docs: cover the judged verification, recall filter and steering actions

- list the two thresholds the judgement points gained (``HARNESS_VERIFIER_SUPPORT_THRESHOLD``, ``MEMORY_RECALL_RELEVANCE_THRESHOLD``) and say that a threshold compares a rating where the point rates instead of asking
- document ``HARNESS_VERIFIER_STRATEGY`` next to the other strategies, and the
  actions the long-run and verification judgements pick
- complete the Harness environment table, which was missing the thresholds the
  previous judgement commit added
- document the recall filter in the long-term memory guide

Change-Id: I7c94a9c0c458ccc14133fc29e3369960a7a9d119

Author: wujiaming-ai

docs/content/docs/framework/memory/long-term/index.en.mdx

@@ -31,6 +31,8 @@ ltm = LongTermMemory(backend="viking", app_name="ltm_demo")
 | `index` | `str` | `""` | Index/collection name for storing memories. Falls back to `app_name`, then `default_app`. |
 | `app_name` | `str` | `""` | The owning application name; used for data isolation and as the `index` fallback. |
 | `user_id` | `str` | `""` | **Deprecated**, kept only for backward compatibility. |
+| `recall_strategy` | `str` | `"off"` | Recall judgement: with `decision`, the decision model drops memories unrelated to the request. |
+| `recall_relevance_threshold` | `float` | `0.5` | Recall relevance threshold: memories judged below it are not returned. |
 
 <Callout type="info">
 Vector backends (`local`, `opensearch`, `redis`) embed memories, which requires `pip install "veadk-python[extensions]"` and an embedding model (env prefix `MODEL_EMBEDDING_`, falling back to `MODEL_AGENT_API_KEY`). `viking`, `mem0`, `openviking`, and `tos_context` are managed services and need no local embedding.
@@ -183,6 +185,7 @@ The default is `MEMORY_SAVE_STRATEGY=threshold`, the two thresholds above. With
 | Unavailable | Falls back to `MIN_MESSAGES_THRESHOLD` / `MIN_TIME_THRESHOLD` |
 
 A skipped turn does not advance the save cursor, so the next accepted judgement writes those events together: nothing is lost. See the [environment variable reference](/references/configuration/environment-variables) for the decision model variables.
+Recall can be judged the same way: with `MEMORY_RECALL_STRATEGY=decision`, `search_memory` rates every returned memory (irrelevant / related but useless / useful background / required) and drops the ones below `MEMORY_RECALL_RELEVANCE_THRESHOLD` (default `0.5`). One judgement covers at most 20 memories, anything beyond that is returned in backend order, and an unavailable judgement keeps every match.
 ## Auto-save Memory Policy
 Configure `auto_save_memory_policy` on `Agent` to decide which events are persisted by automatic long-term-memory saving. If omitted, it is equivalent to `"default"`.
 ```python

docs/content/docs/framework/memory/long-term/index.mdx

@@ -31,6 +31,8 @@ ltm = LongTermMemory(backend="viking", app_name="ltm_demo")
 | `index` | `str` | `""` | 存储记忆所用的索引/集合名。为空时回退到 `app_name`，再为空则用 `default_app`。 |
 | `app_name` | `str` | `""` | 拥有该记忆的应用名，常用作数据隔离与 `index` 的回退值。 |
 | `user_id` | `str` | `""` | **已废弃**，仅为向后兼容保留。 |
+| `recall_strategy` | `str` | `"off"` | 召回判定策略；`decision` 时由判定模型过滤与本次请求无关的记忆。 |
+| `recall_relevance_threshold` | `float` | `0.5` | 召回相关度阈值；判定低于该值的记忆不返回。 |
 
 <Callout type="info">
 向量类后端（`local`、`opensearch`、`redis`）会对记忆做向量化（embedding），需要安装扩展依赖：`pip install "veadk-python[extensions]"`，并配置 embedding 模型（环境变量前缀 `MODEL_EMBEDDING_`，缺省时复用 `MODEL_AGENT_API_KEY`）。`viking`、`mem0`、`openviking` 与 `tos_context` 是托管服务，无需本地 embedding。
@@ -183,6 +185,7 @@ agent = Agent(
 | 判定不可用 | 回落到 `MIN_MESSAGES_THRESHOLD` / `MIN_TIME_THRESHOLD` |
 
 被跳过的轮次不会推进保存游标，后续判定通过时会把这批 event 一起写入，因此不会丢数据。判定模型的环境变量见 [环境变量参考](/references/configuration/environment-variables)。
+检索侧同样可以交给判定模型：`MEMORY_RECALL_STRATEGY=decision` 时，`search_memory` 返回前会逐条给召回片段打相关度（不相关 / 同主题但用不上 / 有用背景 / 必须遵守），低于 `MEMORY_RECALL_RELEVANCE_THRESHOLD`（默认 `0.5`）的会被丢弃；一次判定最多覆盖 20 条，超出的按后端顺序原样保留，判定不可用时保留全部结果。
 ## 自动保存记忆策略
 开发者在 `Agent` 上配置 `auto_save_memory_policy` 来控制自动保存长期记忆时哪些 event 会被写入；不配置时等价于 `"default"`。
 ```python

