econtone-score
/econtone-score
Ocenia angielski dokument z ekonomii lub zarządzania według czterech kontroli i zwraca punktację ważoną, nie zmieniając tekstu.
- Autor
- pykkkkk ↗
- Repozytorium
- EconTone ↗
- Licencja
- Apache-2.0
- Ostatnia zmiana repozytorium
- 10.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 5
- Skrypty
- 2
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 10.11.2026, bez odwołań do wycofanych modeli i API.
Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.
- 2 skrypty w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.
Skrypty: scripts/evidence.py, scripts/score.py
Co robi i kiedy
Ocenia angielski dokument z ekonomii lub zarządzania według czterech kontroli i zwraca punktację ważoną, nie zmieniając tekstu.
Wyzwalacze z opisu
Oryginalny opis (EN)
Scores an English economics / management document against the same four checks the downstream EconTone skills run — terminology and register (econtone-polish), claim strength and structure (econtone-polish), AI-trace risk (econtone-ai-trace), rhythm (econtone-prosody) — and returns a weighted scorecard (40 / 40 / 10 / 10) plus an editing latitude that tells the editors how hard they may press. Scores only: it never edits the text and never suggests rewrites. Load when the user asks "is this ready", "score this draft", "what should I fix first", or before starting a polish run on existing text.
Jak działa
- Hard boundary
- The scorecard (40 / 40 / 10 / 10 = 100)
- No routing: the pipeline always runs
- How hard to edit
- Procedure
- Output
- Honesty
Podgląd SKILL.md
---
name: econtone-score
description: >
Scores an English economics / management document against the same four checks the downstream
EconTone skills run — terminology and register (econtone-polish), claim strength and structure
(econtone-polish), AI-trace risk (econtone-ai-trace), rhythm (econtone-prosody) — and returns a
weighted scorecard (40 / 40 / 10 / 10) plus an editing latitude that tells the editors how hard they may
press. Scores only: it never edits the text and never suggests rewrites. Load when the user asks "is this ready", "score this draft", "what should I fix first",
or before starting a polish run on existing text.
metadata:
pack: econtone
role: triage
---
# EconTone Score — the triage gate
The three downstream skills are thorough and expensive. This one is cheap and fast: it runs the same
four checks, simplified, and tells you which of them the draft actually needs.
## Hard boundary
**Score, do not edit.** Output is a four-row table, one overall number, and an editing latitude. No revised text,
no sentence-level suggestions, no "consider changing X to Y". If the user wants fixes, hand off to the
member that owns them.
## The scorecard (40 / 40 / 10 / 10 = 100)
Each row is a simplified version of what one downstream skill checks.
| # | Row | Owner | What it counts | Points |
|---|---|---|---|---|
| **P1** | Terminology & register | `econtone-polish` | banned glossary variants; register defect hits (hype, ornate, filler) | **40** |
| **P2** | Claims & structure | `econtone-polish` | causal verbs with no identification language; significance with no magnitude; no stated design | **40** |
| **T1** | AI-trace risk | `econtone-ai-trace` | `risk_index` from the detector-proxy scanner | **10** |
| **R1** | Rhythm | `econtone-prosody` | verdict, red markers, WARN count | **10** |
Scoring rule per row: start at full marks and deduct, then scale to the row's weight — the deductions are printed next to the score, so every
number is traceable to a scanner line. A row whose scanner is missing prints `n/a` and is excluded
from the total, and the latitude is computed from the rows that did run.
### What was deliberately left out
The full members check far more than this table does: 32 writing rules, 37 sentence patterns, nine
rhythm metrics, five markers, three trace scanners, seven detector proxies. **The scorecard is not a
summary of those; it is a filter.** If a row scores low, run that member and let it do the detailed
work.
## No routing: the pipeline always runs
**The score does not dispatch.** There are no thresholds and no "skip polish" branch. Every draft
goes through the whole pipeline - polish, then prosody, then the final trace check - because the
score is an orientation aid, not a gate. What the score changes is **how hard the editors press**,
which is the next section.
## How hard to edit
The total feeds an *editing latitude*: how much the downstream members may touch.
```
editing latitude
1.00 | *****************************
0.95 | ***
0.90 | **
0.85 | **
0.80 | *
0.75 | **
0.70 | *
0.65 | *
0.60 | *
0.55 | *
0.50 | *
0.45 | *
0.40 | *
0.35 | *
0.30 | *
0.25 | *
0.20 | *
0.15 | *
0.10 | *
0.05 | *
0.00 | *
+ ---------------------------------------------------
0 5 1
0 50 100
```
Reading it:
| total | latitude | what the editors may do |
|---|---|---|
| ≤ 50 | 100–71% | restructure, cut, re-order - never changing the meaning |
| 50–75 | 71–50% | fix what is wrong; no restructuring; keep sentences that work |
| 75–90 | 50–16% | corrections only: errors, terminology, claims that overstate the design |
| > 90 | < 16% | only outright errors; change nothing else, even if it could read better |
Two rules sit above the curve:
1. **Meaning is invariant.** No edit may change what the author claims, hedges, or concludes. When a
fix requires a different claim, stop and flag it; do not rewrite the claim.
2. **A high score buys restraint, not silence.** Even at 95 the editors fix errors; they just stop
there. Editing for the sake of being seen is a defect, not diligence.
## Procedure
1. Read the file (read-only). Nothing is written.
2. Pick the profile — `body` (default), `abstract`, `intro` — rhythm bands depend on it.
3. Run:
```bash
python scripts/score.py DRAFT.md
python scripts/score.py DRAFT.md --profile abstract --json
```
4. Report the table, the latitude band, and the two lowest rows. Stop.
## Output
```
EconTone score - draft.md (profile: abstract)
P1 Terminology & register 34/40 1 banned variant(s)
P2 Claims & structure 27/40 1 causal verb(s), no identification language; no stated design
T1 AI-trace risk 6/10 risk_index 40 (moderate)
R1 Rhythm 8/10 ADJUST: 0 RED, 0 markers, 5 WARN
--------------------------------------------------------------
TOTAL 75 / 100 (rows with n/a are excluded)
EDITING LATITUDE 71% [minimal]
corrections only: errors, terminology, claims that overstate the design; no stylistic rewriting
(orientation only - nothing was modified; the full pipeline still runs)
```
## Honesty
- A score is a diagnostic, not a verdict on publishability: it sees whether the prose *claims* an
identification, not whether the identification is right.
- **The latitude curve is a design choice, not a measurement.** It says how much *may* be touched, not how
much *should* be. A generous scorer hands out latitude too easily: the deductions have not been calibrated
on a labelled corpus the way the prosody bands were. Until they are, treat a high latitude as permission
with a caveat, not as an instruction to rewrite.
Pliki w paczce (5)
- README.md 3 KB
- README.zh.md 1 KB
- SKILL.md 7 KB
- scripts/evidence.py 7 KB
- scripts/score.py 7 KB
Wersje
W bazie nie ma innych skilli o nazwie econtone-score.
Migawka bazy z 13.11.2026 · wersja paczki 9c69c6c3a1bb (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100econtone-equilibrium
Dobiera kombinację trzech skilli redakcji EconTone: poprawy języka, śladów AI i prozodii, z pierwszeństwem wiarygodności i raportem decyzji.