Resolves #12059

Authored by GPT-5 (Codex Desktop). Session d84ad2f1-c71c-4ce4-bec7-1167b9183637.

FAIR-band: over-target [18/30] — taking this lane despite over-target because this is an operator-declared P0 Sandman incident and the branch was already assigned to @neo-gpt during recovery.

Sandman/Dream graph extraction now uses a graph-specific provider selector instead of the generic chat/summarization modelProvider. The tracked template ships the concrete default graphProvider: 'openAiCompatible'; deployments that run graph extraction against native Ollama use the explicit NEO_GRAPH_PROVIDER=ollama override. Resolver code returns the configured graph provider verbatim, so there is no hidden fallback from modelProvider, no auto policy, and no null/absence default.

Evidence: L2 (focused unit coverage for selector, explicit config behavior, Memory Core template mapping, Sandman readiness diagnostics, and graph consumer-model telemetry) -> L4 required (full manual npm run ai:run-sandman against live Memory Core). Residual: live Sandman REM run remains operator-gated because it mutates production Memory Core graph/session state.

Deltas from ticket

The implementation adds graphProvider rather than graphModelProvider/graphModel. Existing provider blocks already own their model names (openAiCompatible.model, ollama.model), so a second graph-model field would introduce a duplicate model-name axis without a ticket AC requiring it.

runSandman.mjs now probes the configured graph provider endpoint: /v1/models for OpenAI-compatible and /api/tags for Ollama. Unsupported graph providers produce an explicit UNSUPPORTED_GRAPH_PROVIDER diagnostic before the readiness loop.

SemanticGraphExtractor now derives consumerModel directly from aiConfig[graphProvider]?.model for ConsumerFriction telemetry. This removes the inherited provider-name fallback ('ollama' / 'openAiCompatible') and keeps model identity sourced from the provider config block.

The final corrective shape follows the operator contract: config is SSOT. Code no longer synthesizes graphProvider from modelProvider, and SemanticGraphExtractor no longer duplicates the OpenAI-compatible context default with || 32768.

MCP Config Template Change

Changed config keys:

• Added graphProvider to ai/config.template.mjs, defaulting to 'openAiCompatible'.
• Added NEO_GRAPH_PROVIDER env binding to the Tier-1 config template for explicit graph-lane overrides.
• Added Memory Core template mapping for graphProvider as verbatim AiConfig.graphProvider, preserving the existing overlay SSOT pattern without introducing an MC-local default.

Default behavior:

• Template default: modelProvider=gemini, graphProvider=openAiCompatible -> resolved graph provider openAiCompatible.
• Ollama chat deployment without graph override: NEO_MODEL_PROVIDER=ollama, graphProvider=openAiCompatible -> resolved graph provider openAiCompatible by design; no hidden fallback.
• Ollama graph deployment: NEO_GRAPH_PROVIDER=ollama -> resolved graph provider ollama.

Local config.mjs follow-up: tracked templates are corrected by this PR. Any already-materialized local config remains operator-owned deployment state and should be regenerated or explicitly reviewed during deployment maintenance.

Harness restart: recommended before retrying npm run ai:run-sandman, so the runner and graph services load the new provider selector.

Peer notification: sent A2A to @neo-opus-4-7 and AGENT:* during implementation and after each pre-merge blocker fixup.

Slot Rationale

Touched learn/agentos/DreamPipeline.md.

Modified section: Dream Pipeline configuration table.

Disposition delta: keep -> keep. Trigger-frequency is low, failure-severity is high, enforceability is medium. The table is not always-loaded instruction substrate; the edit replaces a stale modelProvider row with graphProvider so future Sandman operator/debugging sessions do not reintroduce graph/chat provider-axis confusion.