docs/content/docs/references/configuration/environment-variables.en.mdx

@@ -79,6 +79,11 @@ Prefix `HARNESS_`, used to attach optional Harness plugins to HarnessApp Runtime
 | `HARNESS_COMPACTION_STRATEGY` | Compaction candidate strategy, `builtin` or `decision`; default `builtin`. |
 | `HARNESS_LONG_RUN_STRATEGY` | Long-run steering strategy, `counter` or `decision`; default `counter`. |
 | `HARNESS_MODE_STRATEGY` | Context mode-block strategy, `keywords` or `decision`; default `keywords`. |
+| `HARNESS_VERIFIER_STRATEGY` | Final-answer verification, `deterministic` or `decision`; default `deterministic`. |
+| `HARNESS_COMPACTION_KEEP_THRESHOLD` | Compaction candidates: a candidate is kept when the judged probability is at or above this value; default `0.5`. |
+| `HARNESS_LONG_RUN_READY_THRESHOLD` | Long-run steering: guidance is injected when the judged probability of being ready is at or above this value; default `0.5`. |
+| `HARNESS_MODE_DECISION_THRESHOLD` | Context mode blocks: a block is injected when the judged probability is at or above this value; default `0.5`. |
+| `HARNESS_VERIFIER_SUPPORT_THRESHOLD` | Final-answer support: the answer fails when the judged support is below this value; default `0.5`. |
 
 The `harness_enhance` block maps to these environment variables when deploying a HarnessApp Runtime. Prefer `harness.yaml` or `veadk agentkit invoke` flags for normal developer workflows; use environment variables for platform integration and container runtimes.
 

docs/content/docs/references/configuration/environment-variables.mdx

@@ -79,6 +79,11 @@ volcengine:
 | `HARNESS_COMPACTION_STRATEGY` | 工具结果压缩候选策略，`builtin` 或 `decision`，默认 `builtin`。 |
 | `HARNESS_LONG_RUN_STRATEGY` | 长任务收尾引导策略，`counter` 或 `decision`，默认 `counter`。 |
 | `HARNESS_MODE_STRATEGY` | 上下文模式块策略，`keywords` 或 `decision`，默认 `keywords`。 |
+| `HARNESS_VERIFIER_STRATEGY` | 最终回答校验策略，`deterministic` 或 `decision`，默认 `deterministic`。 |
+| `HARNESS_COMPACTION_KEEP_THRESHOLD` | 压缩候选保留阈值，默认 `0.5`；判定保留概率低于该值即压缩。 |
+| `HARNESS_LONG_RUN_READY_THRESHOLD` | 长任务收尾阈值，默认 `0.5`；判定可收尾概率高于该值即注入引导。 |
+| `HARNESS_MODE_DECISION_THRESHOLD` | 上下文模式块阈值，默认 `0.5`；判定概率高于该值即注入对应模式块。 |
+| `HARNESS_VERIFIER_SUPPORT_THRESHOLD` | 最终回答支撑度阈值，默认 `0.5`；判定支撑度低于该值即判为失败。 |
 
