Repository navigation
plat 567
| Field | Value |
|---|---|
| State | fixed on main |
| Priority | P2 |
| Product | goals |
| Area | pulse/general |
| Summary | get_pulse_state module view returned 320k to 384k characters; the agent view now caps the ledger and focus history |
A Plan Drift session on Upwork (2026-10-06) read get_pulse_state(view=module) as its first call and received
384,349 characters, about 100k tokens. Its context compacted within minutes (236k to 24k tokens). Earlier calls in the
logs were 185k to 316k. On a copy of Upwork the view measured 320,760 characters: impact_ledger 202,541 (63%),
review_focus_selections 42,880, the reference map 9,191 (3%).
The module view is read by the agent only. It loaded the 100 most recent interventions (plus the last 120 points per
active goal metric, kept on purpose for before/after windows) and 50 focus selections. It now loads the 20 most recent
interventions and 10 focus selections (pulseModuleViewLedgerLimit, pulseModuleViewFocusSelectionLimit); older
rows stay in the database and the UI route is unchanged. The ledger note says only recent rows are shown.
Measured on the Upwork copy: 320,760 to 198,759 characters (-38%).
- The view is still about 200k characters. What remains is history the roles legitimately read: assessments (33k), interventions (30k), observations (44k), the focus history (11k) and the modules block (11k). Cutting more means a summary with pointers to detail views, which changes what Goal Work and Strategic see; design first.
- Confirm live: the size of the next Drift or Technical pass's first
get_pulse_statecall inserver_debug.log(result_length=).
Auto-synced from docs/ on main. Edit there, not here.