feat(plane): tie-break pass on disputed candidates; 2-of-3 agreement

The design revisit after plane-increment-1.json (14/20 agreed, 19/20 in either
pass): a third verify pass on the candidates passes a and b dispute, and agree
decides 2-of-3 where pass c asked (the a/b rule elsewhere; c alone never decides).

- verify: optional OUSAST_INPUT_ONLY restricts the hunt to the named candidates
  (agree's disputed.json); an empty list is done with 0 units and no model client.
- agree: optional OUSAST_INPUT_PASS_C; metrics keep every key and add tiebreak
  {asked, flipped_to_agreed, flipped_to_rejected, rule}; usd_total, hunts and
  declared-site metrics are on the final decision.
- generate: per case <case>-vc and <case>-final appended to validation-46; the
  first 60 entries stay byte-identical, so a rerun keeps their done state.
- Offline replay over the recorded outputs with an empty pass c reproduces the
  recorded agree summaries exactly (14/20 agreed, 23 agreed, 10 disputed).
- Spec: Requirement 6.4, design Measurement, task 8.1.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01EmM1j46qgKMh3zqD1QMwEo

.kiro/specs/ai-service-plane/design.md

@@ -206,6 +206,11 @@ plane applies the recorded triage without a model call, so `cost_per_candidate`
 not judged; the triage share is derived per file from the recorded spend minus the priced hunt usage (about $0.087
 of the $1.01) and labelled as derived in `case.json`. Sites are judged as `declared_sites_matched_agreed` against
 `declared_sites_in_set`, the reference's 16 of 20.
+The tie-break (Requirement 6.4) appends per case `<case>-vc`, `verify` pass c restricted by `OUSAST_INPUT_ONLY`
+to the agree task's `disputed.json` (an empty list is `done` with no model call), and `<case>-final`, `agree` over
+passes a, b and c, which decides 2-of-3 where pass c asked and keeps the a/b rule elsewhere, so pass c alone never
+decides and `metrics.tiebreak` counts the flips; the first 60 entries stay byte-identical, so a rerun keeps their
+`done` state, executes only the 30 new tasks, and pass c's cost adds to `usd_total` for the 6.2 judgement.
 
 ## Data Models
 

.kiro/specs/ai-service-plane/requirements.md

@@ -132,6 +132,9 @@ before the remaining concerns are converted.
    current $0.022 for two passes; otherwise the increment is reported as not met and the design revisited.
 3. The development transcript measure (`benchmarks/dev/token_report.py`) is run at the end of the increment and
    recorded next to the result.
+4. After the first measurement (plane-increment-1.json: 14/20 agreed, 19/20 in either pass), agreement is decided
+   2-of-3 with a third pass run only on candidates passes a and b dispute; the gate of 6.2 is re-evaluated on that
+   decision, and the result is recorded as plane-increment-2.json next to the first.
 
 ### Requirement 7: Routing is a declared concern of the plane
 

.kiro/specs/ai-service-plane/tasks.md

@@ -130,6 +130,15 @@ mypy are clean, and the commit is gated on pytest's own exit code. Script-level
     if not met, revise the design before task 9.
   - _Requirements: 6.1, 6.2, 6.3_
 
+- [ ] 8.1 Tie-break pass and re-measurement
+  - `verify` takes `OUSAST_INPUT_ONLY` (pass c on the disputed candidates only); `agree` takes
+    `OUSAST_INPUT_PASS_C` and decides 2-of-3 where pass c asked, the a/b rule elsewhere (`metrics.tiebreak`).
+  - `validation-46` gains `<case>-vc` and `<case>-final` per case, appended so the first 60 entries keep their
+    state; an offline replay with an empty pass c reproduces the recorded 14/20.
+  - Rerun `validation-46` on ax (only the new tasks run), re-evaluate the 6.2 gate on the final decision and record
+    `benchmarks/independent/plane-increment-2.json` next to the first.
+  - _Requirements: 6.2, 6.4_
+
 - [ ] 9. Handover
   - `ops/README.md`: how to run a Run locally and what an ax deployment would need (Kubernetes, Agent
     Substrate, the runner image, the provider extension); memory note updated; scripts that the tasks replaced

plane/runs/validation-46.yaml

@@ -1,5 +1,5 @@
 # generated by `ousast plane workspaces benchmarks/independent/population-v2.toml --validation-set ~/ousast-results/independent-v2-sets/validation.json --candidates ~/ousast-results/independent-v2-modelsinks --triage ~/ousast-results/independent-v2-batched-check/scans --repos ~/.cache/openultrasast/independent --run validation-46 --runner-image ~/.cache/ousast/ax-src/runner-image`; do not edit, regenerate.
-# 15 cases, 46 candidates (43 kept by the recorded triage); usd ceiling 2.02.
+# 15 cases, 46 candidates (43 kept by the recorded triage); usd ceiling 3.03.
 apiVersion: openultrasast.io/v1alpha1
 kind: Run
 metadata:
@@ -561,3 +561,378 @@ spec:
     - agreed.json
     - disputed.json
     budget: *id015
+  - name: budibase-mongo-template-nosqli-vc
+    task: verify-c-budibase-mongo-template-nosqli
+    dependsOn:
+    - budibase-mongo-template-nosqli-agree
+    inputs:
+      facts: budibase-mongo-template-nosqli-facts/facts.json
+      only: budibase-mongo-template-nosqli-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.06
+      calls: 16
+  - name: budibase-mongo-template-nosqli-final
+    task: agree-budibase-mongo-template-nosqli
+    inputs:
+      pass_a: budibase-mongo-template-nosqli-va/units.jsonl
+      pass_b: budibase-mongo-template-nosqli-vb/units.jsonl
+      pass_c: budibase-mongo-template-nosqli-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0
+  - name: cct-studio-server-cmdi-vc
+    task: verify-c-cct-studio-server-cmdi
+    dependsOn:
+    - cct-studio-server-cmdi-agree
+    inputs:
+      facts: cct-studio-server-cmdi-facts/facts.json
+      only: cct-studio-server-cmdi-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.04
+      calls: 8
+  - name: cct-studio-server-cmdi-final
+    task: agree-cct-studio-server-cmdi
+    inputs:
+      pass_a: cct-studio-server-cmdi-va/units.jsonl
+      pass_b: cct-studio-server-cmdi-vb/units.jsonl
+      pass_c: cct-studio-server-cmdi-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0
+  - name: contextforge-prompt-ssti-vc
+    task: verify-c-contextforge-prompt-ssti
+    dependsOn:
+    - contextforge-prompt-ssti-agree
+    inputs:
+      facts: contextforge-prompt-ssti-facts/facts.json
+      only: contextforge-prompt-ssti-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.04
+      calls: 8
+  - name: contextforge-prompt-ssti-final
+    task: agree-contextforge-prompt-ssti
+    inputs:
+      pass_a: contextforge-prompt-ssti-va/units.jsonl
+      pass_b: contextforge-prompt-ssti-vb/units.jsonl
+      pass_c: contextforge-prompt-ssti-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0
+  - name: froxlor-admin-ip-second-order-sqli-vc
+    task: verify-c-froxlor-admin-ip-second-order-sqli
+    dependsOn:
+    - froxlor-admin-ip-second-order-sqli-agree
+    inputs:
+      facts: froxlor-admin-ip-second-order-sqli-facts/facts.json
+      only: froxlor-admin-ip-second-order-sqli-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.07
+      calls: 16
+  - name: froxlor-admin-ip-second-order-sqli-final
+    task: agree-froxlor-admin-ip-second-order-sqli
+    inputs:
+      pass_a: froxlor-admin-ip-second-order-sqli-va/units.jsonl
+      pass_b: froxlor-admin-ip-second-order-sqli-vb/units.jsonl
+      pass_c: froxlor-admin-ip-second-order-sqli-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0
+  - name: fuxa-tdengine-escape-sqli-vc
+    task: verify-c-fuxa-tdengine-escape-sqli
+    dependsOn:
+    - fuxa-tdengine-escape-sqli-agree
+    inputs:
+      facts: fuxa-tdengine-escape-sqli-facts/facts.json
+      only: fuxa-tdengine-escape-sqli-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.12
+      calls: 8
+  - name: fuxa-tdengine-escape-sqli-final
+    task: agree-fuxa-tdengine-escape-sqli
+    inputs:
+      pass_a: fuxa-tdengine-escape-sqli-va/units.jsonl
+      pass_b: fuxa-tdengine-escape-sqli-vb/units.jsonl
+      pass_c: fuxa-tdengine-escape-sqli-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0
+  - name: lmdeploy-media-url-ssrf-vc
+    task: verify-c-lmdeploy-media-url-ssrf
+    dependsOn:
+    - lmdeploy-media-url-ssrf-agree
+    inputs:
+      facts: lmdeploy-media-url-ssrf-facts/facts.json
+      only: lmdeploy-media-url-ssrf-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.02
+      calls: 8
+  - name: lmdeploy-media-url-ssrf-final
+    task: agree-lmdeploy-media-url-ssrf
+    inputs:
+      pass_a: lmdeploy-media-url-ssrf-va/units.jsonl
+      pass_b: lmdeploy-media-url-ssrf-vb/units.jsonl
+      pass_c: lmdeploy-media-url-ssrf-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0
+  - name: medplum-external-auth-open-redirect-vc
+    task: verify-c-medplum-external-auth-open-redirect
+    dependsOn:
+    - medplum-external-auth-open-redirect-agree
+    inputs:
+      facts: medplum-external-auth-open-redirect-facts/facts.json
+      only: medplum-external-auth-open-redirect-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.02
+      calls: 8
+  - name: medplum-external-auth-open-redirect-final
+    task: agree-medplum-external-auth-open-redirect
+    inputs:
+      pass_a: medplum-external-auth-open-redirect-va/units.jsonl
+      pass_b: medplum-external-auth-open-redirect-vb/units.jsonl
+      pass_c: medplum-external-auth-open-redirect-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0
+  - name: pgadmin-maintenance-sqli-vc
+    task: verify-c-pgadmin-maintenance-sqli
+    dependsOn:
+    - pgadmin-maintenance-sqli-agree
+    inputs:
+      facts: pgadmin-maintenance-sqli-facts/facts.json
+      only: pgadmin-maintenance-sqli-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.45
+      calls: 96
+  - name: pgadmin-maintenance-sqli-final
+    task: agree-pgadmin-maintenance-sqli
+    inputs:
+      pass_a: pgadmin-maintenance-sqli-va/units.jsonl
+      pass_b: pgadmin-maintenance-sqli-vb/units.jsonl
+      pass_c: pgadmin-maintenance-sqli-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0
+  - name: qwed-sympy-eval-vc
+    task: verify-c-qwed-sympy-eval
+    dependsOn:
+    - qwed-sympy-eval-agree
+    inputs:
+      facts: qwed-sympy-eval-facts/facts.json
+      only: qwed-sympy-eval-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.02
+      calls: 8
+  - name: qwed-sympy-eval-final
+    task: agree-qwed-sympy-eval
+    inputs:
+      pass_a: qwed-sympy-eval-va/units.jsonl
+      pass_b: qwed-sympy-eval-vb/units.jsonl
+      pass_c: qwed-sympy-eval-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0
+  - name: reactpress-findall-column-sqli-vc
+    task: verify-c-reactpress-findall-column-sqli
+    dependsOn:
+    - reactpress-findall-column-sqli-agree
+    inputs:
+      facts: reactpress-findall-column-sqli-facts/facts.json
+      only: reactpress-findall-column-sqli-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.02
+      calls: 8
+  - name: reactpress-findall-column-sqli-final
+    task: agree-reactpress-findall-column-sqli
+    inputs:
+      pass_a: reactpress-findall-column-sqli-va/units.jsonl
+      pass_b: reactpress-findall-column-sqli-vb/units.jsonl
+      pass_c: reactpress-findall-column-sqli-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0
+  - name: unleash-addon-webhook-ssrf-vc
+    task: verify-c-unleash-addon-webhook-ssrf
+    dependsOn:
+    - unleash-addon-webhook-ssrf-agree
+    inputs:
+      facts: unleash-addon-webhook-ssrf-facts/facts.json
+      only: unleash-addon-webhook-ssrf-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.03
+      calls: 8
+  - name: unleash-addon-webhook-ssrf-final
+    task: agree-unleash-addon-webhook-ssrf
+    inputs:
+      pass_a: unleash-addon-webhook-ssrf-va/units.jsonl
+      pass_b: unleash-addon-webhook-ssrf-vb/units.jsonl
+      pass_c: unleash-addon-webhook-ssrf-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0
+  - name: winml-cli-serve-cors-vc
+    task: verify-c-winml-cli-serve-cors
+    dependsOn:
+    - winml-cli-serve-cors-agree
+    inputs:
+      facts: winml-cli-serve-cors-facts/facts.json
+      only: winml-cli-serve-cors-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.02
+      calls: 8
+  - name: winml-cli-serve-cors-final
+    task: agree-winml-cli-serve-cors
+    inputs:
+      pass_a: winml-cli-serve-cors-va/units.jsonl
+      pass_b: winml-cli-serve-cors-vb/units.jsonl
+      pass_c: winml-cli-serve-cors-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0
+  - name: wp-directorist-avatar-ssrf-vc
+    task: verify-c-wp-directorist-avatar-ssrf
+    dependsOn:
+    - wp-directorist-avatar-ssrf-agree
+    inputs:
+      facts: wp-directorist-avatar-ssrf-facts/facts.json
+      only: wp-directorist-avatar-ssrf-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.03
+      calls: 8
+  - name: wp-directorist-avatar-ssrf-final
+    task: agree-wp-directorist-avatar-ssrf
+    inputs:
+      pass_a: wp-directorist-avatar-ssrf-va/units.jsonl
+      pass_b: wp-directorist-avatar-ssrf-vb/units.jsonl
+      pass_c: wp-directorist-avatar-ssrf-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0
+  - name: wp-gopay-log-filter-sqli-vc
+    task: verify-c-wp-gopay-log-filter-sqli
+    dependsOn:
+    - wp-gopay-log-filter-sqli-agree
+    inputs:
+      facts: wp-gopay-log-filter-sqli-facts/facts.json
+      only: wp-gopay-log-filter-sqli-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.02
+      calls: 8
+  - name: wp-gopay-log-filter-sqli-final
+    task: agree-wp-gopay-log-filter-sqli
+    inputs:
+      pass_a: wp-gopay-log-filter-sqli-va/units.jsonl
+      pass_b: wp-gopay-log-filter-sqli-vb/units.jsonl
+      pass_c: wp-gopay-log-filter-sqli-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0
+  - name: wp-groundhogg-confirm-redirect-vc
+    task: verify-c-wp-groundhogg-confirm-redirect
+    dependsOn:
+    - wp-groundhogg-confirm-redirect-agree
+    inputs:
+      facts: wp-groundhogg-confirm-redirect-facts/facts.json
+      only: wp-groundhogg-confirm-redirect-agree/disputed.json
+    outputs:
+    - units.jsonl
+    - summary.json
+    budget:
+      usd: 0.05
+      calls: 8
+  - name: wp-groundhogg-confirm-redirect-final
+    task: agree-wp-groundhogg-confirm-redirect
+    inputs:
+      pass_a: wp-groundhogg-confirm-redirect-va/units.jsonl
+      pass_b: wp-groundhogg-confirm-redirect-vb/units.jsonl
+      pass_c: wp-groundhogg-confirm-redirect-vc/units.jsonl
+    outputs:
+    - agreed.json
+    - disputed.json
+    budget:
+      usd: 0
+      calls: 0

