How to read what your twin remembers

Open the panel

On the role home, What your twin remembers sits below the fold under the twin panel. Choosing it opens the panel docked beside the regions (the address gains ?remembers=1; Escape closes it and returns focus to the control). The panel opens from the role home only in this build; opening it from inside a thread is designed. Three regions, in fixed order:

1. In this program: what your twin kept while working with you here (references to records you decided, preferences, vocabulary, rubric adjustments, lessons). Each record shows its kind and date; episodic ones read “Your twin remembers this because you decided it.”
2. About you, everywhere: your preferences, vocabulary and rubric adjustments that travel with you across the organisation’s programs. Nothing here carries a record or passage reference or a program fact.
3. Shared by <role plural> (for example “Shared by medicinal chemists”): methods, vocabulary and curated lessons shared by everyone wearing your workbench role (R5; empty and labelled designed in R4).

The panel reads GET /me/memory?program_id= for the first two regions.

Answer “Remember this?”

When your twin thinks something is worth keeping it proposes a note for this program only (tier 1; the tool memory.propose_note admits no other tier in R4). The note appears under the proposal it came from in the message feed as “Remember this?” with Keep and Not now inline; you read the exact summary before anything is stored. Keep resolves the interrupt under program.agent_memory.manage, stores the record in tier 1 and audits memory.keep; Not now stores nothing. Nothing is kept automatically in R4.

Keep a preference across programs

On a record under In this program, choose Keep across programs. This is your own action and the only way a record travels from a program to “About you, everywhere” in R4; your twin never proposes it (the tool memory.propose_promotion is designed). A one-line inline form expands in the record; confirming calls POST /programs/{program_id}/agent-memory/{record_id}/promote. When the control is disabled it stays focusable and the line under it states why: “Your persona cannot keep notes across programs in this release.” or “Only preferences, vocabulary and rubric adjustments can be kept across programs.” (verified in web/components/workbench/memory-panel.tsx:44-48, 101-111 and web/lib/content.ts:295-296). Only preferences, vocabulary and rubric adjustments travel: an episodic record or a lesson is refused with 422 kind_not_promotable, shown beside the control. The summary is scrubbed of program names, compound and batch identifiers, target names, milestone names, partner names and member names before it moves; the answer MemoryPromoteOut {record, scrubbed_terms_count} tells you how many terms were removed, and the record appears under About you, everywhere. Audited memory.promote.

Suggest a method for your role (R5)

Suggest for the role will expand a one-line form the same way and file a copy stripped of your user id for curation; in this build the control is present, stays focusable, and the line under it reads “Arrives with shared memory in a later release.” (verified in web/components/workbench/memory-panel.tsx:106-112). The tool behind it, memory.propose_promotion, is designed in R4. Two people must agree before it is shared: a curator of role memory accepts and a current holder of your workbench role confirms; both names and the version are shown on the shared record. This control arrives with R5 and is designed in R4.

Forget

Forget is a two-step control: the button turns into “Confirm forget” for a few seconds; Escape, moving away or a click elsewhere reverts it. Confirming calls DELETE /me/memory/{record_id}, audits memory.forget, and the record’s place shows the fixed line “Forgotten. Nothing from this record is used again.” A tombstone prevents the same note from being derived again.