| # Public data dictionary |
|
|
| ## `tb_level` |
| |
| One row represents one validated formal new transport block (TB). The table |
| contains run/pair/controller/UE identity, monotonic transmit and feedback |
| timestamps, MCS/TBS/PRB/layer observations, first-round ACK/NACK/DTX outcome, |
| HARQ result, link-setting command and offline calibrated quantities. |
| `source_file` identifies the sanitized CSV shard from which the row was |
| assembled. |
|
|
| The `p3` split contains the 20→16 step experiments; the `p4` split contains |
| the 22→16→22→16 path experiments. `run_id` and `new_tb_id` identify a TB in a |
| run; `pair_id` links the Frontier and OLLA halves of a matched experiment. |
|
|
| ## Summary tables |
|
|
| - `run_summary`: one row per formal half-run, including first-round BLER, |
| goodput, mean MCS, feedback/decision delay, recovery and validation counts. |
| - `change_summary`: one row per profile change event, including direction, |
| post-change BLER, recovery blocks and right-censoring. |
| - `pair_metrics`: one row per matched pair. Every `diff_*` column is |
| Frontier minus OLLA. |
| - `metric_summary`: paired mean, 95% confidence interval and paired effect |
| size by campaign and metric. |
|
|
| ## Calibration tables |
|
|
| - `p1_calibration_points`: UE03 point/chunk BLER observations used for fitting. |
| - `p1_calibration_curves`: monotone logistic crossing estimates and confidence |
| intervals for MCS 14–26. |
| - `p1_mcs_table`: compact MCS-to-command-crossing lookup. |
| - `p1_fit_checks`: coverage and fit acceptance checks for each MCS. |
| - `p2_evidence`: selected UE/MCS/setting evidence rows. |
| - `p2_crossings`: raw/isotonic sequences and crossing type; `left_censored` or |
| `right_censored` means the crossing was outside the observed setting range. |
| - `p2_checks`: overall three-UE freeze and boundary-coverage checks. |
|
|
| ## Outcome conventions |
|
|
| `first_round_ack=1` means the first transmission was acknowledged. A |
| first-round failure is NACK or DTX. `link_setting_db` is the relative OAI |
| software TX attenuation command; larger values represent a worse effective |
| transmit setting in this experiment. Columns ending in `_derived` are offline |
| calculations, not direct hardware readbacks. |
|
|
