sports-stat-storyteller

name

sports-stat-storyteller

description

Turn a single strange, extreme, or counterintuitive sports statistic into a publication-ready newsletter. Use this skill any time a user notices an odd number in any sport — a pitcher's 67.5 ERA, a batter's .667 average through 5 AB, a QB's perfect passer rating — and wants to understand WHAT it means, WHY it happened mathematically, WHO else has been here before, and HOW LONG recovery will take. Drives an end-to-end workflow: statistical diagnosis, historical search, Bayesian modeling, schedule-mapped projections, broadsheet HTML newsletter. Also handles daily publishing from queue to GitHub Pages, Sunday Edition recaps with fresh data and model reruns, and cross-reference checks. Trigger on "strange stat", "crazy number", "what does this mean", "publish the next issue", or any stat with a small denominator.

Sports Stat Storyteller Skill

Purpose

Convert a single anomalous sports statistic into an educational newsletter that helps fans understand the mathematical structure behind what they're seeing — without requiring statistics knowledge. The canonical workflow is the Paul Skenes Opening Day 2026 story: 5 ER, ⅔ IP, 67.5 ERA → newsletter explaining denominator leverage, historical parallels, Bayesian recovery, and schedule-mapped projections.

---

Tone Selection (REQUIRED before writing each issue)

Before writing any issue, present the user with tone options and ask which voice fits this story. Suggest 2-3 that seem like the best match. The newsletter should feel like it has multiple sports writers, not one voice.

Voice	Style	Best For	Cadence
The Columnist	Dry wit, editorial authority, measured confidence	Analysis, historical comparisons, EO series	Daily default
The Heckler	Irreverent, sarcastic, bar-stool energy, trash talk	Jets misery, bad trades, front office disasters	As needed
The Professor	Patient, empathetic, explains the math gently	Bayesian explainers, Sunday recaps, teaching moments	Sundays, teaching pieces
The Eulogist	Respectful, melancholic, beautiful prose about loss	Player injuries, franchise collapses, end-of-era	As needed
The Fan	Emotional, raw, honest, first-person energy	Drought pieces, playoff heartbreak, loyalty	As needed
Sal	Perfectly worded sarcasm, accumulated wisdom, long-form essay	The big questions that have simmered for weeks	ONCE every 2-3 weeks max

Example prompt to user: "This piece on the Jets' draft history could work as The Heckler (sarcastic roast) or The Eulogist (mourning what could have been). Which voice do you want?"

For the automated publisher (scheduled trigger), default to The Columnist unless the content obviously calls for another voice (e.g., injury = Eulogist, Jets = Heckler). NEVER auto-publish as Sal — Sal only writes when the user explicitly invokes him after a question has been simmering.

---

Sal's Column (Guest Columnist — Special Rules)

Sal is The Sports Page's guest columnist. He's a rare voice — appearing once every 2-3 weeks at most. His pieces are events, not daily filler. Readers should yearn for Sal's column between appearances.

Sal's Persona

• Voice: Perfectly worded sarcasm. Dry. Pointed. Never cruel.
• Posture: Older, more experienced. Has seen it all and remembers.
• Method: Accumulates data for weeks before writing. Reads every footnote.
• Effect: By the end of a Sal column, the reader is laughing AND convinced.
• NOT: Mean-spirited, cheap, quick, or reactive. Never rushed.

Sal's Style Rules

• Long, winding sentences with parenthetical asides (like this one, which exists to prove the point)
• Well-placed em-dashes for rhetorical pivots — the kind that make you slow down
• Occasional all-caps for emphasis, but sparingly (once per piece, maybe twice)
• Dry openings that set a scene before making the point
• Treats the reader as an equal, never condescends
• Cites specific numbers with confidence but no lectures
• Wraps with a one-line verdict that lands like a door closing

Sal's tells: "Ah yes," "As one does," "Of course," "It is said," "One might ask," "It turns out" — deployed with deliberate slowness.

