a failure
|
|
1428
|
+
that outlasts `maxFeedHoldDays`: change retention then stops waiting for the
|
|
1429
|
+
sink, and `runOnce()` reports the retired revisions as `missing` and loads no
|
|
1430
|
+
metadata for them. Their rows, already loaded, stay in the rows table with no
|
|
1431
|
+
revisions row (so the anti-join above keeps returning them and they never
|
|
1432
|
+
appear in `current_rows`); backfill their metadata or delete those rows by
|
|
1433
|
+
`reporting_revision_id`.
|
|
1434
|
+
- **Why the loader is custom.** The alternative is an Iceberg or Delta Lake table
|
|
1435
|
+
behind the object-storage interface, which BigQuery reads as an external table.
|
|
1436
|
+
That path carries no revisions table and no `current_rows` view, so an adopter
|
|
1437
|
+
who takes it owns the leaf-revision query.
|
|
1438
|
+
|
|
1439
|
+
#### How a run loads
|
|
1440
|
+
|
|
1312
1441
|
- Each run follows the change feed and reads every revision's rows through the
|
|
1313
1442
|
store's verified reader. It loads rows first and revision metadata second,
|
|
1314
1443
|
in one load job per table per tick. The sink always adds
|
|
@@ -1445,7 +1574,8 @@ setInterval(() => {
|
|
|
1445
1574
|
Do not discard the plan automatically: its load jobs may have succeeded.
|
|
1446
1575
|
- Query `current_rows`, not the raw rows table. Pacing snapshots are
|
|
1447
1576
|
cumulative, so summing across revisions double-counts. The view keeps each
|
|
1448
|
-
obligation's current revision
|
|
1577
|
+
obligation's current revision, read from the revisions table (see
|
|
1578
|
+
"The two-table contract" above for what is visible while that table lags).
|
|
1449
1579
|
- The sink registers as a change-feed consumer, so `pruneChanges` keeps changes
|
|
1450
1580
|
it has not loaded yet, and `retireExpiredPeriods` keeps the periods it has not
|
|
1451
1581
|
loaded yet, in both cases only while its cursor was saved within
|
|
@@ -131,7 +131,11 @@ service runs for you. See
|
|
|
131
131
|
cannot hold the table indefinitely.
|
|
132
132
|
5. **Warehouse sink.** Apply `REPORTING_WAREHOUSE_SINK_MIGRATION`, create the
|
|
133
133
|
tables, and the `current_rows` view, then schedule `runOnce`. Point analysts
|
|
134
|
-
at `current_rows`.
|
|
134
|
+
at `current_rows`. The sink loads two tables per batch, rows first and then the
|
|
135
|
+
required revisions table that `current_rows` reads through, so rows can be
|
|
136
|
+
visible before their revision metadata and `current_rows` shows only revisions
|
|
137
|
+
whose metadata has loaded (`docs/guides/REPORTING-LEDGER.md`, "The two-table
|
|
138
|
+
contract"). Set the sink's `location` to the dataset's location: replay
|
|
135
139
|
lookups of load jobs need it outside the US and EU multi-regions, and the sink
|
|
136
140
|
warns once per process when it is unset. When upgrading from 15.9.0, stop
|
|
137
141
|
every 15.9.0 sink worker before starting newer ones, then re-apply
|
|
@@ -180,10 +184,13 @@ service runs for you. See
|
|
|
180
184
|
`minAgeSeconds` ago (default 7 days, never less than 1 day, which must stay
|
|
181
185
|
well above the write deadline plus intent expiry); no row set that is not yet
|
|
182
186
|
`pruned` has a chunk at that key and no write intent lists it, at any version;
|
|
183
|
-
and, under the per-object advisory lock
|
|
184
|
-
|
|
185
|
-
|
|
186
|
-
|
|
187
|
+
and, under the per-object advisory lock (the only delete that takes it: a
|
|
188
|
+
commit that adopts an object its own intent did not record takes the same
|
|
189
|
+
lock, so the two cannot interleave), no `live` or `unavailable` row set
|
|
190
|
+
references that exact version and no open intent created it, checked under
|
|
191
|
+
every binding because two bindings can share a bucket and prefix (a retired
|
|
192
|
+
binding kept `readOnly` next to its successor). Retirement, compaction and the
|
|
193
|
+
intent sweep take no lock and hold no connection across their provider deletes.
|
|
187
194
|
`readOnly` bindings are skipped unless `includeReadOnly: true`. A binding
|
|
188
195
|
whose provider has no `list` is reported as `unsupported` (and as a
|
|
189
196
|
maintenance warning), never silently skipped. Everything under the owned prefix
|
|
@@ -294,6 +301,12 @@ Alert on these row-storage signals:
|
|
|
294
301
|
`listCurrentRevisions`. A first read with no cursor can also throw it when a
|
|
295
302
|
prune commits mid-read; retry it. A new warehouse sink stuck on it after such
|
|
296
303
|
a race recovers with `sink.reset()`.
