docs: add release note for Signals ML feature attributes

[Matus Tomlein (matus-tomlein)]

What changed?

Adds a release note at release-notes/new-in-signals-attributes-for-ml-features/ announcing the three attribute-definition additions that came out of AISP-1451:

• time_since_last / time_since_first aggregations (AISP-1446) — recency and tenure, in s/min/h/d, computed at read time against derived_tstamp.
• date_part modifier on timestamp properties (AISP-1448) — extract (hour_of_day, day_of_week, month_of_year) and truncate (active_day, active_week, active_month) families, applied before aggregation and also usable in criteria.
• Optional event filter (AISP-1067) — an empty events list means the attribute is calculated from every event type.

The note is framed as a follow-on to New in Signals: ML training datasets: the dataset builder made attribute definitions the shared contract between training and serving, and these three additions widen what that contract can express.

Why?

The reference documentation for all three features shipped in #1909, but no release note went with it. Per release-notes/_README.md, docs that catch up on functionality that shipped quietly still warrant a note, and this is the follow-up PR for that docs change.

Reviewer guidance

The feature details were taken from the implementation in the Signals monorepo (the aggregation validators, the DatePart literal and its date-time property gating, and the relaxed events constraint) and cross-checked against the reference docs added in #1909, so the two should agree.

Two things worth a second opinion:

• SDK version. The note says snowplow-signals version 0.4.8 or later. 0.4.8 is the first released version whose Attribute model carries time_since_last/time_since_first and time_unit, has date_part on the property models, and drops min_length on events — 0.4.7 has date_part only. Verified by inspecting both wheels from PyPI.
• Optional-filter caveat. The last paragraph of that section notes that an empty event list is accepted as match-all rather than rejected, which is a deliberate behavior change from the previous 422. Worth confirming that's the right level of emphasis for a product-news audience.

Checks run locally: prettier --check passes, and a full docusaurus build succeeds with all three internal links resolving. Vale isn't installed locally, so it hasn't been run.

AI reviews

Claude will automatically review this PR against the docs style guide.

If you have questions or want it to look again at something specific, tag @claude in a comment.

🤖 Generated with Claude Code

[github-actions]

Missing SEO metadata

The following markdown files are missing required metadata fields:

• release-notes/new-in-signals-attributes-for-ml-features/index.md: missing fields: keywords, sidebar_label

Required fields

The file metadata is important for SEO and marketing. All markdown files, except for those with filenames starting with _, should include:

• title: Full, descriptive page title
• sidebar_label: Short title for navigation sidebar (can be the same as the main title)
• description: One to two sentences summarizing the page contents
• keywords: Array of marketing/SEO keywords

Please add the missing metadata.

[greptile-apps]

Greptile Summary

Adds a Signals release note describing time-since aggregations, timestamp date-part modifiers, and optional event filters, with links to the attribute and ML training-dataset references.

• Explains recency and tenure calculations and their read-time behavior.
• Introduces extract and truncate date-part families with composition examples.
• Documents match-all behavior for omitted or empty event filters.
• Provides Console and Python SDK availability guidance.

Confidence Score: 4/5

The batch-group availability claim should be corrected before merging because that group type synchronizes pre-calculated values rather than defining these attributes.

The feature details otherwise align with the reference documentation, but the release note currently directs readers toward an unsupported batch attribute-group configuration.

Files Needing Attention: release-notes/new-in-signals-attributes-for-ml-features/index.md

Important Files Changed

Filename	Overview
release-notes/new-in-signals-attributes-for-ml-features/index.md	The feature descriptions agree with the current attribute reference, but the final availability statement incorrectly extends attribute-definition features to batch attribute groups.

_{Reviews (1): Last reviewed commit: "docs: add release note for Signals ML fe..." | Re-trigger Greptile}

[greptile-apps]

Batch availability is incorrect

When readers apply these attribute-definition features to a warehouse-backed batch group, the group only synchronizes pre-calculated values and performs no attribute calculation, causing them to pursue an unsupported configuration.

Suggested change

	All three are available in Console and in the Python SDK, for both stream and batch attribute groups. Upgrade to `snowplow-signals` version 0.4.8 or later to define them from the SDK.
	All three are available in Console and in the Python SDK for stream attribute groups. Upgrade to `snowplow-signals` version 0.4.8 or later to define them from the SDK.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[claude]

Docs style review

Reviewed against CLAUDE.md and src/pages/style-guide/llm/index.md. Scope is writing quality, style, terminology, structure, frontmatter, and links only.

