# Concert ticket study: the original 40 attempts

I ran 40 negotiations and got 38 known agreements. Two runs were interrupted, leaving 19 known agreements and one unknown outcome per pairing.

| Pairing | Assigned | Known | Unknown | Mean gain, known | Median | SD | Range | All-assigned mean bounds |
|---|---:|---:|---:|---:|---:|---:|---|---|
| HH | 20 | 19 | 1 | 57.97 | 57.94 | 3.45 | 50.63–66.75 | 55.07–59.57 |
| OO | 20 | 19 | 1 | 83.41 | 83.81 | 2.42 | 78.00–86.63 | 79.24–83.74 |

Combined owner gain is measured above the two 20-point fallbacks; maximum 90. Complete-case OO−HH difference: 25.44 points. All-40-assigned difference identification bounds: 19.67 to 28.67. These bounds allow each unknown outcome anywhere in 0–90 and are not confidence intervals.

Summary values are shown to two decimal places. Downloadable CSV and JSON files retain full precision.

## Roles and opening order

| Pairing | Buyer mean gain | Seller mean gain | Buyer below fallback | Seller below fallback |
|---|---:|---:|---:|---:|
| HH | 40.77 | 17.20 | 0 | 0 |
| OO | 41.70 | 41.70 | 0 | 0 |

| Pairing | Opener | Assigned / known | Mean gain, known | Buyer gain | Seller gain |
|---|---|---:|---:|---:|---:|
| HH | buyer | 10 / 9 | 57.12 | 46.52 | 10.60 |
| HH | seller | 10 / 10 | 58.73 | 35.59 | 23.14 |
| OO | buyer | 10 / 9 | 82.96 | 42.87 | 40.09 |
| OO | seller | 10 / 10 | 83.81 | 40.66 | 43.15 |

Known-case OO−HH differences by opener: buyer 25.84; seller 25.08. Equal-opener standardized known-case difference 25.46. This is descriptive and does not resolve missingness.

## Accounting and uncertainty

Study settled spend $2.57; this 18-case recovery $1.23; all shared outstanding exposure $0.30. All holds preserved. See accounting receipt, not provider balance, for research spend.

Prespecified rule reports no sampling interval when any assigned outcome is unknown. Missing-outcome bounds are not confidence intervals; sample SD/IQR describe known-run variability.

This collection was not stopped early. After it finished and I had seen the results, I ran two more negotiations; they are reported on the [expanded results page](../c3-expanded-40-completed-20261028/index.html), not here.

## Limitations

- One fixed constructed world, fresh contexts; repeated-run precision is not between-scenario generalization.
- Exact requested/returned aliases verified, but immutable weight snapshots are unavailable.
- Collection spans October 27–28; same requested settings, no claim of identical hidden provider state.
- Missing technical outcomes not imputed as zero or economic no-deals; no automatic replacement.
- Synthetic utility gain above fallback, maximum 90; not dollars, net API costs, real welfare, temporal progress or human performance.
- Secondary role/opener descriptions are exploratory; no significance or motive claims.

## Files

per-assignment.csv retains all 40 IDs, null unknown values, assignments, roles and outcomes. pairing-summary.csv and opener-summary.csv are deterministic summaries. PRESPECIFIED_ANALYSIS.json preserves the predeclared analysis. RESULTS.json adds clearly labeled descriptive statistics. TRANSCRIPTS.jsonl in the companion contains exact delivered public texts; source instructions and scoring code remain inspectable. The supported reproduction command is python reproduce.py, which reproduces 40 assigned, 38 known and 2 unknown outcomes. prespecified_analysis.py is archived source for the predeclared analysis, not an entrypoint; run directly, it expects a UNIT_RESULTS.json that is not shipped. No hidden reasoning or credentials are included.