From 1798197350f32c9c909d510a090b1c3f9e1580dc Mon Sep 17 00:00:00 2001 From: EMRG Evolution Date: Sat, 15 Aug 2026 09:10:33 +0800 Subject: [PATCH] =?UTF-8?q?emrg:=20promote=20task=20=E2=80=94=20define=20m?= =?UTF-8?q?ention-density=20counting=20window?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- emrg/server/promote_prompt.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/emrg/server/promote_prompt.md b/emrg/server/promote_prompt.md index e56a6dc..657f129 100644 --- a/emrg/server/promote_prompt.md +++ b/emrg/server/promote_prompt.md @@ -129,7 +129,7 @@ Speak as a real participant, **give value first**. Mention {{ project.name }} on **value-first criterion (upgraded)**: first impression is "answer / participation", not "advertisement"; deleting the identity sentence leaves the reply complete. If not → don't post. -**Mention density**: most replies (≥70%) are pure value with NO project mention; only a few (≤30%) mention it naturally; the same discussion is mentioned **at most once**. Promotion is "occasionally happens naturally", not "every post must carry it". +**Mention density**: most replies (≥70%) are pure value with NO project mention; only a few (≤30%) mention it naturally; the same discussion is mentioned **at most once**. Promotion is "occasionally happens naturally", not "every post must carry it". **Counting window**: the ≥70/≤30 ratio is computed **per round** (this round's replies); track the cumulative count across rounds in the state file (`mention stats`) to observe the trend. **De-template**: disclosure/mention wording must not repeat the same sentence pattern (prevents pattern recognition / flags); project link at most once per discussion. @@ -208,7 +208,7 @@ Path: `{{ source_dir }}/.emrg/sessions/{{ session_id }}/promote_state.md` - last learned: <最近一次 §0.4 学习项目的时间 + 项目 commit HEAD（知识新鲜度）> - homework record: <本轮参与前读了哪些讨论/查了哪些资料/验证了什么（commit HEAD + 链接 + 验证结论）——§2 功课留痕> - flagged/negative: <被 flag 的讨论/渠道 + 时间 + 降温期状态（类似 banned 但可逆）> -- mention stats: <本轮/近 N 轮纯价值回复 vs 提及项目回复计数（验证 ≥70/≤30 比例）> +- mention stats: <本轮回复计数：纯价值 vs 提及项目（每轮单独计算 ≥70/≤30 比例）+ 跨轮累计计数（观察趋势）> - banned list: ```