Ollama Pro plan is live (39-model fleet on the same OLLAMA_CLOUD_KEY) and OpenCode Zen was already wired in the gateway but every gpt-oss call site was still pinned to gpt-oss:120b. This PR routes every hot-path call to faster / stronger replacements.
frontier reasoning matters here — fires only after local self-correct fails twice; Zen pay-per-token cost is bounded
Verification
cargo check -p gateway --tests clean
Live probes through localhost:3100/v1/chat:
opencode/claude-opus-4-7 → pong
gemini-3-flash-preview (ollama_cloud) → pong
kimi-k2.6 (ollama_cloud) → pong
deepseek-v3.2 (ollama_cloud) → Pong! 🏓
Notes
kimi-k2:1t still upstream-broken (HTTP 500 on Ollama Pro probe today, matches yesterday memory). Replacement table never picks it.
The Rust execution-loop change requires systemctl restart lakehouse.service to take effect on the running gateway. TS callers reload on next run.
aibridge/src/context.rs still has gpt-oss:{20b,120b} in its window-size lookup table; harmless and kept for callers that pass it explicitly as an override.
Test plan
Auditor cycles within 90s and emits kimi_architect verdict
After systemctl restart lakehouse.service, trigger an overseer escalation (e.g. force a reviewer reject loop) and confirm Langfuse shows provider:opencode model:claude-opus-4-7
Run scrum on a small file and confirm tree-split MAP/REDUCE land via gemini-3-flash-preview
Ollama Pro plan is live (39-model fleet on the same `OLLAMA_CLOUD_KEY`) and OpenCode Zen was already wired in the gateway but every gpt-oss call site was still pinned to `gpt-oss:120b`. This PR routes every hot-path call to faster / stronger replacements.
## Replacement table
| Site | gpt-oss → replacement | Why |
|---|---|---|
| `config/providers.toml` `ollama_cloud` default | `gpt-oss:120b` → `deepseek-v3.2` | newest DeepSeek revision; live-probed `pong` |
| `config/providers.toml` `openrouter` default | `openai/gpt-oss-120b:free` → `x-ai/grok-4.1-fast` | already the scrum LADDER PRIMARY |
| `config/modes.toml` `staffing_inference` | `openai/gpt-oss-120b:free` → `kimi-k2.6` | coding-specialized, on Ollama Pro |
| `config/modes.toml` `doc_drift_check` | `gpt-oss:120b` → `gemini-3-flash-preview` | speed leader for factual checks |
| `tests/real-world/scrum_master_pipeline.ts` tree-split MAP+REDUCE | `gpt-oss:120b` → `gemini-3-flash-preview` | latency-dominated path (5-20× per file) |
| `bot/propose.ts` `CLOUD_MODEL` | `gpt-oss:120b` → `deepseek-v3.2` | same Ollama key, faster |
| `mcp-server/observer.ts` overseer label fallback | `gpt-oss:120b` → `claude-opus-4-7` | matches new overseer model |
| `crates/gateway/src/execution_loop/mod.rs` overseer escalation | `ollama_cloud/gpt-oss:120b` → `opencode/claude-opus-4-7` | frontier reasoning matters here — fires only after local self-correct fails twice; Zen pay-per-token cost is bounded |
## Verification
- `cargo check -p gateway --tests` clean
- Live probes through `localhost:3100/v1/chat`:
- `opencode/claude-opus-4-7` → `pong`
- `gemini-3-flash-preview` (ollama_cloud) → `pong`
- `kimi-k2.6` (ollama_cloud) → `pong`
- `deepseek-v3.2` (ollama_cloud) → `Pong! 🏓`
## Notes
- `kimi-k2:1t` still upstream-broken (HTTP 500 on Ollama Pro probe today, matches yesterday memory). Replacement table never picks it.
- The Rust execution-loop change requires `systemctl restart lakehouse.service` to take effect on the running gateway. TS callers reload on next run.
- `aibridge/src/context.rs` still has `gpt-oss:{20b,120b}` in its window-size lookup table; harmless and kept for callers that pass it explicitly as an override.
## Test plan
- [ ] Auditor cycles within 90s and emits kimi_architect verdict
- [ ] After `systemctl restart lakehouse.service`, trigger an overseer escalation (e.g. force a reviewer reject loop) and confirm Langfuse shows `provider:opencode model:claude-opus-4-7`
- [ ] Run scrum on a small file and confirm tree-split MAP/REDUCE land via `gemini-3-flash-preview`
- [ ] `/v1/usage` reflects new model names
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Ollama Pro plan went live today (39-model fleet on the same
OLLAMA_CLOUD_KEY) and OpenCode Zen was already wired in the gateway
but not consumed. Routing every gpt-oss call site to faster /
stronger replacements:
| Site | gpt-oss → replacement | Why |
|---|---|---|
| ollama_cloud default | gpt-oss:120b → deepseek-v3.2 | newest DeepSeek revision; live-probed `pong` |
| openrouter default | openai/gpt-oss-120b:free → x-ai/grok-4.1-fast | already the scrum LADDER's PRIMARY |
| modes.toml staffing_inference | openai/gpt-oss-120b:free → kimi-k2.6 | coding-specialized, on Ollama Pro |
| modes.toml doc_drift_check | gpt-oss:120b → gemini-3-flash-preview | speed leader for factual checks |
| scrum_master_pipeline tree-split MAP+REDUCE | gpt-oss:120b → gemini-3-flash-preview | latency-dominated path (5-20× per file) |
| bot/propose.ts CLOUD_MODEL | gpt-oss:120b → deepseek-v3.2 | same Ollama key, faster |
| mcp-server/observer.ts overseer label fallback | gpt-oss:120b → claude-opus-4-7 | matches new overseer model |
| crates/gateway/src/execution_loop overseer escalation | ollama_cloud/gpt-oss:120b → opencode/claude-opus-4-7 | frontier reasoning matters here — fires only after local self-correct fails twice; Zen pay-per-token cost is bounded |
Verification:
- `cargo check -p gateway --tests` — clean
- Live probes through localhost:3100/v1/chat:
- `opencode/claude-opus-4-7` → "pong"
- `gemini-3-flash-preview` (ollama_cloud) → "pong"
- `kimi-k2.6` (ollama_cloud) → "pong"
- `deepseek-v3.2` (ollama_cloud) → "Pong! 🏓"
Notes:
- kimi-k2:1t still upstream-broken (HTTP 500 on Ollama Pro probe today,
matches yesterday's memory). Replacement table never picks it.
