p95 was computed over ALL successful rows: cache hits (~10-30ms) set the
baseline, a single lock_wait row (duplicate click waiting on the in-flight
decode of the same VIN) set the spike. On a healthy 3-query window this read
as 'P95 15ms → 2058ms (137× baseline)' (2026-06-11, 4th such false alert).
- p95 now over FRESH decodes only (source='cache', db_hit/redis_positive and
lock_wait rows excluded; null-safe via coalesce)
- both windows must have ≥4 fresh decodes — a 'p95' of fewer is one slow
request, not a tail
- absolute floor raised 1s → 8s (fresh decodes legitimately take seconds;
matches the dashboard's fresh-P95 ok≤8s tone)
Simulated on today's prod data: alert window had 0 fresh decodes → silent.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
/analytics/proxy: Sase proxy_logs telemetrisinden (sase.tr dev 45f7e3f)
provider × servis karnesi (A-F: başarı/ban/transport/latency ağırlıklı),
banlı exit IP / sticky session tablosu, ≥400 status dağılımı (402 = proxy
bakiye), günlük seyir. Tablo prod'a promote edilene dek bilgilendirici boş
durum gösterir.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Lists catalog_issue_reported PostHog events (one-click 'Bildir' on
sase.tr empty/error catalog states, sase.tr dev c972a17): time, reporter
email (distinct_id → Sase users), VIN, vehicle, category, source, and
the page it happened on. Category distribution + unique-reporter KPIs,
7/30/90d ranges, linked from /analytics.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Prod doesn't write pcat_skipped/pcat_transient yet (sase fd78957 is on
dev), so the 2026-06-11 15:31 run still raised false gaps for Audi
(10/10 rows cold-skip), Hyundai (2/3) and VW (1/3 + 2 untraceable cache
echoes). A real pcat answer needs a network round-trip, so pcat <= 10ms
means skipped — exclude those rows. Verified on prod data: the three
false brands drop below threshold, Renault/Fiat/Honda etc. unchanged.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
'No catalog — identified as X' rows produced during a pcat cold-pool
skip, transport blip, or open circuit breaker are not evidence of a
coverage gap; counting them raises false brand-gap insights (Audi,
2026-06-11). Excludes rows carrying the new pcat_skipped /
pcat_transient markers (sase.tr dev fd78957) or pcat_circuit_open.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Raw succeeded/total read 61% on a healthy pipeline because coverage
gaps (no-catalog, unknown VIN) counted as failures, and latency
percentiles mixed ~10ms cache hits with ~5-40s provider chains
(P50 28ms next to P95 25.8s meant nothing).
- partition failures exactly: no_catalog / unknown / real (timeout,
provider) via one CASE bucket
- Başarı = decode health (coverage gaps excluded), Gerçek hata real
failures only, new Kapsama dışı card (katalog yok + unknown VIN)
- P50/P95/P99 now fresh decodes only, cache hits excluded
- evening brief gains realErrorRatePct/decodeHealthPct/noCatalog24h
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
After the success_rate_drop fix, the SAME benign data (one Fiat VIN, "No catalog —
identified as Fiat", retried across a couple users) just moved the false positive to
unknown_vin_spike (live: "Tanınamadı 11.1%→62.5%, 5/8, critical"). A car we DID
identify but lack a catalog for is a coverage gap, not an unrecognized VIN.
- unknown_vins / unknown_vin_users now count only TRULY unrecognized VINs:
exclude identified_no_catalog=1 (the "No catalog — identified as X" case).
- Add distinct_unknown_lookups (deduped) + gate the check on ≥4 distinct
unrecognized lookups so a couple users' uncovered cars can't trip it.
Validated vs prod query_logs: all recent windows truly_unknown=0 → GATED; a real
"tanınamadı" surge (≥2 users, ≥4 distinct unrecognized VINs, rate ≥2× baseline)
still fires. typecheck clean.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Completes acce1fc (which gated volume_spike/unknown_vin_spike for retry noise but
left success_rate_drop on a raw row-rate). Today two 🔴 success_rate_drop alerts
fired (88.9%→60%, 61.9%→33.3%) on a single user retrying one catalogless Fiat VIN
6× in a low-volume window — 0 real decode failures.
