# Engine change-scoped analysis (G2 workstream) Status: **proposed / in progress** (opened 2026-10-11). Owner: engine. Tracks a concrete lever for the engine-detection gap found while driving the G0 baseline. This is G2 ("improve what moves the unseen-repo numbers"), gated on the pool metric; it is NOT part of the G0 record-07 baseline (that runs on the quick-rule tier). ## Problem (measured, not assumed) On real unseen vulnerability changes the engine produces **no finding** — but the diagnosis (live, 300 s long-deadline replays) shows it is usually **not a detection-capability gap**: - **Clean Python injection change** (`aws/bedrock-agentcore-sdk-python`, `install_packages`, introducing, 3-file diff): `wall=303 s`, hit the deadline. The ranker found candidates (`static_rank=1.0`) but **`selected: []`** and every candidate `deferred` with `reason: deadline_exhausted`, `analysis_not_started=deadline_exhausted`, `change_context_incomplete`. **Zero analyses ran.** And the ranked candidates were **repo-wide and unrelated to the push** (`auth.py` access_control/config_secrets), not the changed `install_packages`. The engine is trying to re-audit the whole repo and runs out of budget before analyzing the change. - **Noisy last_touch config_secrets change** (`jenkinsci/metrics-plugin`, 14-file diff): finished in 206 s but `comparison_unknown` + `capability_unavailable` + `eligibility_provenance_mismatch` + `contributor-scan ... projection_unavailable`. Multi-causal; correspondence was one part (fixed separately, commit c39fbb8), capability registry is Task 8.3. ## Hypothesis A pre-push hook should vet **the code this push changed**, not re-scan the whole repository. Seeding/scoping the engine's analysis to the **changed declarations and their dataflow neighborhood** (the `change_sites` are already in the replay spec) should let analysis **actually run within budget** instead of being deferred — which is where detection is being lost, not in Joern's capability. ## Anchors (confirm in increment 1) - `src/openultrasast/rank.py` — the static-evidence ranker (`ranking_mode="evidence"`, `existing-static-evidence-v1`). - `src/openultrasast/mapping.py` — builds the `selected`/`deferred` scope + the budget allocation that defers on `deadline_exhausted`. - `src/openultrasast/push/policy.py`, `push/runner.py` — consume the scope; change_sites arrive via the push input. - `benchmarks/unseen/replay_entry.py` — carries `changed_paths` / `declaration_paths` in the spec. ## Increments (progress tracking) - [x] **0. Cost split confirmed (2026-10-11, timed live replay).** `build_seconds=66.69` (CPG build is cheap), `query_seconds=0.0` (no query ever executed), total 303 s (hit the deadline), **4,095** candidate questions all `analysis_not_started=deadline_exhausted`, `selected: []`, `regions_scanned=0`. So of the 303 s, ~67 s built the CPG and ~236 s went to enumerating/ranking/preparing 4,095 repo-wide candidate questions for a **3-file** change — the budget was spent before a single query ran. The CPG build is NOT the bottleneck; the over-broad candidate set is. Change-scoping is confirmed as the right lever, and because per-candidate prep includes the line-correspondence work, fewer candidates also cuts that cost (the two fixes compound). - [x] **1. Located (2026-10-11).** The over-enumeration is upstream of the scan, not inside it: - `push/runner.py:345-348` — `head_regions = discovery(tip.root, …)` discovers regions across the **whole head tree** (the ~4,095), independent of the change. - `push/runner.py:397-407` — those regions are passed wholesale to `scan_repository(..., change_context=context)`; `change_context` is threaded in but **never scopes the region set**. - `model/scan.py:167` and `:364` — `[_identity(unit, r, f) for r in regions for f in r.families]` enumerates all regions × families; with `order_by_evidence=True` (`runner.py:288/403`) the scan evidence-tiers **every** pair before analysis — that pre-work is the ~236 s measured in increment 0, so the deadline hits at `query_seconds=0`. - Primitive already present: `model/regions.py:279` computes the changed-path set from the context; `compare_targeted_base` (`runner.py:411`) is a precedent for change-targeted work. - **Fix locus (increment 2):** restrict the *analyzed* region set to the change's declarations + dataflow neighbourhood (filter `head_regions` by `context.changed_paths` before the scan, or scope the enumeration inside `scan_repository`). Keep the **CPG whole-repo** so taint queries can still trace sources from unchanged code into the changed regions — only the question set is change-scoped. **Correction (2026-10-11), after a reverted first attempt — two things the naive filter got wrong:** - **The engine already reaches into unchanged code by design, and it is tested.