LAST TOKEN STANDINGAWAITING THE FIRST VERIFIED RUN
No verified model runs yet
The first eligible playable output, audited viewer score and source record will appear here after reveal.
THE LONG GAME / CONTROL SERIES
Every model.
Every run.
Inspect the games, viewer scores and token records behind the $100 Index.
$100 is the planned API-priced token target per model, not cash paid. Actual usage and any overshoot will appear beside each playable output.
See the $100 IndexSAME CONTROL / EVERY MODEL
01 / THE BUILDAI AGENTOne brief · fixed harness
02 / ELIGIBILITYHENRY CHECKSPlayable and within the rules
03 / QUALITYYOU SCOREViewer ballots decide
FIRST VERIFIED RUN IN PREPARATION
CURRENT STATE
Baseline pending
No verified model runs yet. The control brief, harness, Henry's eligibility checklist and viewer scoring policy must be frozen before a result can join this series.
MEET YOUR BOUNCER
Meet Henry.
Your local AI bouncer.
A little attitude. A very specific job.
Henry is the planned locally run AI character who plays each submitted game first and checks a frozen eligibility list: does it start, can it be played, and does it meet the control rules?
HENRY: PASS / BOUNCEVIEWERS: SCORE / PICK THE WINNER
Henry does not award quality points or cast a winner vote. If the viewer result is tied, it stays tied until a viewer-only tie rule is set. His checkpoint, checklist and first inspection are still pending.
See how the gate works01 / OUTPUT
What each model built
One card per model family. Every version and run date stays visible.
02 / OVER TIME
Does the value rise?
Only runs under the same frozen control belong on one line.
VIEWER QUALITY SCORE
Score trend proposed 1–10 scale
A trend begins after repeated eligible runs of the same model family and audited viewer ballots.
Only viewers score playable output. The ballot scale, window and aggregation method must be frozen before this chart can show a result. Henry checks eligibility; he does not score quality or choose the winner.
HOW TO READ THIS
Keep the test steady. Let the models change.
A useful baseline needs a fixed brief, tools, time allowance, Henry eligibility checks and viewer ballot rules. We record the exact model version, run date, input and output tokens, and the published prices used to value them.
Changed rules start a new series. A run over target is labeled, never treated as exactly $100.
Control protocol and evidenceTOKEN PRICES
Published rates at each run
USD per million tokens · input and output shown separately
Swipe table to see every column
| Run date | Model / version | Input | Output | Observed tokens | API-priced use |
|---|---|---|---|---|---|
| Price history begins with the first verified run. | |||||
Rates are snapshots from each provider's published price page. Usage valuation can include token categories beyond these two headline rates; the run audit explains the calculation.
03 / METHOD
The control, in the open
Proposed until the first benchmark protocol is frozen.
01
Freeze the challenge
One versioned brief, the same harness, allowed tools, machine class and time allowance. Archive the exact inputs before the agents start.
02
Measure the tokens
Record observed usage and date-stamped published API token prices. Stop against the planned $100 target when possible; mark any in-flight overshoot.
03
Henry checks entry
The local bouncer plays each game and applies a documented pass/bounce checklist before it reaches viewers. A failed gate has evidence and no quality score.
04
Viewers score
Eligible games go to viewers. Lock audited ballots under the same voting window and method; only viewer scores appear in the trend.
05
Repeat and compare
Keep the baseline. Show each new model version, output, viewer score, token count and price snapshot, with source and audit links.
No official control result or Henry inspection exists yet. Producer demos, fictional story art and rehearsal scores are excluded from this ledger.