assembly-economics/checkpoints
2026-10-13 06:01:06 +00:00
..
2026 Checkpoint 2026-10-13 (light) — activity-core 2026-10-13 06:01:06 +00:00
data Checkpoint 2026-10-13 (light) — activity-core 2026-10-13 06:01:06 +00:00
IndicatorCatalog.md ASSEMBLY-WP-0010-T02 (part): checkpoint protocol, playbook, schedule, catalog v0.1, register; activity definitions 2026-10-12 18:21:55 +02:00
PLAYBOOK.md Checkpoint executor: deterministic runner, sources, tests; definitions route to assembly-checkpoint 2026-10-12 18:45:12 +02:00
predictions.md ASSEMBLY-WP-0010-T02 (part): checkpoint protocol, playbook, schedule, catalog v0.1, register; activity definitions 2026-10-12 18:21:55 +02:00
README.md ASSEMBLY-WP-0010-T02 (part): checkpoint protocol, playbook, schedule, catalog v0.1, register; activity definitions 2026-10-12 18:21:55 +02:00
run_checkpoint.py Checkpoint runner: comparability caveat in assessment prompt 2026-10-12 18:47:18 +02:00
schedule.md Enable checkpoint definitions; one-day baseline window for checkpoint 0 (2026-10-12) 2026-10-12 18:48:12 +02:00
sources.json Checkpoint executor: deterministic runner, sources, tests; definitions route to assembly-checkpoint 2026-10-12 18:45:12 +02:00
test_run_checkpoint.py Checkpoint executor: deterministic runner, sources, tests; definitions route to assembly-checkpoint 2026-10-12 18:45:12 +02:00

Reality checkpoints

Regular comparison of the GADM adoption band and model readings with observed economic, sociological and technical data. Method: research/ScenarioBandMethod.md § 6. Workplan: ASSEMBLY-WP-0010.

Files

File Purpose
schedule.md phase windows; every run checks it first
PLAYBOOK.md step-by-step instructions for the executing agent
IndicatorCatalog.md indicators, jurisdictions, sources, mapping to model quantities
predictions.md append-only register of pre-registered predictions
YYYY/YYYY-MM-DD-<kind>.md one record per checkpoint run

Protocol

  1. Window check. Is today inside an active window in schedule.md for this kind? If not, record an out-of-window no-op and stop.
  2. Collect. Gather the indicators for this kind from IndicatorCatalog.md: light = adoption, investment and sentiment indicators; full = all catalog sections. Use the most recent published value, and record value, unit, reference period, publication date and source URL. Never invent a value: if it isn't found, record not found and the sources tried.
  3. Compare. For every indicator with an entry in predictions.md, locate the observation in the band (which anchor B1–B4 it tracks), and mark it inside or outside the band. Indicators without a registered prediction are recorded as observations only.
  4. Assess. Short notes per jurisdiction: which archetype and band anchor reality currently tracks, notable legislation or sentiment changes, and surprises.
  5. Record. Write checkpoints/YYYY/YYYY-MM-DD-<kind>.md from the template in PLAYBOOK.md. Append the observations to the matching entries in predictions.md. Never edit earlier predictions or observations.
  6. No silent re-tuning. Model changes suggested by a checkpoint become an intake (intakes/intakes.md) or a workplan task. Parameter changes go through models/ParameterGrounding.md.
  7. Complete. Commit and push, then post a State Hub progress event with event_type: assembly_checkpoint and detail: repo, date, kind, path, in_band, out_of_band, not_found.