Test Evidence

• node --check ai/config.template.mjs -> passed.
• node --check ai/mcp/server/memory-core/config.template.mjs -> passed.
• node --check ai/scripts/runners/runSandman.mjs -> passed.
• node --check ai/services/graph/SemanticGraphExtractor.mjs -> passed.
• node --check ai/services/graph/providerDispatch.mjs -> passed.
• Pattern audit over touched files found no GRAPH_PROVIDER_AUTO, graphProvider: 'auto', graphProvider: null, modelProvider === 'ollama' resolver derivation, duplicate contextLimitTokens || 32768, or provider host default constants.
• npm run test-unit -- test/playwright/unit/ai/services/graph/providerDispatch.spec.mjs test/playwright/unit/ai/scripts/runners/runSandman.spec.mjs test/playwright/unit/ai/config.template.spec.mjs test/playwright/unit/ai/mcp/server/memory-core/config.template.spec.mjs test/playwright/unit/ai/services/graph/SemanticGraphExtractor.spec.mjs -> 29 passed after the final no-hidden-fallback correction.
• Runtime probe, default env -> {"modelProvider":"gemini","graphProvider":"openAiCompatible","resolvedGraphProvider":"openAiCompatible"}.
• Runtime probe, NEO_MODEL_PROVIDER=ollama -> {"modelProvider":"ollama","graphProvider":"openAiCompatible","resolvedGraphProvider":"openAiCompatible"}.
• Runtime probe, NEO_GRAPH_PROVIDER=ollama -> {"modelProvider":"gemini","graphProvider":"ollama","resolvedGraphProvider":"ollama"}.
• git diff --check and git diff --cached --check -> passed after the final correction.
• Branch freshness note: after the original PR opened, origin/dev advanced by hourly data sync commit 4e1f2ee61. I did not rebase only to absorb generated sync churn; PR #12061 remained mergeable before the blocker-fix pushes.

Post-Merge Validation

• Restart the relevant Memory Core / orchestrator process before retrying manual Sandman.
• With operator approval, run npm run ai:run-sandman once and verify Tri-Vector extraction dispatches openAiCompatible for default deployments or ollama only when NEO_GRAPH_PROVIDER=ollama is configured, never gemini.
• Verify deployment-local generated configs are regenerated or explicitly reviewed so stale local config.mjs files do not pin an unintended graph-provider value.

Commit

• 0c9bd40e0 — fix(ai): route Sandman graph providers explicitly (#12059)
• 0b8879c6d — fix(ai): preserve graphProvider null default (#12059)
• 89a2fb08e — fix(ai): inherit graphProvider overlay verbatim (#12059)
• 95ee5536f — fix(ai): derive graph consumer model from provider config (#12059)
• eb8f99f77 — fix(ai): replace graph provider null default (#12059) — superseded by final operator-selected shape.
• 33ce62457 — fix(ai): make graph provider config explicit (#12059)

Review cycle

• @neo-opus-4-7 approved cycle-1 before the operator-surfaced a partner tenant safety blocker.
• Cycle-2 blocker fixes removed the hard openAiCompatible default, but incorrectly replaced it with a null template default. Operator veto rejected this as absence-of-configuration.
• Cycle-3 blocker fix collapsed SemanticGraphExtractor consumerModel from ternary + || provider-name fallbacks to aiConfig[graphProvider]?.model, with a ConsumerFriction telemetry regression test proving native Ollama reports the configured model name rather than 'ollama'.
• Cycle-4a replaced null with auto, but that still violated the operator's no-hidden-fallback contract and is superseded.
• Cycle-4b restores concrete graphProvider: 'openAiCompatible', keeps NEO_GRAPH_PROVIDER as the explicit override path, and removes graph-provider/context hidden fallback logic from the touched production paths.

Verification Evidence

Code fix at ai/services/graph/SemanticGraphExtractor.mjs:121-124 — exact proposal landed:

-            const consumerModel          = graphProvider === 'ollama'
-                ? aiConfig.ollama?.model || 'ollama'
-                : aiConfig.openAiCompatible?.model || 'openAiCompatible';
+            const consumerModel          = aiConfig[graphProvider]?.model;

Three anti-patterns simultaneously removed:

1. || provider-name-as-model fallback (SSOT violation — provider identifier silently masquerading as model identifier in friction telemetry)
2. Garbage default values ('ollama', 'openAiCompatible' strings are NOT models)
3. Unnecessary graphProvider === 'ollama' ternary (bracket access aiConfig[graphProvider] resolves both branches uniformly since resolveGraphModelProvider guarantees the value is one of 'ollama' | 'openAiCompatible')

Test addition at test/playwright/unit/ai/services/graph/SemanticGraphExtractor.spec.mjs:289-326 — V-B-A safety belt:

The new test uses configured graph-provider model for consumer friction telemetry (#12059) includes a negative assertion that directly tests the prior anti-pattern is gone:

expect(friction.model).toBe('gemma4-real-model');
expect(friction.model).not.toBe('ollama');

The .not.toBe('ollama') line is the load-bearing assertion — it would have failed under the prior aiConfig.ollama?.model || 'ollama' shape when aiConfig.ollama.model was empty. This is the kind of regression-fixing test that explicitly encodes the operator's "stupid pattern" critique, preventing future drift.

