fastapi/fastapi · Evidence Report

Window2025-09-05 – 2026-08-26

Timezone–

Commits read759

Contributors86

Model calls0

What the evidence shows

8 declared threshold(s) crossed

Thresholds are declared data, not prose. Each finding names its metric and the value that crossed it. None asserts a cause.

critical

Rule: process.review_coverage threshold crossed · value 8.8%

warning

Rule: ai.test_ratio_delta threshold crossed · value -33.9%

warning

Rule: delivery.stale_pr_share threshold crossed · value 40.6%

warning

Rule: people.bus_factor threshold crossed · value 1

warning

Rule: process.direct_push_share threshold crossed · value 32.9%

warning

Rule: temporal.weekend_share threshold crossed · value 24.3%

info

Rule: process.stale_branch_share threshold crossed · value 85.5%

info

Rule: quality.duplication_rate threshold crossed · value 19.4%

Delivery and process

Median time from ready to shipped

12 min

over 91 of 200 pull requests · (lower is generally preferred)

MEASURED46% coverage

Explain

Share of merged changes with a recorded review

8.8%

8 reviewed / 91 merged · (higher is generally preferred)

MEASURED46% coverage

Explain

Slowest tenth of changes take at least

4.1 a contributors

over 91 of 200 pull requests · (lower is generally preferred)

MEASURED46% coverage

Explain

Median wait for the first human review

4.9 h

over 24 of 200 pull requests · (lower is generally preferred)

MEASURED12% coverage

Explain

Releases to the main branch per week

14.95

over 759 of 759 commits · (higher is generally preferred)

PROXY · MED

Explain

Share of changes that needed a fix or revert

0.0%

0 failures / 759 landings · (lower is generally preferred)

PROXY · LOW

Explain

Median time to repair a reverted change

UNAVAILABLE

Unavailable · not collected

Requires: No eligible records in this window.

UnavailableNot zero

What this needs

Changes currently waiting

over 15 of 200 pull requests · (lower is generally preferred)

MEASURED8% coverage

Explain

Volume and cadence

Commits

759

over 759 of 759 commits

MEASURED

Explain

Lines of code added

547,191

over 12,566 of 12,849 file changes

MEASURED98% coverage

Explain

Lines of code removed

166,668

over 12,566 of 12,849 file changes

MEASURED98% coverage

Explain

Net change in codebase size

380,523

over 759 of 759 commits

MEASURED

Explain

Days per week with any activity

3.47

over 759 of 759 commits

MEASURED

Explain

Median commit size

over 755 of 759 commits

MEASURED99% coverage

Explain

Lines changed per active day

4,056

over 759 of 759 commits

MEASURED

Explain

Share of work that is new capability

30.1%

143 feature / 475 classified

PROXY · MED63% coverage

Explain

Agent and cost

7 of 11 metrics in this section are inferred. Each is marked individually; the tier is a property of how the number was obtained, not a defect.

Commits carrying an AI-tool trace (floor)

≥ 0.4%

3 detected / 759 n · a detectable floor, not a share

INFERRED · MED

Explain

Lines arriving in AI-traced commits (floor)

≥ 0.1%

293 detected_lines / 547,191 total_lines · a detectable floor, not a share

INFERRED · LOW

Explain

Contributors with any AI-traced commit

≥ 3.5%

3 adopters / 86 n · a detectable floor, not a share

INFERRED · MED

Explain

Share of AI traces from the single most-used tool

≥ 33.3%

1 top_tool_commits / 3 total_traces · a detectable floor, not a share

INFERRED · MED

Explain

Rework gap between AI-traced and untraced work

≥ 2.2%

5.5 ai_rate / 3.3 human_rate · a detectable floor, not a share · (lower is generally preferred)

INFERRED · LOW

Explain

Test-ratio gap between AI-traced and untraced work

≥ 54.7%

80.9 ai_rate / 26.2 human_rate · a detectable floor, not a share · (higher is generally preferred)

INFERRED · LOW

Explain

Size gap between AI-traced and untraced commits

≥ -54.4%

18 ai_size / 39.5 human_size · a detectable floor, not a share

INFERRED · LOW99% coverage

Explain

AI tool configuration files present

over 1 of 1 repositories

MEASURED

Explain

Quality signals

8 of 12 metrics in this section are proxy. Each is marked individually; the tier is a property of how the number was obtained, not a defect.