|
|
304
|
+
- The warehouse sink's `runOnce()` keeps rejecting and the anti-join in "The
|
|
305
|
+
two-table contract" returns the same revision IDs run after run: rows are in the
|
|
306
|
+
rows table with no row in the revisions table. The revisions load is failing. The plan is retained and the rows are
|
|
307
|
+
already loaded, so fix the cause (usually a column the revisions table lacks) and
|
|
308
|
+
keep running: the retry loads only the revisions and `current_rows` catches up.
|
|
309
|
+
Do not edit the tables by hand and do not `reset()`.
|
|
297
310
|
- A non-empty `missing` from the warehouse sink's `runOnce()`. The sink was down
|
|
298
311
|
longer than `maxFeedHoldDays`, retention stopped waiting for it, and the
|
|
299
312
|
warehouse lacks those revisions. Backfill them or accept the gap, and keep
|
|
@@ -288,6 +288,41 @@ Either way, subsequent runs reuse the cached tokens (auto-refresh on expiry via
|
|
|
288
288
|
|
|
289
289
|
**Output:** JSON with `tracks[].status` (`passed`/`failed`/`skipped`), `steps[]` with diagnostics, and validations per step.
|
|
290
290
|
|
|
291
|
+
### Account-scoped fixtures and negative membership
|
|
292
|
+
|
|
293
|
+
Automatic `seed_product` and `seed_pricing_option` calls use the complete account
|
|
294
|
+
reference used by the storyboard: `context.account`, otherwise the single resolved
|
|
295
|
+
account in its effective sample requests or single-account provisioning request,
|
|
296
|
+
otherwise a single sandbox account fixture. Request overrides, context inputs, and
|
|
297
|
+
the run brand use the same account rules as public requests. Account provisioning
|
|
298
|
+
seeds retain their independent targets; only product and pricing seeds need a
|
|
299
|
+
shared catalog scope. Natural
|
|
300
|
+
keys retain brand countries, operator, operator unit, currency, and timezone. Opaque
|
|
301
|
+
`account_id` references stay opaque. A storyboard that uses several accounts must
|
|
302
|
+
provide an explicit `context.account` for its seeds; unresolved or ambiguous scope
|
|
303
|
+
fails setup rather than seeding a different default account. This is account
|
|
304
|
+
scoping, not a sandbox reset: fixtures for the same account may still coexist.
|
|
305
|
+
|
|
306
|
+
For product-filter checks, exclude a known negative control rather than assuming
|
|
307
|
+
only one product can match. A validation entry can use:
|
|
308
|
+
|
|
309
|
+
```yaml
|
|
310
|
+
check: field_excludes
|
|
311
|
+
path: products[*].product_id
|
|
312
|
+
value: filter_negative_control
|
|
313
|
+
description: The seeded display control is absent from an OLV result
|
|
314
|
+
```
|
|
315
|
+
|
|
316
|
+
Declare `requires: [field_excludes_runner]` at the storyboard root (alongside
|
|
317
|
+
`controller` when needed). Supporting SDKs satisfy this intrinsic requirement;
|
|
318
|
+
older SDKs skip the scenario before seeding instead of silently ignoring an unknown
|
|
319
|
+
check. The matcher has the same scalar, deep-subset, and `allowed_values` semantics
|
|
320
|
+
as `field_contains`, inverted over **every** resolved terminal. Empty arrays pass;
|
|
321
|
+
additional matching products are allowed. Failed checks report the prohibited
|
|
322
|
+
terminal in the `actual` array and `expected.excludes` while preserving the
|
|
323
|
+
seller's response. Diagnostic evidence is bounded to the matching terminal;
|
|
324
|
+
absence is checked across the complete result.
|
|
325
|
+
|
|
291
326
|
### `adcp fuzz` — property-based schema fuzzing
|
|
292
327
|
|
|
293
328
|
Generates schema-valid requests, calls your agent, checks every response under the two-path oracle (schema-valid success *or* well-formed AdCP error envelope with uppercase reason code). Anything else — 500, stack traces, lowercase error codes, token leaks — fails.
|
|
@@ -840,3 +875,12 @@ Random seed on nightly broadens the surface; fixed seed on per-PR keeps reproduc
|
|
|
840
875
|
For creative evaluator security fixtures, see
|
|
841
876
|
[Creative evaluator build probes](CREATIVE-BUILD-PROBE.md) for valid route/input
|
|
842
877
|
preparation, provisioned legacy selectors and explicit coverage gaps.
|
|
878
|
+
|
|
879
|
+
SDK 17 controller registration preserves complete account identity and requires
|
|
880
|
+
`account: { sandbox: true }` on every controller call, including `list_scenarios`.
|
|
881
|
+
Callers that used the AdCP 3.0 account-less form must add this assertion before
|
|
882
|
+
upgrading default MCP registration. `account_id` may accompany the assertion on
|
|
883
|
+
the controller carrier; ordinary public opaque AccountRefs remain `{account_id}`.