plane/tasks/validation-46.yaml

@@ -1738,3 +1738,468 @@ spec:
   workspaces:
   - name: wp-groundhogg-confirm-redirect-inputs
     path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-budibase-mongo-template-nosqli
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: budibase-mongo-template-nosqli-vulnerable
+    path: /workspace/case
+  - name: budibase-mongo-template-nosqli-inputs
+    path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-cct-studio-server-cmdi
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: cct-studio-server-cmdi-vulnerable
+    path: /workspace/case
+  - name: cct-studio-server-cmdi-inputs
+    path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-contextforge-prompt-ssti
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: contextforge-prompt-ssti-vulnerable
+    path: /workspace/case
+  - name: contextforge-prompt-ssti-inputs
+    path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-froxlor-admin-ip-second-order-sqli
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: froxlor-admin-ip-second-order-sqli-vulnerable
+    path: /workspace/case
+  - name: froxlor-admin-ip-second-order-sqli-inputs
+    path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-fuxa-tdengine-escape-sqli
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: fuxa-tdengine-escape-sqli-vulnerable
+    path: /workspace/case
+  - name: fuxa-tdengine-escape-sqli-inputs
+    path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-lmdeploy-media-url-ssrf
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: lmdeploy-media-url-ssrf-vulnerable
+    path: /workspace/case
+  - name: lmdeploy-media-url-ssrf-inputs
+    path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-medplum-external-auth-open-redirect
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: medplum-external-auth-open-redirect-vulnerable
+    path: /workspace/case
+  - name: medplum-external-auth-open-redirect-inputs
+    path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-pgadmin-maintenance-sqli
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: pgadmin-maintenance-sqli-vulnerable
+    path: /workspace/case
+  - name: pgadmin-maintenance-sqli-inputs
+    path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-qwed-sympy-eval
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: qwed-sympy-eval-vulnerable
+    path: /workspace/case
+  - name: qwed-sympy-eval-inputs
+    path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-reactpress-findall-column-sqli
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: reactpress-findall-column-sqli-vulnerable
+    path: /workspace/case
+  - name: reactpress-findall-column-sqli-inputs
+    path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-unleash-addon-webhook-ssrf
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: unleash-addon-webhook-ssrf-vulnerable
+    path: /workspace/case
+  - name: unleash-addon-webhook-ssrf-inputs
+    path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-winml-cli-serve-cors
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: winml-cli-serve-cors-vulnerable
+    path: /workspace/case
+  - name: winml-cli-serve-cors-inputs
+    path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-wp-directorist-avatar-ssrf
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: wp-directorist-avatar-ssrf-vulnerable
+    path: /workspace/case
+  - name: wp-directorist-avatar-ssrf-inputs
+    path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-wp-gopay-log-filter-sqli
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: wp-gopay-log-filter-sqli-vulnerable
+    path: /workspace/case
+  - name: wp-gopay-log-filter-sqli-inputs
+    path: /workspace/inputs
+---
+apiVersion: ax.io/v1alpha1
+kind: Task
+metadata:
+  name: verify-c-wp-groundhogg-confirm-redirect
+  atespace: default
+  annotations:
+    openultrasast.io/model: deepseek-flash
+spec:
+  image: localhost:5001/ousast-runner@sha256:200e335877b52d68ea0bc53b4c9ba49ef2ee19e8a5ec2c874148b969123ef7ab
+  command:
+  - verify
+  env:
+  - name: OUSAST_WORKSPACE_DIR
+    value: /workspace/case/repo
+  - name: OUSAST_INPUT_CANDIDATES
+    value: /workspace/inputs/candidates.json
+  - name: OUSAST_PASS
+    value: c
+  resources:
+    requests:
+      cpu: 250m
+      memory: 384Mi
+    limits:
+      cpu: '1'
+      memory: 1Gi
+  workspaces:
+  - name: wp-groundhogg-confirm-redirect-vulnerable
+    path: /workspace/case
+  - name: wp-groundhogg-confirm-redirect-inputs
+    path: /workspace/inputs

src/openultrasast/plane/generate.py

@@ -233,12 +233,31 @@ def verify_budget(item: CaseInputs) -> dict[str, float | int]:
     return {"usd": math.ceil(HEADROOM * share * 100 - 1e-9) / 100, "calls": max(hunts, 1) * CALLS_PER_HUNT}
 