 `harness_enhance` 配置块会在 HarnessApp Runtime 部署时映射为这些环境变量。推荐开发者优先通过 `harness.yaml` 或 `veadk agentkit invoke` 参数启用，环境变量适合平台集成和镜像运行时。
 

docs/extensions/harness/README.md

@@ -211,6 +211,8 @@ veadk agentkit invoke \
 | `HARNESS_MAX_CONTEXT_CHARS` | `24000` | Context compaction threshold. |
 | `HARNESS_MAX_TOOL_RESULT_CHARS` | `4000` | Tool-result compaction threshold. |
 | `HARNESS_VERIFIER_MODE` | `observe` | Verification behavior: `observe` or `block`. |
+| `HARNESS_VERIFIER_STRATEGY` | `deterministic` | Final-answer verification: `deterministic` or `decision`. |
+| `HARNESS_VERIFIER_SUPPORT_THRESHOLD` | `0.5` | Support rating below which the answer fails. |
 | `HARNESS_STORE_PATH` | unset | Uses a JSONL event store when set. |
 | `HARNESS_COMPACTION_STRATEGY` | `builtin` | Compaction candidates: `builtin` or `decision`. |
 | `HARNESS_LONG_RUN_STRATEGY` | `counter` | Long-run steering: `counter` or `decision`. |
@@ -221,7 +223,7 @@ veadk agentkit invoke \
 
 ## Decision Model Strategies
 
-The three `*_STRATEGY=decision` settings replace a rule with a judgement from
+The four `*_STRATEGY=decision` settings replace a rule with a judgement from
 the configured decision model. They need `DECISION_MODEL_ENABLED=true` and an
 API key; without one, each strategy keeps its rule and logs a warning.
 
@@ -230,6 +232,7 @@ API key; without one, each strategy keeps its rule and logs a warning.
 | `HARNESS_COMPACTION_STRATEGY` | Role and size based compaction candidates | Builtin rules |
 | `HARNESS_LONG_RUN_STRATEGY` | Model-call counter | Counter, forced after the unconditional count |
 | `HARNESS_MODE_STRATEGY` | Precision and artifact keyword markers | Keyword markers |
+| `HARNESS_VERIFIER_STRATEGY` | Completion markers plus a successful-receipt check | Builtin rules |
 
 ### Judgement Thresholds
 
@@ -244,6 +247,13 @@ back to `0.5` for an unusable one.
 | `HARNESS_COMPACTION_KEEP_THRESHOLD` | `0.5` | Keeps more tool output verbatim |
 | `HARNESS_LONG_RUN_READY_THRESHOLD` | `0.5` | Steers a run toward its answer sooner |
 | `HARNESS_MODE_DECISION_THRESHOLD` | `0.5` | Injects the mode block more often |
+| `HARNESS_VERIFIER_SUPPORT_THRESHOLD` | `0.5` | Requires more evidence before the answer passes |
+
+A judgement also picks an action: long-run steering chooses `narrow_scope` /
+`nudge_to_finish` / `force_finish` to shape the injected guidance, and
+verification chooses `retry_tool_call` / `soften_claim` / `drop_claim` /
+`ask_user` to shape the repair instruction. An unusable action keeps the
+default wording while the rating still applies.
 
 ## Compaction Providers
 

docs/extensions/harness/README.zh.md

@@ -203,6 +203,8 @@ veadk agentkit invoke \
 | `HARNESS_MAX_CONTEXT_CHARS` | `24000` | 上下文压缩阈值。 |
 | `HARNESS_MAX_TOOL_RESULT_CHARS` | `4000` | 工具结果压缩阈值。 |
 | `HARNESS_VERIFIER_MODE` | `observe` | 校验行为，支持 `observe` 或 `block`。 |
+| `HARNESS_VERIFIER_STRATEGY` | `deterministic` | 最终回答校验策略：`deterministic` 或 `decision`。 |
+| `HARNESS_VERIFIER_SUPPORT_THRESHOLD` | `0.5` | 最终回答支撑度阈值；判定低于该值即判为失败。 |
 | `HARNESS_STORE_PATH` | 未设置 | 设置后使用 JSONL event store。 |
 | `HARNESS_COMPACTION_STRATEGY` | `builtin` | 压缩候选策略：`builtin` 或 `decision`。 |
 | `HARNESS_LONG_RUN_STRATEGY` | `counter` | 长任务引导策略：`counter` 或 `decision`。 |
@@ -213,13 +215,14 @@ veadk agentkit invoke \
 
