fix: a deleted NodeType gates portal readiness forever (#1391) by rbuergi · Pull Request #1396 · Systemorph/MeshWeaver

rbuergi · 2026-08-13T07:38:07Z

Part of #1391. Independent of #1393 (#1387) — different root, evidence in the issue.

The rule, stated

"No sources matched" is not the same as "sources matched and did not compile." The first is a CONTENT fact and must not gate readiness; the second is a verdict about this image and must.

PreWarmStatus.NoSources exists for exactly the first case and is deliberately non-gating. Almost nothing could reach it.

🚨 And "no sources matched" needs a second witness, which the first revision of this PR got wrong — see Round two below. The sources must have been LOST, not merely absent.

The defect

Both classifiers additionally required d.Sources is { Count: > 0 }:


`DynamicTypePreWarmer.ClassifyCompileFailure`	activation-driven sweep
`NodeTypeBatchBake.StampAndReport`	batch bake

An empty NodeTypeDefinition.Sources does not mean "configuration-only". It means "uses the DEFAULT queries" — namespace:{path}/Source scope:subtree, see CodeQueryResolver.DefaultSources — which is how very nearly every NodeType in a real mesh is authored.

So the conjunct made NoSources unreachable for essentially the whole population. Delete a type's Source/ subtree and it is compiled against nothing, the configuration lambda's resulting CS0246/CS1061 are recorded as an image regression, and a node that no longer exists holds portal readiness on every pod boot.

That is what Edu/Course was doing to memex — failing on 100% of 23 sampled boots, with a diagnostic naming code that no longer exists:

Executed source queries (2):
  - namespace:Edu/Course/Source scope:subtree nodeType:Code
  - namespace:Edu/Course/Test scope:subtree nodeType:Code
Matched Code nodes (0):
  (none) — the configuration lambda cannot reference types because no source
  files were included.
CS0246: The type or namespace name 'CourseContent' could not be found
CS1061: 'LayoutDefinition' does not contain a definition for 'AddCourseLayoutAreas'

Not a new policy — a restored one

SourceSnapshot.cs already documents the intended three-outcome contract, and it is unambiguous:

Established, EMPTY — every query answered and matched nothing. That is a CONTENT fact (the sources were deleted, or the type is configuration-only), so the compile proceeds and any failure classifies as PreWarmStatus.NoSources, which correctly does not gate a rollout.

The extra conjunct had silently drifted from that. A genuinely configuration-only type is folded in on purpose, exactly as that contract says: with no Code nodes at all there is nothing for an image to regress, so a failure in its configuration lambda is content drift for its owner to fix, not a reason to stall everyone else's rollout.

What is deliberately NOT changed

🚨 The gate. This is a classification fix, never "compile errors stop gating". A type whose sources are still present and do not compile keeps a non-empty snapshot and keeps gating.
The two-witness rule (Batch bake's source discovery is truncated by the default query limit (50) — 169/237 types got zero sources in production #1216). The batch still requires BOTH its own resolved snapshot AND the persisted CurrentSourceVersions, so a starved discovery cannot reclassify itself as deleted content.
A NULL snapshot still gates — "the watcher never seeded" is not "the sources are gone".
NodeTypeBatchBake.Assemble's unestablished guard keeps its DeclaresSources exemption. That one is load-bearing and I left it alone on purpose: a legitimately source-less type must not abandon the whole batch, and the persisted snapshot is what tells the two apart at classification time. Fixing it too would make any configuration-only type abandon every batch.

A tested behaviour is being changed, deliberately

ClassifyCompileFailure_NoDeclaredSources_StaysCompileError asserted the old behaviour, on the premise that an empty Sources means the type "compiles from its Configuration alone". That premise is wrong about the domain — it describes a tiny minority as if it were the definition. The test is replaced rather than kept, with the reasoning recorded where the next reader will find it.

Round two: an empty snapshot alone was too coarse

The first revision keyed only on CurrentSourceVersions being empty. CI caught that with three reds in DynamicTypePreWarmerTest, and they were right — this was a defect in my classifier, not in the fixtures.

All three build their "broken" type as Configuration = "config => this is not valid C# at all ((" with no sources. That is not deleted content; it is a type that is broken as authored, and it must gate. Reclassifying it also downgraded its dependents' cascade from the gating UpstreamFailed to the non-gating UpstreamContentBroken — a real weakening of the gate, which is exactly what those tests exist to catch.

