diff --git a/README.md b/README.md
index 7f3fdf6..0c86f29 100644
--- a/README.md
+++ b/README.md
@@ -31,11 +31,12 @@ PersonalityProtect keeps the corpus on disk, measures your cadence, retrieves sh
## How you get voice
1. **Ingest** your LinkedIn export and/or local notes (stays on disk).
-2. **`index-voice`** builds a local retrieval index.
-3. **`build-style-profile`** measures cadence (sentence length, short lines, post length band, banned filler).
-4. **`write --topic --points`** drafts from the brief only; retrieved pieces are rhythm reference.
+2. **`select`** gates the corpus by length (`--min-words`, default 50) and an optional year cap (`--through-year`, default: current year).
+3. **`index-voice`** builds a local retrieval index.
+4. **`build-style-profile`** measures cadence from the selection (sentence length, short lines, post length band, banned filler).
+5. **`write --topic --points`** drafts from the brief only; retrieved pieces are rhythm reference.
-Two channels come out of step 4:
+Two channels come out of step 5:
- **`--channel post`** (default) targets your long-post band, up to the LinkedIn ~3000-character limit (~550 words).
- **`--channel article`** runs outline → sections → stitch, and needs at least five `linkedin_article` pieces in the corpus.
@@ -61,6 +62,7 @@ personality-protect download --format mlx # ~6 GB, once
personality-protect ingest --linkedin ~/path/to/linkedin-export
personality-protect ingest --path ~/path/to/notes --source note
+personality-protect select
personality-protect index-voice
personality-protect build-style-profile
@@ -94,18 +96,23 @@ pip install -e ".[cuda]" # NVIDIA path (optional)
Public docs use **synthetic Contoso / synergy-slop text only**. No personal corpus.
-| `write` (post + article) | `status` | Mark |
+| `write` (post + article) | `status` | Setup |
| --- | --- | --- |
-| 
| 
| 
|
+| | | 
|
```bash
+personality-protect select
personality-protect index-voice
personality-protect build-style-profile
personality-protect write --topic "Contoso pricing" --points "Name one owner."
personality-protect status
```
-Optional smoke tour (no model download; synthetic only):
+Optional smoke tour — runs the write path with a stubbed model call (no download; synthetic Contoso only):
+
+| Smoke tour (`demo`) | Mark |
+| --- | --- |
+| 
| |
```bash
personality-protect demo
@@ -160,13 +167,20 @@ personality-protect ingest --linkedin ~/path/to/linkedin-export.zip
personality-protect ingest --path ~/path/to/notes --source note
```
-### Index and style
+### Select, index and style
+
+`select` is required before `build-style-profile` (the style card reads `selection.json`). It is a length gate plus an optional year cap:
```bash
+personality-protect select
+personality-protect select --min-words 75 --include-undated
+personality-protect select --through-year 2024 # deliberate narrowing only
personality-protect index-voice
personality-protect build-style-profile
```
+Defaults: **≥50 words**, dates through the **current year**. Use `--through-year` when you intentionally want an older slice. Corpus gates: **warn** below 50 selected pieces; **block** below 20 unless `--force`. Holding pieces back from retrieval is separate — `index-voice --holdout-id`, scored by `eval-write-holdout`.
+
Post length targets come from `linkedin_post` pieces (p75/p90), clamped to the LinkedIn ~3000-character band (~550 words).
### Write
@@ -197,12 +211,13 @@ Global flags (most commands): `--profile`, `--home`, `--json`, plus branding `--
| `init` | Create profile under `~/.personality-protect/` |
| `download` | Prefetch quantized MLX or GGUF base |
| `ingest` | Index LinkedIn export and/or local paths |
+| `select` | Gate corpus by length / year — required before `build-style-profile` |
| `index-voice` | Build local voice retrieval index |
| `build-style-profile` | Build cadence / length / banned-filler style card |
| `write` | Draft a post or article (`--channel post\|article`) |
| `eval-write-holdout` | Score write quality on held-out pieces (local receipt) |
| `status` | Show profile state |
-| `demo` | Optional synthetic smoke tour (no download) |
+| `demo` | Optional synthetic smoke tour of the write path (no download) |
| `api` | Loopback HTTP stub |
| `logo` | Print Telivity CLI mark |
| `build-writer-sft`, `train` | Optional LoRA experiments — see [Advanced](#advanced-optional) |
@@ -247,7 +262,7 @@ Keep an adapter only if `eval-write-holdout` shows it beating RAG-alone on held-
### Other experiment commands
-`select`, `filter`, `compare`, `eval`, and the translator-pair commands remain available. They score or rewrite existing text and are not part of the drafting path above.