 ## 判定模型策略
 
-三个 `*_STRATEGY=decision` 开关把一条规则换成判定模型的判定结果，需要 `DECISION_MODEL_ENABLED=true` 与 API Key；没有配置时各自保留原规则并打印告警。
+四个 `*_STRATEGY=decision` 开关把一条规则换成判定模型的判定结果，需要 `DECISION_MODEL_ENABLED=true` 与 API Key；没有配置时各自保留原规则并打印告警。
 
 | 策略 | 被替代的规则 | 判定不可用时 |
 | --- | --- | --- |
 | `HARNESS_COMPACTION_STRATEGY` | 按角色和长度挑选压缩候选 | 内置规则 |
 | `HARNESS_LONG_RUN_STRATEGY` | 仅按模型调用次数计数 | 计数规则，超过强制次数后必定生效 |
 | `HARNESS_MODE_STRATEGY` | 精度/产物关键词匹配 | 关键词匹配 |
+| `HARNESS_VERIFIER_STRATEGY` | 完成类关键词加「有无成功回执」 | 内置规则 |
 
 ### 判定阈值
 
@@ -230,6 +233,9 @@ veadk agentkit invoke \
 | `HARNESS_COMPACTION_KEEP_THRESHOLD` | `0.5` | 更多工具结果原样保留 |
 | `HARNESS_LONG_RUN_READY_THRESHOLD` | `0.5` | 更早把运行推向收尾 |
 | `HARNESS_MODE_DECISION_THRESHOLD` | `0.5` | 更频繁注入模式块 |
+| `HARNESS_VERIFIER_SUPPORT_THRESHOLD` | `0.5` | 要求更充分的证据才放行回答 |
+
+判定还会选动作：长任务引导可选 `narrow_scope` / `nudge_to_finish` / `force_finish` 决定注入的引导文案，最终回答校验可选 `retry_tool_call` / `soften_claim` / `drop_claim` / `ask_user` 决定修复指引；动作不可用时保留默认文案，评级仍然生效。
 
 ## 压缩 Provider
 

veadk/extensions/decisions/README.md

@@ -138,15 +138,18 @@ The judged state and the API key are never logged.
 
 ## Judgement Thresholds
 
-Every decision point keeps its own threshold, compared against the probability
-of "yes" in `[0, 1]`:
+Every decision point keeps its own threshold, compared against the answer of
+its own judgement on `[0, 1]` — the probability of "yes" for a yes/no question,
+the rated position for a rating:
 
 | Decision point | Setting | Default |
 | --- | --- | --- |
 | Compaction candidates | `HARNESS_COMPACTION_KEEP_THRESHOLD` | 0.5 |
 | Long-run steering | `HARNESS_LONG_RUN_READY_THRESHOLD` | 0.5 |
 | Context mode blocks | `HARNESS_MODE_DECISION_THRESHOLD` | 0.5 |
 | Long-term memory saves | `MEMORY_SAVE_WORTH_THRESHOLD` | 0.5 |
+| Final-answer support | `HARNESS_VERIFIER_SUPPORT_THRESHOLD` | 0.5 |
+| Long-term memory recall | `MEMORY_RECALL_RELEVANCE_THRESHOLD` | 0.5 |
 
 Parsing goes through `probability_threshold()`, which **clamps** an
 out-of-range value instead of falling back (`1.5 → 1.0`, `-1 → 0.0`, keeping

veadk/extensions/decisions/README.zh.md

@@ -127,14 +127,17 @@ agent = Agent(name="router", tools=[decision_evaluate])
 