Root cause: success_rate_drop used succeeded/total over raw rows with only a
MIN_CURRENT_VOLUME=5 gate — no distinct-user gate and no coverage-gap exclusion,
unlike the sibling checks. At ~2.6 decodes/15min, one user's retries collapse the
"rate".
Fix — bring it to the same standard:
- Measure REAL (infra) failures only: exclude coverage gaps (no-catalog / unknown
/ unsupported VIN; ~20-35% of traffic, driven by which cars users query, not
decode health — already tracked by unknown_vin_spike).
- Dedupe by DISTINCT (user,vin) lookup so retries count once.
- Gate on ≥2 distinct users + ≥6 decodable lookups; fire on ≥10pp rise in real
failure rate AND ≥20% absolute floor.
- Also added an absolute-volume floor (MIN_VOLUME_SPIKE_ABS=25) to volume_spike,
which fired at now=8 yesterday (below baseline).
Validated against prod query_logs: both alert windows now GATED (real_failed=0);
the 06-10 09:30 real-incident window (7 users, 5/8 real failures, DataImpulse
degradation) still FIRES. typecheck clean.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Night of 2026-06-11 one user retrying a catalogless Toyota VIN tripped
volume_spike + unknown_vin_spike, and ~2.5/window night baselines made
quiet windows fire volume_drop at -100%.
- volume_spike: require >=2 distinct users and >=5 distinct (user,vin)
lookups so same-VIN retries don't count as a spike
- unknown_vin_spike: require unknown failures from >=2 distinct users
- volume_drop: raise expected-per-window floor from 1 to 4
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Recordings tab filters by signal (errors / long >=5m / active >=30 clicks)
with live counts, highlights error sessions, and adds a one-click PostHog
replay player link per recording. Turns the 5k+ archived recordings into a
triage surface for watching problem sessions.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Prisma serialized windowDays as bigint -> make_interval(days => bigint) does
not exist (42883), so detectCatalogCoverageGaps threw on every real call: the
evening brief silently dropped its Sase-DB half and catalog-gap-detect never
created insights. Use a JS Date param (matches the rest of the codebase).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- experiment results compute a two-proportion z-test p-value + 95% CI per
variant vs control (significant = p<0.05); shown in the archive detail view
and used to label the leader honestly.
- evening brief: getExperimentBriefSummary feeds active-experiment results
(leader, lift, p-value, significance) into the brief metrics; prompt tells
the LLM to surface only statistically significant winners, not noisy lifts.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Survey detail now shows the response funnel (shown -> sent -> dismissed) +
per-question distributions (rating/choice bars, open-text list), keyed by
$survey_id -> $survey_response_<questionId> from the archived definition.
Feature-flag/experiment detail shows per-variant exposure -> conversion with
lift-vs-control + a goal-event selector (signup/decode/trial/checkout/payment).
All from posthog_events; no schema change.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
40-page cap captured only the most recent ~13 days. Add a 150ms inter-page
delay (PostHog list API rate-limits a tight loop) and raise the default cap
to 200 pages / 120d window so a run can sweep the full retention.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- groups: posthog-resource-archive now also snapshots group types + group
instances (groups_types/ + groups/?group_type_index) into
posthog_resource_snapshots (change-detected).
- recordings: new posthog-recording-archive job (@*/6h) captures complete
metadata for EVERY recording into posthog_recordings (skipDuplicates over a
rolling 90d window) + MinIO dump — a superset of the insight pipeline's
promoted SessionMeta subset. rrweb blobs still archived by archive-recordings.