- The Rust changes need a `systemctl restart lakehouse.service` to
take effect on the running gateway. TS callers reload on next run.
- aibridge/src/context.rs still has gpt-oss:{20b,120b} in its window-
size lookup table; harmless and kept for callers that pass it
explicitly as an override.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
One-liner: 2 blocking issues: State field rename likely incomplete — opencode_key may not exist on self.state Head SHA:a00e9bb438ac Audited at: 2026-04-28T11:15:02.698Z
dynamic — 1 findings (0 block, 0 warn, 1 info)
ℹ️info — dynamic check skipped — skipped by options
enrichment: 0 bug fingerprints, 0 answers-corpus chunks, prompt avg 12083 chars ⚠️warn — cloud: claim not backed — "OLLAMA_CLOUD_KEY) and OpenCode Zen was already wired in the gateway"
recent failing sigs: none ⚠️warn — scrum-master review for mcp-server/observer.ts — accepted on attempt 7 by openrouter/qwen/qwen3-235b-a22b-2507 (tree-split)
reviewed_at: 2026-04-26T23:27:44.907Z
preview: ```json { "verdict": "fail", "critical_failures": [ { "id": "CF-1", "file": "mcp-server/observer.ts:137", "description": "Execution memory writes ℹ️info — scrum-master review for crates/gateway/src/execution_loop/mod.rs — accepted on attempt 1 by ollama_cloud/kimi-k2:1t (tree-split)
reviewed_at: 2026-04-24T08:21:18.916Z
preview: I'll conduct a forensic audit of this execution loop module against the PRD and change proposal requirements. Given this is a tree-split distillation, I'll only flag contr ℹ️info — core entity sql_policy_check recurs in 2 PRs (types: Function)
count=4 distinct_PRs=2
description: A function that is missing in the code.
description: A function imported from 'node:fs/promises' for writing files
PRs: 8,9 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'error' added in crates/gateway/src/v1/respond.rs but no read-site in the diff — could be placeholder state without a consumer
signature=33f475cd7a5c255e
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'rules' added in crates/gateway/src/v1/truth.rs but no read-site in the diff — could be placeholder state without a consumer
signature=0868549b1fe15baf
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 63 flaggings, conf=0.02): field 'id' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=99ac8803cd15b1a5
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 63 flaggings, conf=0.02): field 'description' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=3e2ed94d1ba06de3
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'playbook_boost_ceiling' added in crates/shared/src/profiles/memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=6fd1860f42d0fa85
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'history_cap' added in crates/shared/src/profiles/memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=066143c1baeb3f7e
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'doc_stale_hours' added in crates/shared/src/profiles/memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=3b5f4b16a3c8b00a
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'auto_retire_on_failure' added in crates/shared/src/profiles/memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=e1cef15c3f560811
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 63 flaggings, conf=0.02): field 'created_at' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=d249d21feebd04ce
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 63 flaggings, conf=0.02): field 'created_by' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=d57b721418f3f088
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'failure_cluster_size' added in crates/shared/src/profiles/observer.rs but no read-site in the diff — could be placeholder state without a consumer
signature=c55cac2e690e8fe6
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'alert_cooldown_secs' added in crates/shared/src/profiles/observer.rs but no read-site in the diff — could be placeholder state without a consumer
signature=b3ddb9a4ec8ab6cb
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'ring_size' added in crates/shared/src/profiles/observer.rs but no read-site in the diff — could be placeholder state without a consumer
signature=ee7e11933fe9a19f
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'forward_to_langfuse' added in crates/shared/src/profiles/observer.rs but no read-site in the diff — could be placeholder state without a consumer
signature=a264dd9c9f22abc2
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'top_k' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=a3e62e18dd4e59b1
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'rerank_top_k' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=94874c682dda210d
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'freshness_cutoff_days' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=99249e0a441ba21c
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'boost_playbook_memory' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=d48c35bffe3863ea
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'enforce_sensitivity_gates' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=886b7e5d93e502b2
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'findings' added in crates/validator/src/lib.rs but no read-site in the diff — could be placeholder state without a consumer
signature=103ad6b6d830a23c
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'elapsed_ms' added in crates/validator/src/lib.rs but no read-site in the diff — could be placeholder state without a consumer
signature=50c24a7a40e41ad7
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'severity' added in crates/validator/src/lib.rs but no read-site in the diff — could be placeholder state without a consumer
signature=f386e2477d18f7d9
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'message' added in crates/validator/src/lib.rs but no read-site in the diff — could be placeholder state without a consumer
scopes: pr-11,pr-12 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'priors' added in crates/vectord/src/pathway_memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=5f5b36d952caf591
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'prior_iter_outcomes' added in crates/vectord/src/pathway_memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=1fdb4d320bcdf203
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'result_summary' added in crates/vectord/src/pathway_memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=5f84c6b830b3ff12
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'parent_version' added in crates/vectord/src/pathway_memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=78f07c2e84a56db6
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'new_version' added in crates/vectord/src/pathway_memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=3631c337796c04a3
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'total_replays' added in crates/vectord/src/pathway_memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=f1426e133dc92ba0
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'successful_replays' added in crates/vectord/src/pathway_memory.rs but no read-site in the diff — could be placeholder state without a consumer
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 36 flaggings, conf=0.03): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "Total: 37 truth tests green (was 32). Workspace warnings still 0."
