Execution Log: reduce the-loop's token consumption

The issue was exploratory, so the loop started at brainstorming. The owner reviewed the brainstorm on PR #41 and directed "implement all suggestions in one PR", collapsing the spec phases into a single implementation increment against the locked brainstorm.

Phase transitions

Phase	Entered	Reviewed/approved by	Notes
brainstorming	2026-07-23	@MadaraUchiha-314 (PR #41)	Root artifact brainstorm.md; locked with owner direction to implement all levers in one PR.
implementation	2026-07-23	—	tokenEconomy config + schema, token-economy.md reference, CLI usage telemetry, external-tools registration, capability doc.
needs-review	2026-07-23	—	Awaiting human review on PR #41.

Progress entries

2026-07-23 — brainstorm locked; all levers implemented in one PR

• Phase: implementation → needs-review
• Did:
  • Corrected the brainstorm's process-vs-tmux re-spawn claim (owner review, line 33) and recorded the owner's resolutions of all six open questions; locked it (approved).
  • Added tokenEconomy to config.schema.json (+ $defs.modelTier), the template config, and the project config: model routing (stage→tier + risk-tier floor), thinking-effort, output-verbosity (with preserve list), progressive disclosure, sub-agent delegation, compaction, telemetry.
  • New reference/token-economy.md (levers, phase→reference loading map, best-practices checklist, absolute guardrail); wired into SKILL.md; added a token-lever note to minimalism.md.
  • Registered caveman + ponytail in .the-loop/external-tools.md (register, not vendor).
  • Implemented best-effort token/cost telemetry: Usage + _usage_from_output in harness/base.py, DispatchResult.usage, per-dispatch logging in the dispatcher.
  • Added the token-economy capability doc + index row.
• Checkpoint/tests:
  • python scripts/validate_config.py → both configs VALID; schema meta-valid.
  • pytest (cli) → 111 passed (7 new usage-parser tests).
  • ruff check cli → All checks passed; pyright on changed files → 0 errors.
  • markdownlint-cli2 on changed docs → 0 issues.
• Next: human review on PR #41.
• Blockers: none. Guardrail honoured — every lever is advisory and gates nothing; the rigor floor (validation/security/tests/paper-trail/review depth) is untouched.

Review cycles

Cycle	Type (self/critic)	Reviewer	Outcome	Link
1	owner (brainstorm)	@MadaraUchiha-314	"implement all suggestions in one PR" + 2 line comments (model routing, tmux)	PR #41

Final validation evidence

• Config schema + both configs validate; schema is a valid draft 2020-12 schema.
• CLI suite green (111 passed) including new telemetry tests; ruff + pyright clean.
• Markdown lint clean across all changed docs.
• All twelve best-practices rows in reference/token-economy.md map to a shipped lever.