When Sal Writes

Sal only writes when ALL of these are true:

1. The question has been in the newsletter ecosystem for 2+ weeks
2. Enough data has accumulated to make a definitive argument
3. The user explicitly invokes Sal ("let Sal take this one")
4. No Sal column has been published in the past 2-3 weeks
5. The topic is a BIG question — not a daily small-sample piece

Sal does NOT write:

• Breaking news (too fresh)
• Sunday recaps (that's The Professor)
• Quick analytical pieces (those are The Columnist)
• Mean-spirited takedowns (those are The Heckler)
• Routine coverage of any kind

Sal's Format

Sal's columns are LONGER than normal issues (expect 600-800 lines):

• Extended scene-setting opening (2-3 paragraphs)
• The question framed with unusual patience
• Multiple layers of data with commentary between them
• Historical parallels that most writers would miss
• A verdict that reframes the question entirely
• Byline reads: "By Sal · Guest Columnist"
• Masthead kicker reads: "Sal's Column · The Sports Page"
• Optional: a small "Sal's Desk" visual element (pull quote in a distinct style)

Sal's Topic Selection

Sal goes after the questions everyone is talking about but nobody is thinking about correctly. Examples of Sal-worthy topics:

• "The WNBA pay gap isn't what you think it is" (after weeks of debate)
• "The Jets aren't cursed. They're just bad at drafting." (after the draft)
• "The A's problem isn't Oakland. It's the owner." (after the move)
• "Coaches don't need 5 years. Year 1 tells you 60%." (pre-season)

Sal is a scalpel, not a hammer. Save him for the cuts that matter.

---

Workflow (run in order)

Step 0 — Sync with GitHub (always, before any other step)

cd <repo path>
git fetch origin main
git status -sb
git pull origin main --ff-only

This is a hard prerequisite for every invocation of this skill on every machine — drafting a new issue, queueing, publishing, Sunday Edition, even just inspecting the queue. The repo is shared across machines and the GitHub Actions autopublish bot pushes commits Mon–Sat at 4:30am ET. A stale local clone will silently produce wrong issue numbers, wrong recap windows, and broken cross-references.

If git pull --ff-only fails: stop and reconcile before doing anything else. Never force-pull or stash blindly. See CLAUDE.md "Step 0 — SYNC WITH GITHUB FIRST" for the full rationale and the 2026-05-10 incident that motivated this rule.

Step 1 — Stat Intake & Diagnosis

When a user presents a stat or event, immediately diagnose:

1. What type of stat is it? Rate (ERA, BA, FG%) vs count (HR, SO) vs composite (passer rating, WAR)
2. What is the denominator? This is the key question. Small denominators weaponize numerators.
3. Direction of extremity: Unexpectedly high? Low? Impossible to sustain in either direction?
4. Baseline: What is typical for this player, this position, this era?

Denominator Leverage Formula:
  rate_stat = numerator / denominator × scale
  sensitivity = d(rate_stat)/d(numerator) = scale / denominator

The smaller the denominator, the more sensitive the rate stat is to any
single event. A pitcher who gives up 5 ER in 2/3 IP has a denominator of
0.67 — enormously sensitive.

Print the diagnosis inline before proceeding.

Step 2 — Historical Search

Use web_search to find:

• Elite performers who had similar anomalous short outings
• Recovery stories: did they bounce back? How fast?
• Any all-time records this event touches (shortest Cy Young start, worst Opening Day ERA, etc.)
• ERA- or performance-adjusted comparisons across eras

Search queries to run (adapt for sport):

"{player name} {event} reaction historical"
"elite {position} terrible short outing career {sport} history"
"{player name} compared to {comparable legend}"
"worst {stat} Opening Day history MLB"

Surface 2–4 historical parallels that illuminate the current event.