 
+def _tiebreak_entries(item: CaseInputs) -> list[dict[str, Any]]:
+    """The tie-break (Requirement 6.4): verify pass c on the candidates a and b disputed, then agree 2-of-3. Fresh
+    budget dicts, so YAML adds no anchor to the entries before them."""
+    case_id = item.case["id"]
+    return [
+        {
+            "name": f"{case_id}-vc", "task": f"verify-c-{case_id}", "dependsOn": [f"{case_id}-agree"],
+            "inputs": {"facts": f"{case_id}-facts/facts.json", "only": f"{case_id}-agree/disputed.json"},
+            "outputs": ["units.jsonl", "summary.json"], "budget": verify_budget(item),
+        },
+        {
+            "name": f"{case_id}-final", "task": f"agree-{case_id}",
+            "inputs": {pass_: f"{case_id}-v{pass_[-1]}/units.jsonl" for pass_ in ("pass_a", "pass_b", "pass_c")},
+            "outputs": ["agreed.json", "disputed.json"], "budget": {"usd": 0, "calls": 0},
+        },
+    ]  # fmt: skip
+
+
 def render(cases: Sequence[CaseInputs], templates: Mapping[str, Task], run_name: str, command: str) -> dict[str, str]:
     """Relative path under the plane root -> file text, for every generated manifest."""
     header = f"# generated by `{command}`; do not edit, regenerate.\n"
     out: dict[str, str] = {}
     tasks: list[dict[str, Any]] = []
     entries: list[dict[str, Any]] = []
+    tiebreak: list[dict[str, Any]] = []
     for item in cases:
         case_id = item.case["id"]
         pinned, inputs = _workspaces(item)
@@ -270,6 +289,11 @@ def render(cases: Sequence[CaseInputs], templates: Mapping[str, Task], run_name:
             "inputs": {"pass_a": f"{case_id}-va/units.jsonl", "pass_b": f"{case_id}-vb/units.jsonl"},
             "outputs": ["agreed.json", "disputed.json"], "budget": free,
         })  # fmt: skip
+        env = {**on_repo, "OUSAST_INPUT_CANDIDATES": f"{INPUTS_PATH}/candidates.json", "OUSAST_PASS": "c"}
+        tasks.append(_task(templates["verify"], f"verify-c-{case_id}", env, repo))
+        tiebreak.extend(_tiebreak_entries(item))
+    tasks.sort(key=lambda doc: 1 if doc["metadata"]["name"].startswith("verify-c-") else 0)  # stable: the old Tasks first
+    entries.extend(tiebreak)  # appended: the first Run's entries stay byte-identical, so a rerun keeps their state
     out[f"tasks/{run_name}.yaml"] = _dump(header, tasks)
     run = {"apiVersion": RUN_API_VERSION, "kind": "Run", "metadata": {"name": run_name}, "spec": {"tasks": entries}}
     total = sum(float(e["budget"]["usd"]) for e in entries)

src/openultrasast/plane/tasks/agree.py

@@ -17,7 +17,13 @@ divide by the candidates before triage when `case.json` gives that count, as the
 Env contract: `OUSAST_INPUT_PASS_A`, `OUSAST_INPUT_PASS_B` (a verify output directory or its `units.jsonl`),
 `OUSAST_INPUT_CANDIDATES`, `OUSAST_OUTPUT_DIR`, optional `OUSAST_INPUT_SITES` (the generator's `case.json`:
 `family`, `sites`, `ranges`, `sites_in_set`, `cost`; or a `population-v2.toml`-shaped TOML with `OUSAST_CASE_ID`
-naming the case). Writes `agreed.json`, `disputed.json`, `summary.json`. Exit 0 done, 2 failed, 3 unfinished.
+naming the case), optional `OUSAST_INPUT_PASS_C` (the tie-break pass: `verify` restricted to the candidates a and
+b disputed). Writes `agreed.json`, `disputed.json`, `summary.json`. Exit 0 done, 2 failed, 3 unfinished.
+
+With pass c the decision is 2 of 3 for every candidate pass c asked about and the a/b rule for the rest, so pass c
+alone never decides; `metrics.tiebreak` counts the candidates asked and the decisions that flipped against a/b,
+and names the rule. The declared-site metrics and `agreed`/`disputed` are on the final decision; `usd_total`,
+`hunts` and the per-hunt means include pass c. Without pass c every number is the two-pass one.
 """
 
 from __future__ import annotations
@@ -122,6 +128,28 @@ def _per_candidate(total: float | None, count: int) -> float | None:
     return None if total is None or not count else round(total / count, 6)
 
