docs(cmo): finalize quality-upgrade wiring, decisions, limits, and cutover evidence

Author: ccross2

ops/cmo-automation/README.md

@@ -16,8 +16,9 @@ What this does today
 6) Build execution report from approved actions (dry-run default)
 
 Current mode
-- Dry-run / assisted only.
-- No autonomous posting in this module yet.
+- Quality-gated assisted automation with optional live execution.
+- Root posts can execute live.
+- Quote-style engagement is subject to X platform eligibility constraints and may fail with 403 when not in-thread/not mentioned.
 
 Structure
 - config/cmo_accounts.yaml
@@ -32,6 +33,7 @@ Structure
 - reports/CMO-AUTOMATION-IMPLEMENTATION-PLAN.md
 - reports/CMO-CUTOVER-48H-RUNBOOK.md
 - reports/CMO-ECOSYSTEM-MARKETING-BRIEF.md
+- reports/CMO-QUALITY-UPGRADE-2026-04-02.md
 
 Quick start
 1) Export credentials (or source ~/.claude/secrets.env)
@@ -63,7 +65,9 @@ Guardrails
 - Repetitive opener patterns are penalized.
 - Per-account recommended caps are enforced by policy review before execution.
 - Voice style enforcement for generated copy: no exclamation marks, no emojis, no em/en dashes.
+- Additional quality constraints: no hashtags, no links in replies, no duplicate target tweet IDs, no duplicate reply patterns.
+- Context-specificity gate blocks generic or low-signal reply candidates.
 
 Notes
-- Generated JSON/analysis artifacts are ignored in git by default.
-- Commit strategy/docs/scripts, not volatile runtime snapshots.
+- Standard practice is to commit strategy/docs/scripts and avoid volatile runtime snapshots.
+- For major cutovers or upgrades, commit representative execution artifacts to preserve auditability of behavior changes.

ops/cmo-automation/reports/CMO-AUTOMATION-IMPLEMENTATION-PLAN.md

@@ -153,7 +153,26 @@ Weekly review output
 
 ---
 
+## Phase 6 — Quality Hardening Upgrade (completed 2026-04-02)
+
+Implemented
+- Policy-level constraints: no hashtags, no links in replies, unique contextual replies.
+- Hydration-level enforcement for copy/style/duplication.
+- Duplicate suppression across both target tweet IDs and reply text patterns.
+- Context-specificity gate and low-signal source filtering.
+- Mission alignment gate wired via operating policy + ecosystem brief.
+
+Measured outcome (latest cycle)
+- 18 approved actions
+- 16 hydrated and ready after quality gating
+- 2 blocked (low signal or duplicate target)
+- 0 hashtag/link/duplicate violations in hydrated replies
+
+Observed platform constraint
+- Live quote-style engagement can fail with HTTP 403 when account is not mentioned or in-thread for the target post.
+
 Status
-- Baseline system scaffold is implemented under:
+- Baseline + quality-hardening scaffold is implemented under:
   ~/cDesign/sovren-website/ops/cmo-automation
-- Ready for day-1 operation and playbook reconstruction.
+- Upgrade details, decision log, limitations, and remaining work:
+  reports/CMO-QUALITY-UPGRADE-2026-04-02.md

ops/cmo-automation/reports/CMO-CUTOVER-48H-RUNBOOK.md

@@ -10,7 +10,7 @@ Decisions captured
 - Automation: hybrid-autonomous, including founder account
 - Targeting: 80% whitelist core + 20% discovery
 - Founder policy: strict deny-list (no unrelated consumer-brand/meme/politics/price-gambling engagement)
-- Fallback on failures/rate-limits: root posts only
+- Fallback on failures/rate-limits/platform eligibility failures: root posts only
 
 ## T-48h (start now)
 1) Run full pipeline and verify outputs:
@@ -34,7 +34,11 @@ Decisions captured
 ## T-0h (Fiverr off)
 1) Disable Fiverr service.
 2) Keep in-house automation active with current caps.
-3) Post-hoc audit sample (minimum 20% of sent actions).
+3) Run execution preflight:
+   - verify no hashtag/link violations
+   - verify no duplicate targets/reply patterns
+   - verify quote eligibility assumptions for candidate targets
+4) Post-hoc audit sample (minimum 20% of sent actions).
 
