docs: drop numeric server error codes from notes and comments

DESIGN.md

@@ -281,9 +281,9 @@ That last case previously produced **no report whatsoever** — the SDK shut dow
 2. `POST /v2/issues?app_token=…` — body = `app_token + access_token + request.json` → `result.endpoint` (presigned URL).
 3. `PUT <presigned>` — raw ZIP bytes, `Content-Length`, no auth header. Success → delete bundle + cached URL.
 
-**Server semantics:** `12003` SimilarCrashExists → drop silently; `12004` TooManySimilar → blacklist those `signatures` locally + drop; `401`/invalid → re-register session + retry; `429`/`503`/network → retry.
+**Server semantics:** the "similar crash exists" verdict → drop silently; the "too many similar" verdict → blacklist those `signatures` locally + drop; `401`/invalid → re-register session + retry; `429`/`503`/network → retry.
 
-**Offline queue + cross-launch retry:** bundles in `<data>/queue/`; sync uploader thread (default; optional async `reqwest`) with backoff `min(30s·2^n, 300s)`, caps export 15 / upload 60. Pending-report list + retry counters persist → next launch re-drives. Crash reports flush synchronously first. A presigned endpoint obtained from `create_issue` is persisted (a `.endpoint` sidecar) when its PUT fails transiently, so a retry **resumes at the PUT** rather than re-creating the issue (which would mint a duplicate or trip server dedup `12003` and drop the bundle un-uploaded). The retry counter (and the abandon-at-cap deletion) advances **only** on backoff-paced attempts; a forced `flush()` re-attempts a queued report but never consumes its durable budget, so an offline flush can't delete a still-deliverable crash bundle.
+**Offline queue + cross-launch retry:** bundles in `<data>/queue/`; sync uploader thread (default; optional async `reqwest`) with backoff `min(30s·2^n, 300s)`, caps export 15 / upload 60. Pending-report list + retry counters persist → next launch re-drives. Crash reports flush synchronously first. A presigned endpoint obtained from `create_issue` is persisted (a `.endpoint` sidecar) when its PUT fails transiently, so a retry **resumes at the PUT** rather than re-creating the issue (which would mint a duplicate or trip server dedup and drop the bundle un-uploaded). The retry counter (and the abandon-at-cap deletion) advances **only** on backoff-paced attempts; a forced `flush()` re-attempts a queued report but never consumes its durable budget, so an offline flush can't delete a still-deliverable crash bundle.
 
 **APM:** separate lighter path `POST /v2/performance/transactions` (one bare transaction, batched ~30 s or realtime).
 

PROGRESS.md

@@ -158,7 +158,7 @@ were pre-verified by inspection against `crash-handler` 0.6.3 + `libc`).
   60-retry cap while offline — forced attempts are decoupled from the counter, so
   a still-deliverable crash bundle can't be deleted in ~1.3 s. Retry now **resumes
   at the presigned PUT** via a persisted `.endpoint` sidecar instead of re-POSTing
-  `create_issue` (no duplicate issues / no `12003`-drop of an un-uploaded bundle).
+  `create_issue` (no duplicate issues / no duplicate-verdict drop of an un-uploaded bundle).
 - **PII (high):** failed-request network entries scrub the URL out of reqwest's
   error `Display` (was leaking full credential-bearing URLs); header redaction is
   unified onto the normalized param matcher (`signature`/`csrf`/`api_key` no
@@ -262,8 +262,8 @@ The defects, in the order they blocked things:
    application errors with `ok:false` and **HTTP 200**, but the transport only
    inspected the envelope for 4xx/5xx. The error degraded to "no endpoint in
    response", the report was dropped, and the server's explanation was thrown
-   away. The known codes `12003`/`12004` were reachable only on a 4xx too — so
-   `12004` skipped the signature blacklist entirely.
+   away. The known duplicate and too-many-similar verdicts were reachable only on a 4xx too — so
+   the latter skipped the signature blacklist entirely.
 3. **Nothing could observe a dropped report** (`1b2a456`). `flush()` returns
    "the queue is empty", and the queue drains on success *and* on permanent
    failure alike — so it returned `true` while the server accepted nothing.

crates/bugsee-core/src/queue.rs

@@ -190,7 +190,7 @@ pub fn next_attempt(report: &QueuedReport) -> i64 {
 /// The cached presigned upload endpoint for a report, if a prior attempt created
 /// the issue but the PUT failed transiently. A retry resumes at the PUT (step 3)
 /// instead of re-POSTing `create_issue` (which would mint a duplicate issue, or
-/// trip server dedup `12003` and drop the bundle without ever uploading it).
+/// trip server dedup and drop the bundle without ever uploading it).
 pub fn endpoint(report: &QueuedReport) -> Option<String> {
     std::fs::read_to_string(report.endpoint_path())
         .ok()
@@ -262,7 +262,7 @@ fn load_blacklist(data_dir: &Path) -> BTreeSet<String> {
         .unwrap_or_default()
 }
 
-/// Persist crash signatures the server told us to stop sending (`12004`).
+/// Persist crash signatures the server told us to stop sending (the too-many-similar verdict).
 ///
 /// Appends only the not-yet-present signatures, each on its own line, under
 /// `O_APPEND`. Appending (rather than read-modify-rewrite of the whole set) is

crates/bugsee-core/src/runtime.rs

@@ -2019,7 +2019,7 @@ fn deliver_queued(
         request_json,
     };
     // Resume at the PUT if a prior attempt created the issue but the PUT failed
-    // (F14): re-POSTing create_issue mints a duplicate issue, or trips 12003 and
+    // (F14): re-POSTing create_issue mints a duplicate issue, or trips the server's dedup and
     // deletes the bundle unuploaded.
     let cached_endpoint = queue::endpoint(report);
     let mut endpoint_out: Option<String> = None;

crates/bugsee-core/src/transport.rs

@@ -55,7 +55,7 @@ pub trait Transport: Send + Sync {
 ///
 /// `cached_endpoint` is a presigned upload URL a prior attempt obtained but
 /// failed to PUT to; when set, delivery resumes at the PUT instead of re-creating
-/// the issue (avoiding a duplicate issue / a `12003` drop of an un-uploaded
+/// the issue (avoiding a duplicate issue / a duplicate-verdict drop of an un-uploaded
 /// bundle). On a *transient* PUT failure the endpoint that should be persisted
 /// for the next retry is written to `endpoint_out`; on any other outcome
 /// `endpoint_out` is left `None` (the caller then clears any stale cache).
@@ -106,7 +106,7 @@ pub fn deliver(
                 return match transport.upload_bundle(&endpoint, &report.zip) {
                     Ok(()) => Ok(()),
                     // Persist the endpoint so the retry resumes at the PUT rather
-                    // than re-POSTing create_issue (duplicate issue / 12003 drop).
+                    // than re-POSTing create_issue (duplicate issue / duplicate-verdict drop).
                     Err(TransportError::Transient(e)) => {
                         *endpoint_out = Some(endpoint);
                         Err(TransportError::Transient(e))