CAMP-BENCH / SEPTEMBER 2026

Trident single-site pilot

3 of 3 planned stages have recorded outcomes across 1 provider. 3 of 3 recorded stages have completed source review; 0 planned stages remain unrecorded.

Scope1 providers / 5 published sessions
Progress3 / 3 stages resolved
New API spending$0.00 / $25 cap
Total economic costUnavailable

Initial extraction / single-site pilot

1 provider · one first attempt per stack
Trident benchmark: recovered facts and elapsed timeVertical axis: established core facts recovered, zero to one hundred percent. Horizontal axis: elapsed minutes, including setup and shutdown. One initial extraction per stack; variability is not measured. Select a system for its numeric details.Camp-Bench / Trident02550751000.02.55.07.510.0Core facts recovered (%)Elapsed time (minutes)CCrawl4AI + Codex100.0% / 1.2 minFFirecrawl Agent100.0% / 4.1 minTTrident100.0% / 8.5 min
Higher recovery and less time are better. One initial extraction per stack on one website; variability is not measured. Identical coordinates share a circular anchor, with square stack marks beside their labels. Total dollar costs are unavailable.
Read the plotSelect a marker or stack name for its exact denominator. All planned attempts count; a failed extraction earns no delivered-fact credit.
100.0% · 40/40
1.2 min
1/1
5/5
0 / 0
100.0% · 40/40
4.1 min
1/1
5/5
0 / 0
100.0% · 40/40
8.5 min
1/1
5/5
0 / 0

Facts recovered is the fraction of independently established core fields delivered correctly. It is not an overall correctness score. Wrong facts, duplicates and complete sessions are shown separately.

Elapsed time includes available setup and shutdown measurements. Failed attempts count. Unfinished groups do not receive a plotted marker.

Three-way tie on core facts

Each stack recovered all 40 established camp facts in its first extraction on this provider. One attempt per stack on one website does not establish broader performance or a general ranking.

Reviewed core fact recovery

120 of 120 established fact opportunities are recovered across the planned workload. The separate source review preserves every frozen core score and all pending or failed-stage denominators.

First-extraction timing

Each selected point records the first onboarding extraction from one stack on the same provider. Refresh and repeated-attempt work is excluded from these comparison points and retained in ancillary accounting.

Cost remains unavailable

Total dollar cost is unavailable because subscription allocation and local compute are unmeasured. Known included-credit use and verified API cash are shown separately.

Source review coverage

3 of 3 recorded stages reviewed; source-supported at observed times. Additional records have separate scope dispositions, while raw extra counts and task-pass results remain unchanged.

Core facts are a subset

Recovering all established camp facts does not certify every business or activity field, current availability, confirmed venues, or complete fees. Trident's native business result is marked partial.

Recorded model calls

Trident recorded 5 model calls; Crawl4AI with Codex recorded 1. Firecrawl internal model-call count is unavailable. These configurations perform different amounts of internal work.

Excluded refresh diagnostic

An additional Trident refresh recorded zero model calls and 271.8 seconds of process wall time. It is preserved in ancillary accounting and excluded from the first-extraction comparison.

Firecrawl billing and resource counters

1 recorded stage with both counters returned 0 billed credits and recorded 106 trace resource credits. These are distinct quantities; resource counters do not establish cash cost.