|
|
884
|
+
The platform wrapper retains its optional-account compatibility shape and trusted
|
|
885
|
+
principal/target sandbox authorization. A caller assertion never grants sandbox
|
|
886
|
+
authority by itself.
|
package/docs/llms.txt
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
# Ad Context Protocol (AdCP)
|
|
2
2
|
|
|
3
3
|
> Generated at: 2026-11-03
|
|
4
|
-
> Library: @adcp/sdk v17.0.0-beta.
|
|
4
|
+
> Library: @adcp/sdk v17.0.0-beta.15
|
|
5
5
|
> AdCP major version: 3
|
|
6
6
|
> Canonical URL: https://adcontextprotocol.github.io/adcp-client/llms.txt
|
|
7
7
|
> Note: the `Library` stamp reflects the package.json version at doc-generation time. The narrative below describes the surface that lands on the next-published minor — including any 6.7 helpers documented here ahead of the release tag.
|
|
@@ -16,8 +16,8 @@ pin using `ADCP_VERSION` inside that tarball.
|
|
|
16
16
|
|
|
17
17
|
| Fact | Value |
|
|
18
18
|
| --- | --- |
|
|
19
|
-
| Checkout `package.json` version | `@adcp/sdk@17.0.0-beta.
|
|
20
|
-
| npm integrity | registry-derived after publication; run `npm view @adcp/sdk@17.0.0-beta.
|
|
19
|
+
| Checkout `package.json` version | `@adcp/sdk@17.0.0-beta.15` |
|
|
20
|
+
| npm integrity | registry-derived after publication; run `npm view @adcp/sdk@17.0.0-beta.15 dist.integrity` |
|
|
21
21
|
| Node.js runtime | `^20.19.0 || >=22.12.0` |
|
|
22
22
|
| Default AdCP wire release | `3.3.0-beta.3` |
|
|
23
23
|
| Maintained wire releases | `v3`, `3.0.0`, `3.0`, `3.0.1`, `3.0.2`, `3.0.3`, `3.0.4`, `3.0.5`, `3.0.6`, `3.0.7`, `3.0.8`, `3.0.9`, `3.0.10`, `3.0.11`, `3.0.12`, `3.0.13`, `3.0.14`, `3.0.15`, `3.0.16`, `3.0.17`, `3.0.18`, `3.0.19`, `3.0.20`, `3.0.21`, `3.0.22`, `3.0.23`, `3.0.24`, `3.0.25`, `3.1.0`, `3.1`, `3.1.1`, `3.1.2`, `3.1.3`, `3.1.4`, `3.1.5`, `3.1.6`, `3.1.7`, `3.1.8`, `3.1.9`, `3.1.10`, `3.1.11`, `3.1.12`, `3.1.13`, `3.1.14`, `3.1.15`, `3.1.16`, `3.1.17`, `3.1.18`, `3.1.19`, `3.1.20`, `3.1.21`, `3.1.22`, `3.1.23`, `3.1.24`, `3.1.25`, `3.1.26`, `3.1.27`, `3.1.28`, `3.1.29`, `3.1.30`, `3.1.31`, `3.2.1`, `3.2`, `3.2.2`, `3.2.3`, `3.2.4`, `3.2.5`, `3.2.6`, `3.2.7`, `3.2.8`, `3.3.0-beta.3`, `3.3-beta.3` |
|
|
@@ -26,8 +26,8 @@ pin using `ADCP_VERSION` inside that tarball.
|
|
|
26
26
|
For a checkout at the published release commit, install the exact package version:
|
|
27
27
|
|
|
28
28
|
```bash
|
|
29
|
-
npm install --save-exact '@adcp/sdk@17.0.0-beta.
|
|
30
|
-
npm view '@adcp/sdk@17.0.0-beta.
|
|
29
|
+
npm install --save-exact '@adcp/sdk@17.0.0-beta.15'
|
|
30
|
+
npm view '@adcp/sdk@17.0.0-beta.15' dist.integrity
|
|
31
31
|
```
|
|
32
32
|
|
|
33
33
|
### Required and optional peers
|
|
@@ -62,8 +62,10 @@ read-only binding is exempt from the requirement, is never chosen for new
|
|
|
62
62
|
writes, and is still used for reads and retirement (`readyRowStorage` still
|
|
63
63
|
probes it, so keep its location reachable). Compaction ignores read-only
|
|
64
64
|
bindings, because nothing can adopt through them. The
|
|
65
|
-
reference-checked deletes
|
|
66
|
-
|
|
65
|
+
reference-checked deletes are unchanged, because header-only revisions still
|
|
66
|
+
share chunks within an obligation. (The per-object lock this note once said was
|
|
67
|
+
unchanged has since been removed from every path but the orphan collector; see
|
|
68
|
+
`docs/guides/REPORTING-LEDGER.md`.)
|
|
67
69
|
|
|
68
70
|
## Renamed fields
|
|
69
71
|
|