Estimation
Choose the project's story point scale, size an item by hand or take Heimdall's suggestion against the items the project already sized, and see where the points go: the sprint's capacity, its planner, the burndown and the velocity.
By the end of this page you will have chosen the scale the project sizes in, put story points on an item by hand, asked Heimdall for a size and taken or dropped it, and know what reads the points afterwards. The points live on the item, in the editor the backlog opens; this page is about the points themselves and the help you get with them.
Before you begin
| To | You need |
|---|---|
| Choose the scale | project:update; the project manager by default |
| Size an item | work-item:update; the project manager, the analysts, the developers and the testers by default |
| Ask Heimdall for a size | work-item:update and ai:use, on a project whose assists are on; never a client guest |
A project sizes in story points or in hours. This page is about points; a project that sizes in hours keeps the Estimate field in minutes beside them, and Heimdall's suggestion is always a step of the point scale.
Choose the scale
On the project's settings page, under the general settings, two choices shape every estimate: the Estimation Unit, story points or hours, and for points the Story Point Scale. Three scales are offered:
| Scale | Steps | When it fits |
|---|---|---|
| Fibonacci | 1, 2, 3, 5, 8, 13, 21 | The default. The gaps grow with the uncertainty, so a big item is not argued down to the point. |
| Linear | 1 to 10 | Teams that want a finer grain on small items and rarely size anything large. |
| T-shirt sizes | XS, S, M, L, XL | Early sizing, before the team has a feel for points; the sums treat these as unsized. |
The scale is read wherever points are suggested: Heimdall answers on it and snaps a stray number to its nearest step. By hand you may type any value of up to ten characters, so agree the scale with the team and keep to it. Changing the scale later does not rewrite the points already on items; it changes what Heimdall answers from then on.
Size an item by hand
Open an item from the backlog (Edit in its row menu, or the pencil in its flyout) and find
Story Points in the Size and Dates fieldset. Type the step the team agreed, such as 5 or
M, and save. The field is optional: an item without points is listed as unsized, counts as nothing
in the sprint's totals, and is still taken into a sprint when you plan it. Beside the points sit the
Estimate in minutes, for projects that size in hours or want both, and Remaining, what is
left in minutes, which the burndown reads.
Points are relative size, not time: how much work, how much uncertainty and how much risk the item carries compared with the items the team already sized. Size at planning time, once the acceptance criteria are written and the Definition of Ready is close; an item sized before it is understood is sized twice.
Ask Heimdall for a size
On a saved item, Suggest with Heimdall sits under the Story Points field (a new item has no button: Heimdall needs the saved record and its criteria). Press it and a panel opens under the row, headed with Heimdall's size and how sure it is, such as 5 points, 82% sure:
- The rationale: two or three sentences that name the sized items Heimdall compared with by their reference and say what makes this item bigger, smaller or the same.
- The anchors: the sized items Heimdall was shown, each with its reference, title and points; the ones Heimdall leaned on carry their points in violet, the rest in grey.
When Heimdall cannot put the item on the scale, the panel is headed No size on this scale and the rationale says what is missing. That is the assist telling you the item is too vague to size, not a failure: write the criteria, then ask again.
What Heimdall is given
The assist has no tools and no memory of earlier turns. It is handed the item as it stands: its kind, reference and title, its story sentence, its description, the steps to reproduce or the outcome sought where the kind has them, any estimate in hours already given, and up to eight of its acceptance criteria. It is handed the project's scale with its steps. And it is handed up to five items the project already sized that resemble this one most, each with its points, its status and how alike it is. Likeness is the share of words the titles and descriptions have in common, with the filler words left out; when fewer than five items resemble it, the list is topped up with the most recently changed sized items, so there is always something to anchor on.
How it answers
Heimdall is told to estimate relative size, to anchor on the nearest items (an item like one sized
at 5 is a 5 unless it is clearly bigger or smaller, and it says which way and why), to answer with
exactly one step as the project writes it, to be more sure when a close twin exists and less sure
when the item is vague or resembles nothing sized, and to name in the rationale the items it
compared with. Its answer is checked before you see it: a number off the scale is snapped to the
nearest step (a 6 on Fibonacci becomes a 5), a size the scale has no step for (an XL on
Fibonacci) becomes no size on this scale, the confidence is held between nought and one, and the
anchors it names are marked as used. The assist runs on the workspace's simple model role and
gives up after a minute rather than keep you waiting.
