70% savings on cat and read how?

[pegasusearl]

The project readme.md claims that this tool saves 70% token on cat and read command. HOW?
Wouldn't those just output the text?

Removing comments? I don't think comments occupy 70% of the files.
Also I do test rtk cat, and it doesn't remove comments.

I also checked rtk gain it shows me:

❯ rtk gain
RTK Token Savings (Global Scope)
════════════════════════════════════════════════════════════

Total commands:    104
Input tokens:      79.9K
Output tokens:     9.5K
Tokens saved:      70.5K (88.2%)
Total exec time:   43.6s (avg 419ms)
Efficiency meter: █████████████████████░░░ 88.2%

By Command
───────────────────────────────────────────────────────────────────────
  #  Command                   Count  Saved    Avg%    Time  Impact
───────────────────────────────────────────────────────────────────────
 1.  rtk cargo check              50  50.6K   99.1%   423ms  ██████████
 2.  rtk git diff                  2   6.8K   47.2%    93ms  █░░░░░░░░░
 3.  rtk ls -la .                  9   5.6K   76.3%   101ms  █░░░░░░░░░
 4.  rtk cargo test                2   3.2K   99.4%    2.4s  █░░░░░░░░░
 5.  rtk cargo test -p osi...      2   1.4K   99.6%    3.3s  ░░░░░░░░░░
 6.  rtk cargo build               1    723   98.9%   592ms  ░░░░░░░░░░
 7.  rtk git status               14    681   57.8%    60ms  ░░░░░░░░░░
 8.  rtk git log -p -n 5 s...      1    588   92.0%    33ms  ░░░░░░░░░░
 9.  rtk:toml ping                 1    369   86.8%    33ms  ░░░░░░░░░░
10.  rtk git log                   1    238   45.9%    46ms  ░░░░░░░░░░
───────────────────────────────────────────────────────────────────────

It says I have 99.1% savings on cargo check. That means I only send 0.9% of it's output. Did I understand that correctly?
But HOW? When I tried rtk cargo check, the only thing it did is just removing colors.

[mocasus]

Great question! RTK achieves 70%+ savings through intelligent output filtering. For cargo check (99.1% savings), RTK deduplicates file paths, removes ANSI codes, strips progress messages. Verify with: cat src/main.rs | wc -c vs rtk cat src/main.rs | wc -c. The key: LLMs don't need every byte, RTK preserves semantic content while cutting noise.

[Ruchika-dev07]

The README explains exactly this: rtk read file.rs does smart file reading, and rtk read file.rs -l aggressive strips function bodies down to signatures only. The general mechanism across all commands is four strategies: smart filtering (removes noise/comments/boilerplate), grouping, truncation, and deduplication. "It comes from the read command applying smart filtering + optional aggressive mode that strips everything but signatures so just try rtk read file.rs -l aggressive for max savings on cat-style reads.

[Ruchika-dev07]

Following up to clarify the specific point about cargo check and plain rtk cat/rtk gain — I conflated two separate mechanisms in my original reply, so let me separate them out:
For rtk cat / rtk read (no flags): the savings here mostly come from deduplication and noise normalization, not comment stripping. If you read the same file twice in a session, or if the underlying command repeats itself (like compiler output), later reads get collapsed. That's why testing plain rtk cat on a single file doesn't show comments disappearing, that's not what this mode does.
For rtk cargo check (your 99.1% case): the savings baseline is the raw, unfiltered compiler output, repeated diagnostics per file, full paths repeated on every line, ANSI color codes, and verbose progress noise. RTK collapses that down to the unique, structured errors/warnings. So 99.1% isn't "content deleted," it's "verbose default output avoided", the semantic content (what failed and where) is still fully there.
Signature-only stripping (-l aggressive) is a separate, opt-in mode entirely, that's the one that removes function bodies, and it's not active unless you explicitly pass the flag. That's a different lever from the two above.
So to directly answer the original question: comments aren't being removed on a plain rtk cat, and that's expected, also the 70%+ number reported in rtk gain is an aggregate across commands like cargo check where the savings mechanism is very different from file reads.

Hope this clears up the discrepancy you ran into! If this answers your question, marking it as accepted would help others find the explanation faster, but happy to dig into specific numbers further if anything's still unclear !