 
+RULE_A_B = "agreed when passes a and b both flagged the candidate (no pass c given)"
+RULE_2_OF_3 = (
+    "a candidate asked in pass c is agreed when at least 2 of passes a, b, c flagged it; a candidate not asked in "
+    "pass c is agreed when passes a and b both flagged it (pass c alone never decides)"
+)
+
+
+def _votes(candidate: str, flagged: Sequence[Mapping[str, Any]]) -> int:
+    return sum(candidate in f for f in flagged)
+
+
+def _pick(candidate: str, flagged: Sequence[Mapping[str, Mapping[str, Any]]]) -> dict[str, Any]:
+    return dict(next(f[candidate] for f in flagged if candidate in f))
+
+
+def _decided(candidate: str, asked_c: set[str], flagged: Sequence[Mapping[str, Any]]) -> bool:
+    """The final decision: 2 of 3 when pass c asked the candidate, else both of a and b."""
+    if candidate in asked_c:
+        return _votes(candidate, flagged) >= 2
+    return candidate in flagged[0] and candidate in flagged[1]
+
+
 def run(
     output_dir: Path,
     candidates: object,
@@ -129,15 +157,28 @@ def run(
     pass_b: Sequence[Mapping[str, Any]],
     *,
     sites: Mapping[str, Any] | None = None,
+    pass_c: Sequence[Mapping[str, Any]] | None = None,
 ) -> dict[str, Any]:
     output_dir.mkdir(parents=True, exist_ok=True)
     family, unique = load_candidates(candidates)
     names = [f"{p}::{fn}" for p, fn, _ in unique]
     lines = {f"{p}::{fn}": ln for p, fn, ln in unique}
-    asked = [{f"{r['path']}::{c[1]}" for r in rows for c in r.get("candidates", [])} for rows in (pass_a, pass_b)]
-    flagged = [{str(f["candidate"]): dict(f) for r in rows for f in r.get("flagged", [])} for rows in (pass_a, pass_b)]
-    agreed, disputed = agreement([list(flagged[0].values()), list(flagged[1].values())])
-    usd_by, turns_by = _per_candidate_cost([*pass_a, *pass_b])
+    passes = (pass_a, pass_b, pass_c or [])
+    asked = [{f"{r['path']}::{c[1]}" for r in rows for c in r.get("candidates", [])} for rows in passes]
+    flagged = [{str(f["candidate"]): dict(f) for r in rows for f in r.get("flagged", [])} for rows in passes]
+    ab_agreed, _ = agreement([list(flagged[0].values()), list(flagged[1].values())])
+    ab = {str(f["candidate"]) for f in ab_agreed}
+    final = {c for c in set(flagged[0]) | set(flagged[1]) | set(flagged[2]) if _decided(c, asked[2], flagged)}
+    agreed = [dict(f, passes=_votes(f["candidate"], flagged)) for f in ab_agreed if f["candidate"] in final]
+    agreed += [dict(_pick(c, flagged), passes=_votes(c, flagged)) for c in sorted(final - ab)]
+    agreed.sort(key=lambda f: str(f["candidate"]))
+    some = (set(flagged[0]) | set(flagged[1]) | set(flagged[2])) - final
+    disputed = [dict(_pick(c, flagged), passes=_votes(c, flagged)) for c in sorted(some)]
+    tiebreak = {
+        "asked": len(asked[2]), "flipped_to_agreed": len(final - ab), "flipped_to_rejected": len(ab - final),
+        "rule": RULE_2_OF_3 if pass_c is not None else RULE_A_B,
+    }  # fmt: skip
+    usd_by, turns_by = _per_candidate_cost([*pass_a, *pass_b, *(pass_c or [])])
     case = dict(sites) if sites else None
     if case is not None and "family" not in case and family:
         case["family"] = family
@@ -147,19 +188,20 @@ def run(
     rows: list[dict[str, Any]] = []
     for name in names:
         in_a, in_b = (name in flagged[0] if name in asked[0] else None), (name in flagged[1] if name in asked[1] else None)
-        finding = flagged[0].get(name) or flagged[1].get(name)
+        in_c = name in flagged[2] if name in asked[2] else None
+        finding = flagged[0].get(name) or flagged[1].get(name) or flagged[2].get(name)
         site = anchor_site(name, finding["site"]) if finding else f"{name.partition('::')[0]}:{lines[name]}:{name.partition('::')[2]}"
         site_match = None
         if assessed and case is not None and isinstance(ranges, Mapping):
             site_match = matches_declared({"site": site, "family": (finding or {}).get("family", family)}, case, ranges)
         cost = usd_by.get(name)
         rows.append({
-            "candidate": name, "site": site, "a": in_a, "b": in_b, "agreed": bool(in_a and in_b), "disputed": bool(in_a) != bool(in_b),
+            "candidate": name, "site": site, "a": in_a, "b": in_b, "c": in_c, "agreed": name in final, "disputed": bool(in_a) != bool(in_b),
             "site_match": site_match, "usd": (None if cost is None else round(cost, 6)), "turns": turns_by.get(name, 0), "finding": finding,
         })  # fmt: skip
     unasked = [r["candidate"] for r in rows if r["a"] is None or r["b"] is None]
     declared = [r for r in rows if r["site_match"]]
-    hunts = [*pass_a, *pass_b]
+    hunts = [*pass_a, *pass_b, *(pass_c or [])]
     usd_total = None if any(r.get("usd") is None for r in hunts) else round(sum(float(r["usd"]) for r in hunts), 6)
     reference: Mapping[str, Any] = (case or {}).get("cost") or {}
     before = int(reference.get("candidates_before_triage") or len(names))
@@ -177,6 +219,7 @@ def run(
         "usd_total_with_recorded_triage": with_triage, "cost_per_candidate_with_recorded_triage": _per_candidate(with_triage, before),
         "turns_per_hunt": _mean([float(r.get("turns") or 0) for r in hunts]),
         "calls_per_hunt": _mean([float((r.get("usage") or {}).get("calls") or 0) for r in hunts]),
+        "tiebreak": tiebreak,
     }  # fmt: skip
     (output_dir / "agreed.json").write_text(
         json.dumps({"family": family, "candidates": rows, "agreed": agreed, "disputed": disputed, "metrics": metrics}, indent=1) + "\n"
@@ -201,7 +244,8 @@ def main() -> int:
         candidates = json.loads(Path(os.environ["OUSAST_INPUT_CANDIDATES"]).read_text())
         sites_path = Path(os.environ["OUSAST_INPUT_SITES"]) if os.environ.get("OUSAST_INPUT_SITES") else None
         sites = load_sites(sites_path, os.environ.get("OUSAST_CASE_ID"))
-        summary = run(output_dir, candidates, pass_a, pass_b, sites=sites)
+        pass_c = load_pass(Path(os.environ["OUSAST_INPUT_PASS_C"])) if os.environ.get("OUSAST_INPUT_PASS_C") else None
+        summary = run(output_dir, candidates, pass_a, pass_b, sites=sites, pass_c=pass_c)
     except Exception:  # noqa: BLE001 -- a crash is `failed` with its traceback
         summary = {
             "status": "failed", "units_done": 0, "units_total": 0, "usd": 0, "calls": 0, "usage": dict.fromkeys(USAGE_FIELDS, 0),
@@ -214,7 +258,18 @@ def main() -> int:
     return EXIT_CODES.get(str(summary["status"]), 2)
 
 
-__all__ = ["LINE_WINDOW", "agreement", "anchor_site", "load_pass", "load_sites", "main", "matches_declared", "run"]
+__all__ = [
+    "LINE_WINDOW",
+    "RULE_2_OF_3",
+    "RULE_A_B",
+    "agreement",
+    "anchor_site",
+    "load_pass",
+    "load_sites",
+    "main",
+    "matches_declared",
+    "run",
+]
 
 
 if __name__ == "__main__":

src/openultrasast/plane/tasks/verify.py

@@ -10,8 +10,10 @@ completes, read back on start: Requirement 2.2). The chat client is wrapped by `
 last -- names the bound Model (Requirement 7.2) and is `done` only when every unit finished.
 
 Env contract: `OUSAST_WORKSPACE_DIR`, `OUSAST_OUTPUT_DIR`, `OUSAST_INPUT_CANDIDATES`, `OUSAST_INPUT_FACTS`
-(optional), `OUSAST_BUDGET_USD`, `OUSAST_BUDGET_CALLS`, `OUSAST_MODEL`, `OUSAST_MODEL_PARAMS`, `OUSAST_PASS`.
-Exit 0 done, 2 failed, 3 unfinished.
+(optional), `OUSAST_BUDGET_USD`, `OUSAST_BUDGET_CALLS`, `OUSAST_MODEL`, `OUSAST_MODEL_PARAMS`, `OUSAST_PASS`,
+`OUSAST_INPUT_ONLY` (optional: a JSON list of `path::function` names, `[path, function, ...]` rows or finding
+objects with a `candidate` -- `agree`'s `disputed.json` -- restricting the hunt to those candidates; the tie-break
+pass c. An empty list is a `done` task with 0 units and no model call). Exit 0 done, 2 failed, 3 unfinished.
 """
 
 from __future__ import annotations
@@ -95,6 +97,34 @@ def load_candidates(payload: object) -> tuple[str | None, list[Candidate]]:
     return family, list(unique.values())
 