** A changed `source` that reaches an unchanged `sink` must still select the unchanged sink's question (`tests/test_model_change_context.py::test_changed_dependency_reaches_unchanged_ranked_question`; `test_change_does_not_replace_ranker_selection` pins that change context must NOT alter the ranker's selection). `_changed_files` feeds `_evidence_pass` only as `contextFilter`/`contextPaths` — which locations to itemise vs summarise — NOT to shrink the tier population. A "changed files only" scope deletes the reach-into-unchanged behaviour. Any scope must keep changed regions **plus the regions they reach** (the affected/`relationships` neighbourhood), not changed files alone. - **The 236 s is now split (2026-10-11, saved artifacts).** `total=291s build=52s query=0s arbitrate=238.75s`, and `query_seconds_by_kind["evidence"]=238.54s` — the evidence-tier pass is **99.9 %** of the arbitrate phase; `affected_context` + `_scope_work` are **0.21 s**. The pass tiered 1,015 pairs (`tier_counts 0:1009, 1:6`) at **~235 ms/pair** (contended; ~120 ms uncontended) and that alone blew the budget — `selected=0`, `deferred=1365` all `deadline_exhausted`. Secondary anomaly: the design assumes ~2.5 ms/pair ("17,700 in 44 s"); here it is 50–100× that, i.e. the evidence query is pathologically slow on a cold large CPG — a separate optimisation from cutting the pair count. - **So increment 2's single target is the evidence pass** (`scan.py:338-359`, `_evidence_pass`). The fork is HOW to cut it without deleting reach-into-unchanged: - (i) **Scope + bound:** order the evidence candidates change-first and bound the pass to a fraction of the budget so analysis queries still run; reached unchanged regions are tiered only while budget lasts. - (ii) **Neighbourhood pass:** tier the changed regions, read the reached regions from the context rows the pass already returns, tier those in a second small pass — exact reach-into-unchanged, more code. - (iii) **Per-pair speedup:** attack the ~235 ms/pair (warm server / batching) rather than the pair count — helps every scan but is a deeper engine change and does not by itself bound a huge repo. **CORRECTION (2026-10-11, after implementing (i) + four live re-verifies — the cost model was wrong).** Scope + bound was built and tested (change-first ordering, adaptive budget-fitted portioning, `_context_pass` scoped to changed files, honest `evidence_pass_bounded` degradation; branch `engine-change-scoped-evidence-pass`). It correctly cut the tiered pair count — but `query_seconds` stayed **0** every time. The live numbers show why: | pairs tiered | run_batch calls | evidence s | build s | |---|---|---|---| | 1015 | 1 | 238 | 52 | | 512 | 1 | 219 | 68 | | 64 | 2 | 220 | 68 | | 4 (probe) | 1 | 160 | 127 | Evidence time is **flat regardless of pair count** → the cost is **per-*call* fixed** (~150–220 s: loading the CPG into the query engine + the census the evidenceOnly batch computes), ~85 ms/pair on top. Pair-count scoping therefore cannot free budget, and portioning re-pays the fixed cost per call (so (i) can be counterproductive). Build is also large and swings 52→127 s — two engines thrashing the 7 GB host (the record-07 baseline shares it). The real levers are now (iii)-shaped, not (i): **warm/persistent query engine or a cached census** to kill the ~150 s per-call fixed cost, and/or a **bigger engine-arm budget** than 300 s; plus a **clean (uncontended) measurement**, because contention alone may be inflating build + fixed cost 2–3×. Scope+bound is NOT merged. **UNCONTENDED measurement (2026-10-11, baseline paused via SIGSTOP + container kill, then resumed).** Shipped engine, one evidence call over 1,015 pairs: `build=48 s`, `evidence=243.79 s`, `query=0`. The evidence pass is **the same ~244 s uncontended as ~238 s contended** → the cost is **INTRINSIC, not contention**; contention only inflated `build` (48 → 127 s) and added run-to-run noise. The model is `evidence ≈ 150 s fixed (CPG load into the query engine + the census the evidenceOnly batch computes) + ~93 ms/pair` (244 = 150 + 1015×0.093). Consequences: - **~200 s of fixed overhead** (`build` + evidence-load) precedes any analysis, so a 300 s budget leaves only ~100 s for everything else. This is why `query=0` even on a clean host. - **Change-scoping still helps, but only by cutting the per-pair tail** (~1,015×93 ms ≈ 95 s): one evidence call over ~30 changed pairs is ~150 s + 3 s instead of 244 s, which would free ~90 s — possibly enough for analysis, but marginal. - **It must be ONE call, not portioned.** The ~150 s is a first-call load, so the adaptive probe attributed it to its 4 probe pairs (per-pair≈37 s), concluded the budget fit ~0 more, and tiered only 4 — the implemented scope+bound is doubly wrong for this cost model. The correct shape: change-first candidates, count-capped, ONE evidence call, `deadline=None`. - **Pre-push viability:** the ~200 s fixed floor means the engine arm cannot be a fast (≤30 s) pre-push check on a repo this size without a warm/persistent session or a cached census. A larger budget (e.g. 600 s) makes the benchmark arm run but is not a realistic pre-push deadline. The lever that matters for the product is killing the ~150 s fixed cost, not scoping pairs. - [ ] **2. Seed analysis from the change.** Prioritize (or restrict first-pass to) the changed declarations plus their call/data neighborhood over repo-wide static rank, within budget. - [ ] **3. Soundness.