- /posthog-archive browser: Groups / Group Types tabs + a Recordings tab.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Read view over posthog_resource_snapshots: type tabs (feature flags,
experiments, surveys, insights, dashboards, annotations, actions) with
resource counts, latest-per-resource table, and a detail page showing the
full lossless payload + change history (version timeline). Nav entry added.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Extends archival beyond events/persons/cohorts. New generic resource
archiver (posthog-resource-archive@*/6h) snapshots surveys, feature flags,
experiments, dashboards, insights, annotations, and actions into
posthog_resource_snapshots — change-detected (hash over non-volatile config),
lossless `data` payload — and dumps a full per-type daily copy to MinIO
(resources/<type>/<date>.json.gz). listResource() pages any PostHog REST
resource. Survey responses + flag calls are already captured as events.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Daily Sase brief aggregating VIN decode health, catalog coverage gaps,
revenue (Stripe/iyzico TRY), users/subscriptions, PostHog behavior + UTM
source, Meta ads (env-gated), and the insight pipeline into one
critical-points-first report.
- Brief model (panel DB): one row/project/day (date, headline, summary,
health score, structured body, raw 360 metrics).
- Panel: getSaseBriefMetrics() (saseDb half) + /api/internal/sase-brief-data.
- Worker: sase-evening-brief job collects panel metrics + PostHog
(hogqlQuery) + Meta (lib/meta, env-gated) + pipeline stats -> DeepSeek
structured brief -> upsert Brief -> Telegram. Scheduled 17:00 UTC
(20:00 Europe/Istanbul).
- UI: /projects/sase/briefs (table of days) + /briefs/[id] (detail:
critical-first, sections by status, recommendations, raw metrics).
Meta dormant until META_ACCESS_TOKEN/META_AD_ACCOUNT_ID set on the worker.
Needs a `briefs` table migration on the panel DB before first run.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Surfaces "vehicle resolved but no parts/categories" — Sase's #1 churn
signal, previously invisible to insights (a prospect emailed support
instead of any insight firing).
Client (PostHog session tagger):
- fetch the catalog_*/category_view_changed/empty_catalog_cta_clicked
event family (was never pulled) and tag parts_render_blocked (P1) /
catalog_empty_result (P2), scoped to the real /dashboard product (demo
excluded). Harden the search-affordance-misuse guard so a real parts
failure is never suppressed. Route both tags to bug_triage.
Server (decode-log catalog gaps):
- new catalog-gap-detect worker job (every 6h) -> panel internal API
/api/internal/catalog-gap-check aggregates "No catalog - identified as
X" query_logs failures by brand -> upserts catalog_coverage_gap
insights (Telegram on P1), respecting founder triage. Add NO_CATALOG
error bucket.
Validated: both apps typecheck clean; tagger smoke 21/21; aggregation
over 90d real data yields 13 brand gaps (Renault/Fiat/Honda P1).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
On SIGTERM/SIGINT the worker only disconnected prisma/redis and exited — the
BullMQ Workers were never closed, so in-flight job locks were never released. On
every deploy the orchestrator killed the worker mid-flight, the next container
saw the half-finished jobs as "stalled", re-ran them, and the dead worker's
pending moveToFinished surfaced as `Missing lock for job
repeat:nightly-refresh:… moveToFinished`, with duplicate nightly-refresh firings
during the container overlap.
- startScheduledJobs / startInsightPipeline / startContentPipeline now return
their Worker so index.ts can close them.
- shutdown() closes all workers FIRST (releases locks, drains in-flight),
bounded by a 15s race so an in-flight job can't block past the orchestrator's
stop grace period, then disconnects prisma/redis.
- nightly worker now uses lockDuration 5min + stalledInterval 60s (matching the
pipeline/content workers); the 30s default could expire during panel-backup
(pg_dump) and trip the same stalled → re-run → Missing-lock cycle.
The scheduler config itself was already correct (nightly@03:00, audit@03:30,
panel-backup@04:00 — verified against the Redis job-scheduler ZSET; next fires
were exactly 03:00/03:30/04:00). tsc --noEmit clean; dedup + tagger smoke 30/30.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Users routinely use the VIN-only /dashboard/search box to look for a part by
name (e.g. "Cam düğme") or fiddle with search/history and rage-click out of
affordance confusion — while nothing is actually broken. The generic
ux_friction / frustrated_session tags turned these pure-rage sessions into
insights (noise; ~40 of the first 67 open insights were exactly this, all
dismissed by the founder).
apps/worker/src/lib/tagger.ts: when a session used the search box
(search_input_focused) but carries NO concrete failure (errorCount=0,
network5xxCount=0, no vin_decode_failed, search_input_validation_failed<3, no
payment_initiated/failed, no checkout_started), do not emit the generic
ux_friction / frustrated_session tags. With no other actionable tag the session
becomes tag-less and tag-sessions discards it → no compress / analyze / insight.