The second witness is LastCompileSucceededAt: a type that has NEVER built cannot have LOST anything. It survives a failure — ApplyCompileFailure clears CompiledSources but never the success stamp — so it is readable exactly where the verdict is formed.

The three failing tests now pass with their assertions and fixtures completely untouched (git diff origin/main...HEAD -- .../DynamicTypePreWarmerTest.cs is empty). The production rule got stricter; nothing was loosened to make a test green.

Verified against production, 11/11, on two independent durable signals

population	`lastCompileSucceededAt`	verdict
`Edu/Course` + 3 × `Doc/*`	non-null	reclassify — their sources genuinely were deleted
6 × `FutuRe/{AmericasIns,AsiaRe,EuropeRe}/{LineOfBusiness,TransactionMapping}`	null	gate — no `sources` key and no `Source/` subdirectory in the repo, yet their lambdas reference `LineOfBusiness` / `AddLineOfBusinessLayoutAreas`. Broken as authored: the real-world twin of the test fixtures
`Type/article`	null	gate — never compiled. The importer derives the node path from the FILENAME while the Code node lands under the `id` (`Type/Article/Source/Article`, capital A)
2 × `Northwind/{AnalyticsCatalog,Product}`	n/a	gate, on the non-empty snapshot as before

mesh_node_history corroborates the same 3/7 split independently — those 7 never recorded a source across 44–47 versions each.

Blast radius: 4 of 16 failures reclassify. memex-cloud keeps 9 gating, memex keeps 1 (BinaryClickerV2/BinaryToggle, null snapshot, correctly gating). The ones left gating deserve to.

Two traps recorded in the code so they are not "simplified" back

THREE shapes of CurrentSourceVersions exist in production, not two: populated, explicitly {}, and ABSENT (SQL NULL, on public.mesh_nodes rows). Only the middle may reclassify. is { Count: 0 } is right by construction — a C# property pattern never matches null — so absent gates. Do not rewrite it as a bare .Count == 0, which throws.
CompiledSources would be the more precise evidence but is useless as history: written only on success and nulled on failure, so every currently-failing type in production lacks it — including the three that demonstrably compiled in June.

Also: the inverse hazard, documented

NodeTypeCompilation.md covered deleting a framework symbol in-mesh code calls. Adding one has the same invisibility: #1386 moved an Article extractor into compiled framework code as MarkdownBody.Of, and the in-mesh callers shipped before the image did — ACME/Article, Cornerstone/Article and Northwind/Article sat at CS0103: The name 'MarkdownBody' does not exist on memex-cloud until the portal self-updated to ci.3435, at which point all three returned to Ok on their own. No action needed; the rule ("the framework half must land first") is now written down.

Verification

Fail-before / pass-after:

	before	after
`ClassifyCompileFailure_DefaultQueriesWithEmptySnapshot_IsNoSources`	FAIL	PASS
`DeletedDefaultQueryType_DoesNotGateReadiness` (end-to-end through the gate)	FAIL	PASS
`ClassifyCompileFailure_DefaultQueriesWithSourcesStillPresent_StaysCompileError`	PASS	PASS
`BrokenDefaultQueryTypeWithLiveSources_StillGatesReadiness` (end-to-end)	PASS	PASS

Both regression guards pass in both states — which is what a guard should do; they exist so the fix cannot widen later.

Plus the round-two boundary, both directions:

	before round two	after
`ClassifyCompileFailure_EmptySnapshotButNeverBuilt_StaysCompileError`	FAIL	PASS
`NeverBuiltSourcelessBrokenType_StillGatesReadiness` (end-to-end)	FAIL	PASS
the three `DynamicTypePreWarmerTest` reds, fixtures UNTOUCHED	FAIL	PASS

67/67 across DynamicTypePreWarmerTest, NodeTypeBakeGateStateTest, SourceDiscoveryUnavailableTest, SourceSnapshotEstablishmentTest, NodeTypeBatchBakeDiscoveryInvariantTest.
DocumentationLinkIntegrityTest + WhatsNewEntryIntegrityTest green.
Release -warnaserror clean: MeshWeaver.Hosting, MeshWeaver.Hosting.Monolith.Test.

