Share your cache-refund card

[ilanbm]

Ran npx cache-refund? Drop your card, your score, or your biggest single miss below. Curious what the spread looks like — subscription and API-billed both welcome.

If a number looked wrong, that's a GitHub issue (there's a priority-lane template for it) — but "huh, interesting" reactions to your own numbers are exactly what this thread is for.

[ilanbm]

cache-refund checkup

Score: 98.5 / 100 — excellent

• Window: last 90 days (567 sessions, 40,911 turns)
• TTL received (last 7d): 1h
• Recoverable ratio: 14.3% (break-even 39.5%)

Leak	Tokens	$-eq	% of write spend
TTL-expiry re-warms (5–60m gaps)	102,194,440	$624.16	7.6%
Session cold starts (>60m or session start) (info)	335,568,686	$2,367.10	28.9%
Model-switch invalidations	63,910,290	$476.77	5.8%
Compaction rewrites	189,887	$1.93	0.0%
Subagent 5m overhead (info)	0	$0.00	0.0%

Biggest single miss: 712K-token re-warm — $8.16-eq in one turn.
Worst day: 2026-06-10 — $49.62-eq leaked.

Verdict: 1h already yours (subscription) ✓ — your 1h cache saved you ~$2,597.76 USD-equivalent (API list rates) vs a 5m world in the last 90 days.
Caching saved you $67,595.16 USD-equivalent (API list rates) vs uncached this window.

methodology: npx cache-refund --explain

[orkeren21]

cache-refund checkup

Score: 90.9 / 100 — solid — a few leaks worth a look

•⁠ ⁠Window: last 90 days (134 sessions, 12,222 turns)
•⁠ ⁠TTL received (last 7d): 5m
•⁠ ⁠Recoverable ratio: 71.6% (break-even 39.5%)

Leak	Tokens	$	% of write spend
TTL-expiry re-warms (5–60m gaps)	226,946,108	$1,297.19	65.3%
Session cold starts (>60m or session start) (info)	54,838,598	$341.43	17.2%
Model-switch invalidations	9,982,945	$62.39	3.1%
Compaction rewrites	11,417	$0.07	0.0%
Subagent 5m overhead (info)	0	$0.00	0.0%

Biggest single miss: 761K-token re-warm — $4.37 in one turn.
Worst day: 2026-07-02 — $100.56 leaked.

Verdict: switching to the 1h TTL saves ~$572.31/30d. Run ⁠ npx cache-refund enable ⁠ to apply.

methodology: npx cache-refund --explain