** Keep cross-file dataflow INTO the changed function (don't scope to literal changed lines); add tests; keep existing push tests green. - [ ] **4. Live re-test.** Re-run the bedrock change: does analysis now RUN and produce a finding on `install_packages` (vs `deferred`)? Re-run metrics-plugin. - [ ] **5. Measure the delta.** Engine-arm sample on the pool (catch/false-alarm + engine-coverage rate) — needs the kube lane (blocked on the AX completion path) for scale. ## Dependencies / notes - Coverage `complete_within_scope` also needs the capability registry (Task 8.3) regardless of this work; the win here is findings moving from `comparison_unknown`/deferred to a decided novelty → advisory engine finding. - Implement directly and verify on the LIVE instrument (a fake-client/unit pass is not evidence the live engine changed); the kube-lane and store endpoints are operator-owned infra, not assumptions to make here. ## Efficiency barrier analysis (post-record-07 priority) **What the record-07 baseline actually measured (verified on the records): regex only.** The default and long_deadline replay arms both hardcode `model="disabled"`, `max_model_calls=0`, `cost_usd=0` — **no LLM is called** — and the engine never completed (`engine_covered=0`, the ~200 s wall). So the measured `catch≈2.5%` is the **quick pattern-rule tier alone**: essentially grep-with-structure. This is the FLOOR, not the product, and it confirms the maintainer's point — pattern rules alone will never be enough, or `grep` would do. The product's detection lives in two tiers that did NOT run: | Tier | Mechanism | Measured cost (uncontended, mid-size repo) | Ran? | Catch | |---|---|---|---|---| | 1 quick rules | regex/pattern over changed lines | ~seconds | yes | 2.5% (floor) | | 2 engine (Joern) | CPG build + taint/dataflow | build 48 s + evidence (150 s fixed + 93 ms/pair) + analysis (unmeasured) | no (walled) | unknown | | 3 LLM adjudication | model judgment / search-with-proof (hunter, fusion, decision engine) | model latency + $/call (unmeasured in the push path) | no (disabled) | unknown | **The barrier:** a releasable net needs tier 2 and/or tier 3 to run and catch within a realistic pre-push window — target **≤60–120 s** (developer patience), hard ceiling 300 s. The analysis is a budget ledger per tier against that window, and the measurements each needs: - **Engine barrier (tier 2).** Fixed floor `build(48) + evidence_fixed(~150) ≈ 198 s` precedes any dataflow answer. Measure (clean host): **E1** split the ~150 s into JVM-start / graph-load / census; **E2** does a warm session with the graph **preloaded at build** eliminate `evidence_fixed` on the first query (the decisive architectural test)?; **E3** analysis (`reachableByFlows`) per-pair cost + warm reuse; **E4** build cost vs repo size. Levers in priority order: warm session + graph preload → cached census → change-scope the per-pair tail. - **LLM barrier (tier 3).** Unmeasured in the push path. Measure: **L1** model round-trip latency per call (DeepSeek) and calls needed per push for the decision/proof stage; **L2** the recall LIFT of tier 3 over tiers 1+2 on the pool — does the LLM catch what regex and dataflow miss, and at what $ and latency per push? - **Deliverable:** a **barrier table** — for budget ∈ {60, 120, 300, 600} s × each tier combination, does it FIT and what is the MEASURED recall/false-alarm on the pool — identifying the minimal viable config for a releasable net. 600 s unblocks the benchmark arm but is not a pre-push deadline; the product target is tiers 2+3 under ~120 s. Sequence: finish G0 (record 07) → **E1/E2 first** (they decide whether the warm session alone crosses the engine barrier) → E3/E4 → L1/L2. Each on a clean host (the session times out contended). The change-scoped branch (`engine-change-scoped-evidence-pass`, unmerged) only trims the tier-2 per-pair tail and is a late add-on, not the lever. ## E1/E2 RESULTS (measured 2026-10-12, CLEAN host, bedrock-agentcore, 3-file change, 300 s budget, head side) Four clean-host replays settle the engine barrier and **correct the earlier "~200 s graph-load wall"**, which was conflated with host contention from the running baseline replay: | Run | build | evidence (taint batch) | dominance | query_s | selected | deferred | |---|---|---|---|---|---|---| | contended (prior) | 48–127 | 243.8 | — | **0** | **0** | 1365 | | disposable (shipped, clean) | 27.9 | 225.1 | 34.8 | 34.8 | **356** | 1009 | | warm session `SESSION=1` (clean) | 32.9 | 228.6 | 25.9 | 25.9 | **356** | 1009 | | session + graph preload (clean) | 30.4 | 223.2 | 32.5 | 32.5 | **356** | 1009 | **Findings (each traced to a measured cause, not assumed):** 1. **The engine arm was walled by CONTENTION, not an intrinsic floor.** The prior `query=0 / selected=0 / all 1365 deferred` was measured while the slice-1 baseline replay contended for the 7 GB host. On a clean host the **shipped (disposable) engine already runs dataflow**: `query_seconds` 0 → ~35 s, `selected` 0 → **356 regions**, dominance flow queries complete. This is the first time the engine arm produced analysis on this change. 