SourceDiscoveryUnavailableTest.StarvedSourceDiscovery_IsNotACompileVerdict_WhileARealCompileErrorStillIs was raised as possibly sharing a root: it does not. Its broken control does create a {path}/Source/code node, so its snapshot is non-empty and this change cannot reach it — green locally throughout.

Confirmed against the durable row — the discriminator is 11 for 11

The one thing that could have made this fix a no-op for the real case is Edu/Course carrying a stale non-empty currentSourceVersions instead of an explicitly empty one. It does not. From edu.mesh_nodes (db memex, read-only SELECT):

SELECT jsonb_typeof(content->'currentSourceVersions'),
       content->'currentSourceVersions',
       (SELECT count(*) FROM jsonb_object_keys(content->'currentSourceVersions'))
FROM edu.mesh_nodes WHERE path='Edu/Course';
--  object | {} | 0

Type object, key count 0 — and sources is SQL NULL, i.e. default queries. So this fix does reclassify it.

The node itself is state=2 (ACTIVE); what is gone is its Source/ subtree — 0 rows in edu.code, and zero rows in edu.mesh_node_history for those paths, so a hard delete rather than a recycle. Edu/_GitSync dates it: the last version with a non-empty map is v282 at 2026-07-11 18:49:45.545+00, matching lastSyncedAt 2026-07-11T18:49:45.618+00 (sha 4ce635b). The GitSync import of the Plugins repo's Edu/ subdirectory removed the sources and left the NodeType node Active. v283→v301 are bake retries against nothing.

Across both portals the discriminator separates cleanly, with the DB LIKE '{path}/Source/%' count and the runtime Matched Code nodes (N) agreeing on every row:

population	keys in `currentSourceVersions`	verdict
`Edu/Course` (memex)	0	content-broken
10 memex-cloud types under `MeshWeaver/samples/Graph/Data/`	0	content-broken
5 memex-cloud types (`*/Article`, `Northwind/{AnalyticsCatalog,Product}`)	2, 2, 2, 26, 2	still gates — genuine `CS0103`/`CS0246` breakage

11 for 11, zero false positives, no overlap. The five REAL failures keep gating, which is the property the regression guards exist to protect.

The third shape, already handled

public.mesh_nodes holds rows whose currentSourceVersions is absent / SQL NULL — a third shape beyond "empty" and "populated". No change needed: d.CurrentSourceVersions is { Count: 0 } is a C# property pattern, which never matches null, so absent ⇒ CompileError ⇒ still gates. That is the intended "the watcher never seeded is not the sources are gone" rule, pinned by ClassifyCompileFailure_DefaultQueriesWithNullSnapshot_StaysCompileError.

Still open on #1391 (not addressed here)

The GitSync importer leaves a NodeType node behind when its sources go away or land at a different path. Edu/Course is one instance; MeshWeaver/samples/Graph/Data/Type/article is another and is a plain platform bug — the importer derives the node path from the FILENAME (article.json) while the Code node lands under the id (Type/Article/Source/Article), so the default query can never match.

This PR stops those from holding readiness hostage. It does not stop the importer creating the condition, and it does not clean up the existing rows. Both are separate, and any data repair is the owner's call to run — nothing was mutated during this investigation (SELECT-only). The repair options are written up on the issue.

