DEV Community

Cover image for PrizePicks API in Python: Only 15% of the Lines on 11 Boards Are Standard Lines
Happy Endpoint
Happy Endpoint

Posted on Originally published at Medium

PrizePicks API in Python: Only 15% of the Lines on 11 Boards Are Standard Lines

A PrizePicks board looks like one number per player and stat: 24.5 points, over or under. On Tuesday 3 November 2026, at about 19:45 UTC, it held 20,436 lines across 21 boards, and on the 11 boards I could summarise (12,004 of those lines; the two largest boards refused a board-wide read) only 15% were that kind of line. The rest were demons and goblins, the harder and easier alternative lines PrizePicks posts next to the standard one.

So I asked a plain question about the structure of the board, not about any game. How are the lines spread across sports, how much of each board is standard, demon or goblin, how far do the alternatives sit from the standard line, and how often does a line move? I pulled the data with Python and counted. This is data analysis of how the board is built, not a betting tip: nothing below says which side of any line to take.

The data

I used the PrizePicks API, which returns the PrizePicks board as JSON, so there is no scraper to build or maintain. Three endpoints did the work. /leagues lists every league with its live projections_count. /board-summary returns one board in numbers, including counts_by_odds_type. And /projections returns a board line by line, 200 per page, with line_score, stat_type, odds_type, allowed_wager_types, trending_count and a group_key that ties a player's lines for one stat in one game together.

I read /leagues and the board summaries on Tuesday 3 November 2026 at about 19:45 UTC, and the full NBA board twice: at 19:45 UTC and again at 22:00 UTC, a day before any of its games started. The NBA board was the following evening's slate: 14 games, 104 players at the first read, 15 stat types. Each line also names its player. I dropped names, photos and logos before saving anything and kept only the player's numeric id, team and position, so nothing below names an athlete.

The code

This reads one league's whole board and prints the three things the article is built on: the odds-type mix, the ladder of alternatives around each standard line, and the pick counts. It needs requests and a key in RAPIDAPI_KEY; you can get one on the PrizePicks API listing on RapidAPI. The NBA board took 8 calls at my first read and 11 at the second, as more lines went up.

import json
import os
import statistics
import time
from collections import Counter, defaultdict

import requests

HOST = "prizepicks-api.p.rapidapi.com"  # PrizePicks API on RapidAPI
BASE = os.environ.get("API_BASE", "https://" + HOST)
LEAGUE = "NBA"


def headers():
    if "API_BASE" in os.environ:
        return json.loads(os.environ.get("API_HEADERS", "{}"))
    return {"x-rapidapi-host": HOST, "x-rapidapi-key": os.environ["RAPIDAPI_KEY"]}


def board(league):
    """Every line on one league's board, 200 per call."""
    lines, page = [], 1
    while True:
        params = {"league": league, "per_page": 200, "page": page}
        r = requests.get(BASE + "/projections", params=params, headers=headers(), timeout=60)
        r.raise_for_status()
        data = r.json()["data"]
        lines += data["projections"]
        if not data["pagination"]["has_more"]:
            return lines
        page += 1
        time.sleep(1.5)


lines = board(LEAGUE)
odds = Counter(x["odds_type"] for x in lines)
print("%d lines: %s" % (len(lines), ", ".join(
    "%s %d (%.0f%%)" % (k, v, v / len(lines) * 100) for k, v in odds.most_common())))

# A player's lines for one stat in one game share a group_key: the standard line and its alternatives
groups = defaultdict(list)
for x in lines:
    groups[x["group_key"]].append(x)
steps, no_standard = defaultdict(list), 0
for rungs in groups.values():
    std = [x["line_score"] for x in rungs if x["odds_type"] == "standard"]
    if not std:
        no_standard += 1
        continue
    for kind, nearest_first in (("goblin", True), ("demon", False)):
        alt = sorted((x["line_score"] for x in rungs if x["odds_type"] == kind), reverse=nearest_first)
        for i, value in enumerate(alt, 1):
            steps["%s %d" % (kind, i)].append((value - std[0]) / std[0] * 100)
print("%d player-stat-game groups, %d with no standard line" % (len(groups), no_standard))
for step in ("goblin 2", "goblin 1", "demon 1", "demon 2", "demon 3"):
    print("%s: median %+.1f%% from the standard line (%d lines)"
          % (step, statistics.median(steps[step]), len(steps[step])))

# trending_count: how many entries have taken a line (missing on most lines)
picks = Counter()
for x in lines:
    picks[x["odds_type"]] += x["trending_count"] or 0
total = sum(picks.values())
for kind, n in picks.most_common():
    print("%s: %s picks (%.0f%%)" % (kind, format(n, ","), n / total * 100))
Enter fullscreen mode Exit fullscreen mode

I ran it at 22:00 UTC, two hours after my first read of the NBA board, and by then the board had grown to 11 pages. It printed:

