---
name: "speed_reading"
description: "Teaches speed reading that keeps comprehension: chunking drills, regression control, skim protocols, and the science of when NOT to speed up. Triggers when the user asks to read faster, speed read, improve reading speed, skim material efficiently, or practice reading drills."
---

# Speed Reading

## Purpose

Teach and run speed-reading practice that increases throughput **without** trading away understanding. Covers three levers — chunking (fewer fixations per line), regression control (fewer wasted re-reads), and technique selection (skim vs. speed vs. close reading) — plus the honest science: faster reading always trades off against comprehension, and some texts must not be sped through. Never promises magic gains; every factual claim in this skill is sourced (see `references/science.md`).

## Workflow

1. **Decide if speed fits.** Consult `references/science.md` "When NOT to speed." Dense technical material, legal text, contracts, poetry, anything the user must recall precisely → route to close reading, not this skill. Say so plainly; refusal to speed is a feature.

2. **Baseline.** Measure the user's normal WPM on a representative text (word count ÷ minutes). Then run the comprehension check: after reading, have them write a one-sentence summary per section. Record both — speed without the comprehension score is meaningless.

3. **Preview (1–2 min).** Before any fast pass: scan headings, first/last sentences of paragraphs, figures, and summaries. This builds the mental framework that lets chunking and skimming actually work — prior familiarity is what makes fast reading effective.

4. **Chunking drill.** Use `bin/chunk.py <file> [--words N]` to reformat practice text into N-word visual chunks (default 4). Read each line as one fixation instead of word-by-word. Run 2–3 short paced passes, then re-read the same text in normal formatting and re-measure WPM + comprehension.

5. **Regression control.** Teach forward-only passes: suppress *redundant* re-reads (mindless looping back), but always allow *strategic* regressions — when a definition, number, or referent can't be resolved, go back. Regressions serve comprehension; suppressing needed ones hurts it. Normal regression rate is 10–15% of eye movements, not zero.

6. **Pick the right gear.** Map goal → technique:
   - Gist / "is this worth reading?" → skim (prioritize informative parts, gloss the rest)
   - Familiar or moderately complex material → speed techniques (chunking + pacer + preview)
   - Learning new complex material → close reading (do not speed)

7. **Verify and stop.** Re-run the one-sentence-summary test. If comprehension dropped below the baseline, slow down and re-read normally — never trade understanding for a higher number silently. Report WPM and comprehension together, always as a pair.

## Output Contract

Every practice session or coaching answer returns:
- **Verdict**: speed mode appropriate or not (and why), for this text and goal
- **WPM before / after** (measured, never claimed)
- **Comprehension score**: one-sentence-summary pass/fail per section
- **Techniques applied**: chunking / regression control / skim protocol / pacer
- **Next step**: which drill to repeat, or a recommendation to drop back to close reading

## Operating Rules

1. **Accuracy rule (non-negotiable):** Every factual claim must be verified via web search or a primary source. Never invent numbers, dates, quotes, or events. Unverifiable = cut it or frame as opinion. No confident unverified claims, ever.
2. **Never promise specific WPM gains.** Research finds doubling or tripling reading speed with equal comprehension is unlikely; eye movements account for at most ~10% of reading time, and the real bottleneck is word recognition and sentence processing. Promise practice, not miracles.
3. **Comprehension is the gate.** If the summary test fails, the session result is "slow down and re-read," never "close enough at 600 WPM."
4. **Regressions are a tool, not a defect.** Suppress only redundant re-reads. Never tell anyone to eliminate regressions entirely — doing so removes the repair mechanism reading relies on.
5. **One-word RSVP claims are out of scope.** Single-word rapid displays eliminate parafoveal preview and block regressions; comprehension reliably declines past natural pace. This skill does not teach or endorse them.
6. **Route honestly.** If the user wants to "speed read" a contract, a paper they're being graded on, or a textbook chapter they're seeing for the first time, redirect to close reading with a reason.
7. Keep SKILL.md operational; technique detail lives in `references/`, formatting logic in `bin/`.