🤖 Generated with Claude Code

"No sources matched" is not the same as "sources matched and did not compile." The first is a CONTENT fact and must not gate readiness; the second is a verdict about this image and must. PreWarmStatus.NoSources exists for exactly the first case and is deliberately non-gating — but almost nothing could reach it. Both classifiers additionally required `d.Sources is { Count: > 0 }`: DynamicTypePreWarmer.ClassifyCompileFailure (activation-driven sweep) NodeTypeBatchBake.StampAndReport (batch bake) An empty NodeTypeDefinition.Sources does NOT mean "configuration-only". It means "uses the DEFAULT queries" — `namespace:{path}/Source scope:subtree`, see CodeQueryResolver.DefaultSources — which is how very nearly every NodeType in a real mesh is authored. So the conjunct made NoSources unreachable for essentially the whole population: delete a type's Source/ subtree and it is compiled against nothing, the configuration lambda's resulting CS0246/CS1061 are recorded as an IMAGE regression, and a node that no longer exists holds portal readiness on every pod boot. That is what Edu/Course was doing to memex — deleted in the 2026-08-12 Edu rework, failing on 100% of 23 sampled boots, with a diagnostic naming code that no longer exists: Executed source queries (2): - namespace:Edu/Course/Source scope:subtree nodeType:Code - namespace:Edu/Course/Test scope:subtree nodeType:Code Matched Code nodes (0): (none) — the configuration lambda cannot reference types because no source files were included. CS0246: The type or namespace name 'CourseContent' could not be found CS1061: 'LayoutDefinition' does not contain a definition for 'AddCourseLayoutAreas' This restores the three-outcome contract SourceSnapshot.cs already documents — an established-but-EMPTY snapshot is a content fact "(the sources were deleted, or the type is configuration-only)" and classifies as NoSources — from which the extra conjunct had silently drifted. A genuinely configuration-only type is folded in on purpose: with no Code nodes there is nothing for an image to regress, so a failure in its configuration lambda is content drift for its owner, not a reason to stall everyone else's rollout. What is deliberately NOT changed: - The GATE. This is a classification fix, never "compile errors stop gating". A type whose sources are still present and do not compile keeps a non-empty snapshot and keeps gating. - The two-witness rule (#1216). The batch still requires BOTH its own resolved snapshot AND the persisted CurrentSourceVersions, so a starved discovery cannot reclassify itself as deleted content. - A NULL snapshot still gates — "the watcher never seeded" is not "the sources are gone". - NodeTypeBatchBake.Assemble's `unestablished` guard keeps its DeclaresSources exemption. That one is load-bearing: a legitimately source-less type must not abandon the whole batch, and the persisted snapshot is what tells the two apart at classification time. The existing ClassifyCompileFailure_NoDeclaredSources_StaysCompileError asserted the old behaviour on the premise that an empty Sources means the type "compiles from its Configuration alone". That premise is wrong about the domain, so the test is replaced rather than kept — with the reasoning recorded where the next reader will find it. Verification (fail-before / pass-after): before — ClassifyCompileFailure_DefaultQueriesWithEmptySnapshot_IsNoSources FAIL DeletedDefaultQueryType_DoesNotGateReadiness FAIL after — 46/46 NodeTypeBakeGateStateTest; 55/55 including SourceDiscoveryUnavailableTest, SourceSnapshotEstablishmentTest, NodeTypeBatchBakeDiscoveryInvariantTest Both regression guards pass in BOTH states, which is what a guard should do. Release -warnaserror clean: MeshWeaver.Hosting, Hosting.Monolith.Test. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Copilot

Pull request overview

This PR fixes how NodeType compile failures are classified during pre-warm/batch bake so that “no sources matched” (deleted or genuinely source-less content) is treated as a non-gating content condition (PreWarmStatus.NoSources), rather than a gating image regression (CompileError). This prevents deleted/default-query NodeTypes from holding portal readiness indefinitely.

Changes:

Adjust DynamicTypePreWarmer.ClassifyCompileFailure to key solely off CurrentSourceVersions being explicitly empty, removing the (incorrect) dependency on NodeTypeDefinition.Sources being non-empty.
Align batch bake classification (NodeTypeBatchBake.StampAndReport) with the activation path so default-query NodeTypes can reach NoSources when their resolved source set is empty and the persisted snapshot corroborates it.
Update/replace regression tests and add an end-to-end gate assertion for deleted default-query NodeTypes; add a WhatsNew entry describing the user-visible fix.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
test/MeshWeaver.Hosting.Monolith.Test/NodeTypeBakeGateStateTest.cs	Replaces the old assumption about empty `Sources` and adds end-to-end gate coverage for deleted vs. broken default-query NodeTypes.
src/MeshWeaver.Hosting/NodeTypeBatchBake.cs	Updates batch bake’s compile-failure classification to treat empty resolved sources as `NoSources` without requiring declared queries, while keeping the two-witness rule.
src/MeshWeaver.Hosting/DynamicTypePreWarmer.cs	Fixes the activation-path classifier so explicitly empty snapshots classify as `NoSources` regardless of whether queries were declared.
src/MeshWeaver.Documentation/Data/WhatsNew/2026-08-13-a-deleted-page-type-no-longer-holds-up-the-portal-starting.md	Documents the behavioral fix as a user-facing “Fix” entry.

