Your four pinned tracks, one column each, matching the sessions you run. Everything else sits on the bench below in declared order. Nudges fire where the labels disagree with that order.
Regina asks a question about their own spreadsheets and gets nothing back.
Documents do not reach the graph intact, so every answer is thinner than the data allows.
A person sees themselves as more than one user β disjointed identity across surfaces.
Content reaches someone who should not see it. Nobody experiences this until it is catastrophic, which is why it loses to everything above and needs a standing allocation instead.
| # | track | open | p0 | p1 | blocked | parked | what breaks without it |
|---|---|---|---|---|---|---|---|
| 5 | person-name-truth | 16 | β | 1 | β | 1 | Regina's people are in the graph under the wrong names, and 40% of her mail is held out of embeddings, extraction and search by a flag no one can explain per-row. Attribution is repaired in production; visibility is not. next: [L1-g2-w13] [person-name-truth] Automated emailβ¦ #4223 |
| 6 | regina-dedup | 10 | β | 3 | β | 1 | Regina's graph was wrongly merged and ingestion frozen for 9 days; the repair is done and ingestion resumed 2026-08-25 β the residual risk is un-verified repair side-effects (stale embeddings, the unsigned OCR pass) and the fold-cost class that caused it. next: [fold-efficiency] the fold's dominant cost isβ¦ #3495 |
| 7 | wire-truth | 3 | β | 1 | β | 1 | What WhatsApp actually carried never reaches us β senders unnamed, replies and mentions lost. next: graph_dedup skips 11 content-node types entirelyβ¦ #3554 |
| 8 | connector-truth | 6 | β | β | β | β | A customer connects a source and it dies days later. 5 of 9 production Gmail connectors are dead now. next: messages.message_hash includes the DERIVED sender name,β¦ #3792 |
| 9 | security | 2 | β | β | β | 1 | Cross-tenant exposure and privilege defects β one tenant's data reachable from another. next: Client end-user read endpoints run as DB superuser (RLSβ¦ #2445 |
| 10 | edge-truth | 5 | β | 1 | β | β | Edges do not mean what they say, so traversal answers are confidently wrong. next: fix(extraction): relationship typing structurally wrongβ¦ #1465 |
| 11 | nightly-nudge | 1 | β | 1 | β | β | Proactive cards reach the wrong people, or nobody. next: [L2-g3-w5] A linked account that resolves to no personβ¦ #3243 |
| 12 | billing | 1 | β | 1 | β | β | Signup cannot complete, or usage is not metered against what a contract promises. next: Meter enterprise AMA questions for billing β Regina'sβ¦ #2915 |
| 13 | retention | 5 | β | β | β | β | Deletion is promised in customer copy and no tenant-wide deleter exists. next: ADR-138 stage 6: the tenant-wide deleter has noβ¦ #2941 |
| 14 | test-truth | 13 | β | β | β | β | Coverage that reads as present and cannot fail β the suite is green about things it never checks. next: [L2-g4-w6] find_or_create's duplicate-key race branchβ¦ #2996 |
| 15 | measurement | 3 | β | β | β | 1 | Nothing tells us when a customer's ingestion has stopped. next: Health check: a connector with no successful sync for 3β¦ #4730 |
| 16 | llm-cost | 10 | β | β | β | β | What providers invoice us is not reconciled against what we think we spent. next: [L1-small-w4] Cost alerts fire into the void ββ¦ #3131 |
| 17 | mr-drain | 8 | β | β | 1 | β | Suppressions and debt accumulate faster than they are drained. next: [Tracker] Per-slice Materializer.add_intent() paths forβ¦ #1341 |
| 18 | doc-truth | 0 | β | β | β | β | The feature catalog describes a product we do not have. |
| 19 | regina-staging-gate | 0 | β | β | β | β | Regina's full-fidelity staging and its gated drain. |
| 20 | deploy-safety | 3 | β | β | β | 1 | A deploy or shared machine silently degrades the running product β the serving tier starved at 7.54% CPU steal until 2026-08-23; the remaining item is finishing the poll cutover so the old cron driver is retired. next: [deploy-safety] a status-board label error turns aβ¦ #3946 |
| 21 | fold-efficiency | 24 | β | 5 | β | β | The cross-source fold exceeds its 3h ceiling without document extraction's help, so new mail goes unscoped when the fold dies before scope_classification β and the doc-needs precompute it feeds is 87% dead weight by rows. next: [fold-efficiency] the fold's dominant cost isβ¦ #3495 |
| 22 | support-brain | 0 | β | β | β | β | Support questions have no answering surface; the console is live but the brain behind it is not. |
| 23 | buildkite-migration | 0 | β | β | β | β | CI ran on a second provider alongside GitHub Actions, at double the maintenance. |
| 24 | answer-truth | 11 | 2 | 1 | β | β | An aggregation question (spend, count, list-every) gets an answer summed from a top-k retrieval sample instead of the full matching set, and states it with confidence. next: [answer-truth] EPIC: answer questions exactly andβ¦ #4266 |
| β | ama-disclose-audience unranked | 5 | β | 1 | β | β | next: Node text and conversation_* spec text drift apartβ¦ #3387 |
| β | ama-disclose-org-membership unranked | 3 | β | β | β | 1 | next: [ama-disclose] disclosure gate keyed on org_membershipβ¦ #3596 |
| β | ama-trust:spreadsheets unranked | 6 | β | β | β | 1 | next: [ama-trust:spreadsheets] AMA refuses to aggregate aβ¦ #3996 |
| β | arabic-retrieval unranked | 32 | β | 8 | β | 4 | next: [arabic-retrieval] cost the tiered-extraction optionβ¦ #4270 |
| β | ask-canary unranked | 3 | β | β | β | β | next: [ask] Per-screen page title and TTI marks belong at theβ¦ #4376 |
| β | client-auth unranked | 2 | β | 1 | β | β | next: Web Google signup can never complete β the OAuthβ¦ #2768 |
| β | conversation-first unranked | 7 | β | β | 1 | β | next: [Tracker] Per-slice Materializer.add_intent() paths forβ¦ #1341 |
| β | conversation-truth unranked | 3 | β | β | β | β | next: [conversation-truth] Should conversation membership beβ¦ #3904 |
| β | identity-dedup-chokepoint unranked | 4 | β | 1 | β | β | next: No choke-point records that two identities are the sameβ¦ #3416 |
| β | mizo-relationships unranked | 1 | β | β | β | 1 | everything is parked |
| β | person-name-truth:low-value unranked | 7 | β | β | β | 1 | next: [L1-g2-w9] [person-name-truth] is_low_value is aβ¦ #3855 |
| β | viewer-account-resolution unranked | 7 | β | 3 | β | β | next: VIEWER-TRUTH: the viewer's display NAME has 5-6β¦ #3173 |