signature=b0235a992c0216ae
checks: kb_query
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 20 flaggings, conf=0.05): field 'reviewer_model' added in crates/gateway/src/v1/respond.rs but no read-site in the diff — could be placeholder state without a consumer
signature=cee27b199d23457f
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 20 flaggings, conf=0.05): field 'status' added in crates/gateway/src/v1/respond.rs but no read-site in the diff — could be placeholder state without a consumer
signature=278f466e54dac13a
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 16 flaggings, conf=0.06): field 'prompt' added in crates/gateway/src/v1/mode.rs but no read-site in the diff — could be placeholder state without a consumer
signature=240451351e1b8cef
checks: static
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 16 flaggings, conf=0.06): field 'notes' added in crates/gateway/src/v1/mode.rs but no read-site in the diff — could be placeholder state without a consumer
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 9 flaggings, conf=0.11): cloud: claim not backed — "Archon-on-Lakehouse integration: gateway side is done. Pi-ai's"
signature=5ef622471680e6a8
checks: inference
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 9 flaggings, conf=0.11): cloud: claim not backed — "sending. Patching Pi locally to override baseUrl works for arch but the"
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 7 flaggings, conf=0.14): cloud: claim not backed — "9 unit tests, all green. Live probe on real shape correctly drops"
scopes: pr-11 ℹ️info — recurring audit pattern (1 distinct PRs, 12 flaggings, conf=0.08): cloud: claim not backed — "commit is the rest of the auditor rebuild it landed alongside)"
signature=bde2b46a5c181555
checks: inference
scopes: pr-11 ℹ️info — recurring audit pattern (2 distinct PRs, 17 flaggings, conf=0.12): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "Replaces the narrow proven-approaches preamble with a unified"
signature=803df5936ad8ebf5
checks: kb_query
scopes: pr-11,pr-12 ℹ️info — recurring audit pattern (1 distinct PRs, 13 flaggings, conf=0.08): cloud: claim not backed — "because scored-runs were deduped against. Fixed by reading"
🛑block — State field rename likely incomplete — opencode_key may not exist on self.state
crates/gateway/src/execution_loop/mod.rs:596
The diff changes self.state.ollama_cloud_keytoself.state.opencode_keyat line 596, but no corresponding struct definition change is shown in the diff. IfGatewayState/Ex`
[grounding: verified at crates/gateway/src/execution_loop/mod.rs:596] ⚠️warn — Hardcoded model string drift between config and code — deepseek-v3.2 vs deepseek-v3.2 naming consistency
config/providers.toml:30
The providers.toml sets default_model = "deepseek-v3.2"but bot/propose.ts usesdeepseek-v3.2as theLH_BOT_MODELfallback. However, the comment in providers.toml mentionsd`
[grounding: verified at config/providers.toml:30] ⚠️warn — Telemetry schema bypass — provider field changed from "ollama_cloud" to "opencode" but Langfuse trace may have provider enum constraints
crates/gateway/src/execution_loop/mod.rs:645
The ChatTracestruct'sproviderfield is set to"opencode"at line 645, but iflangfuse_trace.rsdefines this as an enum with known variants,"opencode" may fail serializ
[grounding: verified at crates/gateway/src/execution_loop/mod.rs:645] ⚠️warn — Silent determinism leak — think: Some(true) preserved for Anthropic model with comment claiming it's "handled in the adapter"
crates/gateway/src/execution_loop/mod.rs:618
The comment at line 617-618 states Anthropic models "reject think(handled in the adapter)" but the code still sendsthink: Some(true). If the adapter's "handling" is to strip
[grounding: verified at crates/gateway/src/execution_loop/mod.rs:618] 🛑block — Load-bearing assumption — opencode::chat module existence not verified in diff
crates/gateway/src/execution_loop/mod.rs:623
The diff changes crate::v1::ollama_cloud::chattocrate::v1::opencode::chatat line 623, but noopencode.rs module creation is shown. If this module doesn't exist or has a di
[grounding: verified at crates/gateway/src/execution_loop/mod.rs:623] ⚠️warn — Default model string in observer.ts doesn't match any config source of truth
mcp-server/observer.ts:772
The fallback string "claude-opus-4-7" at line 772 is hardcoded in observer.ts but doesn't appear in any config file shown in the diff. If the overseer model changes again, this t
[grounding: verified at mcp-server/observer.ts:772] ⚠️warn — Test file uses provider: "ollama_cloud" with model: "gemini-3-flash-preview" — potential provider/model mismatch
tests/real-world/scrum_master_pipeline.ts:1148
At line 1148, the test calls chat({ provider: "ollama_cloud", model: "gemini-3-flash-preview", ... }). The providers.toml comment states the 39-model fleet "includes... gemini-3-
[grounding: verified at tests/real-world/scrum_master_pipeline.ts:1148] ⚠️warn — max_tokens: None for overseer with pay-per-token provider — unbounded cost exposure
crates/gateway/src/execution_loop/mod.rs:615
The chat_reqat line 615 setsmax_tokens: None while the comment at line 605-608 explains the Zen tier is "pay-per-token" and "cost stays bounded" due to low frequency. However
[grounding: verified at crates/gateway/src/execution_loop/mod.rs:615] ℹ️info — OPENROUTER_API_KEY fallback files may not exist in containerized environments
config/providers.toml:45
The auth_fallback_fileslist includes/home/profit/.envand/root/llm_team_config.json. In containerized or CI environments where $HOME differs or these paths are unavailab
Auditor surfaced 2 BLOCKs from kimi_architect; both verified FALSE POSITIVES on 2026-05-02:
opencode_key field rename — kimi flagged self.state.opencode_key as possibly nonexistent. Verified: field exists on GatewayState, multiple call sites already use it on main.
opencode::chat module existence — kimi flagged the module as undeclared in diff. Verified: crates/gateway/src/v1/opencode.rs exists on main, declared at crates/gateway/src/v1/mod.rs:20.
Both flags are the kimi-only-BLOCK pattern called out in feedback_cross_lineage_review.md (only Opus emits trustworthy BLOCKs). Verified by isolating infra/replace-gpt-oss-2026-04-28 in a worktree + running cargo check -p gateway → clean (1 pre-existing unused-var warning, unrelated).