Scoped narrowly to search-box sessions on purpose, to avoid hiding genuine
parts/category bugs. Every concrete signal stays actionable: JS errors
(bug_suspected), 5xx (server_error_impact), payment friction, VIN upstream
provider failures, and ≥3 client validation failures (search_validation_friction).
Smoke: apps/worker/src/lib/tagger.smoke.ts (12/12) covers the misuse case plus
six must-keep cases. tsc --noEmit clean.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The fingerprint hash (even after a7fe80f) still over-splits: the same root
problem produces different tag-sets / first-error / first-failed-endpoint
across sessions, so it hashes to a new fingerprint and creates a duplicate
insight. And generic ux_friction/frustrated_session sessions (category-tree
vs parts-panel vs schema-image) carry no structured signal to separate
sub-problems — only the LLM's reading of the timeline does. Dismissed insights
were only suppressed on an EXACT fingerprint repeat; a near-variant got a fresh
fingerprint and re-opened the theme. Dismissed rows were also excluded from the
fast-path cache, so every recurring session re-ran the LLM for no reason.
Two-layer dedup so (1) duplicates can't be created and (2) nothing similar to a
dismissed/duplicate theme is re-opened:
Layer 1 — fingerprint (or aliased fingerprint) fast-path, no LLM:
- dismissed/duplicate → SUPPRESS (bump occ/lastSeen, keep status)
- validated/shipped → flag REGRESSION (cheap, no re-analysis)
- active & fresh → attach occurrence
- active & stale → fall through to refresh re-analysis (unchanged)
Layer 2 — semantic dedup gate before creating a NEW insight (flash, no new infra):
- build a bounded catalog: all active insights + recently-touched
dismissed/duplicate/validated/shipped (lookback-windowed, capped)
- ask the model whether the session is the SAME underlying problem as a
catalog entry (anti-hallucination: only accept catalog ids; conf >= 0.72)
- on match: alias this fingerprint onto that insight + attach/suppress, so the
next identical session fast-paths in Layer 1 (no LLM, no duplicate row)
- no match: create a new insight as before
Schema: Insight.aliasFingerprints String[] (additive; applied via prisma db push
on deploy). Dedup-gate cost is logged to costLedger as promptTag=dedup_gate.
Tunables: INSIGHT_DEDUP_GATE (default on), INSIGHT_DEDUP_THRESHOLD (0.72),
INSIGHT_DEDUP_SUPPRESS_DAYS (90), INSIGHT_DEDUP_CATALOG_MAX (120).
Pure decision helpers (classifyExistingAction, shouldAcceptMatch) extracted and
covered by a smoke test (apps/worker `pnpm test`, 18/18). tsc --noEmit clean.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Same root cause was producing one Insight row per user because URL carried
vehicleId/categoryId UUIDs and the raw URL fed the fingerprint hash. Result:
6 active "new" insights describing the same kategori-bouncing problem with
slightly different LLM phrasing, none deduplicated, only one (cmpclg7b9 →
sase.tr#76) had been triaged.
apps/worker/src/lib/compress.ts:
- new normalizePath() — collapses UUID / ULID / CUID2 / numeric path segments
to ":id", conservative on plain words. Mirrors what the LLM already sees in
the timeline.
- fingerprint inputs now:
tags (sorted) | normalizePath(url) | normalizeError(errors[0]) | normalizePath(failedEndpoints[0])
- removed header.severity from the hash — severity is a property of the
Insight bucket, not its identity; rage-click counts pushing the same root
cause across P1/P2/P3 was forcing extra rows.
