Skip to content

Commit 7afbca2

Browse files
authored
Merge pull request #222 from ModelMirrorAI/predict/scotus-1001152-evt-petition-disposition-claude-baseline-20260628T172601Z
predict(claude-baseline): scotus/1001152 — evt-petition-disposition
2 parents 3d80c8c + 748a69c commit 7afbca2

3 files changed

Lines changed: 95 additions & 0 deletions

File tree

Lines changed: 16 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,16 @@
1+
{
2+
"case_id": "scotus/1001152",
3+
"event_id": "evt-petition-disposition",
4+
"predictor_id": "claude-baseline",
5+
"engine": "claude-code",
6+
"model": "claude-opus-4-8",
7+
"run_id": "20260628T172601Z",
8+
"created_at": "2026-06-28T17:27:25Z",
9+
"input_snapshot": "data/cases/scotus/1001152/record/snapshots/2026-06-28.json",
10+
"granted": 0,
11+
"probability": 0.02,
12+
"predicted_disposition": "denied",
13+
"confidence": 0.85,
14+
"votes": [],
15+
"reasoning_doc": "reasoning.md"
16+
}
Lines changed: 63 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,63 @@
1+
# Prediction reasoning — scotus/1001152 / evt-petition-disposition
2+
3+
## The legal question
4+
5+
The event `evt-petition-disposition` (kind: `petition`, decision target:
6+
`disposition`) asks how the Supreme Court of the United States will dispose of a
7+
petition for a writ of certiorari in *Roberts v. United States*. "Granted" here
8+
means the Court grants the petition (grants certiorari, i.e. agrees to review the
9+
case); "denied" means it declines review.
10+
11+
## Governing standard
12+
13+
Certiorari is wholly discretionary. Under Supreme Court Rule 10, review "is not a
14+
matter of right, but of judicial discretion," and is granted "only for compelling
15+
reasons" — typically a genuine, entrenched conflict among the courts of appeals
16+
or state courts of last resort, or an important unsettled question of federal
17+
law. The overwhelming default disposition of a cert petition is **denial**:
18+
across recent Terms the Court receives on the order of 7,000–8,000 petitions and
19+
grants roughly 1–2% of them. The grant rate is far lower still for petitions
20+
filed *in forma pauperis* (the unpaid docket), which are granted only a fraction
21+
of one percent of the time.
22+
23+
## Facts from the snapshot that drive the outcome
24+
25+
Predicting only from the latest snapshot
26+
(`record/snapshots/2026-06-28.json`):
27+
28+
- **Court / posture.** `court_id` is `scotus`; the originating court is the U.S.
29+
Court of Appeals for the Fourth Circuit (`original_court_info`). So this is a
30+
cert petition seeking review of a Fourth Circuit judgment.
31+
- **Docket number `01-7733`.** This is an October Term 2001 number, and the high
32+
four-digit serial (`7733`) is characteristic of the Court's *in forma pauperis*
33+
/ pro se docket rather than the paid docket (paid cases carry much lower
34+
serials). IFP petitions have the lowest grant rate of any category.
35+
- **No signals of a grant.** The snapshot shows `date_cert_granted: null`,
36+
`date_argued: null`, and `date_reargued: null`. There is no argument date, no
37+
panel, and no assigned Justice — none of the indicia that accompany a granted,
38+
argued case.
39+
- **No merits-stage activity.** `docket_entries` is empty, `date_terminated` is
40+
null, and there is no nature-of-suit or cause recorded. Nothing in the
41+
point-in-time record suggests the case advanced to plenary review.
42+
- A `clusters` reference is present, but I do not fetch its contents: that would
43+
be acquiring a new case fact beyond the snapshot, which the predictor contract
44+
forbids. (For SCOTUS, an opinion cluster can attach to an ordinary order-list
45+
denial as readily as to a merits decision, so its mere presence is not
46+
probative of a grant.)
47+
48+
## Reasoning behind the probability
49+
50+
The base rate for a SCOTUS cert petition is denial, and every case-specific
51+
signal available in the snapshot points the same way: an IFP/pro-se-style docket
52+
number, no argument or grant date, no panel, and no merits activity. None of the
53+
"compelling reasons" indicia under Rule 10 are visible in the record I am
54+
permitted to use. I therefore predict **denied**.
55+
56+
I set `probability` (P(granted)) at **0.02** — slightly above the raw IFP grant
57+
rate to leave room for the residual uncertainty inherent in reasoning from a thin
58+
snapshot, but firmly on the denial side. `granted = 0`,
59+
`predicted_disposition = "denied"`, with `confidence = 0.85`.
60+
61+
No per-judge votes are predicted: cert denials are typically unsigned order-list
62+
dispositions with no recorded vote breakdown, and the snapshot names no Justices,
63+
so `votes` is empty.
Lines changed: 16 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,16 @@
1+
{
2+
"actor_id": "claude-baseline",
3+
"cache_creation_input_tokens": 19138,
4+
"cache_read_input_tokens": 202090,
5+
"case_id": "scotus/1001152",
6+
"created_at": "2026-06-28T17:26:01Z",
7+
"engine": "claude-code",
8+
"estimated_cost_usd": 0.3732425,
9+
"event_id": "evt-petition-disposition",
10+
"input_tokens": 2717,
11+
"model": "claude-opus-4-8",
12+
"output_tokens": 5560,
13+
"role": "predictor",
14+
"run_id": "20260628T172601Z",
15+
"schema_version": "1.0"
16+
}

0 commit comments

Comments
 (0)