 
+def only_names(payload: object) -> set[str]:
+    """The `path::function` names of an `OUSAST_INPUT_ONLY` list (names, rows, or `agree`'s finding objects)."""
+    rows = payload.get("disputed", []) if isinstance(payload, Mapping) else payload
+    if not isinstance(rows, list):
+        raise ValueError("OUSAST_INPUT_ONLY must be a JSON list (e.g. agree's disputed.json)")
+    names = set()
+    for row in rows:
+        if isinstance(row, Mapping):
+            names.add(str(row["candidate"]))
+        elif isinstance(row, list | tuple):
+            names.add(f"{row[0]}::{row[1]}")
+        else:
+            names.add(str(row))
+    return names
+
+
+def restrict(unique: Sequence[Candidate], only: set[str] | None) -> list[Candidate]:
+    """The candidates named by `only` (all of them when `only` is None)."""
+    return list(unique) if only is None else [c for c in unique if f"{c[0]}::{c[1]}" in only]
+
+
+class _NoModel:
+    """The client of a task with nothing to ask: any call is a defect, never a silent spend."""
+
+    def complete(self, **kw: Any) -> ChatResponse:
+        raise RuntimeError("no model call is expected: the candidate restriction is empty")
+
+
 def callers_of(facts: Mapping[str, Any] | None, path: str, function: str) -> list[dict[str, Any]]:
     """Known call sites of `path::function` from `facts.json`: the task-5 shape (`callers` keyed by
     `path::function`, rows `{"path", "line", "enclosing"}`) or the design's per-file shape (`{path: {"callers":
@@ -224,10 +254,12 @@ def run(
     model: str = "",
     pass_label: str = "a",
     family: str | None = None,
+    only: set[str] | None = None,
 ) -> dict[str, Any]:
     """Hunt every unfinished unit, append `units.jsonl` per hunt, write `summary.json` last; returns the summary.
 
-    `budget` is the meter to run under (built from `client` when absent); `client` is what the meter wraps.
+    `budget` is the meter to run under (built from `client` when absent); `client` is what the meter wraps;
+    `only` restricts the hunt to those `path::function` candidates.
     """
     output_dir.mkdir(parents=True, exist_ok=True)
     log, summary_path = output_dir / "units.jsonl", output_dir / "summary.json"
@@ -235,7 +267,7 @@ def run(
     model = model or getattr(metered, "model", "") or "unbound"
     recorded_family, unique = load_candidates(candidates)
     family = family or recorded_family
-    units = units_of(unique)
+    units = units_of(restrict(unique, only))
     rows = read_units(log)
     done = {unit_key(r["path"], r["candidates"]): r for r in rows if "error" not in r}
     status, reason = "done", ""
@@ -278,6 +310,8 @@ def run(
         "status": status, "units_done": len(done), "units_total": len(units), "usd": usd, "calls": calls, "usage": usage,
         "priced": metered.priced, "model": model, "pass": pass_label, "family": family,
     }  # fmt: skip
+    if only is not None:
+        summary["only"] = len(restrict(unique, only))
     if reason:
         summary["reason"] = reason
     summary_path.write_text(json.dumps(summary, indent=1) + "\n")
@@ -336,13 +370,16 @@ def main() -> int:
         params = json.loads(os.environ.get("OUSAST_MODEL_PARAMS", "") or "{}")
         candidates = json.loads(candidates_path.read_text())
         facts = json.loads(facts_path.read_text()) if facts_path is not None else None
-        client = build_client()
+        only_path = _env_path("OUSAST_INPUT_ONLY", required=False)
+        only = only_names(json.loads(only_path.read_text())) if only_path is not None else None
+        nothing = only is not None and not restrict(load_candidates(candidates)[1], only)
+        client: ChatClient = _NoModel() if nothing else build_client()
         metered = MeteredClient(
             client, prices=params, budget_usd=_env_number("OUSAST_BUDGET_USD", float), budget_calls=_env_number("OUSAST_BUDGET_CALLS", int)
         )
         summary = run(
             workspace, output_dir, candidates, facts, client=client, budget=metered, model=model, pass_label=pass_label,
-            family=os.environ.get("OUSAST_FAMILY") or None,
+            family=os.environ.get("OUSAST_FAMILY") or None, only=only,
         )  # fmt: skip
     except Exception:  # noqa: BLE001 -- a crash is `failed` with its traceback, never a silent zero
         summary = _failed(output_dir, traceback.format_exc()[-2000:], model, pass_label)
@@ -364,6 +401,8 @@ __all__ = [
     "hunt_prompt",
     "load_candidates",
     "main",
+    "only_names",
+    "restrict",
     "run",
     "units_of",
 ]

tests/test_plane_increment.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import hashlib
 import json
 import re
 import shutil
@@ -53,7 +54,7 @@ def test_model_prices_are_the_endpoint_prices() -> None:
 
 def test_validation_run_resolves_and_renders_every_task(tmp_path: Path) -> None:
     run, manifests = reconciler.load_run(RUN)
-    assert len(run.tasks) == 60
+    assert len(run.tasks) == 90
     population = {c["id"]: c for c in tomllib.loads(POPULATION.read_text())["case"]}
     ax_names: set[str] = set()
     for task in manifests.tasks.values():
@@ -69,7 +70,7 @@ def test_validation_run_resolves_and_renders_every_task(tmp_path: Path) -> None:
             for name, sha in workspace.pins.items():
                 case_id = binding.name.removesuffix("-vulnerable")
                 assert (name, sha) == ("repo", population[case_id]["vulnerable"])
-        if entry.name.endswith(("-va", "-vb")):
+        if entry.name.endswith(("-va", "-vb", "-vc")):
             assert model == "deepseek-flash" and entry.budget and entry.budget.usd and entry.budget.calls
             env = {e.name: e.value for e in task.env}
             assert env["OUSAST_PASS"] == entry.name[-1]
@@ -83,6 +84,34 @@ def test_validation_run_resolves_and_renders_every_task(tmp_path: Path) -> None:
         ax_names.add(docs[-1]["metadata"]["name"])
 
 
+# sha256 of the Run's entries as the first measurement ran them (plane-increment-1.json): the text after
+# ``  tasks:`` of plane/runs/validation-46.yaml at c5e6ed3. A rerun keeps their ``done`` state only while they match.
+FIRST_RUN_ENTRIES = "412ad6171cf096d5d7f6b8e7358374077e2beabf5f4dff35a2801d37c341b97e"
+
+
+def test_tiebreak_appends_thirty_tasks_and_keeps_the_first_sixty_byte_identical() -> None:
+    text = RUN.read_text().split("  tasks:\n", 1)[1]
+    first = text.split("  - name: budibase-mongo-template-nosqli-vc\n", 1)[0]
+    assert hashlib.sha256(first.encode()).hexdigest() == FIRST_RUN_ENTRIES
+    run, manifests = reconciler.load_run(RUN)
+    old, new = run.tasks[:60], run.tasks[60:]
+    assert [e.name.rsplit("-", 1)[1] for e in old] == ["facts", "va", "vb", "agree"] * 15
+    assert [e.name.rsplit("-", 1)[1] for e in new] == ["vc", "final"] * 15
+    for vc, final in zip(new[::2], new[1::2], strict=True):
+        case = vc.name.removesuffix("-vc")
+        agree_entry = run.task(f"{case}-agree")
+        assert vc.inputs == {"facts": f"{case}-facts/facts.json", "only": f"{case}-agree/disputed.json"}
+        assert vc.depends_on == (f"{case}-agree",) and vc.budget == run.task(f"{case}-va").budget
+        env = {e.name: e.value for e in manifests.tasks[vc.task].env}
+        assert env["OUSAST_PASS"] == "c" and env["OUSAST_INPUT_CANDIDATES"] == "/workspace/inputs/candidates.json"
+        assert final.task == agree_entry.task, "the final agree binds the same static inputs as the first"
+        assert final.inputs == {**agree_entry.inputs, "pass_c": f"{case}-vc/units.jsonl"}
+        assert set(final.producers) == {f"{case}-va", f"{case}-vb", f"{case}-vc"}
+        for entry in (vc, final):
+            name = reconciler._ax_name(run.metadata.name, entry.name, limit=reconciler.TASK_NAME_LIMIT)
+            assert len(name) <= reconciler.TASK_NAME_LIMIT
+
+
 def test_inputs_carry_the_validation_set() -> None:
     manifests = load_manifests(sorted((PLANE / "workspaces").glob("*-inputs.yaml")))
     assert len(manifests.workspaces) == 15