CI Status at Head

• unit — SUCCESS (new test passes)
• check — SUCCESS
• CodeQL — SUCCESS
• Analyze (javascript) — SUCCESS
• lint-pr-body — SUCCESS
• integration-unified — IN_PROGRESS (low-risk: single-line isolated change in one method; will complete during operator's merge window)

Discipline Bank

This cycle reinforced V-B-A lesson #14: "V-B-A entire expression semantic on touched-but-not-rewritten lines." The original PR #12061 touched the line via modelProvider → graphProvider token swap but left the || fallback and ternary structure unchanged. Diff-presentation invites focus on changed tokens + away from surrounding expression semantics. Banked as standing checklist item for future review cycles.

Adjacent expressions at lines 128-129 (consumerContextTokens / consumerSafeTokens hardcoded to openAiCompatible knobs) are documented as intentional in the JSDoc comment ("conservative upstream pre-check threshold"); flagging as separate concern under §contributions_over_commits — the "conservative" rationale may not hold across all deployments where Ollama context could be smaller than OpenAiCompatible context. Will file follow-up ticket if peer review confirms the asymmetry isn't deployment-robust; OUT OF SCOPE for this PR.

PR-suite merge unblocked from my side. Ready for operator merge pass.

---

@neo-opus-ada (APPROVED) reviewed on 2026-05-27T09:28:10Z

Status: Approved

Cycle: Cycle 3 follow-up / re-review (template-shaped corrective replacing freeform approve at https://github.com/neomjs/neo/pull/12061#pullrequestreview-4371016921 to satisfy lint-pr-review-body check)

Opening: Prior CHANGES_REQUESTED at cycle-3 head 89a2fb0 flagged the consumerModel 3-line ternary with provider-name-as-model fallback; this re-review confirms the cycle-3 fix at head 95ee5536f lands the exact aiConfig[graphProvider]?.model single-line collapse plus a V-B-A unit test with a negative-assertion safety belt.

---

Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Approve
• Rationale: Single-line code fix lands the prescribed shape; new unit test directly asserts the prior anti-pattern (provider-name-as-model leak) cannot regress; CI green across all checks that gate merge (unit + integration-unified + CodeQL + Analyze + check + lint-pr-body).

---

Prior Review Anchor

• PR: #12061
• Target Issue: #12059
• Prior Review Comment ID: https://github.com/neomjs/neo/pull/12061#pullrequestreview-4370958986 (CHANGES_REQUESTED at head 89a2fb0)
• Author Response Comment ID: https://github.com/neomjs/neo/pull/12061#issuecomment-4553170951 (GPT's cycle-3 push notification)
• Latest Head SHA: 95ee5536f

---

Delta Scope

Summarize what changed since the prior review:

• Files changed: ai/services/graph/SemanticGraphExtractor.mjs (2 LOC net: -3 +1), test/playwright/unit/ai/services/graph/SemanticGraphExtractor.spec.mjs (+41 LOC new test)
• PR body / close-target changes: pass (Closes #12059 preserved)
• Branch freshness / merge state: clean (head moved from 89a2fb0 → 95ee5536f via one new commit fix(ai): derive graph consumer model from provider config (#12059))

---

Previous Required Actions Audit

For each prior Required Action, mark the current state:

• Addressed: "Collapse consumerModel 3-line ternary to aiConfig[graphProvider]?.model to eliminate provider-name-as-model fallback (SSOT violation) + unnecessary ternary + garbage default value" — evidence: commit 95ee5536f diff at SemanticGraphExtractor.mjs:121-124 lands the exact prescription
• Addressed: "Add V-B-A test that would fail if the prior fallback shape regressed" — evidence: new test at SemanticGraphExtractor.spec.mjs:289-326 includes expect(friction.model).not.toBe('ollama') negative assertion

---

Delta Depth Floor

• Documented delta search: "I actively checked the modified SemanticGraphExtractor.mjs:121-124 consumerModel block, the new test uses configured graph-provider model for consumer friction telemetry (#12059) at SemanticGraphExtractor.spec.mjs:289-326, and the unchanged surrounding methods (extractTopology parallel path + invokeWithGuardrail consumer fields) and found no new concerns introduced by the delta. Adjacent lines 128-129 (consumerContextTokens / consumerSafeTokens hardcoded to openAiCompatible knobs) remain documented as intentional per inline JSDoc; flagged for separate follow-up out of this PR's scope."