2196 lines: demon 1134 (52%), goblin 616 (28%), standard 446 (20%)
805 player-stat-game groups, 359 with no standard line
goblin 2: median -22.2% from the standard line (112 lines)
goblin 1: median -11.4% from the standard line (201 lines)
demon 1: median +11.8% from the standard line (205 lines)
demon 2: median +37.2% from the standard line (204 lines)
demon 3: median +70.6% from the standard line (185 lines)
standard: 323,004 picks (83%)
demon: 34,497 picks (9%)
goblin: 33,196 picks (8%)
Enter fullscreen mode Exit fullscreen mode

The rest of the NBA numbers below come from the first read, at 19:45 UTC, when the board held 1,479 lines. The two reads agree on the shape: about a fifth to a quarter of the lines standard, the first goblin and the first demon 11% to 16% away from the standard line, and more than four in five counted picks on standard lines.

The group_key is the useful part. It joins player id, stat id, game and period, so every standard, demon and goblin line for one player's points in one game shares it. Without it, the ladder below would mean matching lines by player and stat yourself.

Where the lines are

UCL soccer carried 4,520 PrizePicks lines, the most of any board and three times the NBA's 1,479; CFB, NHL and NFL had 3,397 to 3,741 each.

/leagues listed 141 leagues, and 21 of them had lines that evening: 20,436 in total. UCL soccer had the most, 4,520, or 22.1% of everything. College football (3,741), the NHL (3,505) and the NFL (3,397) came next, and the NBA, whose board held the next evening's 14 games, had 1,479.

Some boards are slices of another one. NHL1P (first period, 817 lines), NFL1H (first half, 443) and NFL1Q (first quarter, 243) carry a parent_name. Counted with their parents, four families hold 81.6% of the board: UCL 22.1%, NHL 21.1%, NFL 20.0% and college football 18.4%. Everything else, from the NBA (1,479) down to UFC (7), shares the remaining 18.4%.

Standard, demon, goblin

The API documentation describes odds_type in three values: standard is the normal line, demon a harder line at a bigger payout, and goblin an easier line at a smaller payout. /board-summary counts them per board. I asked it for the 13 boards with at least 100 lines. It answered for 11 boards with 12,004 lines between them; the two largest, UCL and college football, returned a 503 on every attempt (more on that below), and the 8 boards under 100 lines (175 lines together) were not asked.

Standard lines as a share of each board: CS2 72%, LoL 40%, NFL1Q 28%, NBA 25%, down to NFL first half 8.6%, soccer 0.5% and NHL first period 0%; 15% across the 11 boards.

Across those 12,004 lines, 1,799 were standard (15.0%), 8,088 demons (67.4%) and 2,117 goblins (17.6%). The spread between boards is wide:

Board Lines Standard Demon Goblin Players Lines per player
NHL 3,509 9.3% 79.1% 11.5% 466 7.5
NFL 3,402 17.5% 59.5% 23.1% 372 9.1
NBA 1,479 25.3% 50.1% 24.6% 104 14.2
SOCCER 836 0.5% 86.7% 12.8% 95 8.8
NHL1P 816 0.0% 95.7% 4.3% 223 3.7
TENNIS 589 17.3% 47.5% 35.1% 70 8.4
NFL1H 443 8.6% 79.2% 12.2% 259 1.7
CBB 336 22.3% 51.5% 26.2% 14 24.0
NFL1Q 243 28.4% 68.7% 2.9% 147 1.7
CS2 235 72.3% 14.5% 13.2% 60 3.9
LoL 116 39.7% 31.0% 29.3% 52 2.2

The esports boards lean the other way: CS2 is 72% standard and LoL 40%. The hockey and soccer boards are the opposite: the NHL board had 2,777 demons against 327 standard lines, the soccer board 4 standard lines out of 836, and the NHL first-period board none at all. If you think of a PrizePicks board as a list of standard lines with a few extras, these 11 boards say it is mostly extras.

How far the alternatives sit from the standard line

On the NBA board, the 1,479 lines fall into 651 group_key groups. 374 of them have a standard line, and around it a ladder: up to three goblins below and three demons above. I measured each step against the same group's standard line.

The first demon sits a median 15.7% above the standard line and the first goblin 13.3% below; the second steps are 42.1% above and 29.1% below, the third demon 68.6% above.

The first goblin sits a median 13.3% below the standard line (129 lines) and the first demon 15.7% above (139). The steps then widen: the second demon is 42.1% above, the third 68.6% above, the second goblin 29.1% below. In raw numbers on the Points market (29 groups with a standard line and alternatives), the median goblin steps were 2 and 4 points below the standard line and the median demon steps 2, 7 and 12 points above. Those bases are small: the second goblin rests on 10 groups and the third demon on 20. The order never breaks: in no group was a goblin at or above the standard line, or a demon at or below it.

The other 277 groups (42.5%) have no standard line at all, only demons or goblins. They are concentrated in a few markets: Dunks (56 groups), Double-Double (41), 3-PT Made (40) and Steals (37). Double-Double had no standard line for anyone; in Dunks, 3-PT Made and Steals most players had only alternatives, but some had a standard line (12, 19 and 19 groups).