tests/test_plane_reconciler.py

@@ -448,6 +448,30 @@ def test_unfinished_stops_dependants_and_rerun_skips_done(fake: Fake, tmp_path:
     assert statuses("resume") == {"repo-facts": "done", "verify": "done", "agree": "done"}
 
 
+TIEBREAK = """
+    - name: tiebreak
+      task: verify
+      dependsOn: [agree]
+      inputs: {only: verify/agreed.json, facts: repo-facts/facts.json}
+      outputs: [agreed.json]
+      budget: {usd: 1, calls: 8}
+"""
+
+
+def test_a_rerun_of_a_run_that_gained_tasks_runs_only_the_new_ones(fake: Fake, tmp_path: Path) -> None:
+    manifest = write_run(tmp_path, "grow")
+    assert reconciler.run(manifest, ax=str(fake.ax)) == "done"
+    first = state_of("grow")["tasks"]
+    (fake.root / "ax.log").unlink()
+    fake.script({**SCRIPT, "tiebreak": SCRIPT["verify"]})
+    write_run(tmp_path, "grow", CHAIN + TIEBREAK)  # the same Run, one task appended (the tie-break pass)
+    assert reconciler.run(manifest, ax=str(fake.ax)) == "done"
+    assert [name for _, name in fake.events("apply")] == ["tiebreak"], "done tasks keep their state; only the new task runs"
+    assert statuses("grow") == {"repo-facts": "done", "verify": "done", "agree": "done", "tiebreak": "done"}
+    assert {k: v for k, v in state_of("grow")["tasks"].items() if k != "tiebreak"} == first
+    assert sorted(n for _, n in fake.events("fetched")) == ["tiebreak FACTS 200", "tiebreak ONLY 200"], "inputs of done producers"
+
+
 def test_failed_task_halts_the_run(fake: Fake, tmp_path: Path) -> None:
     script = json.loads(json.dumps(SCRIPT))
     failed = {"status": "failed", "units_done": 0, "units_total": 3, "reason": "402 insufficient balance"}