Step 3 — Bayesian Recovery Model (Python)

For ERA-based recovery (adapt formula for other stats):

import numpy as np

# Career state before bad event
prior_ip  = <career_IP>
prior_er  = <career_ER>
target    = <target_ERA>   # career ERA before disaster

# Post-disaster
cur_ip = prior_ip + bad_ip
cur_er = prior_er + bad_er

# Gamma-Poisson conjugate
# ER/inning ~ Poisson(λ),  λ ~ Gamma(α, rate=β)
alpha = 0.1 + prior_er
beta  = 0.1 + prior_ip   # rate parameterization

N_SIM  = 200_000
ip_mu  = <avg_IP_per_start>   # e.g. 6.5 for SP
ip_sig = 1.2

starts_needed = np.full(N_SIM, 162, dtype=int)
for i in range(N_SIM):
    lam = np.random.gamma(shape=alpha, scale=1.0/beta)
    er, ip = cur_er+0.0, cur_ip+0.0
    for s in range(1, 163):
        ip_s = float(np.clip(np.random.normal(ip_mu, ip_sig), 1.0, 9.0))
        er_s = int(np.random.poisson(lam * ip_s))
        er += er_s;  ip += ip_s
        if er * 9.0 / ip <= target:
            starts_needed[i] = s;  break

rec = starts_needed[starts_needed < 162]
print(f"P(recover this season): {len(rec)/N_SIM:.1%}")
for p in [10,25,50,75,90]:
    print(f"  {p}th pct: {int(np.percentile(rec,p))} starts")

Key insight to always explain: Recovery requires pitching better than the target ERA, not just equal to it. Average future performance locks the ERA above target forever if the damage is large enough.

Step 4 — Schedule Mapping

Look up the team's schedule (use web_search → ESPN or CBS Sports schedule pages) and project starts:

• Identify player's rotation slot (every 4th or 5th game)
• Map percentile starts to calendar dates and opponents
• Flag any notable matchups (division rivals, nationally televised)

Step 5 — Newsletter Generation

Build an HTML artifact in broadsheet newspaper style (see Design System below). Required sections:

1. Masthead — publication name, vol/issue, date, tagline
2. Headline — bold, punchy, includes the extreme number
3. Deck (subheadline) — one sentence capturing the counterintuitive insight
4. Stat card row — 3 key numbers: the extreme stat, current real number, target/prior number
5. Two-column body — narrative lead + "what it actually means"
6. The Math explainer box (dark background, monospace formula block)
7. Recovery chart (SVG, inline) — ERA/stat trajectory by percentile
8. Recovery table — percentile → starts needed → date → opponent
9. Historical parallels — 2–4 stories from search results
10. Pull quote — the key statistical insight in Playfair Display italic
11. "Ask the Stats Desk" AI tool — Claude API-powered textarea at bottom that contextualizes any stat the reader pastes in
12. "Pitch a Story" box — REQUIRED on every piece. Canonical Formspree form that routes reader pitches to Patrick's inbox; he forwards selected ones to GitHub Issues with the story-idea label so Claude reads them when planning queue. See template below — copy verbatim, do not redesign per piece. Inserted INSIDE the .paper div, immediately before <div class="footer">.

Step 6 — Skill Output

Produce two files:

• [topic]_newsletter.html — the artifact (present to user)
• This SKILL.md is the reusable template

---

Design System

Aesthetic direction: editorial broadsheet — aged newsprint, ink, serif type. NOT a tech blog. NOT a sports app.