Smoke (12/12 pass via tsx /tmp/check_fingerprint.ts):
- 6 historical kategori-bouncing URLs → 1 fingerprint
- severity changes don't move the hash
- distinct tag sets / distinct errors still split
Historical rows are untouched — only new compressed sessions get the new
hash. Old near-duplicate insights can be merged manually via the panel.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reproducing insight cmpvfrjgc000114fzc7cdyh66 (a P1 "PL24 timeout" false
positive): trial user typed VW part numbers ("500 907 521", "5Q0 907 521")
into the VIN field on /; client-side regex rejected them with "Geçersiz şase
numarası. 17 karakter olmalı". No provider was called. The pipeline still
tagged the session as `vin_decode_fail_pattern`, routed to `provider_quality`,
and the LLM dutifully invented a PL24 outage.
Root cause spans three files:
1. tagger.ts grouped vin_decode_failed by `provider_attempted ?? source`. When
`provider_attempted` is missing, `source: "landing"` (a UI location) was
treated as a provider name, so a 1-provider set was synthesized and
`vin_decode_fail_pattern` (P1) was emitted.
2. compress.ts formatCustom whitelist excluded `error`, `source`, `vin`. The
LLM therefore never saw "Geçersiz şase numarası" or the offending input.
Pattern 3 mechanical hypothesis told it "check provider health" regardless.
3. prompts.ts pickPromptTag routed any `vin_decode_fail_pattern` straight to
`provider_quality` with no input-quality check, and the v3 system prompt
had no guardrail for client-side validation rejects.
Fix:
- tagger: detect client-side rejects by `error` regex (Turkish + English) and
by VIN shape (length != 17 or contains I/O/Q). When all fails are client
rejects, emit new tag `vin_decode_client_validation_fail` at P3 instead of
`vin_decode_fail_pattern` at P1. Real provider failures now require
`provider_attempted` to be set (no more `source` fallback).
- compress: add `error`, `source`, `vin` to the formatCustom property
whitelist so the LLM can see the actual failure context. Split Pattern 3
into client-reject vs. real-provider-failure branches with distinct
Turkish hypotheses.
- prompts: route `vin_decode_client_validation_fail` to `ux_friction` before
the provider rule. Ship provider_quality v4 with an explicit guardrail
instructing the model to return confidence ≤0.15 and reclassify when the
inlined event properties show client-side rejection.
The seed-runtime upsert path deactivates the active v3 template on next
worker boot and inserts v4 in its place — no manual SQL needed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The sentry-archive job was hardcoded to a single SENTRY_PROJECT slug.
Now that sase has a second Sentry project (sase-web, browser SDK), the
worker has to pull both — otherwise frontend events live only in
Sentry's UI and never reach the panel's sentry_events table for the
behavioral-insight pipeline to pick up.
Changes:
- SENTRY_PROJECT becomes a CSV (e.g. "python,sase-web") parsed once at
module load into PROJECTS[]. Backward-compatible: a single value
keeps working exactly as before.
- Issues are fetched org-wide in a single loop (unchanged endpoint)
but each issue's Sentry project slug is now persisted via the new
`sentrySourceProject` column on sentry_issues.
- Events are project-scoped on Sentry's side, so the job iterates
projects and calls listEventsPage(slug, cursor) for each. Per-project
pagination + dedup is preserved. A project-level failure is recorded
in `error` but doesn't abort the other projects.
- New result shape: `perProject: { [slug]: { fetched, inserted, duplicate } }`
Pipeline log gains a `[python=N/M sase-web=N/M]` breakdown so it's
obvious at a glance which project produced what.
- Schema: nullable `sentrySourceProject` on both SentryIssue and
SentryEvent (+ composite index per projectKey for queries that want
to filter "panel project = sase AND sentry project = sase-web").
Migration is a pure nullable add — `prisma db push` on container
start is safe. Existing rows stay null until next sync touches them.
After merge, set `SENTRY_PROJECT=python,sase-web` on the panel-worker
Coolify app and redeploy.
Mirrors Sentry into our DB before the free tier prunes events (~30d).
- `SentryIssue` (aggregate state, upserted to latest) + `SentryEvent`
(raw occurrences, lossless full payload JSONB, dedup by eventId, cold
dump to MinIO `sentry-archive/{project}/YYYY/MM/DD.jsonl.gz`).
