Schedule token measurement pack
12 October 2026

Purpose
Reproduce the text-size tables in Automation Calculates, AI Interprets, People Decide.
All source data is synthetic. No client schedule or confidential project data is included.

Run
Install Python 3.12 and tiktoken==0.14.0. Run python measure_schedule_tokens.py.
The o200k_base tokenizer vocabulary may download on first use.

What is tested
UTF-8 text and LF line endings, o200k_base token counts, serialized record counts,
unique task identifiers, relationship endpoint existence, and output fixture sizes.
manifest.json records exact byte counts, tokens, hashes and data populations.

What is not tested
P6 import/export compatibility, CPM correctness, real DCMA findings, chatbot accuracy,
or specific Copilot capacity.
The .xer.txt files use selected real XER table and field conventions for a size experiment.
They are not import-ready schedules. Dates, floats, risks and findings are illustrative.

Profiles
1,000 / 2,000 / 5,000 activities with 6 code categories and 20 percent resource assignments.
An additional 5,000-activity profile uses 8 categories and 60 percent resource assignments.
There are 1.4 relationship rows per activity, 60 TASK columns and 1 WBS row per 10 activities.
The article uses the heavier 5,000-activity profile and reports the lighter sensitivity result.

Outputs
The 13-column flat file is an illustrative schema. The 6-column risk table is a size fixture, not a complete
risk model. Exception counts are generated at a fixed artificial density. The briefing
has 14 check placeholders and 50 selected exception references, not a real assessment.