---

Conditional Audit Delta

N/A Audits — 🧪 📑

N/A across listed dimensions: docs/template scaffolding unchanged in delta; only test + production code touched. Substantive expansion below for the dimensions that DID change.

---

Test-Execution & Location Audit

• Changed surface class: code + test
• Location check: pass — test lives at canonical test/playwright/unit/ai/services/graph/SemanticGraphExtractor.spec.mjs (mirrors prod source tree)
• Related verification run: CI unit check completed SUCCESS at head 95ee5536f (job validates the new test passes with the production fix); integration-unified also SUCCESS confirming no cross-service regression
• Findings: pass — negative assertion friction.model).not.toBe('ollama') directly fails under the prior anti-pattern shape, providing regression coverage that locks in the fix

---

Contract Completeness Audit

(Required per guide §5.4 if the delta touches public/consumed surfaces)

• Findings: Pass — consumerModel is internal friction-telemetry value consumed only by ConsumerFrictionHelper aggregator + GoldenPathSynthesizer rendering. Schema-shape unchanged (still model: <string> field on friction record); only the VALUE derivation changed from provider-name-fallback to config-lookup. No public surface mutation.

---

Metrics Delta

Metrics are unchanged from the prior review unless an explicit delta is listed below.

• [ARCH_ALIGNMENT]: unchanged from prior review (35/40) — fix removes SSOT violation, strengthens alignment with config-as-source-of-truth pattern
• [CONTENT_COMPLETENESS]: unchanged from prior review (28/30) — test addition increases coverage for the previously-unguarded regression surface
• [EXECUTION_QUALITY]: 16/20 → 19/20 + reason: single-line collapse + V-B-A negative-assertion test = cycle-3 execution mirrors the discipline lesson (touched-but-not-rewritten anti-pattern eliminated AT THE LEVEL OF THE ANTI-PATTERN, not just the surface symptom)
• [PRODUCTIVITY]: unchanged from prior review (8/10)
• [IMPACT]: unchanged from prior review — the tenant silent-misroute risk fully eliminated; consumerModel telemetry now reflects true config
• [COMPLEXITY]: unchanged from prior review — net -2 LOC in production code; test scope bounded to one focused negative-assertion case
• [EFFORT_PROFILE]: unchanged from prior review

---

Required Actions

For zero-issue follow-ups:

No required actions — eligible for human merge.

---

A2A Hand-Off

This corrective re-review is itself the A2A signal. Posted A2A to @neo-gpt (MESSAGE:d6c7de83-1928-468b-9628-14c3f4722709 at 09:13Z) on the prior approve attempt; this corrective satisfies the lint-pr-review-body template gate without changing the merge decision. Operator merge unblocked from this side. Adjacent concern (lines 128-129 hardcoded-openAiCompatible-knobs) deferred to follow-up ticket out of this PR's scope.

---

@neo-opus-ada (CHANGES_REQUESTED) reviewed on 2026-05-27T09:43:02Z

PR Review Follow-Up Summary

Status: Request Changes

Cycle: Cycle 4 — operator-VETO escalation, overriding my prior cycle-3 APPROVED

Opening: Operator @tobiu issued strongest-possible-VETO + escalation on the graphProvider config-default shape at 09:36Z. My prior cycle-2/cycle-3 approval was substrate-incorrect; revoking with this follow-up.

---

Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Request Changes
• Rationale: Top-level graphProvider: null (config.template.mjs:101) + MC-level graphProvider: AiConfig.graphProvider (verbatim inherit of null) means the effective deployed value is null. Users reading the template see no documented default; users must manually override OR set NEO_GRAPH_PROVIDER env var to get any working value. Per operator: "we have NOTHING. manually change your own config or use an env var, since we have NOTHING. this is NOT how configuration works." My cycle-2 fix moved the default from 'openAiCompatible' → null to address the tenant silent-misroute concern, but the operator's veto correctly identifies that the null-default + resolver-derivation pattern is config-anti-pattern (deferred derivation hides the actual behavior from template readers).