Mergeable=true, compiles, replacement models all live-probed. Merging.
Auditor surfaced 2 BLOCKs from kimi_architect; both verified FALSE POSITIVES on 2026-05-02:
1. **`opencode_key` field rename** — kimi flagged `self.state.opencode_key` as possibly nonexistent. Verified: field exists on `GatewayState`, multiple call sites already use it on `main`.
2. **`opencode::chat` module existence** — kimi flagged the module as undeclared in diff. Verified: `crates/gateway/src/v1/opencode.rs` exists on `main`, declared at `crates/gateway/src/v1/mod.rs:20`.
Both flags are the kimi-only-BLOCK pattern called out in `feedback_cross_lineage_review.md` (only Opus emits trustworthy BLOCKs). Verified by isolating `infra/replace-gpt-oss-2026-04-28` in a worktree + running `cargo check -p gateway` → clean (1 pre-existing unused-var warning, unrelated).
Mergeable=true, compiles, replacement models all live-probed. Merging.
profit
merged commit feb638e4cd into main2026-05-03 03:39:54 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Ollama Pro plan is live (39-model fleet on the same
OLLAMA_CLOUD_KEY) and OpenCode Zen was already wired in the gateway but every gpt-oss call site was still pinned togpt-oss:120b. This PR routes every hot-path call to faster / stronger replacements.Replacement table
config/providers.tomlollama_clouddefaultgpt-oss:120b→deepseek-v3.2pongconfig/providers.tomlopenrouterdefaultopenai/gpt-oss-120b:free→x-ai/grok-4.1-fastconfig/modes.tomlstaffing_inferenceopenai/gpt-oss-120b:free→kimi-k2.6config/modes.tomldoc_drift_checkgpt-oss:120b→gemini-3-flash-previewtests/real-world/scrum_master_pipeline.tstree-split MAP+REDUCEgpt-oss:120b→gemini-3-flash-previewbot/propose.tsCLOUD_MODELgpt-oss:120b→deepseek-v3.2mcp-server/observer.tsoverseer label fallbackgpt-oss:120b→claude-opus-4-7crates/gateway/src/execution_loop/mod.rsoverseer escalationollama_cloud/gpt-oss:120b→opencode/claude-opus-4-7Verification
cargo check -p gateway --testscleanlocalhost:3100/v1/chat:opencode/claude-opus-4-7→ponggemini-3-flash-preview(ollama_cloud) →pongkimi-k2.6(ollama_cloud) →pongdeepseek-v3.2(ollama_cloud) →Pong! 🏓Notes
kimi-k2:1tstill upstream-broken (HTTP 500 on Ollama Pro probe today, matches yesterday memory). Replacement table never picks it.systemctl restart lakehouse.serviceto take effect on the running gateway. TS callers reload on next run.aibridge/src/context.rsstill hasgpt-oss:{20b,120b}in its window-size lookup table; harmless and kept for callers that pass it explicitly as an override.Test plan
systemctl restart lakehouse.service, trigger an overseer escalation (e.g. force a reviewer reject loop) and confirm Langfuse showsprovider:opencode model:claude-opus-4-7gemini-3-flash-preview/v1/usagereflects new model names🤖 Generated with Claude Code
Auditor verdict: 🛑
blockOne-liner: 2 blocking issues: State field rename likely incomplete —
opencode_keymay not exist onself.stateHead SHA:
a00e9bb438acAudited at: 2026-04-28T11:15:02.698Z
dynamic — 1 findings (0 block, 0 warn, 1 info)
ℹ️ info — dynamic check skipped — skipped by options
skipped by optionsinference — 3 findings (0 block, 2 warn, 1 info)
ℹ️ info — pr_audit mode runner completed (model=deepseek-v3.1:671b, consensus=3/3, 4620ms wall-clock)
claims voted: 2parsed runs: 3 / 3enrichment: 0 bug fingerprints, 0 answers-corpus chunks, prompt avg 12083 chars⚠️ warn — cloud: claim not backed — "OLLAMA_CLOUD_KEY) and OpenCode Zen was already wired in the gateway"
at commit:a00e9bb4:4consensus: 3/3 not-backed (resolution: majority_not_backed)cloud reason: diff shows OpenCode added in this PR, not pre-wired⚠️ warn — cloud-flagged gap not in any claim: ?