- `lib/sentry.ts`: read-only client, Link-header cursor pagination,
listIssuesPage / listEventsPage. EU-region aware (SENTRY_API_BASE).
- `sentry-archive` job @hourly: issues upsert + events newest-first with
skipDuplicates, stops once a page is all-duplicates (caught up).
- Config via env: SENTRY_AUTH_TOKEN / SENTRY_ORG / SENTRY_PROJECT /
SENTRY_API_BASE.
Verified live against otolog/python (EU): a test event archived on run 1,
0 inserts / 1 duplicate on run 2 (dedup), issue upserted idempotently.
Note: Sase API currently sends to an inaccessible org's DSN (hardcoded
fallback); repointing SENTRY_DSN to otolog/python is a separate deploy step.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Append-only history of person properties and cohort definitions, since
the free tier overwrites the live person / prunes data. A new row is
written only when the payload changes (hash compare), so the tables stay
a compact change-timeline rather than a daily full copy.
- `PosthogPersonSnapshot` (personId join key to PosthogEvent.personId,
all distinct_ids, properties JSONB, propertiesHash).
- `PosthogCohortSnapshot` (cohortId, name, count, filters, stateHash).
- `listPersons` / `listCohorts` REST helpers (next-pagination) in posthog.ts.
- `stableHash` (sorted-key JSON hash) in hash.ts for change detection.
- `posthog-identity-archive` job wired @03:00 daily. Latest-hash lookup
via one DISTINCT ON query, inserts only changed rows via createMany.
Verified against prod: 242 persons → 242 snapshots on first run, 0 on
immediate re-run (change detection). 0 cohorts currently → graceful no-op.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Live verification surfaced two issues in archive-recordings:
- blob_v2 rejects wide blob-key ranges (a 23-key span 400s; ~5 is fine).
Fetch in contiguous chunks of BLOB_CHUNK (default 10) and concatenate.
- PostHog's snapshot API is aggressively rate-limited; back-to-back
sessions tripped 429. Pace requests (per-session + per-chunk delays)
and stop the run on sustained 429 — unarchived rows keep
rrwebArchivedAt=null and retry next cycle (oldest-first ordering
protects soon-to-be-deleted recordings first). Result now carries
rateLimited.
Verified against prod: 87- and 23-blob recordings reconstruct fully
(no 400), failed=0, graceful 429 stop.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase A of the external observability archive. PostHog Cloud free tier
deletes data on a rolling window (events ~1yr, session recordings ~30d);
this mirrors both into our own infra permanently (hot Postgres + cold
MinIO gzip JSONL).
PostHog raw event archive:
- New `PosthogEvent` model (uuid PK → dedup, full properties JSONB,
hot columns promoted + indexed).
- `posthog-event-archive` job: watermark-paged HogQL pull (cursor in the
existing IngestionWatermark via a "sase:events" stream key), createMany
+ skipDuplicates for idempotency, 365d backfill. Daily closed-day cold
dump to MinIO `posthog-archive/{project}/YYYY/MM/DD.jsonl.gz`.
- `hogqlQuery` helper added to posthog.ts. Timestamp cursor uses
parseDateTimeBestEffort() — ClickHouse 500s on a raw ISO8601 literal
(verified live against the project).
Raw rrweb recording preservation (separate from compress, which only
covers scored sessions and caps blobs):
- `SessionMeta.rrwebArchivedAt` / `rrwebArchiveKey`.
- `archive-recordings` job: every recording within 25d (margin before
30d deletion), full blob range (no 12-cap), gzip → MinIO
`rrweb-archive/{project}/YYYY/MM/DD/{sessionId}.jsonl.gz`.
- `putBuffer` gzip helper in minio.ts.
- Both jobs wired into pipeline.ts (event-archive@*/15min, recordings@*/6h).
- retention.ts untouched → new table + buckets persist forever (the goal).
