Creator decks closeout

Brandon asked for two things: finish the /mtg-creator-decks update, and fix the deck price that stopped appearing in KC Card Finder. The second turned out to be one symptom of a site-wide accident from this morning.

1. The price that stopped appearing

What removed it

strip_kid_mode.py ran at 11:13 to retire kid mode across the site. Its script rule deleted an entire <script> block whenever a kid identifier appeared anywhere inside it, including inside a quoted string or a comment:

def _script(m):
    body = m.group(1)
    return "" if ("toggleKid" in body or "kidifyText" in body) else m.group(0)

On pages where the kid toggle shared one block with the page's own code, the whole application went with the toggle. For three hours nothing appeared broken, because on most pages the deletion was correct.

2. What was actually damaged

The worst hit

Every page that lost code, re-checked against the deployed site

$ python3 verify_live.py        # curls every repaired route and re-parses it
checked 104 repaired pages LIVE
  http 200            : 104/104
  has script          : 104/104
  kid mode gone       : 104/104
  no dead handlers    : 104/104
  total functions live: 922

zero problems across all 104

All 104 re-checked against the live site after the deploy: 104 of 104 at http 200, 104 of 104 carrying script again, 104 of 104 with kid mode gone, 104 of 104 with no button calling a missing function.

3. The repair, and the gate that proves it

The fix recomputes each page from its pristine backup rather than patching the damaged file, so the result is kid mode gone and everything else intact in a single write. The new stripper reads each block through a string, comment and regex mask, and deletes only the kid spans.

Four gates, all green before anything was written

$ python3 repair.py            # dry run, gates enforced in both modes
DRY RUN  scanned=2146  would_change=278  gate_failures=0

pages that get real JavaScript back: 104
   76 fns  marvel-commander-rankings/index.html
   53 fns  chiron-combinations-v4/index.html
   40 fns  commander-marvel/index.html
   38 fns  commander-tracker/index.html
   36 fns  joni/index.html
   36 fns  kc-card-finder/index.html

$ python3 ~/.claude/skills/_shared/strip_kid_mode.py --all   # after the repair
WOULD STRIP 0 files

Three bugs the gates caught in my own first attempt

• A template literal broke the masker. ${JSON.stringify(n).replace(/"/g,'"')} holds a regex containing a double quote. Counting braces alone read that quote as a string opener and swallowed the rest of the file, which made all 36 KC Card Finder functions invisible to the tool meant to protect them.
• The kid component is an IIFE. Removing its inner functions one at a time left the dictionary and helpers behind, so a block that was kid mode entire came out half-deleted.
• Two gate regexes over-matched. onclick="window.open(url)" matched a call to a page function named open, and pages that display onclick="toggleKid()" as documentation text were counted as having a live button.

4. Proof it works, end to end

A real link taken off the creator page, loaded live, with no hand-editing:

https://drwu-htmls.vercel.app/kc-card-finder?deck=20%20Mountain%0A4%20Burst%20Lightning%0A4%20Shock%20(and%207%20more%20lines)

textarea auto-filled     11 lines
priceDeck defined        true
page functions present   5 of 5
kid toggle gone          true
rows priced              11
grand total              $9.08
progress                 "Done. Checked 11 cards across 13 live KC stores."

Cheapest local copies came back named per store: Gamers Getaway KC, Larry's Game Store, MindGames and Magic, Pulp Fiction Comics & Games, Game Cafe.

5. The creator decks ask ledger

closed Row 12 · deck tech, gameplan, sideboard plan

Every deck card now carries a Gameplan block counted off that deck's own list against Scryfall mana value, type line and oracle text. Nothing is written about a deck that was not measured on that deck.

Hand-checked against the raw list: 22 lands, 34 creatures, 4 other spells, average 2.7, 25 at one or two, top 8. Exact match.

The sideboard half is labelled, not asserted

The harvested lists are flat. The board split was discarded when the lists were fetched, so no mainboard/sideboard boundary exists in the data. Where a list runs past 60, those cards are reported as past the first 60 and labelled inferred. The evidence for the inference is stated on the page rather than hidden:

• 83 of the 106 such decks have exactly 15 cards there, which is the legal sideboard size.
• Across them the tail holds 39 lands against 2,988 in the first 60, and 103 of 129 tails hold none.

That pattern is what mainboard-then-sideboard order looks like across the corpus. For any one deck it stays an inference, so the page says so.

capped at the source Row 1 · at least 7 decks per creator

12 of 16 creators clear 7. The four that do not were listed as simply "short", with no reason given. Each is capped by something measurable, and the page now says which:

Strictly Better MtG and CROKEYZ cannot be raised by harvesting harder. One does not publish lists; the other's other decks are not Standard.

6. Left open, and written down

Both are in the shift queue rather than dropped.

7. Files touched