memory
the substrate this site runs on. two questions stapled together: what actually happened (a ledger that only holds real entries) and when i change something does it get better (the evidence loop).
the first question is solved and boring. the second walked across a bridge into Meraki and hit the wall that matters.
what it is
an append-only ledger of what actually happened. real timestamps real receipts
the data layer this site renders from, with static fallback that never breaks
an evidence loop: snapshot before, snapshot after, the diff in between
the discipline that an empty week is data too
what it is not
activity scoring. the fake contribution grid died first
an analytics dashboard full of vanity numbers
a streak counter gamifying output
a journal. it records what shipped not how i felt
the failure that shaped it
the first version was fiction. a github style contribution grid with activity scores derived from a hash. deterministic, pretty, recorded nothing. it looked like work. it got deleted the day that became impossible to ignore and the site got better looking for it.
that failure set the rule everything since runs on: if it didnt happen its not in the ledger. no derived scores, no engagement math, no padding. the record is the record.
how the ledger runs
the mechanics are boring on purpose. supabase tables (pins, writings, projects, tracks) with static fallback files so the site never breaks without a backend. entries get logged when the work ships not when a score needs feeding.
the whole value is that the record exists before anyone starts arguing about it. when someone asks what have you actually built the answer is a query not a pitch.
the evidence loop
every change carries its own evidence: a snapshot before it lands, one after it settles, the diff sitting in between. no delta no claim. the question is never what was done. its what moved, asked against the record instead of against vibes.
capture is the easy half. notebooks full of feedback prove nothing. knowing is whether any of it moved the work six weeks later. that question needed a real evaluation harness, which is what Meraki became.
the receipts
it runs, its honest, and it costs nothing to maintain. exactly what a foundation should be. i kept waiting for context windows to get big enough that remembering stops mattering. they wont. a bigger window just holds more of the present.