+`filter`, `compare`, `eval`, and the translator-pair commands remain available. They score or rewrite existing text and are not part of the drafting path above. (`select` *is* part of the drafting path — see [Select, index and style](#select-index-and-style).)
### Operator script
@@ -268,11 +283,26 @@ Operator checklist: [docs/LAUNCH.md](docs/LAUNCH.md).
```bash
pip install -e ".[dev]"
pytest
-ruff check src tests
+ruff check src tests scripts
```
CI (`.github/workflows/ci.yml`) required checks: `lint`, `test (3.11)`, `test (3.12)`, `sanitize`, `cli-smoke`.
+### Regenerating screenshots
+
+`scripts/shot.py` renders captured ANSI terminal bytes to PNG on a fixed character grid (so Rich box-drawing lines up). Needs `pillow`.
+
+```bash
+export PERSONALITY_PROTECT_HOME=/tmp/shots COLUMNS=94 TERM=xterm-256color
+pip install pillow
+script -qec "personality-protect --logo off demo" /dev/null \
+ | python3 scripts/shot.py docs/images/cli-demo.png "personality-protect demo"
+script -qec "personality-protect --logo off status" /dev/null \
+ | python3 scripts/shot.py docs/images/cli-status.png "personality-protect status"
+```
+
+Do not regenerate `docs/images/cli-shipped.png` off Apple Silicon — it shows a real `write` against MLX weights.
+
---
## License
diff --git a/docs/images/cli-demo.png b/docs/images/cli-demo.png
index 08a211b..d9fe7ed 100644
Binary files a/docs/images/cli-demo.png and b/docs/images/cli-demo.png differ
diff --git a/docs/images/cli-setup.png b/docs/images/cli-setup.png
new file mode 100644
index 0000000..872c736
Binary files /dev/null and b/docs/images/cli-setup.png differ
diff --git a/docs/images/cli-status.png b/docs/images/cli-status.png
index 2335e4a..faf7c8b 100644
Binary files a/docs/images/cli-status.png and b/docs/images/cli-status.png differ
diff --git a/scripts/shot.py b/scripts/shot.py
new file mode 100755
index 0000000..e29c880
--- /dev/null
+++ b/scripts/shot.py
@@ -0,0 +1,272 @@
+#!/usr/bin/env python3
+"""Render captured ANSI terminal bytes to a PNG on a fixed character grid.
+
+Usage:
+ script -qec "personality-protect --logo off demo" /dev/null \\
+ | python3 scripts/shot.py docs/images/cli-demo.png "personality-protect demo"
+
+Needs pillow. Fixed cell size keeps Rich box-drawing aligned.
+"""
+
+from __future__ import annotations
+
+import re
+import sys
+from pathlib import Path
+
+from PIL import Image, ImageDraw, ImageFont
+
+# Fixed grid — match existing docs shots (≈10×20 cell, macOS chrome).