2. **The evidence/taint batch (~225 s) is INVARIANT to session and to graph preload** (225.1 / 228.6 / 223.2 across the three clean runs). Warm session and preload-at-build do **not** reduce it. So the dominant cost is intrinsic to the taint batch, not graph I/O. 3. **Graph-load is cheap — it is NOT the wall.** The `dominance` query is a full cold query *including* `importCpg`, and it costs only ~35 s. So `importCpg` is tens of seconds, not ~150 s. **This corrects the earlier note** that attributed the ~150 s fixed to "first query JVM-start + graph-load + census." The ~160 s fixed lives inside `taint.sc`'s own per-batch flow work (seed/sink memo fill over the distinct `(family, file)` combinations it walks), paid lazily as the first pairs process — independent of the raw pair count. 4. **The warm session is not the lever.** It starts fine on a clean host (the contended `session_startup_timeout` was contention-only), but gives ~no benefit here: subsequent queries (dominance) are already cheap disposable, and the evidence pass is a single batch call, so there is nothing for the session to amortise. 5. **The barrier, restated:** `build(~30) + evidence(~225) ≈ 255 s` of a 300 s budget, so 1009/1365 regions defer as `deadline_exhausted`; only 356 are analysed. The lever is **reducing the taint batch's work** — change-scope the pair/family set so the batch fits budget — NOT session, preload, or a bigger JVM. The parked branch (`engine-change-scoped-evidence-pass`) is therefore **relevant after all, on a clean host**, provided it scopes the family/file set `taint.sc` walks (not merely the pair list); its adaptive portioning still needs replacing with a fixed cap (per-call-fixed cost, not per-pair). 6. **Side note:** the `base`-side CPG build failed (`cpg_build_failed`, 0.9 s) in the preload run — a separate intermittent base-revision build issue; the `head` side built and analysed normally. Track separately. **What E1/E2 did NOT answer (the real G2 question):** on a clean host the engine selects 356 regions — *does it CATCH vulns regex misses?* That needs an **engine-on unseen pilot** (a handful of pool repos replayed with the engine arm enabled on a clean host) scored against record 07's regex-only floor (catch 3/92). That is the decisive next experiment, ahead of E3/E4/L1/L2. ## ENGINE-ON PILOT RESULTS (measured 2026-10-12, CLEAN host, 5 public pool repos, vuln change each, 300 s) Five public repos picked for flow-shaped families and small size (engine's best chance to finish and catch): | repo | fam | size | REGEX on vuln | ENGINE finding | cov | vuln region ranked | best taint | selected/deferred | wall | |---|---|---|---|---|---|---|---|---|---| | masci/banks | path | 2f | ✗ | **none** | incomplete | yes | sink found, no source (score 0) | 281 / **0** | 236 s | | vitejs/launch-editor | injection | 1f | ✗ | **none** | incomplete | yes | tier1 score2.0 sink (wrong fam) | 13 / 38 | 261 s | | getzep/graphiti | injection | 5f | ✗ | **none** | incomplete | yes | injection score 0 | 882 / **0** | 304 s | | aws/bedrock-agentcore | injection | 3f | ✗ | **none** | incomplete | yes | — | 356 / 1009 | 305 s | | miguelgrinberg/python-socketio | deser | 10f | **✓ (n=2)** | **none** | incomplete | yes | **tier1 score2.0 sink found** | 136 / **0** | 257 s | **Result: regex 1/5, engine 0/5.** On a clean host, with budget to spare on most, the shipped engine **added no catches over the regex floor.** **Why (measured, traced — CORRECTED after inspecting `head.findings`, not ranker evidence):** 1. **The budget hypothesis is REFUTED.** banks, graphiti and socketio **finished their region queue** (deferred=0) and still emitted no finding. Change-scoping to fit budget would not have changed the outcome. 2. **The engine produced ZERO vuln-family findings (path/injection/deser) on all 5.** Actual `head.findings`: banks 1 (config_secrets), socketio 2 (config_secrets), launch-editor / graphiti / bedrock **0 findings each**. The dataflow/taint tier traced **no source→sink flow** for the vuln families (`taint_evidence_not_applicable`; e.g. banks found the `path.open` sink but **no source reaches it**). **Correction:** the tier=1 / score=2.0 "sink" seen in the ranker's `normalized_evidence` is the RANKER's static sink candidate, NOT a confirmed taint finding — an earlier draft of this section misread it as detection partially working. It was not. 3. **`ambiguous_line_correspondence` is a downstream RED HERRING for these vulns.** It fires on all 5, but there are **no vuln-family findings to anchor**, so a region-level witness fallback (the proposed "Option A") would change nothing here. Witness/anchoring only touched the config_secrets findings that did appear. 4. **The LLM adjudication tier was DISABLED** (`model_calls=0`, `cost_usd=0`, `model_assistance={}` on all 5). The rungs reached: `suspicion=0, model_corroborated=0, model_entailed=0–2, execution_confirmed=0`. The tier designed to confirm a vuln the static dataflow cannot trace **never ran**. 