/* Core palette */
--ink:    #1a1208;   /* near-black, slightly warm */
--cream:  #f5f0e8;   /* paper background */
--aged:   #e8e0cf;   /* outer background */
--rust:   #b83a1e;   /* accent / bad stats */
--steel:  #2c4a6e;   /* accent / good stats / explainer BG */
--gold:   #c9962a;   /* AI tool accent */
--muted:  #6b5e4a;   /* secondary text */

/* Typography */
/* Display: Playfair Display — headlines, pull quotes */
/* Body: Libre Baskerville — editorial copy */
/* Data/labels: Roboto Mono — stats, datelines, captions */

/* Load from Google Fonts: */
/* Playfair+Display:ital,wght@0,400;0,700;0,900;1,400;1,700 */
/* Libre+Baskerville:ital,wght@0,400;0,700;1,400 */
/* Roboto+Mono:wght@400;600 */

Component patterns:

• Masthead: centered, triple-bordered, vol/issue/date line
• Stat cards: 3-column grid, val class for big number, lbl for mono label
• Pull quote: double-ruled top/bottom, Playfair italic centered
• Explainer box: background: var(--steel), white text, .math code block
• Recovery chart: SVG, hand-drawn, labeled percentile curves
• AI Tool: background: var(--ink), gold accents, Anthropic API call

Recovery chart SVG guide:

• viewBox="0 0 680 260"
• Y-axis: ERA (0–4), scale = 52.5px per ERA unit, y = 220 - ERA*52.5
• X-axis: starts (0–22+), scale = ~27px per start, x = 54 + start*27
• Draw 3 curves for 10th/25th/50th percentile
• Target ERA: green dashed line
• Current ERA: rust dashed line
• Opening-Day dot: rust filled circle at origin

---

Statistical Concepts to Always Explain

The Denominator Problem

Rate stats are ratios. Small denominators amplify single events catastrophically:

ERA = ER × 9 / IP
sensitivity = 9 / IP   ← explodes as IP → 0

Translate: "5 ER in ⅔ IP = 5 ER in 0.2% of career innings moved ERA by 7%"

Stabilization Thresholds (approximate)

Stat	PA/IP for stability
ERA	50 IP (seasonal), 150+ IP (reliable)
Batting avg	150 PA
K%	70 PA
BB%	120 PA
FG% (hoops)	100 attempts
QB rating	100+ pass attempts
Save %	30+ save opportunities

Recovery Math

Deficit = 9×current_ER − target_ERA×current_IP
Units gained per start = target_ERA×IP_start − 9×ER_start
  → positive only when start_ERA < target_ERA

Bayesian Intuition

The Gamma-Poisson model treats a player's true rate as uncertain (Gamma prior) and updates based on all observed events (career + bad game = posterior). Future starts are drawn from this posterior — meaning the model respects both "he's historically great" and "something weird just happened."

---

Historical Parallel Template

For each historical parallel, structure as:

1. Setup: Who, when, what happened (1–2 sentences)
2. The number: What was the extreme stat
3. The math: Why it looked so bad
4. The recovery: How long, what it took, outcome

Canonical Parallels (for pitchers)

Dwight Gooden, April 1984 (second career start) Second career start at Wrigley: 3⅓ IP, 6 ER. Told his father he "may not be ready yet." He was. Gooden went on to finish 17-9 that season, then produced a 1.53 ERA in 1985 — arguably the greatest single season since the Deadball Era. The bad start is now a footnote.

Doc Gooden's 1985 ERA started at 4.50 After his first start of 1985, Gooden's ERA stood at 4.50. His second start was a complete-game shutout with 10 Ks. His ERA never went above 2.00 for the rest of the year. The opening stumble became invisible.

Roger Clemens, 1986 Opening Day Coming off back-to-back Cy Young seasons, Clemens occasionally had brutal short outings that inflated his ERA briefly before his workload diluted them. His career ERA of 3.12 weathered every individual disaster because the denominator — 4,916 IP — made any single start a rounding error.

Pedro Martínez, 2000 AL Pedro's 1.74 ERA in 2000 was so dominant that even after rough outings his ERA barely twitched — because the denominator was large. By contrast, a comparable rate in his 2nd career start would have shown an ERA >30.

