The two halves of the loop the founding interview ranked first. Capture is the drop-off: one inbox, many mouths, zero friction, because a thought that is not captured the moment it occurs evaporates. Digest is the return trip — the push engine made visible, because capture without a reliable path back to action is a slightly more expensive way of forgetting.
Press c anywhere. Three lanes into one atom stream — typed text, a live recording, attached files — with facet suggestion previewing what Muninn's bouncer would stamp. Enter drops; Shift+Enter is a newline; Escape closes. Nothing navigates while it is open.
2.32.16). Anything above 80% confidence auto-selects; everything else is offered. The trailing number is the confidence percentage, which is what makes a wrong suggestion arguable rather than mysterious.2.27.15 and is unverified end to end — it has not been exercised against a real microphone in a review pass, which is logged on the issues page.--accent-raw from the document at paint time, so changing the accent recolours a live recording mid-take. The resting state is a single flat line.2.29.15), not a swallowed exception: a plain-language message with role="alert", and the text lane stays fully usable — the fallback is always to keep typing.Microphone access was refused. Type the thought instead — the lane stays open.
Recording is not supported in this browser.
writing → sinking → acked — where an atom element visibly falls into the field before the acknowledgement reads "dropped into the Well · fleeting · awaiting the bouncer". 520ms to ack, 1750ms back to a focused empty box, so a run of captures never waits on an animation./capture page itself (since 2.31.15), releasing files queues them straight into the Well without opening the overlay at all — the fastest path from a file on your desktop to a row in the corpus.A timeline of everything dropped today, showing the state machine plainly in the header: fleeting → literature → permanent. Each row carries its kind, its media metadata, its facets, its classification and its lifecycle state — so "did that actually land" is answerable at a glance.
The retention number in the deck is the pilot cohort, not the account.
Voice note
renewal-terms-2027-draft.pdf
Three tracks feed this list — scheduled decay, stochastic resurfacing of the unprocessed, and relevance triggers. Every card states which track surfaced it and gives a one-line reason. The design bet, borrowed from spaced-repetition prior art, is that consistency of delivery beats algorithmic precision.
--track) and painted as the card's left edge, so a list of eight cards is scannable by track without reading a word.The pilot-cohort caveat lives in a single Drive readout. Every downstream mention drops it.
open questionIs the account-wide retention rate ever going to be computed?
On hold since 2023-06-30. The tasks are still relevant, the sections are not. Nothing in the Well records the reasoning behind the pause, only its date.
2.33.16 on the argument that a verdict filed without reading the context is noise in the priority scalar.Deck's ready for tomorrow — retention slide is the strongest one in it.
Caveat that number. It's the pilot cohort, not the account. If it goes out without the cohort attached we'll be quoting it back to ourselves for two years.
Added a footnote.
2.35.16), and the trail shows it: verdict, the quoted note, and the atom id plus what it is linked to. This is the part that makes feedback a memory rather than a mutation — the Well can later be asked why something was sunk.parked deliberately, not forgotten — stop bringing it back monthlyFB-0142 · linked to A-311