5. **Ranking also under-promotes the vuln region** (tier 0 for most families; banks path family stuck low). So even where a sink exists, the region is not analysed deeply enough to attempt a flow. 6. **Measurement caveat:** the pool's SZZ last-writer change definition yields large/moved diffs; a cleaner probe uses tight vuln-introducing diffs. Secondary to (2)/(4). **Reprioritised engine levers (CORRECTED — anchoring is NOT the lever):** - **#1 the LLM adjudication tier (disabled in the push path) — now the primary lever.** regex+dataflow = 1/5 with the model OFF; the static tiers produce no vuln-family findings on these shapes. The product thesis ([[search-with-proof]]) is that the LLM tier catches what regex+dataflow miss. The pilot is the empirical case to turn it on: **L1** model latency + calls/push, **L2** recall LIFT on the same 5 (and beyond). Costs money — stage it (5–10 sizing → ~10–20 pilot → full only on lift). - **#2 taint source/sink coverage + region ranking** so the dataflow recognises these vuln sources and promotes the vuln region (banks path: sink found, source unrecognised; graphiti injection: score 0). - **#3 change-anchoring robustness** — real, but only matters AFTER the static/LLM tiers produce vuln-family findings to anchor. Deprioritised until (1)/(2) yield findings. **Release read:** the pre-push net is NOT releasable on detection. Clean host, generous budget, LLM off: regex 1/5, engine 0/5. The static tiers (regex + Joern dataflow) do not detect these unseen vulns; the decisive untested lever is the **LLM adjudication tier**. Next: an **LLM-tier pilot (L1/L2)** on these 5 repos — does enabling the model produce vuln-family findings regex+dataflow miss, and at what $/latency. ## ROOT CAUSE of engine 0/5 + the "#2 guided by #1" workstream (2026-10-12) **Why the taint tier traced no flow (verified in code + evidence):** the pilot repos are **libraries**, whose untrusted input arrives as **public-API function parameters**, not HTTP requests. The Python taint **source** vocabulary (`ruleset/semantic/python.toml`) is entirely web/CLI: Flask/Django `request.*`, `sys.argv`, `sys.stdin`, `input(`. There is no source for a library's own API parameter. The engine *does* have a parameter-as-source mode (`_parameters_are_input`, `scan.py`), but it fires **only for recognised entry points** (`_is_entry_point` and `entry_kind != "function"`) — the pilot vuln regions carry `entry:false`, so their parameters are **deliberately not** sources. That restriction is intentional: treating every function parameter as a source caused a flood of false SQL-injections on WordPress/PMPro (documented in the `_parameters_are_input` docstring). So **the static engine cannot safely widen its sources to library APIs without reintroducing that false-alarm explosion** — `banks` path: sink `path.open` found, `source_local/near=false`; `graphiti`/`bedrock` injection: no source. **This is the precise seam for the maintainer's "#2 guided by #1" directive.** The static taint cannot *safely* decide which library parameters are trust boundaries; the **LLM can** reason about which public-API surface is attacker-reachable (intent, not pattern). So: - **#1 (LLM) guides:** on a changed library region the engine ranks but cannot source, the LLM adjudicates whether a parameter is an untrusted trust boundary and names the source→sink it believes exists (search-with-proof). - **#2 (taint) traces + proves:** the engine then runs the flow from the LLM-named source to the known sink, confirming (or refuting) with a traced path — a proof, not the LLM's say-so. Net: catch without the static FP blowup, and every catch carries a dataflow witness. - **Feedback:** where the LLM repeatedly names the same source shape (e.g. a registry-key → path family), promote it to a static fact in `python.toml` guarded by the entry/shape conditions that keep WordPress quiet. **Proposed first increment (pending budget confirmation — LLM calls cost money):** wire the disabled LLM tier on the 5 pilot regions in *LLM-guided-source* mode and measure: (a) does it name the right trust-boundary source per vuln, (b) does the taint engine then trace source→sink and emit a vuln-family finding, (c) catch lift over regex (1/5) and false-alarm cost on the ordinary changes. Stage it (5–10 sizing → ~10–20 → full on lift). This makes "#2 guided by #1" the concrete G2 detection workstream; budget/change-scoping and change-anchoring stay deprioritised until it yields findings. ## BANKS SIZING + "2 guided by 1" validated (2026-10-12, budget-approved, DeepSeek) Sized the LLM tier on banks (the clearest pilot miss) with the project's own DeepSeek client (`deepseek-flash`). **Infra + cost (sizing answer):** DeepSeek reachable, model id `deepseek-flash` valid. **~$0.00016–0.00025 per call, ~1.1 s latency.