A task can have the right owner, valid JSON and all required fields—and still invent a commitment. Work through six excerpts from public Rust compiler-team minutes, then take the sources, policy and reviewable project to your own workflow.
Not a benchmark or a current task list. Sources are real historical documents; responses, policy choices and judgments are agent-authored with shared context. The first-pass examples deliberately expose plausible mistakes; they are not outputs from independent model runs. Revised acceptance is self-review, not proof of model improvement, human benefit or Rust team endorsement.
Use this policy as a review aid; do not automatically assign the extracted tasks. A promise to try requires effort-only wording. A start date belongs in the task, not the deadline field. General requests and useful suggestions remain outside a commitments-only register unless accepted. The final register contains four historical action rows across two excerpts; four other excerpts correctly remain empty under the stated policy.
These are agent task choices, not a universal interpretation of meeting language. Some teams exclude tentative commitments entirely or track requests separately. Agree on that distinction first. A null owner means no explicitly named owner was established; it is not permission to assign the speaker. Dates here describe 2019 intent, not work still outstanding today.
Both authored sets pass structure on all six cases. The first pass passes expected fields on two cases but only one is self-accepted. R01 is the important gap: its owner, ID and deadline all pass while its task exaggerates the promise. The revised set passes expected fields and is self-accepted on all six. There are six paired cases, but these counts describe an editorial exercise—not accuracy estimates or causal prompt comparisons.
The revision supplements the existing commitment policy rather than silently changing the original fictional pack. Exact refinements:
Historical-source scope: evaluate what is explicitly open in the excerpt at the time of that meeting, not what remains open today. Do not use a linked discussion or today's date to fill gaps. A commitment to try promises effort, not successful completion: preserve that qualification. Explicit commitments by several individually named people are separate actions and may share the same source_id. A collective plan explicitly adopted in the supplied context can be an action, but a collective pronoun does not establish a named owner. Preserve a stated start date in the task; it is not a due date unless the note also states a deadline. A general request to read, a wish for a guide or a suggestion to collect a list is not an accepted assignment. Source metadata and license notices are not meeting notes.
R05 also needs the preceding settled-arrangement paragraph to resolve “this idea”. It is supplied verbatim in both variants; the revision does not get extra source evidence. No linked discussion was used to fill a gap.
The meeting commenced with a call for involvement in the [RLS 2.0 working group][rls2] led by
[@matklad][matklad] as sharing expertise between the RLS 2.0 effort and rustc would be helpful.
There is documentation (videos and guides), issues with `E-mentor` and other resources available
for those interested. [@davidtwco][davidtwco], [@pnkfelix][pnkfelix] and [@Xanewok][Xanewok]
mentioned that they would endeavour to find some time to dig into RLS 2.0.
Authored first pass
{
"actions": [
{
"source_id": "N1",
"task": "Complete an investigation of RLS 2.0.",
"owner": "@davidtwco",
"due": null
},
{
"source_id": "N1",
"task": "Complete an investigation of RLS 2.0.",
"owner": "@pnkfelix",
"due": null
},
{
"source_id": "N1",
"task": "Complete an investigation of RLS 2.0.",
"owner": "@Xanewok",
"due": null
}
]
}
Self-review: reject. Reject despite passing structure and expected fields: 'complete an investigation' strengthens a tentative commitment to try to find time into a promised deliverable. Under this exercise's chosen policy the three named people each committed to effort, so separate rows may share N1. Another organization might exclude soft commitments; agree on that policy before using this task.
Editorial revision
{
"actions": [
{
"source_id": "N1",
"task": "Endeavour to find some time to dig into RLS 2.0.",
"owner": "@davidtwco",
"due": null
},
{
"source_id": "N1",
"task": "Endeavour to find some time to dig into RLS 2.0.",
"owner": "@pnkfelix",
"due": null
},
{
"source_id": "N1",
"task": "Endeavour to find some time to dig into RLS 2.0.",
"owner": "@Xanewok",
"due": null
}
]
}
Self-review: accept. Accept under the explicit effort-only policy: each named volunteer retains the qualification 'endeavour to find some time', with no invented deliverable or deadline. This is not evidence any investigation happened. The general invitation and the RLS 2.0 leader are not additional assigned actions.
Both [@nikomatsakis][nikomatsakis] and [@mw][mw] wanted there to be a simple guide to follow for
creating out-of-tree crates - this should include links to examples and things to copy-and-paste.
Authored first pass
{
"actions": [
{
"source_id": "N1",
"task": "Create a simple out-of-tree crates guide with examples and copy-and-paste material.",
"owner": "@nikomatsakis",
"due": null
}
]
}
Self-review: reject. Reject: wanting a guide specifies a desired outcome, not an accepted task or named author. Assigning it to one speaker invents responsibility. The nonempty array also fails the expected empty-register check.
Editorial revision
{
"actions": []
}
Self-review: accept. Accept the empty register: two people want a guide, but the excerpt supplies no acceptance, assigned author or adopted action item. Empty does not mean the idea lacks value.
R03 · Suggested inventory is not an accepted assignment
[@davidtwco][davidtwco] suggested that we would need to collate a list of things that would need to
be added to a repository (licenses, code of conduct, etc.) and things that would need set up (bors,
highfive, etc).
Authored first pass
{
"actions": [
{
"source_id": "N1",
"task": "Collate repository requirements and setup steps for licenses, code of conduct, bors and highfive.",
"owner": "@davidtwco",
"due": null
}
]
}
Self-review: reject. Reject: the sentence reports a suggestion that 'we would need' an inventory. It does not say the speaker accepted responsibility. Task plausibility is not evidence of a commitment.
