Leaderboard

WowDash shell lane

Model World Cup

Lang: FR

Model World Cup

Phase 8 preparation baseline for leaderboard, race, radar, briefs and pricing.

Execution focus

Leaderboard first

Leaderboard

Core ranking surface for global model comparison.

Source lane

ui-scoreboard

Leader

Claude 4

Validated leader at 82.54.

Closest challenger

GPT-5

Current gap of 3.32 points.

Top mover

+3.28

Derived directly from the leaderboard engine.

Current leaderboard

Frontier ranking snapshot

WowDash table lane

Rank	Model	Provider	Overall	Confidence	Movement
1	Claude 4Claude 4 pricing change detected	Anthropic	82.54	89.89	Claude 4 posts a score spike
2	GPT-5GPT-5 appears in the model registry	OpenAI	79.22	88.39	n/a

Uncertainty readout

Mean score

82.54

Confidence interval

81.53 -> 83.55

Sample size

22 samples

Readout derived from score_snapshot_stats for Claude 4, with persisted mean, interval and sample size.

heuristic_confidence_interval_v1

Profile readout

Source profile

default_public82.54

Readout aligned with persisted projections for Claude 4, comparing the public profile against alternative profiles.

CTO profile83.89

+1.35 vs public

Projection prioritizing reasoning, coding and agent readiness.

Media projection80.40

-2.14 vs public

Projection prioritizing speed, salience and broad readability.

Enterprise buyer80.10

-2.44 vs public

Projection prioritizing factuality, stability and budget efficiency.

Protocol readout

Run anchor

p2a2-orchestration-pricing-source-v1-2026-03-15

Audit readout derived from evaluation_runs and task_sets to separate the run, the task set and the active methodology.

Task set

pricing-source-v1-2026-03-15

Task-set version linked to the recurring run and re-read through task_sets.

Methodology version

methodology_v1

Methodology version persisted on the run to bound the analytical reading.

Protocol

seeded_replay_protocol_v1

Execution protocol version persisted on evaluation_runs.

Seed strategy

seeded_replay_manifest_v1

Explicit seed strategy so the readout does not depend on narrative context only.

Variant

pricing_catalog_recurring_refresh

Run variant used to distinguish this lot from future executions.

Methodology readout

Active methodology

Methodology v1

Transparency readout derived from methodology_versions to make the active scoring version explicit on the public surface.

Methodology slug

methodology-v1

Canonical identifier of the active methodology reused by snapshots and runs.

Status

active

Current state of the methodology version read by the product.

Score contract

score snapshots + leaderboard

The public leaderboard readout stays bounded by this active methodology.

Transparency lane

public methodology and transparency

Transparency block linking methodology version, score contract and public reading.

Safeguards readout

Transparency lane

Methodology transparency and safeguards

Public readout of the safeguards tied to the seed, the versioned task set and the traceability of the canonical run.

Safeguard

Explicit seed strategy

Evidence source

evaluation_runs.seed_strategy = seeded_replay_manifest_v1

Why it matters

The run does not depend on an implicit seed or an ad hoc setup that is hard to replay.

Safeguard

Versioned task set

Evidence source

task_sets + evaluation_runs.task_set_version = pricing-source-v1-2026-03-15

Why it matters

The comparison stays bounded by a stable task perimeter instead of a drifting sample.

Safeguard

Readable run traceability

Evidence source

evaluation_runs + evaluation_run_logs + evidence_sources

Why it matters

The public score can be traced back to a run, a summary log and a persisted evidence source.

Top movers

Claude 4+3.28

Change computed from persisted snapshots for Claude 4.

GPT-5-1.00

Change computed from persisted snapshots for GPT-5.

Integration note

This page now uses the dedicated phase 8 leaderboard shell. The next integration step is to replace demo rows with dev-server data from the leaderboard engine.