** Sub-cent per push even at 8 candidates × a few regions. The GLM/abliteration fallback (`ABLITERATION_*`) currently errors `HTTP 422` on the `thinking` request param — a client fix (don't send `thinking` to that endpoint) is needed before it can be used; DeepSeek works today so it is not blocking. **"2 guided by 1" validated as a mechanism — the LLM gives the correct verdict WHEN given the real sink + source context:** - Asked about `_load` (the pool's SZZ-labelled site, 135–141): `vulnerable=false` — *correct*, `_load` reconstructs paths from its own index, not the sink. - Asked about the REAL sink `from_prompt_path`/`set()` line 44 (`path / f"{prompt.name}…"; write_text`) with the context that `prompt.name` can be request-derived: **`vulnerable=true`** — *"prompt.name is joined into a filesystem path and written without sanitization… path traversal or arbitrary file write."* A correct catch. **Two blockers to realising it in the live pipeline (both now concrete):** 1. **Context, not single-shot.** The model only catches with the real sink + caller/source context. A single-shot question on the ranker's region (which pointed at `_load`, and at the benign local-glob `_scan`) does not. This requires the **agentic search-with-proof hunter** (explores callers/entry points to establish the trust boundary), not the single CANDIDATE_QUESTION path. That is more calls → the real cost driver to size next. 2. **Corpus-label soundness.** The pool's banks label is **mislocated**: `head_lines=[[30],[104],[138–141]]`, `function=_load` — the real sink is line **44**, outside head_lines±10 and a different function. So (a) even a correct finding at line 44 would NOT score as a catch under `location_match`, and (b) the sink pre-exists in base (the change did not touch it), so a change-scoped pre-push hook **correctly stays quiet** — the pool even annotates this bias ("delta hook can correctly stay quiet"). **Part of the pilot's engine 0/5 is this SZZ mislocation, not a detection failure.** **Revised next steps (budget approved, stay staged):** - **Validate the 5 pilot labels first** (cheap, LLM-assisted corpus audit): for each, is the real sink inside the changed lines, and is it change-introduced vs pre-existing-in-base? Only sound, change-located vulns are fair targets. This separates true detection gaps from SZZ artifacts before spending on detection runs. - **Then run the AGENTIC hunter** (search-with-proof) on the sound targets and size its per-vuln call count/cost; measure catch + false-alarm (on ordinary changes) vs regex 1/5. The single-shot path is insufficient; the hunter is the real "#1". - Feed confirmed source shapes (e.g. library param → path join) back as guarded static facts (#2). ## Prompt evolution is a CORE lever for the LLM arm (maintainer directive 2026-10-12) **"What is key is that we evolve the prompts as well"** — and **"GEPA or comparable; this is an evolving field."** The bedrock hunter result makes this concrete: the agentic hunter ran (64 model calls, $0.0076) and reasoned carefully, but with its **current prompts** it flagged adjacent concerns (network-PUBLIC default, download_file) and never named the actual `install_packages` command injection — a *prompting/targeting* miss, not a cost or capability limit. So the LLM arm needs a **prompt-optimisation loop** that reflects on misses/low-confidence outputs, evolves the prompt, and retries — built on **DSPy** (its GEPA / MIPRO optimisers), kept **pluggable** behind an interface because the field moves fast (GEPA, MIPRO, TextGrad, successors). This aligns with [[learned-decisions-not-hand-tuning]] (learned decisions + local memory, not hand rules). It applies to BOTH LLM uses here — the search-with-proof detector and the corpus-label auditor. **The optimisation objective is to learn WHAT HINTS/ANNOTATIONS the model needs** to reach the real sink — not just reword the prompt. bedrock needed the hint "`install_packages` runs `pip install` with caller-supplied package names → command injection"; banks needed "`prompt.name` is request-derived and joined into a filesystem path → traversal". DSPy optimises which evidence/hints to feed the signature so the model targets the true sink instead of adjacent noise. **Source those hints from the framework knowledge the project ALREADY ships** — the semantic facts taxonomy (`ruleset/semantic/*.toml`, `ruleset/frameworks.toml`) encodes sources/sinks/sanitizers per framework WITHOUT user annotations ([[php-detection-is-engine-only]] zero-annotation value). Where framework knowledge covers the API, feed it as a hint; where it does not (library APIs like banks/bedrock), the LLM proposes the trust boundary and, on repeated confirmation, it graduates into the facts (the #2 feedback loop). No new annotation burden on the user. This is now a first-class part of the "#2 guided by #1" workstream, not a deferred nicety. (DeepSeek `deepseek-flash` is the working default; GLM 5.3 "cyber" model is future-spec.) ## POOL-P1 LABEL AUDIT (LLM, 2026-10-12) — 47% of labels are sound Widened the banks label-mislocation check across all **49 public** pool vulns via DeepSeek (advisory + head code + pool label → sound/unsound). Full artifact: `benchmarks/measurements/2026-10-12-pool-p1-label-audit/`. - **label_sound = 23/49 (47%)**; 53% unsound (25 pre-exist in base, 