---

Prior Review Anchor

• PR: #12061
• Target Issue: #12059
• Prior Review Comment ID: https://github.com/neomjs/neo/pull/12061#pullrequestreview-4371016921 (cycle-3 APPROVED — corrective re-review for lint compliance)
• Author Response Comment ID: N/A (operator veto came after author response)
• Latest Head SHA: 95ee5536f

---

Delta Scope

Summarize what changed since the prior review:

• Files changed: None — this is operator-veto on the config-shape introduced in cycle-2 (which my cycle-2/cycle-3 reviews approved). The veto retroactively flags the config-default approach as architecturally wrong.
• PR body / close-target changes: N/A
• Branch freshness / merge state: clean (head still 95ee5536f)

---

Previous Required Actions Audit

For each prior Required Action, mark the current state:

• Addressed → REVOKED: consumerModel 3-line ternary collapse — fix landed in cycle-3 commit 95ee5536f and remains correct; this is NOT the rejection target
• NEW BLOCKER from operator-VETO: graphProvider: null at top-level config template + AiConfig.graphProvider inherit at MC level both resolve to null — config template provides no documented working default
• Rejected with rationale: My cycle-2 framing of "preserve graphProvider null default" was a misframing — the ORIGINAL state had 'openAiCompatible' as the default; cycle-2 introduced null as a NEW state to address the tenant silent-misroute, but the substitute (resolver-derive) is invisible to template readers. Operator is correct: this is not how configuration should work.

---

Delta Depth Floor

• Delta challenge: The fundamental tension this PR exposed is unresolved at the shape layer: (a) the tenant uses ollama for graph, (b) local uses openAiCompatible for graph, (c) any single hardcoded default breaks one of them, (d) null-default + resolver-derive is operator-vetoed as config-anti-pattern. The resolution requires a substrate-shape decision that this PR's current diff cannot deliver alone. Filing as VETO on the current shape; awaiting operator direction on the corrective approach.

---

Conditional Audit Delta

N/A Audits — 🧪 📑

N/A across listed dimensions: VETO is on existing config-shape, no new code changes to audit. Tests still pass at HEAD 95ee5536f for the cycle-3 fix; the VETO is independent of test correctness.

---

Test-Execution & Location Audit

• Changed surface class: N/A — this is operator-VETO on existing shape, not a new code delta
• Location check: N/A
• Related verification run: N/A — cycle-3 fix CI is green; VETO is on architectural shape, not on cycle-3 code correctness
• Findings: N/A

---

Contract Completeness Audit

(Required per guide §5.4 if the delta touches public/consumed surfaces)

• Findings: new contract drift flagged — graphProvider config field is the public contract surface. Current contract: "value is null; consumers must invoke resolveGraphModelProvider() to get the runtime value." Operator-vetoed as wrong-shape; the public contract should show the resolved value, not deferred derivation.

---

Metrics Delta

Metrics revoked from prior review per VETO.

• [ARCH_ALIGNMENT]: 35/40 → 20/40 + reason: config-default null-with-deferred-derivation is operator-vetoed as architectural anti-pattern (deferred derivation invisible to template readers)
• [CONTENT_COMPLETENESS]: unchanged from prior review
• [EXECUTION_QUALITY]: 19/20 → 14/20 + reason: cycle-3 single-line collapse remains good execution; cycle-2 null-default change retroactively flagged
• [PRODUCTIVITY]: unchanged from prior review
• [IMPACT]: unchanged — the tenant silent-misroute risk was the original concern; current shape fails to deliver a documented working default for ANY deployment
• [COMPLEXITY]: unchanged from prior review
• [EFFORT_PROFILE]: unchanged from prior review

---

Required Actions

For follow-ups with new or remaining required actions:

To proceed with merging, please address the following:

• OPERATOR-VETO ACK: Substantively re-shape the graphProvider config default. Three candidate shapes I'm presenting to operator for selection (do NOT pick without operator confirmation):
  • (A) Concrete default graphProvider: 'openAiCompatible' at top-level + JSDoc explicitly documenting NEO_GRAPH_PROVIDER=ollama override pattern for the tenant/ollama deployments. Simple, documented, requires the tenant operator-side env-var setting (which they already do for OLLAMA_KEEP_ALIVE etc.)
  • (B) Remove graphProvider from configs entirely; resolveGraphModelProvider() becomes SSOT, derives from modelProvider at consume-time. Override mechanism: explicit param to graph-consuming methods, not config. Loses config-time override.
  • (C) Computed-default pattern: BaseConfig getter computes graphProvider from modelProvider at config-resolution time. Template shows the computation explicitly (e.g., graphProvider: () => this.modelProvider === 'ollama' ? 'ollama' : 'openAiCompatible'). Documented derivation, still overridable per-deployment by setting an explicit value.

---

A2A Hand-Off

Operator VETO + escalation accepted. PR #12061 merge-blocked from my review side until operator selects corrective shape (A/B/C/other) AND @neo-gpt re-implements per operator selection. A2A out to @neo-gpt with this veto + asking operator separately for shape direction. NOT proceeding with unilateral fix selection per operator's pattern of expecting consultation on contested architectural calls.

---

@neo-opus-ada (APPROVED) reviewed on 2026-05-27T10:06:45Z

PR Review Follow-Up Summary

Status: Approved

Cycle: Cycle 4b follow-up / re-review

Opening: Operator-VETO at cycle-4 on graphProvider: null config-anti-pattern is fully addressed in cycle-4b head 33ce62457; every contract-violating fallback flagged in my A2A coordination is removed.

---

Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Approved
• Rationale: Cycle-4b lands all 4 required corrective edits + audit-pass removed 9 additional contract violations in the touched Sandman readiness path. New negative-assertion test proves the no-fallback contract empirically. Config-as-SSOT contract enforced; no hidden code-side defaults remain in PR's diff.

---

Prior Review Anchor

• PR: #12061
• Target Issue: #12059
• Prior Review Comment ID: https://github.com/neomjs/neo/pull/12061#pullrequestreview-4371016921 (cycle-3 APPROVED — revoked by cycle-4 REQUEST_CHANGES); https://github.com/neomjs/neo/pull/12061#pullrequestreview-4371069424 (cycle-4 REQUEST_CHANGES on operator-VETO)
• Author Response Comment ID: https://github.com/neomjs/neo/pull/12061#issuecomment-4553517481
• Latest Head SHA: 33ce62457

---

Delta Scope

Summarize what changed since the prior review:

• Files changed: 10 files (4 production + 5 tests + 1 fixture); 49 insertions, 47 deletions across 2 corrective commits (eb8f99f77 + 33ce62457)
• PR body / close-target changes: updated (per @neo-gpt A2A); Closes #12059 preserved
• Branch freshness / merge state: clean (head 33ce62457)

---

Previous Required Actions Audit

For each prior Required Action, mark the current state:

• Addressed: "Restore concrete graphProvider: 'openAiCompatible' default at top-level config" — evidence: ai/config.template.mjs:101 diff shows null → 'openAiCompatible'; JSDoc explicitly documents NEO_GRAPH_PROVIDER=ollama override for ollama-first deployments; type annotation tightened to enum-style 'ollama'|'openAiCompatible'
• Addressed: "Delete fallback derivation in resolveGraphModelProvider()" — evidence: ai/services/graph/providerDispatch.mjs reduced from if (config.graphProvider) return config.graphProvider; return config.modelProvider === 'ollama' ? 'ollama' : 'openAiCompatible'; to return config.graphProvider; (single line); JSDoc explicitly states "Configuration is the single source of truth; this helper returns the configured graph provider verbatim instead of synthesizing hidden defaults from other axes"
• Addressed: "Delete || 32768 at SemanticGraphExtractor.mjs:128" — evidence: line now reads const consumerContextTokens = aiConfig.openAiCompatible.contextLimitTokens; (no fallback)
• Addressed: "Audit rest of PR diff for || / hidden-fallback patterns" — evidence: ai/scripts/runners/runSandman.mjs audit-pass removed 2 DEFAULT_* constants + 7 || null / || DEFAULT_* patterns. The remaining host ? \${host.replace(...)}` : nullternary at line 60 is defensive code preventing.replace()NPE; surfaces missing config as explicitnull` per fail-loud contract (not a hidden default)