Rework rate: new code rewritten within 21 days

3.3%

18,243 reworked_lines / 547,191 added_lines · (lower is generally preferred)

PROXY · MED

Explain

Share of new code that is tests

26.3%

143,762 test_lines / 547,191 total_lines · (higher is generally preferred)

PROXY · MED

Explain

Share of code changes that updated tests too

39.3%

123 coupled / 313 code_commits · (higher is generally preferred)

PROXY · MED41% coverage

Explain

Share of all changes landing in the busiest 10 files

8.9%

10 top_n / 63,470 top_changes · (lower is generally preferred)

MEASURED

Explain

Share of changes too large to review properly

11.5%

23 large / 200 n · (lower is generally preferred)

PROXY · MED

Explain

Share of functions long enough to be hard to follow

5.0%

251 long / 4,978 n · (lower is generally preferred)

PROXY · MED

Explain

Deepest nesting in the worst tenth of functions

over 4,978 of 4,978 functions measured · (lower is generally preferred)

PROXY · MED

Explain

Share of code that is copy-pasted

55.8%

54,519 duplicated_lines / 97,778 scanned_lines · (lower is generally preferred)

PROXY · MED

Explain

People

Active contributors

over 86 of 86 commit authors

MEASURED

Explain

Bus factor: people holding half the code

over 86 of 86 contributors · (higher is generally preferred)

MEASURED

Explain

Share of the code written by one person

43.0%

306,838 top_lines / 713,859 total_lines · (lower is generally preferred)

MEASURED

Explain

Share of work from people new to this codebase

49.0%

350,037 newcomer_lines / 713,859 total_lines

MEASURED

Explain

Share of files only one person has ever touched

38.6%

1,514 single_owner / 3,925 files · (lower is generally preferred)

MEASURED

Explain

Share of contributors still active in the last 90 days

10.5%

9 recent / 86 total · (higher is generally preferred)

MEASURED

Explain

Commits per month

Counts · observed intervals only

Share of merges with a review

Rate · gaps are never bridged

Commits with an AI trace (floor)

A detectable floor · only tools that left a trace are counted

Proportion and completeness

Numerator and denominator always adjacent

Merged changes carrying a review

Collection completeness

Evidence and verification

33 measured · 17 proxy · 7 inferred

Each state is carried three times over: in words, in a pattern or stroke, and in placement. Any single channel can fail without losing the meaning: print, colour vision, or a high-contrast mode.

Measured · solid

Counted directly from a source record.

Proxy · 45° hatch

An estimator standing in for the thing itself. Always carries a caveat.

Inferred · stipple

Derived from detectable traces only. Reads as a floor, never a share.

Partial · 135° hatch

Opposite orientation to proxy, so the two never read as one another.

Unavailable · blank

A stated reason and the capability required. Never a zero.

Suppressed · absent

No value, no plotted position, no tooltip, no cell in any download.

Derivation · process.review_coverage

Formula	8 merged pull requests with a review / 91 merged = 8.8%
Sample	n = 91 of population 200 pull requests · 45.5% coverage
Excluded	109 · pull requests not merged (open or closed unmerged)
Assumption	Only GitHub review records count. Approval given in a plain comment, over chat, or in person is invisible, so a team with a different review habit will score unfairly low — verify against the listed examples before treating a low number as a finding.
Strongest evidence	Rank 1: 1936.0 lines merged unreviewed.

Coverage and limitations

Repositories collected	6
Repositories failed	0
Lines excluded as generated	555,383
Pull request collection truncated	6 repository(-ies). Every pull-request metric inherits that ceiling.