16 sink outside changed lines). Cost $0.038. - By family (sound/total): injection 10/13, access_control 5/8, untrusted_destination 4/7, deserialization 1/2, config_secrets 1/4, **path 1/7, output_encoding 1/8**. Injection labels mostly sound; path/output_encoding mostly mislocate or pre-exist. - **record 07's floor (catch 3/92) is scored against a corpus ~half unfair to a delta hook** — on the ~53% pre-existing/mislocated vulns a change-scoped hook correctly stays quiet. The honest catchable denominator is the sound subset; record 07 needs this caveat + a sound-subset recomputation. - **Fair pilot re-score:** only bedrock + graphiti (injection) of the 5 are sound; the engine missed both — the real detection gaps. The 23 sound targets are the fair local eval set for "#2 guided by #1". - Next: run the agentic search-with-proof hunter on the sound targets (start bedrock + graphiti), size its per-vuln call count/cost, measure catch + false-alarm. Then recompute record 07 on the sound subset. ## MODEL HUNTER on the 2 sound targets (2026-10-12) — first LLM catch the static tiers missed Ran `ousast scan --mode deep` (DeepSeek `deepseek-flash`, model layer + tool_hunter, 60-call/8-hotspot bound) on bedrock + graphiti, the two sound pilot targets the static engine missed. Artifact: `benchmarks/measurements/2026-10-12-hunter-sound-targets/`. - **hunter 1/2 vs static engine 0/2.** graphiti: **HIT** — `falkordb_driver.py:207`, tag `family:cypher-injection`, rationale traces `build_fulltext_query` concatenating user input into a Cypher string at line 278 (inside the labelled function, location-matches). bedrock: **miss** — 12 suspicions on adjacent concerns, never named the `install_packages`→`pip` command injection. - **Cost is a non-issue:** ~$0.007/repo, 64 model calls for a whole-repo deep scan. - **Everything stopped at `suspicion`** — the proof arm did not elevate to corroborated/entailed/execution_confirmed. Release-grade BLOCK needs *proven* findings; measuring the proof arm's lift is the next depth. - bedrock's miss is the concrete case for the DSPy prompt/hints loop above. Next: (a) measure the proof arm (can it elevate graphiti's suspicion to proven?); (b) stand up the DSPy hint loop and re-run bedrock; (c) widen the hunter to the 23 sound targets for a catch rate; (d) then a change-scoped + proven run for the release-relevant shape. ## STRATEGIC PIVOT: core problem = unprovable reachability; FIX THE EVALUATION TARGET FIRST (2026-10-12) **Core problem (named with the maintainer):** the tool can SPOT dangerous ops but cannot prove they are ATTACKER-REACHABLE — the trust boundary lives outside the changed library code, in the caller/application a change-scoped scan does not see; and the project rightly won't emit unproven alarms. Recall → ~0 from unprovable reachability, not blindness. Cost/latency/engine-wall/corpus-labels were all downstream noise. **Chosen bet (of three — fix-target / advisory-with-conditions / build-prover): FIX THE EVALUATION TARGET FIRST.** A pre-push hook runs inside a developer's APPLICATION (entry points present, request→sink in-repo), but the pool is library-heavy, so we measured the hardest/wrong case. **Shape classification (LLM, 49 public vulns, $0.033 — `benchmarks/measurements/2026-10-12-pool-p1-shape-classification/`):** reachability_in_repo True only 12/49 (24%); 37/49 are pure libraries. A legitimate catch needs sound label AND reachability in-repo → **5/49 (~10%)**: graphiti, desktopcommandermcp, malla, document-merge-service, mcp-remote. **Shape predicted the hunter result** (graphiti app-shaped→caught; bedrock library→missed). So record-07's 3/92 is scored against a ~90%-illegitimate denominator — the near-zero catch is largely a TARGET MISMATCH. **Next:** deep scan (regex/engine/hunter) on the 5 fair targets → per-tier catch rate on the corrected set; then stand up an app-shaped eval set from known-vulnerable apps (VAmPI/NodeGoat/DjanGoat/vibe-py) and recompute the floor. Deferred bets (advisory-with-conditions, reachability prover) revisited once the fair-target numbers land. ## FAIR-TARGET RESULT: detection 3/5 on app-shaped+sound (vs ~0 library-heavy) — target-fix validated (2026-10-12) Deep scan on the 5 app-shaped+sound targets (`benchmarks/measurements/2026-10-12-hunter-fair-targets/`): - **3/5 detected**: graphiti (hunter, Cypher injection), mcp-remote (hunter, url-scheme injection), dcmcp (ENGINE overlay — "outbound.request reached from parameter" on readFileFromUrl; the param-as-source mode fired because an entry point is present). **2/5 missed**: malla (DOM/template XSS) and dms (Jinja SSTI) — both in families families.toml DOCUMENTS as limited (template-mediated escaping). - vs the library-heavy pilot's ~0 → **the target mismatch was real and material.