 ## 判定阈值
 
-每个判定点各自持有阈值，比较的都是「是」的概率在 `[0, 1]` 上的取值：
+每个判定点各自持有阈值，比较的都是各自判定结果在 `[0, 1]` 上的取值（是否类问题是
+「是」的概率，评分类问题是加权位置）：
 
 | 判定点 | 阈值 | 默认 |
 | --- | --- | --- |
 | 压缩候选 | `HARNESS_COMPACTION_KEEP_THRESHOLD` | 0.5 |
 | 长任务引导 | `HARNESS_LONG_RUN_READY_THRESHOLD` | 0.5 |
 | 上下文模式块 | `HARNESS_MODE_DECISION_THRESHOLD` | 0.5 |
 | 记忆落库 | `MEMORY_SAVE_WORTH_THRESHOLD` | 0.5 |
+| 最终回答支撑度 | `HARNESS_VERIFIER_SUPPORT_THRESHOLD` | 0.5 |
+| 长期记忆召回 | `MEMORY_RECALL_RELEVANCE_THRESHOLD` | 0.5 |
 
 解析统一走 `probability_threshold()`：越界的值**夹紧**而不是回落（`1.5 → 1.0`、
 `-1 → 0.0`，保留「永不生效 / 总是生效」的原意，回落会把行为整个翻转）；`NaN`

veadk/extensions/harness/README.md

@@ -104,18 +104,32 @@ each one keeps its rule and logs a warning.
 | Compaction candidates | `HARNESS_COMPACTION_STRATEGY=decision` | Role and size based candidate selection | Builtin rules |
 | Long-run steering | `HARNESS_LONG_RUN_STRATEGY=decision` | Model-call counter | Counter, and always after `unconditional_after_model_calls` |
 | Context mode blocks | `HARNESS_MODE_STRATEGY=decision` | Precision and artifact keyword markers | Keyword markers |
+| Final-answer support | `HARNESS_VERIFIER_STRATEGY=decision` | Completion markers plus a successful-receipt check | Builtin rules |
 
 Every judgement asks for the probability of "yes" in `[0, 1]`, and each point
 keeps its own threshold: raising one point's bar does not raise the others',
-because the same probability costs each point a different thing. Each setting
-accepts a `HARNESS_ENHANCE_`-prefixed alias, clamps out-of-range values, and
-falls back to `0.5` for an unusable one.
+because the same answer costs each point a different thing. Two points rate
+instead of asking yes/no, and their thresholds compare that rating. Each
+setting accepts a `HARNESS_ENHANCE_`-prefixed alias, clamps out-of-range
+values, and falls back to `0.5` for an unusable one.
 
 | Threshold | Setting | What a high value means |
 | --- | --- | --- |
 | Compaction candidates | `HARNESS_COMPACTION_KEEP_THRESHOLD=0.5` | Keeps more tool output verbatim |
 | Long-run steering | `HARNESS_LONG_RUN_READY_THRESHOLD=0.5` | Steers a run toward its answer sooner |
 | Context mode blocks | `HARNESS_MODE_DECISION_THRESHOLD=0.5` | Injects the mode block more often |
+| Final-answer support | `HARNESS_VERIFIER_SUPPORT_THRESHOLD=0.5` | Requires more evidence before the answer passes |
+
+Two strategies also choose an action instead of only crossing a threshold, and
+the action shapes what the plugin injects:
+
+| Strategy | Action it picks | Effect |
+| --- | --- | --- |
+| Long-run steering | `narrow_scope` / `nudge_to_finish` / `force_finish` | Replaces the injected guidance with the one that fits the trajectory |
+| Final-answer support | `retry_tool_call` / `soften_claim` / `drop_claim` / `ask_user` | Fills the repair instruction the caller hands back to the model |
+
+An action that names no known option keeps the default wording; the rating it
+came with is still used.
 