Schema applies via the existing `prisma db push` on web start (additive:
new table + nullable columns).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Click a VIN in the list to open a modal with the full decode internals:
- Proxy times per provider (pcat/emex/pl24/vin-api) from query_logs.timings
- Race/lock status (lock_wait ms + cache_source='lock_wait' → "aynı VIN
paralelde çözülüyordu"), candidate_pick, pl24 circuit, aborted
- Candidate counts (pcat_car_count, emex_candidate_count), result_kind, wmi
- User behaviour for that vehicle, correlated from PostHog custom events in
the panel DB by user_id + vehicle_id: categories opened (parts_panel_viewed)
and OEM copied (oem_code_copied + distinct codes)
New: getVinDecodeDetail/getVinUserBehavior (cross-DB: sase RO + panel),
GET /api/sase/query-log/[id] (auth-guarded), client VinDetailButton modal.
Both queries verified against prod data.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Business page returned 500: getTrialFunnel's `ORDER BY CASE bucket ...`
referenced the SELECT-list alias inside an expression, which Postgres
can't resolve (error 42703 "column bucket does not exist"). Wrap the
GROUP BY in a subquery so `bucket` is a real column the ORDER BY can use.
Verified against prod: original errors, fixed query returns ordered buckets.
- Remove the "Çözülen / Çözülemeyen VIN'ler" dashboard buttons — the VIN
list already has a success/error status filter, so they were redundant.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Systematize Faz 4 + close three observability gaps:
- WMI opportunity radar (getWmiOpportunities): group query_logs by
substring(vin,1,3) = WMI, rank low-success WMIs by distinct-user demand.
brandMatched=0 flags fully-uncovered manufacturer codes. Surfaced on the
Trends page (#wmi) + a dashboard header shortcut. Replaces the manual Faz 4.
- Per-source p50: add percentile_cont(0.50) to getProviderAttempts and
getProviderDeepStats; show P50·P95 on the dashboard and provider drill-down.
- unknown_vin_spike anomaly: track "tanınamadı" rate in the 15min/baseline
windows, fire when it jumps >=2x baseline (or surges from ~0), and attribute
the dominant failing source (+aborted count) in the Telegram message. Runs
in the existing 5min anomaly cron; worker/telegram unchanged (generic type).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Add two header buttons on the VIN decode dashboard linking to the
existing /vins list pre-filtered by success=true / success=false,
carrying the current time-range window via the `from` param so the
list matches the dashboard view. Shows actual VIN codes (not counts).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Approved drafts publish via a single n8n `publish-content` webhook; the panel
POSTs the effective content (founder edits merged over generated body) and
awaits a synchronous Respond-to-Webhook result (sp.semih.ai is Tailscale-only,
so we avoid an n8n→panel callback). n8n routes by channel.
- lib/n8n.ts: publishToN8n client (X-Content-Secret header, timeout, tolerant
result parsing: ok|success + publishedUrl|url|postUrl|permalink)
- publishDraft server action: approved|failed → publishing → published(+url) /
failed(+error), audit-logged; effective content = bodyJson + founderEdits
- DraftCard: "Yayınla" / "Yeniden yayınla" button + publishing state
- docs/n8n: importable publish-content workflow (Webhook → Switch → LinkedIn /
HTTP-blog → Respond) + runbook (contract, panel envs, LinkedIn OAuth setup,
blog endpoint = sase.tr POST /blog/posts/internal Bearer)
Needs panel-web envs N8N_PUBLISH_WEBHOOK_URL + N8N_WEBHOOK_SECRET. Publish is a
graceful no-op (clear error) until those are set and the n8n workflow exists.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three coupled fixes that together stop frustrated sessions from being
mislabelled as `power_user_path` and make click timelines actually
diagnosable.
1) Counter propagation (compress.ts, compress-sessions.ts):
compress now returns rageClickCount / network5xxCount / network4xxCount
and writes them back to SessionMeta. Previously these stayed at 0
forever because PostHog recording metadata doesn't expose them and
nothing updated the row after compress ran.
2) Tag rule hardening (tagger.ts, compress-sessions.ts):
- power_user_path v1 fallback now requires rageClickCount===0; without
this guard any auth user with 20+ clicks (rage clusters included)
was labelled a power user.
