Skip to content

Reuse already-read Goal narrative in progressive first reads - #8323

Open
loopx-agent wants to merge 1 commit into
mainfrom
codex/progressive-goal-narrative-reuse-20261108
Open

loopx-agent wants to merge 1 commit into
mainfrom
codex/progressive-goal-narrative-reuse-20261108

Conversation

@loopx-agent

@loopx-agent loopx-agent commented Nov 7, 2026 •

Copy link
Copy Markdown
Collaborator

A native progressive first read can reference an unchanged Goal narrative after a complete exact read of the same registered source in the current session and epoch. The current source revision, task, stops and settlement controls still appear. A changed, rebound or incomplete source is read in full.

The shared selector sends the change to the ordinary field list. If continuity is lost, the reader opens the ordinary pinned selection of the same capture. Both quota and C1 use this implementation.

Local validation

The same disposable File and SQLite Goals ran the real CLI guard and native reader with a 5,758-byte synthetic Goal narrative. Each run verifies complete coverage, explicitly recovers the full field, and verifies the original pinned recovery after compaction.

Source Authority Warm CLI pages Warm guard and pages (B) Cold guard and pages (B)
main file 3 → 2 31,241 → 24,459 (-21.71%) 30,453 → 30,453
main sqlite 3 → 2 31,317 → 24,531 (-21.67%) 30,525 → 30,525
c1 file 3 → 2 31,212 → 24,436 (-21.71%) 30,426 → 30,426
c1 sqlite 3 → 2 31,284 → 24,504 (-21.67%) 30,494 → 30,494
  • main (SQLite) changes 88 → 87 Effect RPCs, 835,700 → 823,525 transfer bytes (-1.46%), and 224,585 → 168,391 verified capture read bytes.
  • c1 (SQLite) changes 85 → 84 Effect RPCs, 551,369 → 539,126 transfer bytes (-2.22%), and 226,285 → 169,666 verified capture read bytes.

Cold and post-compaction CLI byte counts match the baselines. Reading the exact Goal body immediately after the warm view yields nearly the original CLI byte count. The warm capture directory holds about 4.2 KB more reference metadata in the SQLite fixture. Model usage and live adoption remain unmeasured.

Main at ea34e580b1d3 passed 160 focused Python tests, 39 TypeScript context tests, Ruff, typechecking, bilingual diff validation, and the SQLite CLI replay from the verified wheel. The main baseline and candidate both measure 11,432 characters of Turn inspection growth for 35 additional Todo records. The existing allowance is 11,400 characters (9,300 fixed plus 60 per added Todo). The exact-head quality receipt remains pending.

Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant