Fix cache hit percentage calculation

[dylanneve1]

Summary

• Cache hit % was calculated as cache_read / (input + cache_read + cache_write), which incorrectly included cache writes in the denominator — diluting the metric
• Now calculates as cache_read / (input + cache_read), which accurately reflects how much of the readable prompt was served from cache
• Fixed in both the agent response logging (claude-sdk/index.ts) and the /status command display (commands.ts)

Test plan

• Run /status and verify cache hit % looks reasonable
• Check agent logs show corrected cache percentages

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Updates cache hit percentage reporting to avoid diluting the metric by excluding cache-write tokens from the denominator.

Changes:

• Recomputes cache hit % as cache_read / (input + cache_read) in Telegram status output.
• Recomputes cache hit % as cache_read / (input + cache_read) in backend agent logging.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
src/frontend/telegram/commands.ts	Adjusts cache hit % calculation used in the /status display.
src/backend/claude-sdk/index.ts	Adjusts cache hit % calculation used in agent response logs.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated no new comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[claudiusthebot]

Review from heartbeat #44 (2026-05-03)

The fix here is correct and still needed — both locations still have the old denominator on main. But the branch is stale and needs updating.

What changed since April 5: The SDK refactor in v1.9.0 (PR merged after this was filed) split src/backend/claude-sdk/index.ts into 8 focused modules. The cache hit % calculation that was in the old index.ts is now in src/backend/claude-sdk/handler.ts (~line 217):

// Current main (handler.ts) — still wrong:
const totalPrompt =
  state.sdkInputTokens + state.sdkCacheRead + state.sdkCacheWrite;
const cacheHitPct =
  totalPrompt > 0 ? Math.round((state.sdkCacheRead / totalPrompt) * 100) : 0;

The commands.ts fix is still directly applicable — that file hasn't changed structurally and the bug is at the same location.

To land this:

1. Rebase fix-cache-hit-percentage onto current main
2. Drop the claude-sdk/index.ts hunk (file no longer exists — it's now a barrel re-export only)
3. Apply the same fix to src/backend/claude-sdk/handler.ts instead
4. Keep the commands.ts fix as-is

The fix logic is identical in both places: rename totalPrompt → cacheTotal, drop + cacheWrite from the denominator. Six lines total changed.

Confirmed cacheWrite should not be in the denominator: it represents tokens being written to cache on this call — they're not "readable input served from cache." Including them deflates the metric any time the cache is being warmed for the first time.

APPROVE — once rebased.

[claudiusthebot]

Filed the rebased fix as #107 on a fresh branch. Same fix, applied at the post-SDK-refactor file locations (handler.ts instead of the old claude-sdk/index.ts; commands.ts slightly different lines now). 1635/1635 tests, tsc clean, lint 0 errors.

Suggest closing this one in favour of #107 — left this PR alone rather than force-resetting your branch.