location: ?kb_query — 132 findings (0 block, 15 warn, 117 info)
ℹ️ info — KB: 101 recent scenario runs, 210/291 events ok (fail rate 27.8%)
most recent: ?recent failing sigs: none⚠️ warn — scrum-master review for
mcp-server/observer.ts— accepted on attempt 7 byopenrouter/qwen/qwen3-235b-a22b-2507(tree-split)reviewed_at: 2026-04-26T23:27:44.907Zpreview: ```json { "verdict": "fail", "critical_failures": [ { "id": "CF-1", "file": "mcp-server/observer.ts:137", "description": "Execution memory writesℹ️ info — scrum-master review for
crates/gateway/src/execution_loop/mod.rs— accepted on attempt 1 byollama_cloud/kimi-k2:1t(tree-split)reviewed_at: 2026-04-24T08:21:18.916Zpreview: I'll conduct a forensic audit of this execution loop module against the PRD and change proposal requirements. Given this is a tree-split distillation, I'll only flag contrℹ️ info — core entity
sql_policy_checkrecurs in 2 PRs (types: Function)count=4 distinct_PRs=2description: A function that is missing in the code.PRs: 0,11ℹ️ info — core entity
mkdirrecurs in 2 PRs (types: Function)count=3 distinct_PRs=2description: A function imported from 'node:fs/promises' for creating directoriesPRs: 8,9ℹ️ info — core entity
gatewayrecurs in 2 PRs (types: Constant,System Component,Software Component)count=3 distinct_PRs=2description: A component in the system responsible for handling initial access and communication.PRs: 0,8ℹ️ info — core entity
execute_queryrecurs in 2 PRs (types: Function,function)count=3 distinct_PRs=2description:PRs: 0,11ℹ️ info — core entity
writeFilerecurs in 2 PRs (types: Function)count=2 distinct_PRs=2description: A function imported from 'node:fs/promises' for writing filesPRs: 8,9ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'error' added in crates/gateway/src/v1/respond.rs but no read-site in the diff — could be placeholder state without a consumer
signature=33f475cd7a5c255echecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'rules' added in crates/gateway/src/v1/truth.rs but no read-site in the diff — could be placeholder state without a consumer
signature=0868549b1fe15bafchecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 63 flaggings, conf=0.02): field 'id' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=99ac8803cd15b1a5checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 63 flaggings, conf=0.02): field 'description' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=3e2ed94d1ba06de3checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'playbook_boost_ceiling' added in crates/shared/src/profiles/memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=6fd1860f42d0fa85checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'history_cap' added in crates/shared/src/profiles/memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=066143c1baeb3f7echecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'doc_stale_hours' added in crates/shared/src/profiles/memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=3b5f4b16a3c8b00achecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'auto_retire_on_failure' added in crates/shared/src/profiles/memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=e1cef15c3f560811checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 63 flaggings, conf=0.02): field 'created_at' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=d249d21feebd04cechecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 63 flaggings, conf=0.02): field 'created_by' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=d57b721418f3f088checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'failure_cluster_size' added in crates/shared/src/profiles/observer.rs but no read-site in the diff — could be placeholder state without a consumer
signature=c55cac2e690e8fe6checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'alert_cooldown_secs' added in crates/shared/src/profiles/observer.rs but no read-site in the diff — could be placeholder state without a consumer
signature=b3ddb9a4ec8ab6cbchecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'ring_size' added in crates/shared/src/profiles/observer.rs but no read-site in the diff — could be placeholder state without a consumer
signature=ee7e11933fe9a19fchecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'forward_to_langfuse' added in crates/shared/src/profiles/observer.rs but no read-site in the diff — could be placeholder state without a consumer
signature=a264dd9c9f22abc2checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'top_k' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=a3e62e18dd4e59b1checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'rerank_top_k' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=94874c682dda210dchecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'freshness_cutoff_days' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=99249e0a441ba21cchecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'boost_playbook_memory' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=d48c35bffe3863eachecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'enforce_sensitivity_gates' added in crates/shared/src/profiles/retrieval.rs but no read-site in the diff — could be placeholder state without a consumer
signature=886b7e5d93e502b2checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'findings' added in crates/validator/src/lib.rs but no read-site in the diff — could be placeholder state without a consumer
signature=103ad6b6d830a23cchecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'elapsed_ms' added in crates/validator/src/lib.rs but no read-site in the diff — could be placeholder state without a consumer
signature=50c24a7a40e41ad7checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'severity' added in crates/validator/src/lib.rs but no read-site in the diff — could be placeholder state without a consumer
signature=f386e2477d18f7d9checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'message' added in crates/validator/src/lib.rs but no read-site in the diff — could be placeholder state without a consumer
signature=c8843a8a24d7b00achecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (2 distinct PRs, 54 flaggings, conf=0.04): TODO/FIXME/XXX/HACK comment added in mcp-server/role_scenes.ts
signature=c74394177c67a49dchecks: staticscopes: pr-11,pr-12ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'priors' added in crates/vectord/src/pathway_memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=5f5b36d952caf591checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'prior_iter_outcomes' added in crates/vectord/src/pathway_memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=1fdb4d320bcdf203checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'result_summary' added in crates/vectord/src/pathway_memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=5f84c6b830b3ff12checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'parent_version' added in crates/vectord/src/pathway_memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=78f07c2e84a56db6checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'new_version' added in crates/vectord/src/pathway_memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=3631c337796c04a3checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'total_replays' added in crates/vectord/src/pathway_memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=f1426e133dc92ba0checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): field 'successful_replays' added in crates/vectord/src/pathway_memory.rs but no read-site in the diff — could be placeholder state without a consumer
signature=51da88b79b07925bchecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 21 flaggings, conf=0.05): todo!() macro call in tests/real-world/scrum_master_pipeline.ts
signature=1d64852d2b0a95c0checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 36 flaggings, conf=0.03): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "Total: 37 truth tests green (was 32). Workspace warnings still 0."