tests/test_plane_tiebreak.py

@@ -0,0 +1,150 @@
+"""The tie-break of the first increment's revisit (ai-service-plane Requirement 6.4, task 8.1): `verify` restricted
+to the candidates passes a and b disputed (`OUSAST_INPUT_ONLY`), and `agree` deciding 2-of-3 where pass c asked."""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+import pytest
+from test_plane_verify import CANDIDATES, FACTS, FLASH, HuntClient, _two_passes, _workspace
+
+from openultrasast.plane.budget import MeteredClient
+from openultrasast.plane.tasks import agree, verify
+from openultrasast.tool_hunter import CLIENT_ENV
+
+SITES = {"id": "case", "family": "injection", "sites": ["app.py::run", "db.py::query"], "ranges": {"app.py": [[40, 41]]}}
+
+
+def _verify(root: Path, out: Path, client: HuntClient, only: set[str] | None, label: str = "c") -> dict:
+    metered = MeteredClient(client, prices=FLASH)  # type: ignore[arg-type]
+    return verify.run(root, out, CANDIDATES, FACTS, client=client, budget=metered, model="deepseek-chat", pass_label=label, only=only)
+
+
+def test_only_restricts_the_hunt_to_the_named_candidates(tmp_path: Path) -> None:
+    client = HuntClient({"app.py::other"})
+    summary = _verify(_workspace(tmp_path), tmp_path / "c", client, {"app.py::other", "gone.py::f"})
+    assert summary["status"] == "done" and summary["units_total"] == 1 and summary["only"] == 1 and summary["pass"] == "c"
+    prompts = client.prompts()
+    assert prompts and all("`other`" in p and "`run`" not in p and "db.py" not in p for p in prompts)
+    rows = [json.loads(line) for line in (tmp_path / "c" / "units.jsonl").read_text().splitlines()]
+    assert [r["candidates"] for r in rows] == [[["app.py", "other", 5]]]
+    assert [f["candidate"] for f in rows[0]["flagged"]] == ["app.py::other"]
+
+
+def test_only_names_reads_disputed_json_rows_and_names() -> None:
+    disputed = [{"candidate": "a.py::f", "site": "a.py:1:f", "passes": 1}]
+    assert verify.only_names(disputed) == {"a.py::f"}
+    assert verify.only_names({"disputed": disputed}) == {"a.py::f"}
+    assert verify.only_names([["b.py", "g", 3], "c.py::h"]) == {"b.py::g", "c.py::h"}
+    with pytest.raises(ValueError):
+        verify.only_names("a.py::f")
+
+
+def test_an_empty_restriction_is_done_with_zero_units_and_no_call(tmp_path: Path) -> None:
+    client = HuntClient({"app.py::run"})
+    summary = _verify(_workspace(tmp_path), tmp_path / "c", client, set())
+    assert summary["status"] == "done" and (summary["units_done"], summary["units_total"], summary["calls"]) == (0, 0, 0)
+    assert client.calls == [] and (tmp_path / "c" / "units.jsonl").read_text() == ""
+
+
+def test_main_with_an_empty_only_list_needs_no_model_client(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
+    root, out = _workspace(tmp_path), tmp_path / "out"
+    (tmp_path / "candidates.json").write_text(json.dumps(CANDIDATES))
+    (tmp_path / "disputed.json").write_text("[]\n")
+    for name in ("DEEPSEEK_API_KEY", "OPENROUTER_API_KEY", CLIENT_ENV, "OUSAST_CLIENT"):
+        monkeypatch.delenv(name, raising=False)
+    env = {
+        "OUSAST_WORKSPACE_DIR": str(root), "OUSAST_OUTPUT_DIR": str(out), "OUSAST_INPUT_CANDIDATES": str(tmp_path / "candidates.json"),
+        "OUSAST_INPUT_ONLY": str(tmp_path / "disputed.json"), "OUSAST_MODEL": "deepseek-chat", "OUSAST_PASS": "c",
+    }  # fmt: skip
+    for key, value in env.items():
+        monkeypatch.setenv(key, value)
+    assert verify.main() == 0
+    summary = json.loads((out / "summary.json").read_text())
+    assert (summary["status"], summary["units_total"], summary["calls"], summary["only"]) == ("done", 0, 0, 0)
+
+
+def _pass_c(tmp_path: Path, root: Path, disputed: Path, flag: set[str]) -> list[dict]:
+    only = verify.only_names(json.loads(disputed.read_text()))
+    _verify(root, tmp_path / "pass-c", HuntClient(flag), only)
+    return agree.load_pass(tmp_path / "pass-c")
+
+
+def _final(tmp_path: Path, a: Path, b: Path, c: list[dict] | None) -> dict:
+    return agree.run(tmp_path / "final", CANDIDATES, agree.load_pass(a), agree.load_pass(b), sites=SITES, pass_c=c)
+
+
+def test_a_and_c_agreeing_flips_a_disputed_candidate_to_agreed(tmp_path: Path) -> None:
+    root, a, b = _two_passes(tmp_path)  # a: run, query; b: run, other -> disputed: other (b), query (a)
+    first = agree.run(tmp_path / "agree", CANDIDATES, agree.load_pass(a), agree.load_pass(b), sites=SITES)
+    c = _pass_c(tmp_path, root, tmp_path / "agree" / "disputed.json", {"db.py::query", "app.py::run"})
+    summary = _final(tmp_path, a, b, c)
+    report = json.loads((tmp_path / "final" / "agreed.json").read_text())
+    rows = {r["candidate"]: (r["a"], r["b"], r["c"], r["agreed"]) for r in report["candidates"]}
+    assert rows == {
+        "app.py::run": (True, True, None, True),  # not asked in c: the a/b rule
+        "app.py::other": (False, True, False, False),  # b alone
+        "db.py::query": (True, False, True, True),  # a + c
+    }
+    assert [f["candidate"] for f in report["agreed"]] == ["app.py::run", "db.py::query"]
+    assert [f["candidate"] for f in report["disputed"]] == ["app.py::other"]
+    metrics = summary["metrics"]
+    assert metrics["tiebreak"]["asked"] == 2 and metrics["tiebreak"]["flipped_to_agreed"] == 1
+    assert metrics["tiebreak"]["flipped_to_rejected"] == 0 and "2 of passes a, b, c" in metrics["tiebreak"]["rule"]
+    assert metrics["agreed"] == 2 and metrics["disputed"] == 1
+    assert first["metrics"]["declared_sites_matched_agreed"] == 1 and metrics["declared_sites_matched_agreed"] == 2
+    assert metrics["hunts"] == first["metrics"]["hunts"] + 2 and metrics["usd_total"] > first["metrics"]["usd_total"]
+    assert set(first["metrics"]) - {"tiebreak"} <= set(metrics)
+
+
+def test_b_and_c_agreeing_flips_to_agreed_and_a_lone_a_is_rejected(tmp_path: Path) -> None:
+    root, a, b = _two_passes(tmp_path)
+    agree.run(tmp_path / "agree", CANDIDATES, agree.load_pass(a), agree.load_pass(b), sites=SITES)
+    c = _pass_c(tmp_path, root, tmp_path / "agree" / "disputed.json", {"app.py::other"})
+    report_metrics = _final(tmp_path, a, b, c)["metrics"]
+    report = json.loads((tmp_path / "final" / "agreed.json").read_text())
+    assert [f["candidate"] for f in report["agreed"]] == ["app.py::other", "app.py::run"]
+    assert [f["candidate"] for f in report["disputed"]] == ["db.py::query"]
+    assert report_metrics["declared_sites_matched_agreed"] == 1 and report_metrics["tiebreak"]["flipped_to_agreed"] == 1
+
+
+def test_without_pass_c_the_a_b_rule_and_numbers_are_unchanged(tmp_path: Path) -> None:
+    _, a, b = _two_passes(tmp_path)
+    before = agree.run(tmp_path / "ab", CANDIDATES, agree.load_pass(a), agree.load_pass(b), sites=SITES)["metrics"]
+    empty = _final(tmp_path, a, b, [])["metrics"]
+    assert before["tiebreak"] == {"asked": 0, "flipped_to_agreed": 0, "flipped_to_rejected": 0, "rule": agree.RULE_A_B}
+    assert empty["tiebreak"]["rule"] == agree.RULE_2_OF_3 and empty["tiebreak"]["asked"] == 0
+    assert {k: v for k, v in empty.items() if k != "tiebreak"} == {k: v for k, v in before.items() if k != "tiebreak"}
+
+
+def test_pass_c_alone_never_decides(tmp_path: Path) -> None:
+    root = _workspace(tmp_path)
+    a, b = tmp_path / "pass-a", tmp_path / "pass-b"
+    for out in (a, b):
+        _verify(root, out, HuntClient({"app.py::run"}), None, label=out.name[-1])
+    c = tmp_path / "pass-c"
+    _verify(root, c, HuntClient({"app.py::other"}), None)  # asked about everything, flags only what a and b rejected
+    metrics = _final(tmp_path, a, b, agree.load_pass(c))["metrics"]
+    report = json.loads((tmp_path / "final" / "agreed.json").read_text())
+    assert [f["candidate"] for f in report["agreed"]] == ["app.py::run"]  # 2 of 3 without c's vote
+    assert [f["candidate"] for f in report["disputed"]] == ["app.py::other"]  # 1 of 3
+    assert metrics["tiebreak"] == {"asked": 3, "flipped_to_agreed": 0, "flipped_to_rejected": 0, "rule": agree.RULE_2_OF_3}
+
+
+def test_agree_main_reads_pass_c(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
+    root, a, b = _two_passes(tmp_path)
+    agree.run(tmp_path / "agree", CANDIDATES, agree.load_pass(a), agree.load_pass(b), sites=SITES)
+    _pass_c(tmp_path, root, tmp_path / "agree" / "disputed.json", {"db.py::query"})
+    (tmp_path / "candidates.json").write_text(json.dumps(CANDIDATES))
+    (tmp_path / "case.json").write_text(json.dumps(SITES))
+    env = {
+        "OUSAST_OUTPUT_DIR": str(tmp_path / "final"), "OUSAST_INPUT_PASS_A": str(a / "units.jsonl"),
+        "OUSAST_INPUT_PASS_B": str(b / "units.jsonl"), "OUSAST_INPUT_PASS_C": str(tmp_path / "pass-c" / "units.jsonl"),
+        "OUSAST_INPUT_CANDIDATES": str(tmp_path / "candidates.json"), "OUSAST_INPUT_SITES": str(tmp_path / "case.json"),
+    }  # fmt: skip
+    for key, value in env.items():
+        monkeypatch.setenv(key, value)
+    assert agree.main() == 0
+    metrics = json.loads((tmp_path / "final" / "summary.json").read_text())["metrics"]
+    assert metrics["agreed"] == 2 and metrics["tiebreak"]["flipped_to_agreed"] == 1