Take or drop it
Use 5 writes the step into the Story Points field and closes the panel; nothing is saved until you save the item, so you can still change the value or cancel the editor. Discard closes the panel and leaves the field as it was. The panel belongs to the item it was asked for: open another item and it is gone. Every suggestion is a metered call on the team's AI budget; ask once per item, and ask again after the item changed.
A suggestion, not a verdict
The team's number wins. Heimdall sees the words of the record, not the code behind it; when the developers know the ledger is a mess, the 8 they give is the right one, and the suggestion is the prompt for that conversation.
All the fields
What happens next
- The points sit on the item and travel with it: the backlog shows them in the row, the sprint's flyout adds them up under its capacity, the plan dialog and Heimdall's sprint planner count them when a scope is picked, the burndown follows what is left, and the velocity a sprint closes with is the points of the items it finished.
- A change of points writes a version of the item, by whoever made it; a size taken from Heimdall is saved by you, so the version carries your name.
- Over the API and the MCP server the points are the item's
story_points, and a suggestion is a call of its own that writes nothing.
Heimdall on this page
Suggest with Heimdall under the points is this page's assist. The rest of Heimdall's help on an item (drafting, improving and making the description and the criteria testable) is on the inline assists page, and the panel reads and changes items when you ask.
From the API and MCP
Work items API
Read and write an item's points as story_points, and ask for a suggestion under the item's estimate suggestions endpoint.
suggest_estimate
An agent asks Heimdall for a size, reads the reasons back to the person, and writes the points they settle on through update_work_item.
update_work_item
Write the points an agent and the person agreed on, through the item's own field.
work_item.* events
A change of points goes out with the item's other changes as work_item.updated.
Permissions
| Action | Permission |
|---|---|
| See an item's points | work-item:view-any |
| Size an item | work-item:update |
| Ask Heimdall for a size | work-item:update and ai:use; assists on; never a client guest |
| Choose the scale | project:update |
Frequently asked
Why did Heimdall answer "No size on this scale"? It answered a size the project's scale has no step for, which it does when the item is too vague to place: no criteria, a one-line description, nothing sized that resembles it. Read the rationale, write what is missing and ask again.
Can it size in hours? No. The suggestion is always a step of the point scale, because that is what the sprint's capacity and the velocity count. The Estimate in minutes stays yours to type.
Does it change the item? Never. Only Use followed by a save writes the points, and the version written carries your name. An agent that asks through the MCP server gets the suggestion as data and writes the points through the item's own update tool on the person's word.
The anchors look odd. Why these items? Likeness is the share of words the titles and
descriptions have in common, so an item titled with what it does finds its siblings and an item
titled Fix finds nothing. When nothing resembles the item, the most recently sized items are
shown instead and Heimdall's confidence drops; the words of the record are the fix.
Who sees the button? Staff who may change items on the project and use the assists, on a project whose assists are on. A client guest never sees it, and a project whose assists are off hides it for everyone.
Does Heimdall remember its last answer? No. Each ask is a fresh, metered call, which is why the panel stays open until you take or drop it; the AI ledger keeps the record of the call.
Related
- Backlog: the editor the points live in, and the rest of its fields.
- Sprints: the capacity the points fill, and Plan with Heimdall.
- Sprint planner: how Heimdall proposes a scope from the points.
- Project settings: the estimation unit and the scale.
- Usage and budgets: where every suggestion is filed.
How is this guide?
Last updated on
Sprints
Plan a sprint with its dates, goal and capacity, pull its scope from the backlog, put it to the team for the commitment that starts it, follow it while it runs, close it with its velocity and carry the unfinished work on.
Definition of Ready and Done
The two checklists every work item carries: the templates the project writes once, the lists each item ticks on its own, and the gates that hold an item out of a sprint or out of Done until the required entries are ticked.