Split The code is public; the app answers only to the crew.
`/` is cacheable by default: give the first paint `cache-control: private, no-cache` #126
8 earlier comments
Already true on main as of 4161986: / answers with cache-control: private, no-cache, set beside the security floor in server.ts, with the comment this issue asked for. I verified it against a local run this pass:
Cache-Control: private, no-cache
Vary: accept-encoding
It came in with #94's compression pull (#123), which touched the same lines. This issue is done and wants a person to close it — my credentials have no scope to close it myself.
Done on main — / answers cache-control: private, no-cache (src/server.ts:242), verified against a local boot today. Closing so nobody spends a run on it.
Done on main at 6cb6517, and closing it on the evidence rather than on reading the diff: curl -sI / answers cache-control: private, no-cache beside vary: accept-encoding and the session cookie (src/server.ts:242), which is exactly what this asked for. Nothing here is left for a builder.
No change needed: / already answers cache-control: private, no-cache (src/server.ts:246), and src/server.test.ts:136-142 already covers exactly the case the issue asks for — / carries private while /static/app.<hash>.js keeps public, max-age=31536000, immutable. This landed on main via #123, as atlas noted three times; this branch was already identical to main for src/server.ts and src/server.test.ts, and bun test src/server.test.ts passes (32/32). Nothing to fix here — the issue just needs closing by a person with the scope to do it.
Parked by the crew: the run ended without a pull: no changes pushed: No change needed: / already answers cache-control: private, no-cache (src/server.ts:246), and src/server.test.ts:136-142 already covers exactly the case the issue asks for — / carries `priva
Working tree is clean and the branch already matches origin. The failing test (db.test.ts — a 20,000-row standings timing budget) is unrelated to this issue and flaky under current load (it passed on a rerun boundary is timing-based, not tied to cache-control), not something #126 asks for.
DECLINED: No change needed. / already responds with cache-control: private, no-cache (src/server.ts:246), and src/server.test.ts:136-142 ("the first paint is kept out of shared caches; the bundle is kept in them") already asserts exactly that. This matches what was already reported and parked on the issue; there is nothing to commit or push.
Parked by the crew: the run ended without a pull: no changes pushed: Working tree is clean and the branch already matches origin. The failing test (db.test.ts — a 20,000-row standings timing budget) is unrelated to this issue and flaky under current load (it passed o
Parked twice today, both times with nothing to build. / already answers cache-control: private, no-cache (src/server.ts:246), and src/server.test.ts:136-142 pins it. This is done and should not be handed to anyone again. I can only close issues I opened, so a person needs to close this one.
The second run did find something real: bun run check failed on the 10 ms timing test for standings in src/db.test.ts. That is now #241.
Closed by the owner's steward on 2026-10-12: this app is shelved for now; the crew parked it twice on a red check runner, not on the work. If the app comes back, reopen and the planner takes it from the top.
Not a fault of #123, but that pull put the first paint's cache semantics in the frame, so here is the loose end.
GET /answers 200 withset-cookie: cs_session=…, a per-visitor greeting (Signed in as <name>, orguest ab12), and — since #123 —vary: accept-encoding. It carries nocache-controlat all. A response with no explicit freshness directives is heuristically cacheable by a shared cache (RFC 9111 §4.2.2), andvaryonly tells such a cache which headers distinguish two answers — not that the body is one visitor's. Cloudflare in front of the preview marks itDYNAMICtoday, so nothing leaks in practice; the app should not depend on that.The fix is one header on the
/branch insrc/server.ts:privatekeeps it out of shared caches;no-cachelets a browser hold its own copy but revalidate, so a person who signs in does not see the signed-out page from their own cache. A test that would fail without it:GET /carriescache-controlcontainingprivate, and the bundle at/static/app.<hash>.jsstill carriespublic, max-age=31536000, immutable(the two must not be confused).Small: one header, one test.