Falessi 2010 (the 2006 experiment); van Heesch 2013
50 students, effectiveness under changed requirements; four student teams over seven weeks
How a proposal is structured
PEP 12 (Goodger 2002), Rust RFC template (Rust 2026b), KEP template (Kubernetes 2026), Google design doc, Swift proposal template (Swift 2026)
none found
descriptive comparison of templates only
How a proposal is judged
Rust final comment period (Rust 2026a), Steering Council (Warsaw 2000), SIG approvers (KEP template), Go weekly review group (Go 2026b), Swift review manager and workgroup (Swift 2026)
Kondo 2025 (Go)
1,095 classic proposals, coded decline reasons
Whether people adopt ADRs
Nygard's template in open source
Buchgeher 2023
repository mining
Why rationale is not captured
cost, effort
Lee 1997; Tang 2006; Ahmeti 2024; Tofan 2014 asks for the right amount
argument, survey of 81, one action study
How much research exists on documenting decisions
n/a (a research record)
Tofan 2014: 83% of 144 papers propose documentation; 4% use experiments
systematic mapping, 2002-2012
Which rationales drive decisions
decision drivers in MADR 2024
Weinreich 2015; Borowa 2023
25 interviews; 63 answers and 13 interviews
Which documentation practice practitioners write about
ADR templates: 12 of 19 grey sources (Harhio 2022)
none found
review of grey sources, 2016 onward
Adoption counts, descriptive, taken 2026-10-28
Count
Value
Caveat
merged Rust RFC files
652
numbers reach 3999 because they follow pull request numbers
PEP files
746
highest is 8107
KEP folders with a kep.yaml
643
Go design documents in the proposal repository
91
proposals themselves are issues; Kondo 2025 counted 1,095 classic ones
RFC Editor index
numbers to 9999
a script count gave 8,995 numbered entries, 122 marked Not Issued; the gap was not investigated
repositories tagged adr
875
self-tagged, includes tools
repositories tagged architecture-decision-records
488
overlap with the line above unknown
stars: adr-tools, MADR, the ADR collection repository
5,747; 2,595; 17,342
interest, not use
The counts come from public repository listings and the GitHub API; the API search totals are approximate.
A count of records says nothing about whether they were read (GitHub 2026).
What the templates share
Part
Nygard
MADR
PEP
Rust RFC
KEP
Google doc
HashiCorp (example only)
status
yes
optional
header
no
yes
no
yes
context or motivation
yes
yes
yes
yes
yes
yes
yes
goals and non-goals
no
drivers
no
no
yes
yes
no
decision or proposal
yes
yes
specification
reference-level
design details
design
proposal
alternatives
no
options, pros and cons
rejected ideas
rationale and alternatives
alternatives
alternatives
abandoned ideas
consequences or drawbacks
yes
optional
backwards compatibility
drawbacks
risks, drawbacks
cross-cutting
no
open questions
no
no
open issues
unresolved questions
no
no
no
"no" means the section is not among those the template or guide lists. The table is read from the templates (Nygard 2011; MADR; PEP 12; Rust RFC template; KEP template; Ubl 2020; HashiCorp 2026, example only).
Among these seven, only the Rust RFC template asks for prior art and only the KEP template asks for a production readiness review.
Nygard 2011 has no alternatives section; MADR added one, which participants in Nogueira 2026 saw as better for rationale depth and slower to fill in.
ADR guidance treats an accepted record as immutable: write a new record that supersedes the old (GDS 2026; Microsoft 2026 ADR; AWS 2026c, example only).
PEPs carry a change history (PEP 12) and published RFDs may be updated (Oxide RFD 1, Frazelle 2026).
Upper bounds (achieved)
Rationale documentation raised decision effectiveness under changed requirements in a lab (Falessi 2010).
Decision viewpoints made student teams more systematic in weighing options, in one multiple-case study (van Heesch 2013).
ADR template comparison has a first controlled experiment (Nogueira 2026).
Research volume on documenting decisions is mapped (Tofan 2014) and the grey side has a first review (Harhio 2022).
Decline reasons for design proposals have a coded taxonomy (Kondo 2025).
ADR use has been counted in open source (Buchgeher 2023; Santana 2026).
Lower bounds (not yet shown)
This search did not find a study where reviewers judge a design proposal in two document forms.
This search did not find a study that ties a proposal template to the quality of the decision made.
No evidence was found that Nygard's or MADR's shape helps a maintainer recover why, as opposed to helping the writer write.
No study of the effect of the "rejected alternatives" section on repeated debate was found.
Bratthall 2000 and Karsenty 1996 may hold relevant results; their text was not read.
Q/A sites (searched 2026-10-28) hold no report of how a proposal is judged (Q4) or whether a team keeps ADRs (Q5).