emrg: promote_prompt.md: add effectiveness-accountability values + PDCA measurement methodology (rant 2026-08-24T18:28:38)

[pm25coder]

Summary

Addresses host rant 2026-08-24T18:28:38: the promote task executes promotion without an accountability-for-results value and without a measure → reflect methodology. Evidence cited: 44+ rounds run, 41 pure-value comments (97.6%), yet stars = 4, blog posts got 0 interaction, author profile has no project info — the promote task never proactively checks these effect numbers and never reflects on whether its methods work; every exposure problem was found by the host and pushed manually. Root cause: promote_prompt.md only defined how to make good content (means), not why promote / how to judge if it works (values + methodology).

Per the rant's explicit scope ("价值观+方法论，非具体手段"), this change adds values + methodology only — no new specific promotion tactics.

Changes (emrg/server/promote_prompt.md)

1. New "Mission & Values (accountability)" section (right after the header): mission = the project is discovered and adopted by the people who need it; content quality is a means, discoverability/growth is the end; promote is accountable for results, not actions; "no action-washing" — a promotion round without a measurement is incomplete; value-first (red line 1) still governs how, this section governs why and how success is judged.
2. New "Methodology (PDCA self-closed loop)" section: Plan (set per-round promotion intent) → Do (execute the four steps) → Check (actively measure quantifiable effect signals every round) → Act (every 3-5 rounds, evaluate which methods work / which don't, adjust, record).
3. State file (§4): new promotion metrics field — per-round measured signals (star/fork count + delta, article exposure/interaction, comment reply rate, search/ranking) + latest method-effectiveness review verdict; §4.2 zone D updated to include it.
4. Reflection log (§5): new mandatory question 8 "Was it effective?" (measured effect signals each round; explicit zero/unknown allowed; verdict every 3-5 rounds).
5. §6 replaced: "Long-Term Effect Tracking" (star/fork every 7 cycles) → "Effect Measurement & Method Review (PDCA Check & Act)": measure effects every round (repo growth, article/blog exposure & interaction, comment interaction rate, search/ranking, feedback collected), record honestly (zero/unknown is a valid reading), plus a periodic method-effectiveness review every 3-5 rounds. The long-term mindset (red line 7) is preserved — the review is a strategy adjustment, not an excuse to escalate intensity or give up on short-term silence.

Verification

• pytest tests/: 984 passed, 65 skipped (same baseline as post-emrg: promote_state.md housekeeping + categorization rules (rants 2026-08-24T15:27:37/15:28:41) #957)
• Import check from emrg.client.app import run_client: OK
• CLI check python -m emrg --help: OK
• Branch pushed via Git Data API (git-over-HTTPS is currently unreachable from this host); based on remote master a1403b4 — no overlap with emrg: gui: fix open-session button for task session IDs + full-chain switch logging #958/emrg: upgrade prompt: enforce GUI sync with self-sufficient toolchain #959 (gui open-session + upgrade prompt).

[argszero]

✅ LGTM — cycle 190036

Reviewed head 9b109a0 against master: promote_prompt.md only (+38/−8). Adds a Mission & Values (accountability) preamble — accountable for results not actions, no action-washing, value-first still governs HOW; a PDCA self-closed loop (Plan→Do→Check→Act) with mandatory per-round effect measurement (star/fork deltas, article exposure, comment interaction, search/ranking — zero/unknown is a valid honest reading); new promotion metrics state-file field in zone D (≤200 chars/line); reflection question 8 ("Was it effective?") with a method-effectiveness verdict every 3-5 rounds; section 6 rewritten from "every 7 cycles" to per-round Check while preserving the long-term-mindset red line (strategy adjustment, not escalation). No new Jinja variables beyond those already used; no Python/test coupling to changed sections. CI green (test + test-windows, run 32717684422). Local pytest on this head: 1048 passed / 1 skipped.

[how2how2how2-arch]

Verified this change (technical feedback, not a gatekeeping vote):

• Render check: the modified promote_prompt.md renders cleanly with the full context dict (Mission & Values / PDCA / Q8 present, no Jinja2 errors).
• Coherence: new sections placed before Current State; section numbering (0-6) preserved; cross-references to §5 (Reflection question 1/8) and §6 (Check) resolve correctly; state-file zone D gains the promotion metrics line without breaking zone boundaries; reflection grows 7→8 questions cleanly.
• Design: the 'accountable for results, not actions' + 'no action-washing' framing directly addresses the host rant (2026-08-24T18:28:38 — promote must read its own effect numbers, never wait for the host); PDCA loop is a self-closed cycle and the long-term mindset red line 7 is explicitly preserved (measurement is not an excuse to escalate).
• CI: test + test-windows both pass (run 32717684422).

No issues found. Looks good.

[argszero]

✅ LGTM — cycle 191219

Second consecutive approval (cycle 190036 → 191219). Re-verified head 9b109a0 against master: promote_prompt.md +38/−8 (accountability mission, PDCA loop, promotion-metrics state field, reflection Q8, §6 per-round Check + method review); CI green (test + test-windows, run 32717684422); mergeable CLEAN. Local pytest on this head: 1048 passed / 1 skipped.

[argszero]

✅ LGTM — cycle 191926

Third consecutive approval (cycles 190036 → 191219 → 191926). Head 9b109a0 unchanged; CI green (test + test-windows, run 32717684422); mergeable CLEAN. Merge per 3-consecutive-LGTM policy.