Suppressed comments (1)

src/MeshWeaver.Hosting/DynamicTypePreWarmer.cs:832

The XML doc cref Graph.Configuration.SourceSnapshot doesn’t resolve (no such namespace). Also, SourceSnapshot is internal to MeshWeaver.Graph.Configuration, so a cross-assembly <see cref> may not bind even with the correct namespace; using <c>SourceSnapshot</c> avoids broken cref warnings.

    ///
    /// <para>This restores the three-outcome contract <see cref="Graph.Configuration.SourceSnapshot"/>
    /// already documents — an established-but-EMPTY snapshot is a content fact "(the sources were
    /// deleted, or the type is configuration-only)" and classifies as <c>NoSources</c> — from which

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

+    /// does not mean "configuration-only" — it means <b>"uses the DEFAULT queries"</b>
+    /// (<c>namespace:{path}/Source scope:subtree</c>, see
+    /// <see cref="Graph.Configuration.CodeQueryResolver.DefaultSources"/>), which is how very
+    /// nearly every NodeType in a real mesh is authored. Requiring declared queries therefore made


…lt (#1391) The member doc still said 'the type's DECLARED source queries', which is the premise the classifier fix removes. Restates it as the effective queries and records why the declared-only condition must not come back. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

github-actions · 2026-08-13T07:55:08Z

Test Results (shard 0)

723 tests ±0 723 ✅ ±0 10m 43s ⏱️ - 1m 3s
10 suites ±0 0 💤 ±0
10 files ±0 0 ❌ ±0

Results for commit 6346405. ± Comparison against base commit b073c6b.

♻️ This comment has been updated with latest results.

github-actions · 2026-08-13T07:56:12Z

Test Results (shard 3)

11 files ±0 11 suites ±0 5m 15s ⏱️ -34s
1 767 tests ±0 1 662 ✅ ±0 105 💤 ±0 0 ❌ ±0
2 074 runs +5 1 969 ✅ +5 105 💤 ±0 0 ❌ ±0

Results for commit 6346405. ± Comparison against base commit b073c6b.

♻️ This comment has been updated with latest results.

github-actions · 2026-08-13T07:56:26Z

Test Results (shard 5)

1 908 tests +2 1 716 ✅ +2 5m 10s ⏱️ -38s
11 suites ±0 192 💤 ±0
11 files ±0 0 ❌ ±0

Results for commit 6346405. ± Comparison against base commit b073c6b.

♻️ This comment has been updated with latest results.

github-actions · 2026-08-13T07:56:44Z

Test Results (shard 4)

1 975 tests +16 1 967 ✅ +16 8m 33s ⏱️ +6s
11 suites ± 0 8 💤 ± 0
11 files ± 0 0 ❌ ± 0

Results for commit 6346405. ± Comparison against base commit b073c6b.

♻️ This comment has been updated with latest results.

github-actions · 2026-08-13T07:58:28Z

Test Results (shard 2)

2 411 tests +2 2 410 ✅ +2 6m 25s ⏱️ -17s
12 suites ±0 1 💤 ±0
12 files ±0 0 ❌ ±0

Results for commit 6346405. ± Comparison against base commit b073c6b.

♻️ This comment has been updated with latest results.

github-actions · 2026-08-13T07:58:51Z

Test Results (shard 1)

1 474 tests +21 1 474 ✅ +21 7m 34s ⏱️ +26s
11 suites ± 0 0 💤 ± 0
11 files ± 0 0 ❌ ± 0

Results for commit 6346405. ± Comparison against base commit b073c6b.

This pull request removes 1 and adds 22 tests. Note that renamed tests count towards both.

MeshWeaver.Hosting.Monolith.Test.NodeTypeBakeGateStateTest ‑ ClassifyCompileFailure_NoDeclaredSources_StaysCompileError

MeshWeaver.Data.Test.PatchStringSpliceTest ‑ AFullBaseStringCarriedWithASplice_StillGetsTheThreeWayRebase
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ ALiveValueThatIsNoLongerAString_RefusesRatherThanWritingTheMarker
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ ANearTotalRewrite_StaysOnTheWholeValuePath
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ ASmallString_StaysOnTheWholeValuePath
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ ASpliceWithNoBaseSignal_IsRefused_NeverAppliedAtAnUnverifiedOffset
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ AgainstTheBaseItWasComputedFrom_TheSpliceReproducesTheFullValueExactly
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ AnAppendToABigString_EncodesAsASplice_NotTheWholeValue
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ ExtractBaseValues_ShipsAFingerprintForASplicedLeaf_NotTheWholeOldString
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ ExtractBaseValues_StillShipsTheWholeOldValueForANonSplicedLeaf
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ ManySequentialSplices_ReproduceTheWholeStreamedText
…

