Cache frontier review ledger

This folder is the recurring review artifact for the cache frontier operating plan. Each entry answers the three weekly questions from evidence:

What cache-frontier value did we use ourselves this week?
What can a new person demo this week?
What is the next missing witness or product surface?

Entries should cite the command output they were derived from and keep Track 1 WITNESSED kernel reuse separate from Track 2 OBSERVED provider-dollar savings. The ranked default-enablement backlog lives in DEFAULT-ENABLEMENT-NEXT-50.md; it keeps pure fak/API work, O(1) context/query, provider-cache preservation, and external-engine adapters on separate evidence planes. Each review has two artifacts:

a human-readable markdown note under reviews/YYYY-MM-DD.md;
one machine-readable row appended to review-ledger.jsonl with the same evidence fields.

+-----------------------+     +----------------------------------+
| fak cachevalue review | --> | reviews/YYYY-MM-DD.md (markdown) |
+-----------------------+     +----------------------------------+
            |                 +----------------------------------+
            +---------------> | review-ledger.jsonl (one row)    |
                              +----------------------------------+

Generate new dated artifacts from the ledgers rather than hand-writing them:

go run ./cmd/fak cachevalue review \
  --since 2026-06-22 \
  --date 2026-06-29 \
  --source-markdown reviews/2026-06-29.md \
  --append-ledger docs/cache-frontier/review-ledger.jsonl \
  --markdown-out docs/cache-frontier/reviews/2026-06-29.md

Use --json without --append-ledger to inspect the row before mutating the append-only ledger.

Entries

2026-06-29 cache-frontier review

Findings

Standalone research findings (not the generated weekly review) live alongside the entries when a specific question is investigated from evidence:

2026-07-04 — cache-read vs rate-limit burn on OAuth seats (#2183): the managed-cache passive-on-OAuth rule was assumed on the dollar axis and never measured on the rate-limit axis; the correlation is currently unmeasurable because the limit-burn and cache-read sides of the join are not co-persisted.