---

Pitch a Story — Reader Idea Form (REQUIRED on every piece)

This box belongs INSIDE the .paper div, immediately before the <div class="footer">. It is the canonical Formspree form: clean inline inputs, gold accent, GitHub link as a sidecar option. Reader submissions hit Patrick's inbox via Formspree; he routes the good ones into GitHub Issues with the story-idea label so Claude can read them when planning queue. Copy the HTML below verbatim — do not redesign per piece.

  <!-- STORY IDEAS INBOX -->
  <div id="pitch" style="border:2px solid var(--gold);background:var(--card);padding:1.5rem 1.8rem;margin:1.8rem 0">
    <h3 style="font-family:'Playfair Display',serif;font-size:1.1rem;font-weight:700;color:var(--gold);margin-bottom:.3rem">Pitch a Story</h3>
    <p style="font-size:.88rem;color:var(--muted);margin-bottom:1rem">Noticed a weird stat? Saw something that doesn&rsquo;t add up? Send it in. The best ideas become issues.</p>
    <form id="pitch-form" action="https://formspree.io/f/xykbazvr" method="POST" style="display:flex;flex-direction:column;gap:.6rem">
      <input type="hidden" name="_subject" value="New Story Pitch &mdash; The Sports Page">
      <input type="text" name="_gotcha" style="display:none" tabindex="-1" autocomplete="off">
      <input type="text" name="name" id="pitch-name" placeholder="Your name" required style="font-family:'Libre Baskerville',serif;font-size:.9rem;padding:.5rem .7rem;border:1px solid var(--div);background:var(--cream);color:var(--ink);outline:none">
      <input type="text" name="sport" id="pitch-sport" placeholder="Sport (MLB, NFL, NHL, CFB, WNBA, etc.)" style="font-family:'Libre Baskerville',serif;font-size:.9rem;padding:.5rem .7rem;border:1px solid var(--div);background:var(--cream);color:var(--ink);outline:none">
      <textarea name="idea" placeholder="What&rsquo;s the stat or story? Why is it weird, interesting, or important?" required rows="3" style="font-family:'Libre Baskerville',serif;font-size:.9rem;padding:.5rem .7rem;border:1px solid var(--div);background:var(--cream);color:var(--ink);outline:none;resize:vertical"></textarea>
      <div style="display:flex;gap:.8rem;align-items:center">
        <button id="pitch-btn" type="submit" style="background:var(--gold);color:var(--ink);font-family:'Roboto Mono',monospace;font-size:.72rem;font-weight:600;letter-spacing:.12em;text-transform:uppercase;border:none;padding:.6rem 1.4rem;cursor:pointer">Submit Idea</button>
        <span style="font-family:'Roboto Mono',monospace;font-size:.6rem;color:var(--muted)">or <a href="https://github.com/pem725/the-sports-page/issues/new?title=Story+Idea:&labels=story-idea&body=Sport:%0A%0AStat+or+story:%0A%0AWhy+it%27s+interesting:" style="color:var(--steel)">post on GitHub</a></span>
      </div>
    </form>
    <div id="pitch-ok" style="display:none;padding:.8rem 0;font-family:'Roboto Mono',monospace;font-size:.78rem;color:var(--green);font-weight:600;letter-spacing:.08em">Sent! We read every pitch. The weird ones become issues.</div>
    <div id="pitch-err" style="display:none;padding:.8rem 0;font-family:'Roboto Mono',monospace;font-size:.78rem;color:var(--rust);font-weight:600;letter-spacing:.08em">Something went wrong. Try the GitHub link instead.</div>
  </div>
  <script>
  document.getElementById('pitch-form').addEventListener('submit',async function(e){
    e.preventDefault();
    var btn=document.getElementById('pitch-btn'),ok=document.getElementById('pitch-ok'),err=document.getElementById('pitch-err');
    btn.disabled=true;btn.textContent='SENDING\u2026';ok.style.display='none';err.style.display='none';
    try{
      var r=await fetch(this.action,{method:'POST',body:new FormData(this),headers:{'Accept':'application/json'}});
      if(r.ok){ok.style.display='block';this.reset();}
      else{err.style.display='block';}
    }catch(x){err.style.display='block';}
    btn.disabled=false;btn.textContent='SUBMIT IDEA';
  });
  </script>

Formspree endpoint f/xykbazvr is the production endpoint for The Sports Page; do not change it. Patrick monitors the inbox and routes selected pitches to GitHub Issues (manually or via Gmail filter rules) so they land where Claude reads them.

