fix(implications): move per-call target to end of prompt to enable prompt caching

[fashionmascherine-svg]

Problem

IMPLICATION_PROMPT places the per-call {target_title} before the large, run-static {group_titles_text} block and the instruction body. Prompt caching keys on the longest common prefix, so a target that changes every call makes the whole ~35–44k-token static section uncacheable — every implications call is billed full input.

Fix

Move the ## TARGET EVENT block to the end of the prompt (after the available-events list and the instructions). The static prefix (events list + instructions) becomes a stable, cacheable prefix across all calls in a run. No change to the task semantics or output schema.

Measured (DeepSeek deepseek-chat / v4-flash)

Same input, controlled before/after run:

	before	after
input cache-hit ratio	~0.2%	~40% aggregate (~80% on the implications step alone)
run cost (57-group tag)	baseline	~−40%

• Larger tags benefit more (the static prefix scales with the number of groups).
• Output unchanged in order of magnitude — identical coverage distribution on the re-run.
• The validation prompt is untouched, which is why the aggregate cache-hit sits below the per-step figure.

Diff

Two hunks in backend/core/steps/implications.py:

 IMPLICATION_PROMPT = """Find ONLY logically necessary relationships between prediction market events.
 
-## TARGET EVENT:
-"{target_title}"
-
 ## AVAILABLE EVENTS:
 {group_titles_text}
@@ ## CRITICAL RULES:
 5. When in doubt, LEAVE IT OUT
+
+## TARGET EVENT
+Find the necessary relationships between the TARGET event below and the AVAILABLE EVENTS listed above. Use the exact titles from the AVAILABLE EVENTS list.
+
+TARGET: "{target_title}"
+
+Output ONLY the JSON described in OUTPUT FORMAT above.
 """

Summary by CodeRabbit

• Improvements
  • Enhanced market-event implication analysis with clearer relationship-matching instructions.
  • Improved precision by requiring exact event titles and strictly formatted JSON output.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: bf46ddbc-84d9-4877-bbd7-e5e1ec9f243d

📥 Commits

Reviewing files that changed from the base of the PR and between 65aea7a and 2c813ab.

📒 Files selected for processing (1)

• backend/core/steps/implications.py

---

📝 Walkthrough

Walkthrough

The implication prompt is reorganized so target-event instructions appear later, require exact titles from available events, use a stricter target format, and require only the specified JSON output.

Changes

Implication prompt update

Layer / File(s)	Summary
Target event instruction flow backend/core/steps/implications.py	The earlier target-event block is removed, and a later section adds exact-title relationship instructions, a tightened target placeholder, and JSON-only output requirements.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Suggested reviewers: smypmsa

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly matches the main change: moving the per-call target block to the end of the prompt to improve caching.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}