What is already correct: frontmatter matches release-notes/_README.md (title, description, date, category, components — all from the allowed value lists); four H2s, all sentence case and imperative; no H1; every heading and list preceded by prose; Signals, Console (no article), Profiles Store, and Python SDK all styled correctly; version 0.4.8 rather than v0.4.8; Oxford commas; straight quotes. All three internal links (/docs/signals/attributes/attributes/, /docs/signals/ml-training-datasets/, /release-notes/new-in-signals-ml-training-datasets/) resolve to existing files, and the trailing-slash form matches the convention used across release-notes/ (per _README.md, which asks for paths like /docs/signals/). No external URLs were added or modified.

Four things worth changing.

1. Passive voice (style guide, Writing style: "Use active voice")

Several of the load-bearing explanatory sentences are passive:

"The value is computed when you read the attribute, not when the event arrives"

"Out-of-order events can't push the result below zero, as it's clamped at 0."

"the attribute is calculated from every event Signals processes"

"A request that sends an empty event list is read as matching all events rather than rejected"

Suggested fixes, naming the actor (Signals) in each case:

• "Signals computes the value when you read the attribute, not when the event arrives"
• "Signals clamps the result at zero, so out-of-order events can't push it negative."
• "Signals calculates the attribute from every event it processes"
• "Signals reads a request with an empty event list as matching all events rather than rejecting it"

2. "visitor" instead of "the user" (style guide, Writing style: use "the user" for the reader's end users)

"A model scoring a visitor mid-session sees how long that visitor has actually been idle."

The rest of the note correctly says "a user" ("a user's page views", "how many separate days a user has been active"), and "visitor" appears nowhere else in docs/signals/. Suggest: "A model scoring a user mid-session sees how long that user has actually been idle."

3. Trailing periods on single-sentence list items (style guide, Lists: "No period at the end unless the item contains multiple sentences")

"time_since_last measures the duration since the most recent matching event, which is the recency signal behind re-engagement, churn, and session-freshness features."

Both bullets in that list are single sentences, so drop the final period from each.

4. Conversational asides in place of a technical tone (CLAUDE.md: "Use professional, technical tone")

Two spots read as opinion rather than description:

"A raw timestamp is close to useless as a model feature, because every value is distinct."

"it gets in the way of the whole-pipeline features that models often want, where enumerating every schema is both tedious and wrong the moment someone adds a new event."

Suggested rewrites:

• "A raw timestamp works poorly as a model feature, because every value is distinct."
• "...where enumerating every schema is laborious and goes out of date as soon as someone adds a new event."

Minor

• Tense. "Signals recently gained a training dataset builder" and "until now it wasn't expressible" sit against the style guide's "Stay in the present tense. Never use 'currently', 'now'". Release notes across this directory do use "now" for what changed, so treat that as a judgment call, but "recently" adds nothing and could be dropped: "Signals gained a training dataset builder, which turns...".
• Numeral consistency. "push the result below zero, as it's clamped at 0" mixes both forms in one sentence; pick one (the style guide's spell-out rule points at "zero").

[Copilot]

Pull request overview

Adds a new Product news release note announcing three recently shipped Signals attribute-definition enhancements that make common ML features easier to express (recency/tenure aggregations, date_part modifiers for timestamp properties, and an optional event filter).

Changes:

• Introduces a new release note page for the Signals attribute-definition updates relevant to ML feature engineering.
• Summarizes time_since_last / time_since_first, date_part, and match-all semantics for events=[] (and related SDK guidance).
• Provides links to the related prior release note and reference documentation.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	documentation	34ea665	Commit Preview URL Branch Preview URL	Aug 27 2026, 02:20 PM

[jborlase-snowplow]

Left a couple comments on naming

[jborlase-snowplow]

Rather than whole-pipeline, should we call it 'All event features'?

[jborlase-snowplow]

What does this last bit mean after "Worth knowing..."?

[jborlase-snowplow]

Little harsh?

[jborlase-snowplow]

Lets just call this "Recency, tenure, and time-of-day attributes"