+CELL_W = 10
+CELL_H = 20
+PAD_X = 14
+PAD_Y = 12
+TITLE_H = 36
+COLS = 94
+
+BG = (18, 20, 24)
+TITLE_BG = (40, 42, 48)
+TITLE_EDGE = (48, 50, 56)
+FG = (230, 232, 236)
+DIM = (140, 146, 156)
+BOLD = (255, 255, 255)
+CYAN = (120, 210, 220)
+GREEN = (120, 200, 140)
+YELLOW = (220, 190, 100)
+RED = (230, 120, 110)
+MAGENTA = (200, 140, 200)
+BLUE = (120, 160, 230)
+
+# CSI / OSC / charset noise from `script` / Rich
+_ANSI_OSC = re.compile(r"\x1b\][^\x07\x1b]*(?:\x07|\x1b\\)")
+_ANSI_CSI = re.compile(r"\x1b\[[0-9;?]*[ -/]*[@-~]")
+_ANSI_OTHER = re.compile(r"\x1b[@-Z\\-_]")
+_ANSI_CHARSET = re.compile(r"\x1b[()][0-9A-Za-z]")
+
+
+def _font(size: int = 15) -> ImageFont.FreeTypeFont | ImageFont.ImageFont:
+ candidates = [
+ "/usr/share/fonts/truetype/jetbrains-mono/JetBrainsMono-Regular.ttf",
+ "/usr/share/fonts/truetype/dejavu/DejaVuSansMono.ttf",
+ "/System/Library/Fonts/Menlo.ttc",
+ ]
+ for path in candidates:
+ if Path(path).is_file():
+ return ImageFont.truetype(path, size=size)
+ return ImageFont.load_default()
+
+
+def _sgr_color(code: int, fg: bool) -> tuple[int, int, int] | None:
+ table = {
+ 30: (60, 64, 72),
+ 31: RED,
+ 32: GREEN,
+ 33: YELLOW,
+ 34: BLUE,
+ 35: MAGENTA,
+ 36: CYAN,
+ 37: FG,
+ 90: DIM,
+ 91: RED,
+ 92: GREEN,
+ 93: YELLOW,
+ 94: BLUE,
+ 95: MAGENTA,
+ 96: CYAN,
+ 97: BOLD,
+ }
+ if not fg:
+ return None
+ return table.get(code)
+
+
+def parse_ansi(data: bytes) -> list[list[tuple[str, tuple[int, int, int], bool]]]:
+ """Return rows of (char, fg, bold) cells."""
+ text = data.decode("utf-8", errors="replace")
+ text = text.replace("\r\n", "\n").replace("\r", "\n")
+ text = _ANSI_OSC.sub("", text)
+ text = _ANSI_CHARSET.sub("", text)
+
+ rows: list[list[tuple[str, tuple[int, int, int], bool]]] = [[]]
+ fg = FG
+ bold = False
+ i = 0
+ while i < len(text):
+ ch = text[i]
+ if ch == "\x1b":
+ m = _ANSI_CSI.match(text, i)
+ if m:
+ seq = m.group(0)
+ i = m.end()
+ if seq.endswith("m"):
+ body = seq[2:-1]
+ parts = [int(x) for x in body.split(";") if x.isdigit()] if body else [0]
+ j = 0
+ while j < len(parts):
+ code = parts[j]
+ if code == 0:
+ fg, bold = FG, False
+ elif code == 1:
+ bold = True
+ if fg == FG:
+ fg = BOLD
+ elif code == 2:
+ fg = DIM
+ elif code == 22:
+ bold = False
+ if fg == BOLD:
+ fg = FG
+ elif 30 <= code <= 37 or 90 <= code <= 97:
+ color = _sgr_color(code, True)
+ if color:
+ fg = color
+ elif code == 39:
+ fg = BOLD if bold else FG
+ elif code == 38 and j + 1 < len(parts):
+ mode = parts[j + 1]
+ if mode == 2 and j + 4 < len(parts):
+ fg = (parts[j + 2], parts[j + 3], parts[j + 4])
+ j += 4
+ elif mode == 5 and j + 2 < len(parts):
+ # 256-color: approximate via grayscale/primary buckets
+ n = parts[j + 2]
+ if n < 16:
+ basic = [
+ (0, 0, 0),
+ (205, 0, 0),
+ (0, 205, 0),
+ (205, 205, 0),
+ (0, 0, 238),
+ (205, 0, 205),
+ (0, 205, 205),
+ (229, 229, 229),
+ (127, 127, 127),
+ RED,
+ GREEN,
+ YELLOW,
+ BLUE,
+ MAGENTA,
+ CYAN,
+ BOLD,
+ ]
+ fg = basic[n]
+ elif n < 232:
+ c = n - 16
+ r = (c // 36) * 51
+ g = ((c // 6) % 6) * 51
+ b = (c % 6) * 51
+ fg = (r, g, b)
+ else:
+ v = 8 + (n - 232) * 10
+ fg = (v, v, v)
+ j += 2
+ else:
+ j += 1
+ j += 1
+ continue
+ m2 = _ANSI_OTHER.match(text, i)
+ if m2:
+ i = m2.end()
+ continue
+ i += 1
+ continue
+ if ch == "\n":
+ rows.append([])
+ i += 1
+ continue
+ if ch == "\t":
+ spaces = 4 - (len(rows[-1]) % 4)
+ rows[-1].extend([(" ", fg, bold)] * spaces)
+ i += 1
+ continue
+ if ch == "\x08":
+ if rows[-1]:
+ rows[-1].pop()
+ i += 1
+ continue
+ if ord(ch) < 32:
+ i += 1
+ continue
+ rows[-1].append((ch, fg, bold))
+ i += 1
+
+ # Drop trailing empty rows from script(1) noise
+ while rows and not rows[-1]:
+ rows.pop()
+ return rows
+
+
+def _strip_script_noise(rows: list[list[tuple[str, tuple[int, int, int], bool]]]):
+ """Drop typescript headers / bare prompts that `script` sometimes emits."""