signature=b0235a992c0216aechecks: kb_queryscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 20 flaggings, conf=0.05): field 'reviewer_model' added in crates/gateway/src/v1/respond.rs but no read-site in the diff — could be placeholder state without a consumer
signature=cee27b199d23457fchecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 20 flaggings, conf=0.05): field 'status' added in crates/gateway/src/v1/respond.rs but no read-site in the diff — could be placeholder state without a consumer
signature=278f466e54dac13achecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 16 flaggings, conf=0.06): field 'prompt' added in crates/gateway/src/v1/mode.rs but no read-site in the diff — could be placeholder state without a consumer
signature=240451351e1b8cefchecks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 16 flaggings, conf=0.06): field 'notes' added in crates/gateway/src/v1/mode.rs but no read-site in the diff — could be placeholder state without a consumer
signature=b69738e705604f02checks: staticscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 19 flaggings, conf=0.05): cloud: claim not backed — "|
tests/real-world/scrum_master_pipeline.ts| 9-rung ladder (Ollama Cloud tier + OpenRouter rescue"signature=edbd5e67a0387680checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 19 flaggings, conf=0.05): cloud: claim not backed — "- iter 4 (post fixes): 4.00/10 (+0.44 — fixes landed on the 3 files that were touched)"
signature=07975499ecb4dbaachecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 19 flaggings, conf=0.05): cloud: claim not backed — "Auto-applier dry-run: 5 eligible files → 1 green commit-ready, 2 build-red reverts (auto-reverte"
signature=396c0aecd5527dccchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 19 flaggings, conf=0.05): cloud: claim not backed — "- Applier dry-run file-write bug — currently writes files even in dry-run mode, should revert after "
signature=21a58eacd406823cchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 13 flaggings, conf=0.08): cloud: claim not backed — "docs: SCRUM_MASTER_SPEC timeline — productization wave + verified live state"
signature=69c0a41f864a8f71checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 12 flaggings, conf=0.08): cloud: claim not backed — "fixed in
107a682)."signature=a8f0c146f272e49fchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 12 flaggings, conf=0.08): cloud: claim not backed — "new response/request struct shipped through
/v1/*was flagged as"signature=6e86ff852cb12613checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 13 flaggings, conf=0.08): cloud: claim not backed — "Verified live: kb_preamble grew 416 → 727 chars after wiring third"
signature=510660f02f9f0bbbchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 10 flaggings, conf=0.10): cloud: claim not backed — "Verified end-to-end: Pi CLI (
pi --print --provider openrouter)"signature=9d031dcf7ca12a86checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 10 flaggings, conf=0.10): cloud: claim not backed — "landed a clean 1902-token request through
/v1/chat/completions,"signature=5c212551c59da7a6checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 39 flaggings, conf=0.03): cloud: claim not backed — "Verified end-to-end:"
signature=8a8db031b06a6e73checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 9 flaggings, conf=0.11): cloud: claim not backed — "Archon-on-Lakehouse integration: gateway side is done. Pi-ai's"
signature=5ef622471680e6a8checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 9 flaggings, conf=0.11): cloud: claim not backed — "sending. Patching Pi locally to override baseUrl works for arch but the"
signature=bae54948bedaad89checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 9 flaggings, conf=0.11): cloud: claim not backed — "SDK path (langchain-js / agents / patched Pi) works today."
signature=6e2a531a69c2d96echecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 7 flaggings, conf=0.14): cloud: claim not backed — "wired in this phase."
signature=05a943a19b1faf8dchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 7 flaggings, conf=0.14): cloud: claim not backed — "default, force_mode override works + validates, bogus modes return"
signature=436acd155eb1a1efchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 6 flaggings, conf=0.17): cloud: claim not backed — "- what landed today (observer fix, HANDOVER.md render, relevance filter)"
signature=2d214359173d8a9bchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 7 flaggings, conf=0.14): cloud: claim not backed — "9 unit tests, all green. Live probe on real shape correctly drops"
signature=c1d8f1ed53857de9checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 5 flaggings, conf=0.20): cloud: claim not backed — "contact_1; reported_cost is integer-string. Fixed filter (was rejecting"
signature=192a0a8fd2ef6466checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 5 flaggings, conf=0.20): cloud: claim not backed — "Verified live (4/4 ops):"
signature=7488287a5b8dae2echecks: inferencescopes: pr-11⚠️ warn — recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "Replaces the narrow proven-approaches preamble with a unified"
signature=653816a556f59908checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 2 flaggings, conf=0.50): cloud: claim not backed — "to openrouter/x-ai/grok-4.1-fast — proven to emit precise semantic"
signature=867a82fc9a8cddc8checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 2 flaggings, conf=0.50): cloud: claim not backed — "What landed:"
signature=fdedfb3591acccbechecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 2 flaggings, conf=0.50): cloud: claim not backed — "Verified end-to-end: smoke loop iter 1 — observer returned
cycleon"signature=65b1ebd75e1e9560checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 12 flaggings, conf=0.08): cloud: claim not backed — "9 schemas + foundation types + 48 tests passing in 502ms:"
signature=b6a0b605c0c43adfchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "Invariants enforced (proven by tests + real run):"
signature=bbf22f2213563a24checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "- Failure propagation: any stage validation.passed=false → overall_passed=false"
signature=4d9a026e152a0eb2checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "[x] drift detection works (severity ok|warn|alert)"
signature=d4eee21dd8099cc1checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "Bug surfaced + fixed during testing: module-level evidenceCache"
signature=bc26dee12aaf6ec7checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "BUG SURFACED + FIXED:"
signature=e537ffe25b037bdbchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "surface — synthetic tests passed before the run, real data"
signature=eca2e6fd2a54798cchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "has source_file + line_offset + sig_hash + recorded_at. Proven."
signature=a6a47725b125e835checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "Invariants enforced (proven by tests + real-data audit):"
signature=3cfcf147c85d37bechecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "- Idempotent: second run on identical input writes 0 rows (proven on"
signature=06f3761914b49df2checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "- Bit-stable: identical input produces byte-identical output (proven"
signature=73800b46527b8362checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "Phase 2 done-criteria all met:"
signature=9ac4e9c2aa8300fbchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "✓ Provenance round-trip proven on real sampled rows"
signature=004eec535740ee24checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 13 flaggings, conf=0.08): cloud: claim not backed — "Both runs passed the structural validation gate (length, no hedges,"
signature=d09d0c3898a9ca4fchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 13 flaggings, conf=0.08): cloud: claim not backed — "- Retrieval is jaccard keyword. Works at 446 corpus, scale via"
signature=c653622096fa3a12checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "probes; multi-hour outage). deepseek is the proven drop-in from"
signature=a45738cd08b2edd4checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 12 flaggings, conf=0.08): cloud: claim not backed — "- serde-derived struct exemption (commit
107a682shipped this; this"signature=7537eb77270d7bd4checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 12 flaggings, conf=0.08): cloud: claim not backed — "commit is the rest of the auditor rebuild it landed alongside)"
signature=bde2b46a5c181555checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (2 distinct PRs, 17 flaggings, conf=0.12): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "Replaces the narrow proven-approaches preamble with a unified"
signature=803df5936ad8ebf5checks: kb_queryscopes: pr-11,pr-12ℹ️ info — recurring audit pattern (1 distinct PRs, 13 flaggings, conf=0.08): cloud: claim not backed — "because scored-runs were deduped against. Fixed by reading"
signature=33b25ee5b60df8ffchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 13 flaggings, conf=0.08): cloud: claim not backed — "Phase 8 done-criteria (per spec):"
signature=5b6fb5cbd5d3e476checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 12 flaggings, conf=0.08): cloud: claim not backed — "The Phase 43 scaffolds (FillValidator, EmailValidator) shipped with"
signature=8364565c3e958055checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 12 flaggings, conf=0.08): cloud: claim not backed — "Pre-existing issue (NOT from this change, NOT fixed here):"
signature=45ccbf946cc4c6e0checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 12 flaggings, conf=0.08): cloud: claim not backed — "6 missing fields.
