> For the complete documentation index, see [llms.txt](https://docs.cortex.aic.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.cortex.aic.io/workflow-guides/analyse/indicator-watches.md).
# Indicator Watches
**Open this area:** In the CORTEX menu, select **Assurance & governance → Pulse & audit**.
Define exact judgement change conditions, distinguish absent and stale observations, and retain deliberate alert follow-up.
**Who uses this:** Analyst, Reviewer, Requirement owner.\
**Open in CORTEX:** `/pulse/watches`.\
**In-app guide:** `/help?topic=indicator-watches`.
**Expected result:** Evidence-backed observations and accountable feedback inform reassessment without automatic confidence, closure or tasking decisions.
## Before you start
* Select the correct organisation and workspace. Current Analysis read and cited evidence/source access are required; authors also need Analysis write. Personal subscriptions require Pulse read.
* Choose the exact retained analytical revision and judgement. The indicator freezes that judgement and profile, even if the analysis later changes.
* Decide what can be measured, its units, threshold, horizon, expiry and maximum observation age. Record how the expected change would affect uncertainty.
## What to do
{% stepper %}
{% step %}
## Define the condition
From analytical detail, open Define or review judgement indicators.
Choose the exact revision and judgement, state observable evidence and measurement units, choose at least/at most/exactly equal and enter the threshold.
Describe the expected change and effect on uncertainty, provide UTC horizon/expiry and retain your reason.
{% endstep %}
{% step %}
## Approve the watch deliberately
Open the indicator, review its exact context and choose Approve or pause.
Record an approval reason for the current unexpired condition. A revised threshold or horizon needs fresh approval; historical observations and evaluations remain attached to their original revision.
{% endstep %}
{% step %}
## Enable a personal subscription
Choose Manage my personal watch and enable it with a reason and an interval of 15–1,440 minutes.
Delivery begins from that decision and only for later observations. Enabling or resuming does not replay historic alerts; other people must enable their own subscriptions.
{% endstep %}
{% step %}
## Record a manual observation
On the indicator, enter a measured value, its observed UTC time and explanation.
Use the shared evidence picker to cite 1–4 exact retained passages. The evidence must fit the analysis handling.
Retain the observation, then evaluate explicitly or let the scheduled watch check it.
{% endstep %}
{% step %}
## Interpret the evaluation
No observation, stale observation and condition not met are distinct outcomes. Before-horizon, expired, paused and awaiting-approval conditions are also explicit.
The latest observation is chosen by observed time, so late reporting of an older observation does not displace newer evidence.
None of these outcomes revises confidence.
{% endstep %}
{% step %}
## Triage a delivered alert
In Pulse, open My indicator watches and the relevant watch.
The directory pages through currently accessible watches without hidden-item counts; Refresh checks access again.
Review the exact condition revision, observation explanation and linked evidence history.
Retain actionable, dismissed or false-positive feedback with a reason. Earlier feedback remains available if you later add another decision.
{% endstep %}
{% step %}
## Prepare deliberate follow-up
Optionally retain a follow-on requirement proposal.
Prepare this requirement proposal opens the normal Direction editor with reviewable text and source handling; access is checked again before submission.
Review permissions, question, ownership, scope and handling, then explicitly save.
Alternatively open the analysis for deliberate amendment and its normal review/assurance steps.
{% endstep %}
{% endstepper %}
## Check the result
* Does the indicator refer to the intended exact judgement revision?
* Is the measure observable and supported by accessible exact evidence?
* Has missing or stale data been distinguished from a negative observation?
* Have false-positive feedback and revised thresholds preserved their earlier history?
* Is any follow-up still subject to its ordinary approval and tasking authority?
## If you get stuck
Why did no alert arrive?
Check current access, condition approval, horizon and expiry, subscription state and interval, observation freshness, and whether the observation was recorded after the latest approval and subscription enablement.
Only a condition-met evaluation can create an alert. Historic or duplicate observations are not replayed as new alerts.
Why is my watch paused or unavailable?
A condition may have expired, been revised without approval or been paused; access may have changed or a retained history limit may have been reached.
Return to the indicator and review current authority and evidence. Do not treat missing alerts as evidence of safety.
Why did my watch page not change?
A temporary read failure leaves the last confirmed page visible so you can retry with Refresh, Previous or Next.
If current access is lost, the protected list clears; restore the required permissions and refresh before continuing.
Hidden watches never create an empty intermediate page or a visible count.
Why did saving fail?
Check required reasons, UTC times, numeric bounds, distinct evidence passages and handling.
Concurrent changes require reload and deliberate reconciliation; unsaved drafts remain protected.
An expired condition needs a new approved revision.
Does false-positive feedback remove the alert?
No. Feedback, its reason and any proposal are retained.
The alert no longer counts as awaiting feedback, but its evaluation and earlier decisions remain historical.
Threshold tuning creates a new condition revision.
Does preparing a requirement authorise collection?
No. It opens a reviewable draft, preserves the source handling and checks access again.
Normal Direction permissions, deliberate submission and subsequent collection/authority controls still apply.
## Responsibility and limits
* Every scheduled evaluation and inbox read rechecks current grants, workspace membership and cited evidence/source access. Lost access hides protected alerts, titles and counts and pauses delivery. Restoration requires deliberate resumption.
* Only in-app delivery is implemented. Stable evaluation keys and durable Mongo claims prevent duplicate logical alerts after restart, while retaining attempts. No email, SMS or automatic connector is enabled.
* Indicators retain up to 32 revisions, 64 watch decisions, 128 observations, 256 distinct evaluations and 4 MiB. Subscriptions retain up to 4,096 attempts, 128 alerts, 64 decisions, 256 feedback entries and 4 MiB. Capacity preserves history and requires a replacement watch.
## Reference and scope
The reviewed PHIA, intelligence-cycle and futures requirements inform this workflow. A measured threshold is an observable condition, not a predictive person or threat score and not an investigative authority.
## Related workflows
* [Develop and independently review an assessment](broken://pages/79db57a6c6245feef3fb58e4fe6d4a507b2a74a0)
* [Explain confidence, changed judgements and dissent](broken://pages/b5f23381c9372db8261edb88339424566cabc7f2)
* [Review reporting origins, independence and corrections](broken://pages/519a6dd9104e5b592aab17a02265642b8ecd221a)
* [Use Pulse and inspect retained activity](file:///7397340/assurance/pulse.md)
* [Define and manage an intelligence requirement](file:///7397340/direction/requirements.md)
* [Plan collection and record observed results](file:///7397340/collection/collection.md)
* [Plan analytical assurance and obtain separate sign-off](broken://pages/1ff25928b8eecd677eef29b9b4c97fb68c33a32a)