+ cleaned = []
+ for row in rows:
+ line = "".join(c for c, _, _ in row).strip()
+ if line.startswith("Script started") or line.startswith("Script done"):
+ continue
+ cleaned.append(row)
+ return cleaned
+
+
+def render(
+ rows: list[list[tuple[str, tuple[int, int, int], bool]]],
+ title: str,
+ cols: int = COLS,
+) -> Image.Image:
+ rows = _strip_script_noise(rows)
+ # Prepend a synthetic prompt line matching existing docs shots
+ prompt = [
+ (">", CYAN, True),
+ (" ", FG, False),
+ ]
+ prompt.extend((c, BOLD if c != " " else FG, c != " ") for c in title)
+ # If the capture already starts with the command, don't double it
+ first = "".join(c for c, _, _ in rows[0]).strip() if rows else ""
+ if not first.startswith(">") and title not in first:
+ rows = [prompt] + rows
+
+ width = PAD_X * 2 + cols * CELL_W
+ height = TITLE_H + PAD_Y * 2 + max(1, len(rows)) * CELL_H
+ img = Image.new("RGB", (width, height), BG)
+ draw = ImageDraw.Draw(img)
+
+ # Title bar
+ draw.rectangle([0, 0, width, TITLE_H], fill=TITLE_BG)
+ draw.line([(0, TITLE_H), (width, TITLE_H)], fill=TITLE_EDGE)
+ for x, color in ((18, (255, 95, 86)), (38, (255, 189, 46)), (58, (39, 201, 63))):
+ draw.ellipse([x, 12, x + 12, 24], fill=color)
+ font_title = _font(13)
+ tw = draw.textlength(title, font=font_title)
+ draw.text(((width - tw) / 2, 10), title, fill=DIM, font=font_title)
+
+ font = _font(15)
+ y = TITLE_H + PAD_Y
+ for row in rows:
+ x = PAD_X
+ for ch, color, _bold in row[:cols]:
+ draw.text((x, y - 1), ch, fill=color, font=font)
+ x += CELL_W
+ y += CELL_H
+ return img
+
+
+def main(argv: list[str]) -> int:
+ if len(argv) < 2:
+ print(
+ "usage: shot.py OUT.png [TITLE]\n"
+ " reads ANSI bytes from stdin",
+ file=sys.stderr,
+ )
+ return 2
+ out = Path(argv[1])
+ title = argv[2] if len(argv) > 2 else out.stem
+ data = sys.stdin.buffer.read()
+ rows = parse_ansi(data)
+ img = render(rows, title=title)
+ out.parent.mkdir(parents=True, exist_ok=True)
+ img.save(out, format="PNG", optimize=True)
+ print(f"wrote {out} ({img.size[0]}x{img.size[1]})")
+ return 0
+
+
+if __name__ == "__main__":
+ raise SystemExit(main(sys.argv))
diff --git a/src/personality_protect/cli.py b/src/personality_protect/cli.py
index 0910dec..d149265 100644
--- a/src/personality_protect/cli.py
+++ b/src/personality_protect/cli.py
@@ -459,7 +459,8 @@ def select_cmd(
console.print(f"[red]{gate_error}[/red]")
raise typer.Exit(2)
console.print(f"Saved: {paths.selection_path}")
- console.print("Next: personality-protect train")
+ console.print("Next: personality-protect index-voice")
+ console.print("Then: personality-protect build-style-profile")
@app.command("build-style-profile")
@@ -1104,8 +1105,8 @@ def on_progress(info: dict) -> None:
if result.notes:
console.print(result.notes)
if result.status == "ok":
- console.print("Next: personality-protect filter --text '…'")
- console.print("Or: personality-protect compare --synthetic slop_branding")
+ console.print("Next: personality-protect eval-write-holdout --out receipt.json")
+ console.print("[dim]Keep this adapter only if it beats RAG-alone on the holdout.[/dim]")
@app.command("filter")
@@ -1414,20 +1415,42 @@ def demo_cmd(
),
as_json: bool = typer.Option(False, "--json"),
) -> None:
- """Optional synthetic mock tour (not the shipped mlx/llama path)."""