cargo check --workspace(production) is green;"signature=a0a5b3a346e7ff2achecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 12 flaggings, conf=0.08): cloud: claim not backed — "cargo check --workspace green"
signature=e3ede3f4dd10e6fcchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 12 flaggings, conf=0.08): cloud: claim not backed — "coding endpoint is text-only; passing a [{type,text},...] array"
signature=2e448df73d047858checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 12 flaggings, conf=0.08): cloud: claim not backed — "- appendMetrics join with "..": works for current path; deferred."
signature=a5290686a5f8e175checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 12 flaggings, conf=0.08): cloud: claim not backed — "verified by grep. 2 of 10 actionable bugs landed. The other 8 were"
signature=e42b45244e926181checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "The kimi.rs adapter (provider=kimi) stays wired as a fallback when"
signature=93d264d27b1458d8checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 12 flaggings, conf=0.08): cloud: claim not backed — "Verification (all green):"
signature=717c35314b69dd00checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 27 flaggings, conf=0.04): cloud: claim not backed — "LANDED:"
signature=bafae3bec6618043checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "cargo check -p aibridge green"
signature=f86bca6368049083checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 11 flaggings, conf=0.09): cloud: claim not backed — "cargo build --release -p gateway green"
signature=36b58ab7f2d3d4a4checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 9 flaggings, conf=0.11): cloud: claim not backed — "LH_AUDITOR_KIMI_MAX_TOKENS env override still works (forces value"
signature=06e8b0edcf74d87echecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 2 flaggings, conf=0.50): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "The kimi.rs adapter (provider=kimi) stays wired as a fallback when"
signature=dbddb1ed194d8797checks: kb_queryscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 8 flaggings, conf=0.13): cloud: claim not backed — "in comments; the shipped regex requires
fetch(...)or"signature=5477158c3900b392checks: inferencescopes: pr-11⚠️ warn — recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "Tests: 27 pathway_memory tests green (was 18). New tests:"
signature=34382324bf7c2a55checks: kb_queryscopes: pr-11⚠️ warn — recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "proven review pathways."
signature=edb960def1660954checks: kb_queryscopes: pr-11⚠️ warn — recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "not wired yet; probation + success_rate gates alone enforce safety"
signature=9709ea20bf9668d1checks: kb_queryscopes: pr-11⚠️ warn — recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "PathwayMemoryStats. 18/18 tests green."
signature=44b4f9260f0abe84checks: kb_queryscopes: pr-11⚠️ warn — recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "Team extract results (fields wired, callers not yet)"
signature=1dea433cb02225d5checks: kb_queryscopes: pr-11⚠️ warn — recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "Total: 28 truth tests green (was 24), 7 new queryd policy tests gre
signature=28a52763ffbbad41checks: kb_queryscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 5 flaggings, conf=0.20): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "s
signature=7a04d1b89c00775achecks: kb_queryscopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 6 flaggings, conf=0.17): cloud: claim not backed — "Verified live (current synthetic data):"
signature=784c3977a0ac9158checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (2 distinct PRs, 13 flaggings, conf=0.15): cloud: claim not backed — "Verified live:"
signature=7965ec7d350eac32checks: inferencescopes: pr-11,pr-12ℹ️ info — recurring audit pattern (1 distinct PRs, 6 flaggings, conf=0.17): cloud: claim not backed — "→ 2 not configured (claude, gemini — keys not wired)"
signature=4d287c2223dfa02fchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 6 flaggings, conf=0.17): cloud: claim not backed — "docs: PHASES tracker — mark Phases 42/43/44/45 complete"
signature=c93d122e037f690dchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 6 flaggings, conf=0.17): cloud: claim not backed — "already shipped in prior sessions: DocRef struct, doc_drift module,"
signature=b07f0e2e03242f00checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 6 flaggings, conf=0.17): cloud: claim not backed — "Verified end-to-end with kimi-k2.6 via ollama_cloud:"
signature=34f2f34a4e910deachecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 6 flaggings, conf=0.17): cloud: claim not backed — "validation) and converge on validation-passing artifacts in 1-2"
signature=358fcc5514ec6edfchecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 6 flaggings, conf=0.17): cloud: claim not backed — "Verified live (gateway + 500K worker snapshot):"
signature=2402a57ca395b1bachecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 6 flaggings, conf=0.17): cloud: claim not backed — "Green'")"
signature=c04797074f5ddef4checks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 5 flaggings, conf=0.20): cloud: claim not backed — "Pre-existing manual reset still works (state edit + daemon"
signature=ada2afb623940b0achecks: inferencescopes: pr-11ℹ️ info — recurring audit pattern (1 distinct PRs, 5 flaggings, conf=0.20): cloud: claim not backed — "Verified compile (bun build auditor/index.ts → green). Daemon"
signature=9a5144617ca8e3dbchecks: inferencescopes: pr-11⚠️ warn — recurring audit pattern (2 distinct PRs, 4 flaggings, conf=0.50): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "T
signature=db2e3357f7c64ebachecks: kb_queryscopes: pr-11,pr-12⚠️ warn — recurring audit pattern (2 distinct PRs, 5 flaggings, conf=0.40): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "p