Every metric

All metrics in this scope

Metric	Identifier	Value	Evidence kind	Confidence	Coverage	State
Median time from ready to shipped	delivery.lead_time_p50	12 min	measured	high	46%	Observed
Slowest tenth of changes take at least	delivery.lead_time_p90	4.1 a contributors	measured	high	46%	Observed
Median wait for the first human review	delivery.time_to_first_review_p50	4.9 h	measured	high	12%	Observed
Releases to the main branch per week	delivery.deploy_frequency_per_week	14.95	proxy	medium	100%	Observed
Share of changes that needed a fix or revert	delivery.change_failure_rate	0.0%	proxy	low	100%	Observed
Median time to repair a reverted change	delivery.mttr_hours	–	proxy	low	–	Unavailable
Changes currently waiting	delivery.wip_open_prs	15	measured	high	8%	Observed
Share of open changes stalled over 30 days	delivery.stale_pr_share	6.7%	measured	high	8%	Observed
Commits	velocity.commits_total	759	measured	high	100%	Observed
Lines of code added	velocity.lines_added	547,191	measured	high	98%	Observed
Lines of code removed	velocity.lines_removed	166,668	measured	high	98%	Observed
Net change in codebase size	velocity.net_lines	380,523	measured	high	100%	Observed
Days per week with any activity	velocity.active_days_per_week	3.47	measured	high	100%	Observed
Median commit size	velocity.commit_size_p50	39	measured	high	99%	Observed
Lines changed per active day	velocity.lines_per_active_day	4,056	measured	high	100%	Observed
Share of work that is new capability	velocity.feature_work_share	30.1%	proxy	medium	63%	Observed
Share of commits that are trivial	velocity.trivial_commit_share	29.1%	proxy	medium	100%	Observed
Rework rate: new code rewritten within 21 days	quality.rework_rate	3.3%	proxy	medium	100%	Observed
Share of new code that is tests	quality.test_line_ratio	26.3%	proxy	medium	100%	Observed
Share of code changes that updated tests too	quality.test_coupling_rate	39.3%	proxy	medium	41%	Observed
Share of all changes landing in the busiest 10 files	quality.hotspot_concentration	8.9%	measured	high	100%	Observed
Share of changes too large to review properly	quality.large_change_share	11.5%	proxy	medium	100%	Observed
Share of functions long enough to be hard to follow	quality.long_function_share	5.0%	proxy	medium	100%	Observed
Deepest nesting in the worst tenth of functions	quality.max_nesting_p90	4	proxy	medium	100%	Observed
Share of code that is copy-pasted	quality.duplication_rate	55.8%	proxy	medium	100%	Observed
Share of lines that are comments	quality.comment_density	8.4%	proxy	medium	100%	Observed
Commits carrying an AI-tool trace (floor)	ai.detected_commit_share	0.4%	inferred	medium	100%	Observed
Lines arriving in AI-traced commits (floor)	ai.detected_line_share	0.1%	inferred	low	100%	Observed
Contributors with any AI-traced commit	ai.adopter_share	3.5%	inferred	medium	100%	Observed
Share of AI traces from the single most-used tool	ai.tool_concentration	33.3%	inferred	medium	100%	Observed
Rework gap between AI-traced and untraced work	ai.rework_delta	2.2%	inferred	low	100%	Observed
Test-ratio gap between AI-traced and untraced work	ai.test_ratio_delta	54.7%	inferred	low	100%	Observed
Size gap between AI-traced and untraced commits	ai.commit_size_delta	-54.4%	inferred	low	99%	Observed
AI tool configuration files present	ai.config_artifact_count	0	measured	high	100%	Observed
Share of AI suggestions actually accepted	ai.native_acceptance_rate	–	measured	high	–	Unavailable
Lines accepted from AI suggestions (measured)	ai.native_lines_accepted	–	measured	high	–	Unavailable
How far the commit-trace floor understates measured AI use	ai.floor_understatement	–	measured	medium	–	Unavailable
Active contributors	people.active_contributors	86	measured	high	100%	Observed
Bus factor: people holding half the code	people.bus_factor	2	measured	high	100%	Observed
Share of the code written by one person	people.top_contributor_share	43.0%	measured	high	100%	Observed
Share of work from people new to this codebase	people.newcomer_share	49.0%	measured	high	100%	Observed
Share of files only one person has ever touched	people.knowledge_silo_share	38.6%	measured	high	100%	Observed
Share of contributors still active in the last 90 days	people.contributor_retention	10.5%	measured	high	100%	Observed
Share of merged changes with a recorded review	process.review_coverage	8.8%	measured	high	46%	Observed
Share of changes merged by their own author	process.self_merge_share	57.1%	measured	high	46%	Observed
Median change size	process.pr_size_p50	29	measured	high	98%	Observed
Share of reviews done by one person	process.reviewer_load_top_share	32.5%	measured	high	100%	Observed
Share of work that bypassed pull requests	process.direct_push_share	92.4%	proxy	medium	100%	Observed
Share of branches abandoned over 90 days	process.stale_branch_share	72.7%	measured	high	100%	Observed
Branches	process.branch_count	22	measured	high	100%	Observed
Share of work saved outside working hours	temporal.after_hours_share	36.0%	proxy	medium	100%	Observed
Share of work saved at weekends	temporal.weekend_share	22.0%	proxy	medium	100%	Observed
Busiest hour of the day	temporal.peak_hour	10:00–11:00 local	measured	high	100%	Observed
How bursty the work pattern is	temporal.burstiness	2.68x	proxy	low	100%	Observed
Share of automated checks that failed	ci.failure_rate	30.7%	measured	high	96%	Observed
Median time for automated checks to finish	ci.duration_p50_minutes	0.6 min	measured	high	100%	Observed
Automated check runs per week	ci.runs_per_week	3.78	measured	high	96%	Observed

Provenance. Computed from fastapi/fastapi, window [2025-09-05, 2026-08-26). 0 language-model calls. 555,383 lines excluded as generated before any metric was computed.

Redaction 1 · privacy policy 1.1.0.

No contributor is named, ranked or ordered anywhere in this document. That capability does not exist in the system that produced it.