+ """Optional synthetic smoke tour of the write path (no model download)."""
_banner_from_ctx(ctx, json_mode=as_json)
result = run_demo(home=home, draft=draft)
if as_json:
typer.echo(json.dumps(result, indent=2, ensure_ascii=False))
return
console.print(
- "[bold]Mock tour complete[/bold] "
- "(synthetic data + mock adapter only — not the shipped product path)."
+ "[bold]Smoke tour complete[/bold] "
+ "(synthetic Contoso corpus; the model call inside write is stubbed)."
)
console.print(
- "[dim]Shipped path: train --backend mlx|cuda → filter/compare with llama|mlx.[/dim]"
+ "[dim]Product path: ingest → select → index-voice → "
+ "build-style-profile → write.[/dim]"
)
- console.print(f"Ingested: {result['ingested']} Selected: {result['selected']}")
+ console.print(
+ "[dim]For a real draft: download --format mlx, then "
+ "write --topic … --points … on Apple Silicon.[/dim]"
+ )
+ console.print(
+ f"Ingested: {result['ingested']} Selected: {result['selected']} "
+ f"Indexed: {result['indexed']}"
+ )
+ console.print(
+ f"Style card: {result['style_pieces']} pieces, "
+ f"{result['banned_ai_filler']} banned filler phrases"
+ )
+ console.print("")
+ exemplars = result.get("write_exemplars") or []
+ console.print(
+ f"write channel={result['write_channel']} "
+ f"adapter={result['write_adapter']} model=stub "
+ f"exemplars={len(exemplars)}"
+ )
+ console.print(result["written"])
+ console.print("")
+ console.print("[dim]Legacy mock train/filter (not the drafting path):[/dim]")
console.print(f"Train: {result['train_status']} via {result['train_backend']}")
console.print("")
console.print("[bold]Draft[/bold]")
diff --git a/src/personality_protect/config.py b/src/personality_protect/config.py
index b4f6560..4d22ded 100644
--- a/src/personality_protect/config.py
+++ b/src/personality_protect/config.py
@@ -5,6 +5,7 @@
import json
import os
from dataclasses import asdict, dataclass, field
+from datetime import datetime, timezone
from pathlib import Path
from typing import Any
@@ -26,7 +27,14 @@
DEFAULT_BASE_MODEL = DEFAULT_MLX_MODEL
DEFAULT_MIN_WORDS = 50
-DEFAULT_THROUGH_YEAR = 2024
+
+# `select` caps the corpus at this year. It must NOT exclude current writing:
+# a fixed past year silently selects zero pieces for anyone whose corpus is
+# recent, and the style profile is built from the selection. The writer SFT
+# path already bypasses the year gate for exactly this reason
+# (see run_build_writer_sft). Use --through-year to narrow deliberately.
+# Holdouts are a separate, explicit mechanism (--holdout-id).
+DEFAULT_THROUGH_YEAR = datetime.now(timezone.utc).year
DEFAULT_PROFILE = "default"
DEFAULT_VOICE_MODE = "rag"
DEFAULT_WRITE_ADAPTER: str | None = None
diff --git a/src/personality_protect/demo.py b/src/personality_protect/demo.py
index c9f0f22..38de785 100644
--- a/src/personality_protect/demo.py
+++ b/src/personality_protect/demo.py
@@ -9,10 +9,31 @@
from personality_protect.filter import filter_draft
from personality_protect.ingest import run_ingest
from personality_protect.select import run_select
+from personality_protect.style_profile import run_build_style_profile
from personality_protect.train import run_train
+from personality_protect.voice_index import build_voice_index
+from personality_protect.write import run_write
DEMO_PROFILE = "demo"
+# Synthetic brief for the demo's write step. Contoso only — never personal text.
+DEMO_TOPIC = "Contoso Ledger exceptions queue"
+DEMO_POINTS = "- Name one owner\n- Keep the rollout boring"
+
+# The demo never downloads weights, so the model call is stubbed. Everything
+# around it — retrieval, style card, trim, guards — is the real write path.