♻️ This comment has been updated with latest results.

github-actions · 2026-08-13T07:59:13Z

Test Results

66 files ± 0 66 suites ±0 43m 41s ⏱️ - 2m 1s
10 258 tests +41 9 952 ✅ +41 306 💤 ±0 0 ❌ ±0
10 565 runs +46 10 259 ✅ +46 306 💤 ±0 0 ❌ ±0

Results for commit 6346405. ± Comparison against base commit b073c6b.

This pull request removes 1 and adds 42 tests. Note that renamed tests count towards both.

MeshWeaver.Hosting.Monolith.Test.NodeTypeBakeGateStateTest ‑ ClassifyCompileFailure_NoDeclaredSources_StaysCompileError

MeshWeaver.Data.Test.PatchStringSpliceTest ‑ AFullBaseStringCarriedWithASplice_StillGetsTheThreeWayRebase
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ ALiveValueThatIsNoLongerAString_RefusesRatherThanWritingTheMarker
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ ANearTotalRewrite_StaysOnTheWholeValuePath
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ ASmallString_StaysOnTheWholeValuePath
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ ASpliceWithNoBaseSignal_IsRefused_NeverAppliedAtAnUnverifiedOffset
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ AgainstTheBaseItWasComputedFrom_TheSpliceReproducesTheFullValueExactly
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ AnAppendToABigString_EncodesAsASplice_NotTheWholeValue
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ ExtractBaseValues_ShipsAFingerprintForASplicedLeaf_NotTheWholeOldString
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ ExtractBaseValues_StillShipsTheWholeOldValueForANonSplicedLeaf
MeshWeaver.Data.Test.PatchStringSpliceTest ‑ ManySequentialSplices_ReproduceTheWholeStreamedText
…

♻️ This comment has been updated with latest results.

…ST (#1391) Three DynamicTypePreWarmerTest failures on CI were caused by my diff, and they were RIGHT. This is the classifier's fault, not the fixtures'. An empty CurrentSourceVersions has two very different causes, and the previous commit conflated them: Sources were DELETED (Edu/Course) → content verdict, must NOT gate NEVER had sources, own Configuration is broken → authored defect, MUST gate All three fixtures are the second kind — `Configuration = "config => this is not valid C# at all (("` with no sources — so reclassifying them to NoSources also downgraded their dependents' cascade from the gating UpstreamFailed to the non-gating UpstreamContentBroken. That is a real weakening of the gate, which is precisely what those tests exist to catch. Their assertions are untouched. The second witness is LastCompileSucceededAt: a type that has NEVER built cannot have LOST anything. It survives a failure — ApplyCompileFailure clears CompiledSources but never the success stamp — so it is readable exactly where the verdict is formed. Verified against production, 11/11, on two independent durable signals: Edu/Course + 3 × Doc/* non-null → reclassify (their sources WERE deleted) 6 × FutuRe/{region}/* null → gate. No `sources` key and no Source/ subdirectory in the repo, yet their lambdas reference LineOfBusiness / AddLineOfBusinessLayoutAreas. Broken as authored — the real-world twin of the test fixtures. Type/article null → gate. Never compiled: the importer derives the node path from the FILENAME while the Code node lands under the id (capital A). 2 × Northwind (live sources) n/a → gate, on the non-empty snapshot as before. mesh_node_history corroborates the same 3/7 split — those 7 never recorded a source across 44-47 versions each. Blast radius: 4 of 16 failures reclassify; memex-cloud keeps 9 gating, memex keeps 1 (BinaryClickerV2/BinaryToggle, null, correctly gating). Two traps recorded in the code so they are not "simplified" back: - THREE shapes of CurrentSourceVersions exist in production, not two: populated, explicitly {}, and ABSENT (SQL NULL, on public.mesh_nodes rows). Only the middle may reclassify. `is { Count: 0 }` is right BY CONSTRUCTION — a C# property pattern never matches null — so absent gates. Do not rewrite it as a bare `.Count == 0`, which throws. - CompiledSources would be the more precise evidence but is useless as history: written only on success and nulled on failure, so every currently-failing type in production lacks it, INCLUDING the three that demonstrably compiled in June. Also documents the inverse of the deletion hazard in NodeTypeCompilation.md: ADDING a framework symbol that in-mesh source calls has the same invisibility. #1386 moved an Article extractor to MarkdownBody.Of and the in-mesh callers shipped before the image, leaving three types at CS0103 in prod until the portal self-updated. The framework half must land first. Verification: 67/67 across DynamicTypePreWarmerTest (all three previously-red, fixtures UNCHANGED), NodeTypeBakeGateStateTest, SourceDiscoveryUnavailableTest, SourceSnapshotEstablishmentTest, NodeTypeBatchBakeDiscoveryInvariantTest. Doc link + What's New integrity green. Release -warnaserror clean: Hosting, Hosting.Monolith.Test. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