Where the picks go

trending_count is, in the documentation's words, how many entries have taken a line. On the NBA board it was set on 297 of the 1,479 lines, and those 297 add up to 291,778 picks.

Standard lines are 25% of the NBA board and take 87% of the counted picks; goblins are 25% of lines and 7% of picks, demons 50% of lines and 6% of picks.

Standard lines are 25.3% of the NBA board and take 87.1% of the counted picks: 254,200, against 19,927 on goblins (6.8%) and 17,651 on demons (6.0%). Half the board is demons, and they draw one pick in seventeen. Part of that is coverage: 217 of the 374 standard lines (58%) carry a count, against 40 of the 742 demons and 40 of the 363 goblins. But per counted line the gap holds too: a median of 393 picks on a standard line, against 219 on a demon and 213 on a goblin. All ten of the most-picked lines on the board were standard lines, seven of them on Points.

How often a line moved in two hours

Between my two reads of the NBA board, 2.2 hours apart and well before the first tip-off, the board changed more than I expected. It grew from 1,479 to 2,196 lines: 907 lines were new and 190 had gone. The games were the same 14; the board was filling in, with 7 more players and more alternative lines (518 of the new lines were demons, 294 goblins, 95 standard).

Of the 1,289 lines that were on the board both times, 84 had a different line_score the second time: 6.5%. The standard lines moved most, 45 of 351 (12.8%), then the demons, 39 of 616 (6.3%). Not one of the 322 goblins moved. The moves were small and mostly upward: 65 up, 19 down, and 63 of the 84 by exactly one unit. Pts+Rebs+Asts lines moved most often (28 times), then Points (16), Pts+Rebs (11) and Rebs+Asts (10).

Two things make that count harder than it looks:

  • A move can arrive as a new line. 52 of the new lines sat in the same group_key and odds_type slot as a line that had disappeared (44 slots), each with a number the gone line did not have. Matching on id alone misses them; count them as moves and the number changes rise from 84 to as many as 136.
  • updated_at is not a move signal. It changed on 1,001 of the 1,289 lines, and only 84 of those had a new number. Every line that moved did have a new updated_at, so it narrows the search, but you still have to compare line_score yourself.

Two hours on one evening is a small window. What it does show is that a single read of the board is a snapshot that can go stale within hours: in this 2.2-hour window, standard lines moved most.

Pitfalls in the PrizePicks data

One player, one stat, many lines. A median over "the lines" mixes the ladder into it. On the NBA board the median Points line was 17.0 across all Points lines, 15.5 for the standard ones only, 19.5 for demons and 11.5 for goblins. Filter on odds_type first, or group by group_key, before you say anything about where a line is set.

Never average line_score across stat types. The unit is whatever the stat is. /board-summary reports a median line of 9.5 for the whole NBA board and 19.5 for the NFL, which mixes points with rebounds and passing yards with tackles. It is a property of the stat mix, not of the players.

A missing standard line is not missing data. 42.5% of NBA groups had no standard line at all, only demons or goblins. Joining demons to "their" standard line will silently drop those groups.

Under is mostly a standard-line option. allowed_wager_types says which way a line can be played. On the NBA board 336 of 374 standard lines were under_or_over, against 71 of 1,105 demons and goblins; 961 of those were over only. 108 lines had no value at all, and the documentation is explicit that a missing value is not a third bucket, so do not fold it into either side.

trending_count is null on most lines, and it starts at about 100. 1,182 of 1,479 NBA lines had none, and the smallest count on the board was 102. So a null looks like "fewer than about a hundred picks", not zero, and pick shares are shares of the picks on the lines that pass that bar.

Counts move between calls. /leagues said NHL 3,505 and NFL 3,397; the board summaries a few minutes later said 3,509 and 3,402. The board is live, so record when you read it and compare like with like.

The biggest boards can be refused on a board-wide read. /board-summary returned 503 for UCL and college football on every attempt, with pauses between them. Both messages said the whole board was not available and pointed to narrower reads: player_id or external_game_id on the projection endpoints, or /projections-by-ids. I left both boards out of the odds-type mix rather than estimate them, so the 15% covers 58.7% of all lines on the board, and the share on the full board could differ.

Sub-boards are separate leagues. NHL1P, NFL1H and NFL1Q have their own league value and their own counts. A total per sport needs parent_name; a total per league double counts nothing but undercounts the sport.

What I would do next

Read the NBA board every 15 minutes from the time lines go up until the first game starts, so each move has a time and a size, and see whether standard lines move before or after their pick counts jump. And repeat the odds-type count on a quiet day and a busy one, because 11 boards on one Tuesday evening is a snapshot, not a pattern.

The data came from the PrizePicks API. If you would like PrizePicks boards read at fixed times and sent to you as CSV or JSON files, contact us.

Top comments (0)