+_DEMO_STUB_DRAFT = (
+ "The Contoso Ledger exceptions queue is not a tooling problem.\n\n"
+ "Name one owner.\n\n"
+ "Keep the rollout boring. Boring is what survives contact with a Monday."
+)
+
+
+def _demo_generate(_messages, **_kwargs) -> str:
+ """Canned generator for the demo — no model, no download."""
+ return _DEMO_STUB_DRAFT
+
+
# Inline fallback if package data missing
_FALLBACK_DOCS = {
"post_2023_craft.md": """Date: 2023-06-12
@@ -98,7 +119,7 @@ def run_demo(
home: Path | None = None,
draft: str | None = None,
) -> dict:
- """Full synthetic pipeline: init → ingest → select → mock train → filter."""
+ """Full synthetic pipeline: init → ingest → select → write path → mock train/filter."""
paths, config, _ = init_profile(DEMO_PROFILE, home=home, force=True)
corpus = ensure_demo_corpus(paths.cache_dir / "demo_corpus")
added, _ = run_ingest(paths, local=[corpus], source_hint="demo")
@@ -108,6 +129,17 @@ def run_demo(
through_year=2024,
include_undated=True,
)
+ voice = build_voice_index(paths)
+ style, _style_path = run_build_style_profile(paths)
+ written = run_write(
+ DEMO_TOPIC,
+ DEMO_POINTS,
+ paths,
+ k=2,
+ channel="post",
+ use_adapter=False,
+ generate_fn=_demo_generate,
+ )
train = run_train(
paths,
backend="mock",
@@ -129,6 +161,15 @@ def run_demo(
"ingested": added,
"selected": len(selected),
"selection_summary": selection.summary,
+ "indexed": voice["indexed"],
+ "style_pieces": style["stats"]["pieces"],
+ "banned_ai_filler": len(style["banned_ai_filler"]),
+ "write_topic": DEMO_TOPIC,
+ "write_channel": written["channel"],
+ "write_adapter": written["adapter"],
+ "write_exemplars": written["exemplar_ids"],
+ "write_stubbed_model": True,
+ "written": written["text"],
"train_status": train.status,
"train_backend": train.backend,
"adapter_dir": train.adapter_dir,
diff --git a/tests/test_demo_pipeline.py b/tests/test_demo_pipeline.py
index 903821a..aa873d2 100644
--- a/tests/test_demo_pipeline.py
+++ b/tests/test_demo_pipeline.py
@@ -106,6 +106,58 @@ def test_detect_backend_mock():
assert detect_backend("mock") == "mock"
+def test_demo_runs_write_path(tmp_path: Path):
+ """Demo must exercise index-voice → style profile → write, not just train/filter."""
+ result = run_demo(home=tmp_path)
+ assert result["indexed"] >= 1
+ assert result["style_pieces"] >= 1
+ assert result["banned_ai_filler"] >= 1
+ assert result["write_channel"] == "post"
+ assert result["write_adapter"] == "none"
+ assert result["write_stubbed_model"] is True
+ # Guards must have passed — run_write raises/flags rather than return junk.
+ assert "Contoso" in result["written"]
+
+
+def test_style_profile_requires_selection(tmp_path: Path):
+ """README quick start must keep `select` before `build-style-profile`."""
+ import pytest
+
+ from personality_protect.config import get_paths, init_profile
+ from personality_protect.style_profile import run_build_style_profile
+
+ init_profile("t", home=tmp_path)
+ with pytest.raises(FileNotFoundError):
+ run_build_style_profile(get_paths("t", home=tmp_path))
+
+
+def test_select_default_includes_current_year(tmp_path: Path):
+ """A corpus written this year must not select to zero under bare `select`."""
+ from datetime import datetime, timezone
+
+ from personality_protect.config import DEFAULT_THROUGH_YEAR
+ from personality_protect.models import Piece
+ from personality_protect.select import filter_pieces
+
+ this_year = datetime.now(timezone.utc).year
+ assert DEFAULT_THROUGH_YEAR >= this_year
+
+ recent = Piece(
+ id="r1",
+ source="linkedin_post",
+ text="word " * 200,
+ year=this_year,
+ word_count=200,
+ )
+ kept = filter_pieces(
+ [recent],
+ min_words=50,
+ through_year=DEFAULT_THROUGH_YEAR,
+ include_undated=False,
+ )
+ assert [p.id for p in kept] == ["r1"]
+
+
def test_api_health(tmp_path: Path):
import threading
import urllib.request