…es' into fix/1391-nosources-default-queries

Copilot AI lite review requested due to automatic review settings August 13, 2026 07:38

Copilot started reviewing on behalf of rbuergi August 13, 2026 07:38

View session

Copilot AI reviewed Aug 13, 2026

View reviewed changes

rbuergi mentioned this pull request Aug 13, 2026

Edu/Course and BinaryClickerV2/BinaryToggle are at CompileError on the running production image #1391

Closed

rbuergi and others added 3 commits August 13, 2026 13:15

Merge branch 'main' into fix/1391-nosources-default-queries

e130a22

Merge remote-tracking branch 'origin/fix/1391-nosources-default-queri…

6346405

…es' into fix/1391-nosources-default-queries

rbuergi merged commit cd8b09a into main Aug 13, 2026
19 checks passed

This was referenced Aug 13, 2026

A NodeType row can sit at compilationStatus='Compiling' indefinitely — nothing reconciles a non-terminal compile state #1462

Closed

fix(#1462): a stranded Compiling claim is re-driven by the sweep that already runs #1593

Merged

rbuergi deleted the fix/1391-nosources-default-queries branch August 19, 2026 10:33

rbuergi mentioned this pull request Sep 6, 2026

Release 3.0.0, then open 3.1.0 (tracking) #3355

Closed

6 of 26 tasks

Navigation Menu

Repository navigation

fix: a deleted NodeType gates portal readiness forever (#1391) - #1396

fix: a deleted NodeType gates portal readiness forever (#1391)#1396
rbuergi merged 5 commits into
mainfrom
fix/1391-nosources-default-queries

rbuergi commented Aug 13, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

github-actions Bot commented Aug 13, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Aug 13, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Aug 13, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Aug 13, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Aug 13, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Aug 13, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Aug 13, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

rbuergi commented Aug 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

The rule, stated

The defect

Not a new policy — a restored one

What is deliberately NOT changed

A tested behaviour is being changed, deliberately

Round two: an empty snapshot alone was too coarse

Verified against production, 11/11, on two independent durable signals

Two traps recorded in the code so they are not "simplified" back

Also: the inverse hazard, documented

Verification

Confirmed against the durable row — the discriminator is 11 for 11

The third shape, already handled

Still open on #1391 (not addressed here)

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

github-actions Bot commented Aug 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Test Results (shard 0)

Uh oh!

github-actions Bot commented Aug 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Test Results (shard 3)

Uh oh!

github-actions Bot commented Aug 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Test Results (shard 5)

Uh oh!

github-actions Bot commented Aug 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Test Results (shard 4)

Uh oh!

github-actions Bot commented Aug 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Test Results (shard 2)

Uh oh!

github-actions Bot commented Aug 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Test Results (shard 1)

Uh oh!

github-actions Bot commented Aug 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Test Results

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

rbuergi commented Aug 13, 2026 •

edited

Loading

github-actions Bot commented Aug 13, 2026 •

edited

Loading

github-actions Bot commented Aug 13, 2026 •

edited

Loading

github-actions Bot commented Aug 13, 2026 •

edited

Loading

github-actions Bot commented Aug 13, 2026 •

edited

Loading

github-actions Bot commented Aug 13, 2026 •

edited

Loading

github-actions Bot commented Aug 13, 2026 •

edited

Loading

github-actions Bot commented Aug 13, 2026 •

edited

Loading