Filing a ticket no longer starts it working automaticallyFiling a ticket no longer starts it working automatically
September 1, 2026 · A ticket filed as a plan finally stays put, Telegram's buttons work and are locked to my chat, and a won match stopped reporting itself as a draw.My assistant stopped auto-starting work I'd told it to leave alone, its chat bot's buttons now work and answer only to me, and a game app stopped calling wins draws.
last updated
Bigbrain
Linear treats delegation as dispatch: hand a ticket to a role and it starts a session, whether a human meant that or not. Four different places in bigbrain-core used to do this on their own initiative: a five-minute sweep that pulled unheld issues into Triage, a filing path that delegated on creation, an operations sweep, a session's own follow-up. On 2026-09-01 the sweep caught 21 tickets I had filed as a plan and explicitly marked not to start, within one tick of their assignee guard lifting, and ran three Triage sessions and a Delivery session against them before anyone noticed. All four routes are gone now except the one that matters: when a blocker closes, its chained successor still moves to Triage automatically, because at that point the decision about who should own it is real work, not a guess. Everything else gets filed unheld, and stays that way until I assign it.
Two smaller gaps closed on the same webhook. The Telegram handler only ever read the text of a message, so a button press (the reminder list's dismiss/mute controls) arrived and was silently dropped; nothing was listening for it. And the shared secret Telegram sends back to verify a request matches for every chat the bot is a member of, not just mine, so a check that passed proved only that Telegram sent it, not that I did. Both are fixed together: callback queries are now parsed and handled, and every branch (chess, the milestone question, ticket filing) is gated on chat ID before it runs, failing closed if that ID is ever unset. The gate logs which chat got refused, without logging what it said.
The same day added the other half of a thread from JIO-707: a ticket filed with no milestone now gets asked about it over Telegram, one at a time, and the answer is recognised (a number, a milestone name, or "skip") before it falls through to being read as a new ticket titled "3". Each question expires after 24 hours and can be listed or answered from the command line if Telegram isn't where you happen to be.
Reckreation
From the deployed app: "Why is it saying I drew 0-0 after I won the casual set?" The confirmation screen was reading its score from two boxes meant only for score-only entries. Sets logged by tapping a winner per game, the ordinary way, left those boxes at their default of zero and zero, and the screen dutifully reported a draw in the largest text on the page. The scoring rule itself lived in a React component that the test suite has no way to run, which is why a build could pass while every normal win displayed wrong. The fix moves that rule into a plain function the tests can reach, and changes what the confirmation reads: instead of recomputing a score, it now reports the score that was actually sent to the server, the same value from the same computation, once.
Session setup got simpler alongside it: a ranked night now separates "ranked" from "how do we pair people up" as two independent questions instead of one, a pairing that's happened before is a single tap to log again, and a free-form ranked night only draws a matchup card when someone actually asks for one. A session no longer has to carry a card it doesn't need.
I run a personal assistant system that files and works tickets for me. Part of it used to jump the gun: a background check ran every five minutes and swept up any ticket that wasn't explicitly claimed, even ones I'd filed as a plan and marked "don't start this." On 2026-09-01 it grabbed 21 of those within a minute of an unrelated guard lifting, and set three work sessions running on tickets I hadn't approved. I removed every place in the system that could start work on its own initiative like that, except the one case where it's supposed to happen: when a blocker ticket finishes and its follow-up becomes ready, that follow-up still routes automatically, because at that point routing it is the actual job.
Two related gaps closed in the chat bot the same day. It has buttons now, for dismissing or muting a reminder, but the code receiving messages was only built to read text, so a button press vanished into nothing. And the check confirming a message really came from me was weaker than it looked: it only proved Telegram sent it, not that it came from my own chat, since Telegram sends the same proof to any chat the bot is part of. Both are fixed together, and every feature now checks whose chat a message came from first. Separately, a ticket filed without a milestone attached now gets asked about over the chat bot instead of just logged, one question at a time.
On my deployed pickup-game app: after winning a match, the confirmation screen would say you'd drawn it 0-0. It was reading the wrong stored value, one meant only for a rarely used manual-entry mode, so a normal win always displayed wrong even though the saved record was correct. It's fixed, and the fix also made that scoring rule testable, which it wasn't before. Setting up a session also got quicker: repeat pairings are one tap, and a matchup card only generates when someone actually wants one.