Research methods · Release 2026-07-28.3

Manual validation, without pretending unfinished review is a result

The sampling frame is ready. Human annotation, independent dual coding, adjudication, accuracy estimates, and agreement statistics have not yet been completed.

No accuracy result is available yet

This artifact is a review assignment frame, not a validation result. Accuracy, recall, precision, and agreement remain unavailable until independent human annotations and adjudication are complete.

Deterministic sampling frame

All ten planned domains are ready for review

10domains sampled
3,864review units
773units requiring two independent reviews
0of 4,637 required reviews completed

Deterministic SHA-256 bottom-k sampling within chamber, year, and domain-specific outcome strata. Each observed stratum receives a minimum allocation before remaining slots are distributed proportionally. The sample is bound to databasebfdbe2ce37a5a159….

Review packet

What will be checked

Ready for human review

Chamber quorum

sitting

Population
1,563
Sample
400
Strata
40
Dual-coded target
80
Reviews complete
0 / 480
Unresolved disagreements
0

Planned metrics: accuracy, class-specific precision, class-specific recall.

Results: not yet reviewed

Ready for human review

Floor attendance

member-sitting

Population
83,647
Sample
400
Strata
48
Dual-coded target
80
Reviews complete
0 / 480
Unresolved disagreements
0

Planned metrics: precision, recall, confusion matrix.

Results: not yet reviewed

Ready for human review

Committee attendance

member-source account

Population
51,429
Sample
400
Strata
33
Dual-coded target
80
Reviews complete
0 / 480
Unresolved disagreements
0

Planned metrics: precision, recall, status agreement.

Results: not yet reviewed

Ready for human review

Media-to-meeting linkage

media asset

Population
7,678
Sample
400
Strata
40
Dual-coded target
80
Reviews complete
0 / 480
Unresolved disagreements
0

Planned metrics: selected-target accuracy.

Results: not yet reviewed

Ready for human review

Motions and choices

printed committee action

Population
20,038
Sample
400
Strata
53
Dual-coded target
80
Reviews complete
0 / 480
Unresolved disagreements
0

Planned metrics: detection accuracy, member identity accuracy, choice accuracy.

Results: not yet reviewed

Ready for human review

Member contributions

printed attribution clause

Population
264
Sample
264
Strata
31
Dual-coded target
53
Reviews complete
0 / 317
Unresolved disagreements
0

Planned metrics: detection accuracy, attribution accuracy, role accuracy.

Results: not yet reviewed

Ready for human review

Committee assignments

membership event

Population
2,381
Sample
400
Strata
81
Dual-coded target
80
Reviews complete
0 / 480
Unresolved disagreements
0

Planned metrics: identity accuracy, role accuracy, date accuracy.

Results: not yet reviewed

Ready for human review

Meeting reconciliation

canonical meeting and retained source accounts

Population
5,459
Sample
400
Strata
62
Dual-coded target
80
Reviews complete
0 / 480
Unresolved disagreements
0

Planned metrics: match precision, match recall, outcome accuracy.

Results: not yet reviewed

Ready for human review

Transcript assertions

candidate transcript segment

Population
1,024,457
Sample
400
Strata
17
Dual-coded target
80
Reviews complete
0 / 480
Unresolved disagreements
0

Planned metrics: text accuracy, contextual validity, publication-scope accuracy.

Results: not yet reviewed

Ready for human review

Speaker attribution

reviewed transcript segment

Population
1,024,457
Sample
400
Strata
26
Dual-coded target
80
Reviews complete
0 / 480
Unresolved disagreements
0

Planned metrics: identity accuracy, speaker-boundary accuracy.

Results: not yet reviewed

Publication gate

How a result becomes publishable

  1. Sample. Freeze deterministic, stratified units against one release.
  2. Review. Inspect retained official evidence using controlled labels.
  3. Double-code. Independently review at least 20 percent without sharing judgments.
  4. Adjudicate. Preserve both original decisions and document the final resolution.
  5. Calculate. Publish accuracy, class metrics, confidence intervals, and agreement.
  6. Bind. Generate the website and manuscript table from the same frozen result.

The downloadable sample records the release binding, deterministic seed, sampling method, strata, assignments, and current review state. The complete annotation and adjudication protocol is maintained with the frozen publication bundle.