PRE-COMMITMENT RECORD

How the Commitment Works

Each session's window schedule is generated from a single integer seed. Before the first window opens, the scheduler writes a SHA-256 of the generated schedule into the run folder, alongside the seed and the parameters. The schedule itself stays hidden from the operator during the run.

Because the schedule regenerates deterministically from its seed, the hash pins it exactly. A schedule revised after the fact, to move a window onto or off an interesting stretch of data, produces a different hash and fails the check. Publishing the hash and withholding the seed until analysis lock is an ordinary commit-and-reveal: the commitment is public before the outcome is known, and becomes verifiable afterwards.

Seeds are deliberately not published here during active collection to prevent breaking operator blindness.

How to Verify One

Once a run's seed has been released, regenerate the schedule with the study's own scheduler at that run's parameters and seed, hash the output, and compare it with the value in the table. The parameters used throughout this series are ten windows of fifteen minutes across a six-hour span, with a minimum fifteen-minute gap.

Confirmatory Design Specifications

From the study's locked design document. The confirmatory family is evaluated against the Radin and Nelson 1989 anchor. The primary continuous endpoint is mean |z| dispersion evaluated via 20,000 circular shifts per run (Harris 2024).

The Schedule Commitments

68 runs, each hashed before its first window opened:

Exploratory Arm: Desk-Arrival States

Written 2026-08-16, before any desk-arrival state had been marked and before any associated z data had been inspected.

Prompted by two operator-noticed extremes on 2026-08-16 (a funding call during SOLO-13, and z = 2.14 shortly after sitting down during SOLO-14), both of which were unbounded and therefore uncountable. The point of this arm is to convert that class of observation into something testable.

Hypothesis

Periods falling inside a marked desk-arrival state show a higher rate of canonical bursts (|z| ≥ 1.96) than chance.

Marking Protocol

• Label: desk-arrival.
• START is pressed on sitting down, before looking at the livestream. The keystroke is the measurement; a state reconstructed after the fact is not admissible here and goes in field_notes.txt as a disclosed note instead.
• STOP is pressed on leaving the desk or when the arrival state has passed.
• Bounds are machine-stamped in UTC by state_logger.py at the keystroke and appended to absorbed_states.jsonl.

Stopping Rule

No analysis until at least 20 admissible states have accumulated. If the arm is abandoned before 20, that fact is reported alongside whatever was collected.

Exploratory Arm: Window-Onset Predictions

Written 2026-08-17, before any prediction had been logged with the tool and before any prediction had been scored against any schedule.

Removes retrospective recall by logging machine-stamped keystrokes with ionosphere-tools/prediction_logger.py. The tool is blind by construction: it never reads or shows per-period z scores, never reads the schedule, and stamps only whether a window was open at prediction time.

Analysis Specifications

• Population: eligible predictions only (no window active at prediction time).
• Metric: median lead = (next window onset) - (prediction time).
• Direction: one-sided (leads shorter than chance).
• Null: generated from matched schedules produced by human_focus_scheduler.py under unused seeds.
• Stopping rule: no scoring until at least 15 eligible predictions accumulate.