 They need a configured decision model; see
 [decisions](../decisions/README.md) for the `DECISION_MODEL_*` variables. A
@@ -125,7 +139,8 @@ Assembling plugins in code selects the same strategies as arguments instead of
 environment variables: `compaction_config=ToolResultCompactorConfig(strategy="decision")`,
 `context_config=HarnessInvocationContextConfig(mode_strategy="decision")`, and
 `long_run_strategy="decision"` / `long_run_ready_threshold=0.5` on
-`HarnessExtension`. Passing an `env` mapping instead makes the environment
+`HarnessExtension`, plus `verifier_config=FinalResponseVerifierConfig(strategy="decision", support_threshold=0.5)`.
+Passing an `env` mapping instead makes the environment
 variables the only source, as `HarnessExtension.from_env()` does.
 
 ## Direct Module Usage

veadk/extensions/harness/README.zh.md

@@ -99,23 +99,36 @@ harness_enhance:
 | 压缩候选 | `HARNESS_COMPACTION_STRATEGY=decision` | 按角色和长度挑选压缩候选 | 内置规则 |
 | 长任务引导 | `HARNESS_LONG_RUN_STRATEGY=decision` | 仅按模型调用次数计数 | 计数规则，并在 `unconditional_after_model_calls` 后强制生效 |
 | 上下文模式块 | `HARNESS_MODE_STRATEGY=decision` | 精度/产物关键词匹配 | 关键词匹配 |
+| 最终回答校验 | `HARNESS_VERIFIER_STRATEGY=decision` | 完成类关键词加「有无成功回执」 | 内置规则 |
 
 判定返回的是「是」在 `[0, 1]` 上的概率，每个点各自持有阈值：同一个概率落在不同点上
-代价不同，所以调高一个点的门槛不会抬高其它点。三个设置都接受 `HARNESS_ENHANCE_`
-前缀的别名，越界的值会被夹紧，不可用的值回落到 `0.5`。
+代价不同，所以调高一个点的门槛不会抬高其它点。两个点返回的是评级而不是是否，阈值
+比较的就是该评级。每个设置都接受 `HARNESS_ENHANCE_` 前缀的别名，越界的值会被夹紧，
+不可用的值回落到 `0.5`。
 
 | 阈值 | 开关 | 值调高意味着 |
 | --- | --- | --- |
 | 压缩候选 | `HARNESS_COMPACTION_KEEP_THRESHOLD=0.5` | 更多工具结果原样保留 |
 | 长任务引导 | `HARNESS_LONG_RUN_READY_THRESHOLD=0.5` | 更早把运行推向收尾 |
 | 上下文模式块 | `HARNESS_MODE_DECISION_THRESHOLD=0.5` | 更频繁注入模式块 |
+| 最终回答校验 | `HARNESS_VERIFIER_SUPPORT_THRESHOLD=0.5` | 要求更充分的证据才放行回答 |
+
+两个策略除了过阈值还会选动作，动作决定注入内容：
+
+| 策略 | 可选动作 | 影响 |
+| --- | --- | --- |
+| 长任务引导 | `narrow_scope` / `nudge_to_finish` / `force_finish` | 用贴合当前轨迹的引导替换固定文案 |
+| 最终回答校验 | `retry_tool_call` / `soften_claim` / `drop_claim` / `ask_user` | 组装交回主模型的修复指引 |
+
+判定返回未知动作时保留默认文案，同一次判定里的评级仍然生效。
 
 策略依赖已配置的判定模型，环境变量见 [decisions](../decisions/README.zh.md)。判定失败会回落到上表规则，不会让运行失败。
 
 用代码装配插件时，同样的选择通过参数传入，而不是环境变量：
 `compaction_config=ToolResultCompactorConfig(strategy="decision")`、
 `context_config=HarnessInvocationContextConfig(mode_strategy="decision")`、
-`HarnessExtension(long_run_strategy="decision", long_run_ready_threshold=0.5)`。
+`HarnessExtension(long_run_strategy="decision", long_run_ready_threshold=0.5)`、
+`verifier_config=FinalResponseVerifierConfig(strategy="decision", support_threshold=0.5)`。
 一旦传入 `env` 映射，就以环境变量为唯一来源（`HarnessExtension.from_env()` 即这种形态）。
 
 ## 直接使用模块