** On an app-shaped repo the engine itself detects (parameter is a recognized source once the entry point is in-repo). - **But everything stops at `suspicion`/`static_corroboration`** — none proven. Cost ~$0.007/repo. **Strategy verdict:** bet #1 (fix the eval target) is validated as NECESSARY and MATERIAL but NOT SUFFICIENT. Detection appears on fair targets; release-grade BLOCK still needs (a) proof-arm elevation (suspicion→proven) and (b) family coverage (template XSS/SSTI) + the DSPy hint loop for misses. Next: recompute a floor on an app-shaped sound subset / real app corpus (VAmPI, NodeGoat, DjanGoat, vibe-py), and measure the proof arm's suspicion→proven lift. ## PROOF-ARM MEASUREMENT on the 3 fair-target hits (2026-10-12) — push-time lift 0/3, structural Measured whether the proof arm can elevate the 3 fair-target suspicions (graphiti, mcp-remote, dcmcp) to `exploit_demonstrated`. - **Correction:** the search arm's `FAMILY_ORACLES` (search/demo.py) differs from families.toml's `verifier`. All 3 hits HAVE a runtime oracle: injection→(sql,command), untrusted_destination→(ssrf). (deserialization/ access_control/config_secrets have NONE — empty tuple → `no_oracle`.) So dcmcp IS provable in principle (earlier "unprovable" was a families.toml misread). - **But proving is a runtime differential:** `search/verify.verify()` requires (1) an AUTHORED declarative demo (start the app on a port, run exploit steps, fill oracle placeholders: sql→database_path, command→marker_dir, ssrf→callback_url), (2) BOTH head AND base runnable with OFFLINE deps (`pip --no-index`, `npm ci --offline`), (3) a sandbox differential "owned effect in 3/3 affected runs and 0/3 safe runs". Local host has a sandbox (`isolation_mode: userns`); the plane supplies a task-boundary dispatcher. - **The scan/push path reaches only `suspicion`/`static_corroboration`** (verification.py levels) — there is NO code path from a scan suspicion to `exploit_demonstrated`, and no caller wires the search Coordinator from the push flow. So **proof-arm lift at push time = 0/3**: not because the vulns are unreal, but because proving is a per-target, offline-build, runtime, differential operation that cannot run inside a pre-push hook. (Prior per-pair demo attempts were mostly inconclusive on this exact wiring — app startup, offline deps, oracle plumbing.) **Strategic consequence (resolves the three-bet fork):** proof/BLOCK is an ASYNC capability (CI / plane / triage with authored demos + a runnable sandbox), NOT a pre-push one. At push time the honest ceiling is corroborated suspicion. So: - **Bet #2 (advisory-with-conditions) is the honest PRE-PUSH product**: surface corroborated suspicions ("change introduces a flow that looks like X; reachable if untrusted Y reaches it"), let the dev / coding-agent (who has the app context) judge — matching the "oversee coding agents" framing and the suspicion-rung reality. - **Bet #3 (the proof arm) is the ASYNC BLOCK lane**: run `search/verify` on the plane/CI for the suspicions worth proving, with authored (eventually DSPy-authored) demos. Worth an empirical run on one tractable target (mcp-remote) to confirm the arm demonstrates on a real unseen vuln — a bounded, separate effort. ## BANKS FIXED-SEARCH MVP (2026-10-12) — "no wrong targets, only wrong searching" demonstrated Record: `benchmarks/measurements/2026-10-12-banks-fixed-search-mvp/`. Target: banks (the "library" that gave the engine ZERO path findings in the pilot). Same target, same CPG — only the search changes. - Injected the two annotations a per-target LLM steer would supply: **parameter-as-source** (`parameter_sources=True`, so `prompt.name` counts as untrusted) + the **missing file-write sink** (`write_text`/`read_text` added to path sinks). - **Baseline (shipped search): 0 resolved `prompt → write_text` flows** — its 5 flow rows are all census/ evidence (`sink=None`), matching the pilot's zero path findings. - **Fixed search: the engine resolves `prompt → path-join → prompt_file.write_text(prompt.raw)` in `from_prompt_path`** (`source=prompt`), plus a second `write_text` sink in `_scan`. These appear ONLY with the annotations. - **Thesis demonstrated on a concrete case:** banks was not a wrong target; the search was unannotated. The fix needed exactly **two** inputs — the param-source and the file-write sink — which are precisely what the per-target LLM steer (DSPy-optimised, framework-seeded) has to learn to supply. This is the first measured instance of the "#2 guided by #1" loop closing on an unseen target. - Caveats: (1) one row shows `source=path` not `prompt` — the param-source fires broadly, a precision knob for the steering loop; (2) this is a TRACE, not the runtime differential proof (proof stays the async BLOCK lane). **Instrument note / filed bug:** the first MVP used the single-query `cpg.run("taint", …)` path and returned 0/0 — an INSTRUMENT FAILURE, not a result (`cpg query taint failed` on every query). Root cause filed in `bugs/taint-sc-single-query-signature-skew.md`: `request_params()` emits 5 keys (`guards`, `fixedOrigin`, `originAnchors`, `quotedSanitizers`, `fieldParameterSources`) that `taint.sc`'s `exec` signature does not declare, so both single-query sub-routes (CLI `--param` and session named-arg) reject them. The batch path (`run_batch` → `requestsFile` JSON) tolerates them, which is why the pilot scans ran and this MVP used it.