If a future piece is missing this box, run scripts/migrate_pitch_box.py to backfill cleanly — the script removes any older variant and inserts the canonical Formspree form before <div class="footer">.

---

Ask the Stats Desk — AI Tool Template

<div class="stats-desk">
  <div class="stats-desk-head">Ask The Stats Desk</div>
  <div class="stats-desk-sub">Drop any strange stat. Get context.</div>

  <!-- Quick examples -->
  <div class="desk-examples">
    <span class="desk-example" onclick="fillExample(this)">Batter is .667 after 3 AB</span>
    <span class="desk-example" onclick="fillExample(this)">Kicker 0-for-3 (0% hit rate)</span>
    <!-- add more examples relevant to the newsletter's sport -->
  </div>

  <textarea id="stat-input" placeholder="Paste your weird stat here..."></textarea>
  <button onclick="askDesk()">▶ Explain the Context</button>

  <div id="desk-response"></div>
</div>

<script>
async function askDesk() {
  const input = document.getElementById('stat-input').value.trim();
  if (!input) return;
  const resp = await fetch("https://api.anthropic.com/v1/messages", {
    method: "POST",
    headers: { "Content-Type": "application/json" },
    body: JSON.stringify({
      model: "claude-sonnet-4-20250514",
      max_tokens: 1000,
      system: `You are The Stats Desk — a sharp sports statistics explainer.
Given a weird sports stat, explain:
1. WHY the number looks so extreme (denominator, leverage, small sample)
2. What it would take to sustain it — anchor to reality
3. Historical context: has anyone been here before?
4. What it actually MEANS vs what it DOESN'T mean
5. Verdict: red flag / noise / genuinely impressive?
Style: editorial, dry wit, flowing paragraphs, no bullet lists. 2-4 paragraphs.`,
      messages: [{ role: "user", content: input }]
    })
  });
  const data = await resp.json();
  const text = data.content?.find(b=>b.type==='text')?.text || "No response.";
  document.getElementById('response-text').textContent = text;
  document.getElementById('desk-response').classList.add('visible');
}
</script>

---

Quality Checklist (before presenting artifact)

• Pitch a Story box (Formspree form) inserted INSIDE .paper before the footer
• Headline includes the extreme number
• Denominator math is shown explicitly
• At least 2 historical parallels found via web search
• Bayesian model run with ≥100,000 simulations
• Recovery percentiles mapped to actual schedule dates
• P(never recover this season) stated explicitly
• "Ask the Stats Desk" AI tool functional with Anthropic API call
• All assumptions stated in footer
• Broadsheet aesthetic: Playfair Display + Libre Baskerville + Roboto Mono
• Cross-reference check: Grep for "Issue #", "Issue No.", and "See Issue" — verify every reference points to an ALREADY PUBLISHED issue using reader-facing numbers (not internal file numbers). If referencing an unpublished piece, use "coming soon" instead of a number.