signature=56eae12a688ef046checks: kb_queryscopes: pr-11,pr-12⚠️ warn — recurring audit pattern (2 distinct PRs, 5 flaggings, conf=0.40): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "n
signature=786001b0aa188f3echecks: kb_queryscopes: pr-11,pr-12⚠️ warn — recurring audit pattern (2 distinct PRs, 5 flaggings, conf=0.40): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "P
signature=0fad08efa8aeb9ffchecks: kb_queryscopes: pr-11,pr-12⚠️ warn — recurring audit pattern (2 distinct PRs, 5 flaggings, conf=0.40): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "T
signature=959bf54aa5a6f4f1checks: kb_queryscopes: pr-11,pr-12⚠️ warn — recurring audit pattern (2 distinct PRs, 5 flaggings, conf=0.40): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): cloud: claim not backed — "T
signature=33c112ea77f3c532checks: kb_queryscopes: pr-11,pr-12⚠️ warn — recurring audit pattern (2 distinct PRs, 3 flaggings, conf=0.67): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 distinct PRs, 3 flaggings, conf=0.33): recurring audit pattern (1 d
signature=a13abf1d9c7baf17checks: kb_queryscopes: pr-11,pr-12ℹ️ info — recurring audit pattern (1 distinct PRs, 2 flaggings, conf=0.50): cloud: claim not backed — "Verified end-to-end via playwright on devop.live/lakehouse:"
signature=93c83d5251ba2048checks: inferencescopes: pr-12ℹ️ info — recurring audit pattern (1 distinct PRs, 2 flaggings, conf=0.50): cloud: claim not backed — "Verified end-to-end on devop.live/lakehouse/profiler:"
signature=d177549dfa18accbchecks: inferencescopes: pr-12kimi_architect — 9 findings (2 block, 6 warn, 1 info)
🛑 block — State field rename likely incomplete —
opencode_keymay not exist onself.statecrates/gateway/src/execution_loop/mod.rs:596The diff changesself.state.ollama_cloud_keytoself.state.opencode_keyat line 596, but no corresponding struct definition change is shown in the diff. IfGatewayState/Ex`[grounding: verified at crates/gateway/src/execution_loop/mod.rs:596]⚠️ warn — Hardcoded model string drift between config and code —
deepseek-v3.2vsdeepseek-v3.2naming consistencyconfig/providers.toml:30The providers.toml setsdefault_model = "deepseek-v3.2"but bot/propose.ts usesdeepseek-v3.2as theLH_BOT_MODELfallback. However, the comment in providers.toml mentionsd`[grounding: verified at config/providers.toml:30]⚠️ warn — Telemetry schema bypass —
providerfield changed from"ollama_cloud"to"opencode"but Langfuse trace may have provider enum constraintscrates/gateway/src/execution_loop/mod.rs:645TheChatTracestruct'sproviderfield is set to"opencode"at line 645, but iflangfuse_trace.rsdefines this as an enum with known variants,"opencode"may fail serializ[grounding: verified at crates/gateway/src/execution_loop/mod.rs:645]⚠️ warn — Silent determinism leak —
think: Some(true)preserved for Anthropic model with comment claiming it's "handled in the adapter"crates/gateway/src/execution_loop/mod.rs:618The comment at line 617-618 states Anthropic models "rejectthink(handled in the adapter)" but the code still sendsthink: Some(true). If the adapter's "handling" is to strip[grounding: verified at crates/gateway/src/execution_loop/mod.rs:618]🛑 block — Load-bearing assumption —
opencode::chatmodule existence not verified in diffcrates/gateway/src/execution_loop/mod.rs:623The diff changescrate::v1::ollama_cloud::chattocrate::v1::opencode::chatat line 623, but noopencode.rsmodule creation is shown. If this module doesn't exist or has a di[grounding: verified at crates/gateway/src/execution_loop/mod.rs:623]⚠️ warn — Default model string in observer.ts doesn't match any config source of truth
mcp-server/observer.ts:772The fallback string"claude-opus-4-7"at line 772 is hardcoded in observer.ts but doesn't appear in any config file shown in the diff. If the overseer model changes again, this t[grounding: verified at mcp-server/observer.ts:772]⚠️ warn — Test file uses
provider: "ollama_cloud"withmodel: "gemini-3-flash-preview"— potential provider/model mismatchtests/real-world/scrum_master_pipeline.ts:1148At line 1148, the test callschat({ provider: "ollama_cloud", model: "gemini-3-flash-preview", ... }). The providers.toml comment states the 39-model fleet "includes... gemini-3-[grounding: verified at tests/real-world/scrum_master_pipeline.ts:1148]⚠️ warn —
max_tokens: Nonefor overseer with pay-per-token provider — unbounded cost exposurecrates/gateway/src/execution_loop/mod.rs:615Thechat_reqat line 615 setsmax_tokens: Nonewhile the comment at line 605-608 explains the Zen tier is "pay-per-token" and "cost stays bounded" due to low frequency. However[grounding: verified at crates/gateway/src/execution_loop/mod.rs:615]ℹ️ info —
OPENROUTER_API_KEYfallback files may not exist in containerized environmentsconfig/providers.toml:45Theauth_fallback_fileslist includes/home/profit/.envand/root/llm_team_config.json. In containerized or CI environments where$HOMEdiffers or these paths are unavailab[grounding: verified at config/providers.toml:45]Metrics
Lakehouse auditor · SHA
a00e9bb4· re-audit on new commit flips the status automatically.Auditor surfaced 2 BLOCKs from kimi_architect; both verified FALSE POSITIVES on 2026-05-02:
opencode_keyfield rename — kimi flaggedself.state.opencode_keyas possibly nonexistent. Verified: field exists onGatewayState, multiple call sites already use it onmain.opencode::chatmodule existence — kimi flagged the module as undeclared in diff. Verified:crates/gateway/src/v1/opencode.rsexists onmain, declared atcrates/gateway/src/v1/mod.rs:20.Both flags are the kimi-only-BLOCK pattern called out in
feedback_cross_lineage_review.md(only Opus emits trustworthy BLOCKs). Verified by isolatinginfra/replace-gpt-oss-2026-04-28in a worktree + runningcargo check -p gateway→ clean (1 pre-existing unused-var warning, unrelated).Mergeable=true, compiles, replacement models all live-probed. Merging.