Editorial revision
{
"actions": []
}
Self-review: accept. Accept: no agreed assignment appears in the excerpt. Do not follow linked discussions to invent a later acceptance; that would be a different source set.
- [davidtwco](https://rust-lang.zulipchat.com/#narrow/stream/131828-t-compiler/topic/steering.20meeting.202019.2E03.2E22.20.2358850/near/161450922)
has, [after numerous attempts](https://rust-lang.zulipchat.com/#narrow/stream/131828-t-compiler/topic/steering.20meeting.202019.2E03.2E22.20.2358850/near/161451008), managed to produce a have a working branch
that unblocks RFC 2008.
Authored first pass
{
"actions": []
}
Self-review: accept. Accept: the working branch is an accomplished result in this excerpt, not a new promise to finish or merge RFC 2008. No follow-up is supplied.
Editorial revision
{
"actions": []
}
Self-review: accept. Accept the unchanged empty register. 'Unblocks RFC 2008' does not establish that the RFC was merged, nor an assignment to merge it. Empty cases provide a needed control against always inventing follow-ups.
R05 · A collective rollout plan has a start date, not a due date
While the basic idea of the proposal was popular, we wound up changing
some of the details for how it would run. We settled ultimately on
**reusing the slot for the existing steering meeting** (Friday
mornings, Boston time) but making it **weekly**.
Read the exact meeting excerpt
[We plan to flesh out some more details and put this idea into action
starting May 3rd (the next steering meeting).](https://rust-lang.zulipchat.com/#narrow/stream/131828-t-compiler/topic/steering.20meeting.202019.2E04.2E12.20.2358850/near/163198505)
Self-review: reject. Reject: 'starting May 3rd' is a planned start, not a deadline to finalize the whole design. The context says the basic arrangement was settled, supporting a collective adopted plan, but no named individual/team is explicitly responsible; owner remains null. Expected due:null also fails.
Editorial revision
{
"actions": [
{
"source_id": "N1",
"task": "Flesh out more details and put the regular design-meeting idea into action starting May 3rd (the next steering meeting).",
"owner": null,
"due": null
}
]
}
Self-review: accept. Accept under the supplied context: preserve the planned start in task, keep due null and do not resolve the collective 'we' to a named owner. The phrase 'regular design-meeting' resolves 'this idea' using the explicitly included context paragraph, not external discussion. This is historical intent, not a claim that rollout occurred.
R06 · An invitation to read is not confirmed acceptance
- [nikomatsakis](https://rust-lang.zulipchat.com/#narrow/stream/131828-t-compiler/topic/steering.20meeting.202019.2E04.2E12.20.2358850/near/163192566)
has opened a [PR with a draft
RFC](https://github.com/rust-lang/compiler-team/pull/52) on the
compiler-team repository. This RFC describes adding a new level
("compiler team contributors") to recognize people who have been
contributing regularly. It also tries to document a few things about
members etc. Please give it a read!
Self-review: reject. Reject: 'Please give it a read!' is an invitation, not an adopted unassigned Action or confirmed acceptance. The opened PR is already reported complete. Assigning no owner does not cure an invented commitment.
Editorial revision
{
"actions": []
}
Self-review: accept. Accept: neither the completed PR opening nor the unaccepted invitation creates an open assigned action in this excerpt. A workflow that tracks requests as well as commitments would need a separate category, not silently weaken this policy.
Reproduce, inspect, adapt
Read the source excerpts and the policy disagreements above. Download the provenance manifest with exact quotations, context, SHA-256 hashes and pinned source URLs.
Download the unrun project or preview it in the Test Bench. It has no responses or judgments. Both variants see exactly the same source input. The B policy is intentionally informed by this exercise: use unseen notes for any later evaluation.
Run prompts with your own model or chat tool. PromptKB itself makes no model call. Expected records and review notes are not in the rendered prompts; source attribution and the license notice are.
Projects use format 5 for structure contracts. The browser stores nothing unless you enable Remember; starter loading replaces only the current tab after confirmation. Download current work first. Attribution travels in each case input because the current project format has no separate source metadata fields; that increases prompt length but keeps standalone files attributed. Do not strip the included license when redistributing excerpts.
Sources and permission
Excerpts from Rust compiler-team steering-meeting minutes, by the Rust compiler-team contributors. Repository MIT permission notice is included unchanged. No affiliation, endorsement or current assignment is implied. The upstream README reports that its website was retired in April 2025; these are historical repository documents, not current team policy.
Each quote is an exact contiguous Markdown excerpt from a pinned source file. N1 labels, source metadata, task policy, expected fields, responses and review notes are PromptKB agent additions. Evaluate the excerpt at its stated historical meeting time, not as a claim about work still outstanding today. Linked discussions are outside this exercise and were not used to fill gaps. The six selections are not the complete meetings, a representative dataset or independently reviewed ground truth.
PromptKB's added prompts, expected fields, authored responses and review commentary may be reused and adapted, including commercially, without attribution, as-is without warranties. That grant does not replace the upstream MIT conditions on source excerpts. Source hashes identify the downloaded bytes; they do not authenticate the truth or completeness of the minutes.
Read the upstream MIT permission notice
Permission is hereby granted, free of charge, to any
person obtaining a copy of this software and associated
documentation files (the "Software"), to deal in the
Software without restriction, including without
limitation the rights to use, copy, modify, merge,
publish, distribute, sublicense, and/or sell copies of
the Software, and to permit persons to whom the Software
is furnished to do so, subject to the following
conditions:
The above copyright notice and this permission notice
shall be included in all copies or substantial portions
of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF
ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED
TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A
PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT
SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR
IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
DEALINGS IN THE SOFTWARE.