---

Adapting to Other Sports

Sport	Rate Stat	Denominator	Bayesian Model
Baseball (pitching)	ERA	IP	Gamma-Poisson on ER/inn
Baseball (hitting)	BA, OBP	PA	Beta-Binomial on hits/PA
Basketball	FG%, 3P%	Attempts	Beta-Binomial
Football	Passer Rating	Pass attempts	Beta-Binomial on comp, TD, INT
Football	FG%	Attempts	Beta-Binomial
Soccer	xG conversion	Shots	Beta-Binomial
Hockey	SV%	Shots faced	Beta-Binomial

For Beta-Binomial:

# Prior: Beta(α, β) where α, β derived from career stats
# Update: observed successes/failures → Beta(α+successes, β+failures)
# Posterior predictive: Beta-Binomial
alpha_post = career_successes + prior_alpha
beta_post  = career_failures  + prior_beta
p_samples  = np.random.beta(alpha_post, beta_post, N_SIM)

---

Example Trigger Phrases

This skill should fire whenever you see phrases like:

• "that's a crazy [stat]" / "what does that number mean"
• "how bad is [X]" / "how good is [X]" for a rate stat
• "he'll never recover" / "how long to fix his ERA"
• "worst [stat] in history" / "best ever through [N games/starts]"
• "is [small-sample stat] real"
• Any stat with explicit small sample: "through 3 starts", "in his first AB"

---

Publishing Workflow (for automated and manual publishing)

Repository Structure

index.html              <- Homepage with issue archive
publish.sh              <- One-command publish script
assets/                 <- Logo, banner, favicon, QR code
published/              <- LIVE issues (linked from index.html)
queue/                  <- Ready to publish (NOT linked)
reserve/                <- Evergreen, no specific date

How to Publish an Issue (step by step)

1. Select the article from queue/. Priority:

   • Timely pieces first (breaking news, game reactions)
   • Then analytical pieces (series parts, trade analyses)
   • Sunday = publish the Sunday Edition recap

2. Determine reader issue number: Count <div class="issue"> blocks in index.html. The next issue is that count + 1. This is the READER-FACING number — NOT the internal filename number.

3. Update the HTML file:

   • Find all occurrences of Issue No. [TBD] or Issue No. XX (usually 2: datebar and footer)
   • Replace with the correct reader number: Issue No. 14
   • Update the date if it has a placeholder or old date

4. Cross-reference check (REQUIRED):

   grep -n "Issue #\|See Issue" queue/FILENAME.html
   

   Every reference must point to an ALREADY PUBLISHED issue using reader-facing numbers. If it references an unpublished piece, change to "coming soon" or remove the reference.

5. Add entry to index.html: Insert a new issue block at the TOP of <div class="issues">, BEFORE all existing issues. Use this exact HTML pattern:

   <div class="issue">
     <div class="issue-num">NUMBER</div>
     <div class="issue-body">
       <div class="issue-date">Month Day, Year</div>
       <div class="issue-hed"><a href="published/FILENAME.html">HEADLINE</a></div>
       <div class="issue-deck">One-sentence summary.</div>
       <div class="issue-tags">
         <span class="tag mlb">SPORT</span>
         <span class="tag">TOPIC</span>
       </div>
     </div>
   </div>
   

   Available tag classes: .mlb, .nfl, .nhl, .cfb (colored), or plain .tag.

6. Move the file:

   git mv queue/FILENAME.html published/FILENAME.html
   

7. Commit and push:

   git add index.html published/FILENAME.html
   git commit -m "Publish Issue #N: brief description"
   git push
   

Sunday Edition Workflow

The Sunday Edition MUST be built fresh on Sunday morning — never from a pre-built draft. Follow these steps:

Step 1 — Gather fresh data (REQUIRED before writing anything):

• Use web search to pull current MLB standings, team records, and player stats
• Check injury report updates (especially for players we've been tracking, e.g., Soto calf)
• Get current NHL standings and playoff picture
• Check NFL draft news if draft season
• Get any other relevant current stats for teams we cover

Step 2 — Rerun prediction models with updated data: For EVERY prediction made in the past week's issues, rerun the relevant model:

• Beta-Binomial BA projections: update with current AB/hits
• Bayesian injury estimates: update with actual recovery timeline
• Season win projections: update with current W-L record
• Any other quantitative prediction: pull actual outcome and compare