- New frustrated_session tag (rageClickCount>=3, P2) for sustained
friction beyond a single cluster.
- compress-sessions re-runs tagSession+scoreSession after writing the
fresh counters, so the corrected tags land on the row.
3) Click target enrichment (compress.ts):
describeTarget used to read tagName/attributes off MouseInteraction
events — fields that don't exist on rrweb type=3 source=2 — so every
click rendered as `[el]`. Now compress builds a node map from
FullSnapshot (type=2) and keeps it in sync with mutations
(type=3 source=0), then resolves clicks via `d.id`. Clicks now show
real tag + visible text / aria-label / name / data-testid.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The HYPOTHESES block was empty on most sessions because the only rules
covered backend failure patterns (5xx + rage). Add 7 behavioral patterns
derived from canonical custom events — these give the LLM analyzer
concrete starting points instead of reasoning from raw timeline alone.
Patterns (all conservative — fire only on clear evidence):
1. Catalog evaluated, no checkout (parts_panel_viewed>=3 + on upgrade
page + !checkout_started)
2. Fresh trial + successful decode (happy onboarding)
3. VIN decode failure pattern (failures>=2 or fallback>=1)
4. Rage click on subscription UI (upgrade friction)
5. Cancel flow entered without resume (churn risk)
6. Trial urgency banner ignored (banner_viewed without CTA)
7. Search opened but no decode attempted (funnel drop)
The previous 500-token threshold fired on virtually every authenticated
session because rrweb masks inputs at source and stripQuery() removes
URL params before sanitize() runs — so 0 matches is the normal case,
not an anomaly. Confirmed on session 019e455e: 9m32s authed session
with ~1093 tokens, 0 matches, no actual PII leak.
- Raise broad threshold 500→2000 tokens
- Add precise leak detector: any unmasked \`input [tag] "text"\` line
in the compressed timeline AND authed AND 0 matches → alert. This
catches the actual failure mode (PostHog source-side masking broken)
instead of guessing from total token volume.
- Sync UI warning threshold to match (500→2000).
inferSubscriptionTier only checked subscription_tier|tier|plan_tier|plan,
but Sase.tr's setPeopleProperties writes subscription_plan. Result: every
authed user showed tier=— in /insights/sessions/* and the tagger's
userValue boost (full=1.0, brand=0.7) collapsed to the 0.5 default.
- Add subscription_plan to the key list (priority first since it matches
Sase.tr's actual property name)
- Extract into shared extractTierFromProps(props) helper
- In tag-sessions, re-evaluate tier from enrichment.userProperties and
backfill when null. Tag step runs ~2min after ingest, by which point
PostHog has the user's post-signup $set props that weren't on the
recording row at ingest time.
Drop successful (2xx/3xx) asset and telemetry GETs from the semantic
timeline — these were eating MAX_LINES (80) before any user-behavior
signal could surface. In a 9m32s session we observed only the first
1:18 reached the LLM input because PostHog /collect/* pings filled
the budget. Failures (4xx/5xx) are still emitted as signal.
Filtered prefixes: /collect/, /flags, /array/, /static/, /assets/,
/api/surveys, /css2. Also raise MAX_LINES 80→120.
Add optional { sessionId } to POST /api/insights/reprocess so a single
session can be re-run end-to-end without resetting a whole batch.
For string filter values (q, success, sort, etc.), the URL query string
came out empty — array values (provider, status with multi-select)
worked. Root cause was a brace-less if/for/if chain:
if (Array.isArray(v)) for (const item of v) if (item) usp.append(k, item);
else usp.set(k, String(v));
JS associates the `else` with the inner `if (item)`, not the outer
`if (Array.isArray(v))`. So when v is a string, the outer if is false,
nothing runs, usp stays empty, and the URL becomes just "?". The status
pills and the "Ara" search button both looked broken because their
patches were string-typed.
Same buildHref in users/_query.ts already had braces, which is why
the user list search was unaffected.
Add the missing braces. Also drops the temporary console.log
diagnostics from the VIN filter bar.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