 ## T+24h and T+48h
 1) Rebuild playbook report and trend deltas.

ops/cmo-automation/reports/CMO-QUALITY-UPGRADE-2026-04-02.md

@@ -0,0 +1,127 @@
+# CMO Quality Upgrade — Decisions and Completion Notes
+
+Date: 2026-04-02
+Scope: X CMO automation quality and policy hardening
+Status: Upgrade implemented with documented platform constraint
+
+## Summary
+This upgrade moved the pipeline from generic high-volume engagement behavior to strict quality-gated engagement with policy-enforced copy constraints and duplicate prevention.
+
+Implemented outcomes:
+- No hashtags in generated copy
+- No links in replies
+- No duplicated reply text patterns in a run
+- No duplicate target tweet IDs in a run
+- Context-specific reply requirement with low-signal filtering
+- Mission-aligned relevance gate in policy
+
+Latest dry-run before live:
+- 18 approved actions
+- 16 hydrated and ready
+- 2 blocked (low_signal_source or duplicate_target_tweet)
+- 0 hashtag violations, 0 link violations, 0 duplicate reply patterns
+
+Live execution result:
+- 3 root posts executed successfully
+- Most quote-style replies failed with X API 403 due conversation/mention restriction
+
+## Decisions made
+
+### D1. Enforce hard copy constraints at policy + generator layer
+Decision:
+- Added policy flags for no hashtags, no links in replies, and unique contextual replies.
+- Enforced these during hydration, not just review.
+
+Rationale:
+- Guarantees style consistency even if queue inputs vary.
+- Prevents low-quality copy from reaching execution stage.
+
+Trade-offs:
+- Fewer candidate actions survive gating.
+- More blocked actions in noisy cycles.
+
+Expected benefits:
+- Higher output quality floor.
+- Lower reputational risk from spam-like formatting.
+
+### D2. Deduplicate both content and targets
+Decision:
+- Blocked duplicate target tweet IDs and duplicate reply text patterns within a run.
+
+Rationale:
+- Repetition is visible and degrades account quality perception.
+
+Trade-offs:
+- Reduced throughput when candidate pools are narrow.
+
+Expected benefits:
+- Distinct engagement footprint.
+- Better perceived intent and originality.
+
+### D3. Add context-specificity gate and low-signal filter
+Decision:
+- Reply generation requires either recognizable topic signal or meaningful source anchor.
+- Low-signal content (giveaway/airdrop-like noise) is blocked.
+
+Rationale:
+- Generic responses were the primary failure mode.
+
+Trade-offs:
+- Higher block rate on weak timelines.
+
+Expected benefits:
+- Replies are more defensible and relevant.
+- Better alignment with founder and brand voice requirements.
+
+### D4. Keep quote-workaround execution path, document platform constraint
+Decision:
+- Retained quote execution mode where direct reply constraints apply.
+- Documented X API restriction observed during live run.
+
+Rationale:
+- Maintains one executable path while policy and quality gates improve.
+
+Trade-offs:
+- Quote eligibility depends on conversation relationship and can fail at runtime.
+
+Expected benefits:
+- Immediate operability for eligible targets.
+- Clear visibility into where platform constraints block execution.
+
+## Drawbacks and known limitations
+- X platform restriction: quote attempts can fail with HTTP 403 when account is not mentioned or in-thread.
+- Current pipeline does not preflight quote eligibility before live execution.
+- Some context anchors are concise and can still read mechanically in edge cases.
+- Candidate discovery quality depends on `x-cli tweet search` result quality and available context.
+
+## Remaining work to fully complete upgrade
+1) Add quote eligibility preflight
+- Before execution, verify target is quote-eligible for the posting account.
+- Auto-skip or reroute ineligible targets.
+
+2) Add fallback action strategy per failed reply
+- If quote ineligible, either:
+  - switch to root-post reserve queue, or
+  - store as manual-review candidate.
+
+3) Add stronger context semantic scoring
+- Replace lightweight heuristics with a richer contextual scorer.
+- Require topic + claim + value-add structure.
+
+4) Add per-account command routing/auth isolation
+- Ensure execution uses correct account credentials/session context for each action.
+
+5) Add post-execution telemetry loop
+- Track executed vs failed by reason and adapt candidate selection weights.
+
+## Files changed in this upgrade window
+- config/operating_policy.json
+- scripts/hydrate_approved_queue.py
+- reports/CMO-ECOSYSTEM-MARKETING-BRIEF.md
+- reports/cmo-hydrated-queue.json
+- reports/cmo-hydrated-queue.md
+- reports/cmo-execution-report.json
+- reports/cmo-execution-report.md
+
+## Recommended next checkpoint
+- Implement quote eligibility preflight and fallback routing, then rerun live with target success threshold >= 80% for ready actions.