Step 3 — Score each prediction:

• HIT: Our prediction/model matched the outcome within a reasonable margin
• MISS: We were materially wrong — own it, explain why
• PARTIALLY HIT: Direction was right, magnitude was off (or vice versa)
• PENDING: Not enough data yet to judge — state what we're still watching

Step 4 — Write the Sunday Edition:

• Recap each issue from the week with a one-line summary and link
• Present the prediction scorecard with FRESH numbers (not stale pre-built data)
• Honest self-assessment: over-reactions and under-reactions
• What we learned, what we'd do differently
• Preview of next week's planned content

Step 5 — Publish:

• Follow the standard 7-step publish workflow
• This issue should be LONGER than regular issues

Key rule: If a Sunday Edition skeleton exists in queue/ from an earlier session, treat it as a TEMPLATE ONLY. Replace all statistics, scores, and predictions with fresh data pulled that morning. Never publish stale numbers on Sunday.

Issue Numbering Rules

• Reader-facing numbers are sequential: #1, #2, #3... assigned at publish time
• Internal filenames keep their creation-order numbers (e.g., 010-soto-calf.html was published as reader Issue #7)
• NEVER expose internal file numbers to readers
• Count existing published entries in index.html to determine the next number

Content Tiers

Tier	Shelf Life	When to Publish
Timely	Hours to 1 day	Immediately — goes stale fast
Analytical	1-2 weeks	Fill gaps between timely pieces
Evergreen	Indefinite	Slow news weeks, pre-season
Sunday Edition	Must be fresh	Sunday mornings only

Editorial Rules

• ONE statistical question per issue
• No holiday specials — sports only
• Do NOT label who roots for which team — cover the teams naturally
• Milestone celebrations at Issue #50, #100, etc.
• No basketball content
• Goal: 500 total issues over ~2 years

Team Priorities (in order)

1. Notre Dame football (always, year-round)
2. College football (during season; off-season only for ND)
3. NY Mets (during season)
4. NY Rangers (when doing well)
5. NY Jets + Raiders, Bills, Seahawks (extended family)
6. Any player with a remarkably great or terrible performance

Sport-Season Calendar

Pieces should track the live sports calendar. Each month has a different mix of "in season" content; pieces about dormant sports must justify their timing.

Month	Live cycles
Jan	NFL playoffs / Super Bowl buildup; CFP National Championship; hot stove finishes
Feb	Super Bowl; CFB Signing Day (1st Wed); MLB spring training opens mid-month
Mar	MLB spring training; March Madness; MLB Opening Day = newsletter anniversary
Apr	MLB Opening Week; NBA/NHL playoffs; Masters; NFL Draft (last week)
May	MLB regular; Kentucky Derby (1st Sat); Indy 500 (Memorial Day); NBA/NHL conf finals
Jun	Stanley Cup Final; NBA Finals; US Open golf; College World Series
Jul	MLB All-Star Break; Wimbledon; Open Championship; MLB trade deadline (end of month)
Aug	NFL preseason; CFB opens (last week); MLB pennant races
Sep	NFL Week 1; CFB conference play; MLB stretch; US Open tennis; WNBA playoffs
Oct	MLB playoffs / World Series; NFL mid; CFB mid; NHL/NBA open
Nov	NFL stretch; CFB rivalry week (Thanksgiving); MLB awards; CFP rankings begin
Dec	NFL playoff picture; CFB bowls / CFP first round; MLB Winter Meetings

Tentpole annual events that warrant pre-planned issues: MLB Opening Day (anniversary), First Day of Spring Training, Super Bowl Sunday (eve + morning-after), March Madness Selection Sunday, Masters Sunday, NFL Draft Round 1, MLB All-Star Break, MLB Trade Deadline, World Series Game 1, CFB Rivalry Week, CFP National Championship.

Full guidance in CLAUDE.md → "Sport-Season Calendar". Use that as the authoritative version.