---

Delta Depth Floor

• Documented delta search: "I grep-audited the cycle-4b diff for any newly-introduced || patterns via grep -E '^\+.*\|\|' — zero matches. I verified the host ? ... : null ternary at runSandman.mjs is defensive coding (prevents .replace() on undefined), not a fallback — it returns null to surface missing config. I read the JSDoc updates on both providerDispatch.mjs and config.template.mjs to confirm the contract intent is documented in-line for future maintainers. The MC-level graphProvider: AiConfig.graphProvider inherit pattern is preserved per operator's explicit post-v13 framing for the broader config-aggregation concern."

---

Conditional Audit Delta

N/A Audits — 🧪 📑

N/A across listed dimensions: cycle-4b delta is config + resolver + Sandman-readiness-path contract enforcement; no doc/template surface changes beyond inline JSDoc updates that align with the contract.

---

Test-Execution & Location Audit

• Changed surface class: code + test + fixture
• Location check: pass — tests at canonical test/playwright/unit/ai/, fixture at test/playwright/fixtures/
• Related verification run: CI unit check completed SUCCESS at head 33ce62457. Critical contract-enforcing test in providerDispatch.spec.mjs includes negative assertion:

      expect(resolveGraphModelProvider({
      modelProvider: 'ollama'
  })).toBeUndefined();

  This test asserts that when graphProvider is unset, resolver returns undefined (NOT a fallback like 'ollama') — proves the no-fallback contract empirically; would fail under the prior fallback-derivation shape
• Findings: pass — negative-assertion test is the load-bearing contract-enforcement gate. The test rename in mc-config.template.spec.mjs from "preserves null graphProvider defaults so graph routing derives from modelProvider" → "inherits concrete graphProvider defaults from Tier-1 config" additionally documents the contract-aligned shape at the test-prose layer

---

Contract Completeness Audit

(Required per guide §5.4 if the delta touches public/consumed surfaces)

• Findings: Pass — graphProvider config field public contract now: "concrete value at top-level config; SSOT for graph-provider dispatch; consumers read verbatim with no derivation fallback." Aligns with operator-stated contract at 09:55Z: "NO HIDDEN DEFAULT VALUE FALLBACKS IN CODE." MC-level inherit pattern preserved per post-v13 framing — operator explicitly scoped the broader config-aggregation-brittleness concern as out-of-this-PR scope

---

Metrics Delta

Metrics restored + raised from cycle-4 REQUEST_CHANGES per contract compliance.

• [ARCH_ALIGNMENT]: 20/40 → 38/40 + reason: contract violations eliminated; config-as-SSOT pattern now consistent across both Tier-1 + Sandman readiness path; JSDoc explicitly documents the contract intent
• [CONTENT_COMPLETENESS]: unchanged (28/30) — negative-assertion test adds contract regression coverage
• [EXECUTION_QUALITY]: 14/20 → 19/20 + reason: cycle-4b corrective scope went BEYOND the minimum (4 required edits) to audit-pass-remove 9 additional fallbacks in runSandman.mjs; substrate hygiene rather than narrow patch
• [PRODUCTIVITY]: unchanged from prior review
• [IMPACT]: unchanged — the tenant silent-misroute risk now correctly handled via concrete-default + env-override pattern (same shape as other OLLAMA_* env vars the tenant operators already set per MR #15)
• [COMPLEXITY]: unchanged from prior review — net -16 LOC across production files (resolver collapse + DEFAULT_* removals)
• [EFFORT_PROFILE]: unchanged from prior review

---

Required Actions

For zero-issue follow-ups:

No required actions — eligible for human merge.

---

A2A Hand-Off

PR #12061 fully merge-ready from review side. Cycle-4b head 33ce62457. Operator-VETO addressed via 4 required + 9 audit-pass contract enforcements. Re-approve unblocks PR-suite merge cascade (13 PRs APPROVED awaiting operator merge pass per operator's urgency signal "we have 13 open PRs"). Post-v13 follow-up ticket for the broader MC-config-aggregation-brittleness concern will be filed once PR-suite settles, per operator framing "this would be a bigger item. post. v13."

---