Local-first document agent

Nebula Writer is your document agent.

Create polished, editable work from the files already in your project.

Download free See workflows

Nebula WriterFileEditViewHelp

Saved

123456789

Nebula Research OperationsReport 26-05 · 14 May 2026

Q2 pilot evaluation

Partner onboarding pilot: outcomes and scale recommendation

Executive briefPrepared for Product Council

Executive summary

The partner-led cohort completed onboarding more often and reached review readiness faster than both internal pilots. Results support a controlled 30-day expansion with named owners and weekly evidence checks.

DecisionApprove a 30-day partner-led expansion with two safeguards.

91%
completion: 13.9 min
median review: n = 188
evidence items

Partner demo leads completion across cohorts

Figure 1 · Completion rate by cohort, Q2 2026

Target ≥ 85%

Partner demo91%

Pilot B88%

Pilot A84%

Research review79%

The partner cohort exceeded the scale threshold while maintaining the largest evidence base.

Figure 2 · Facilitated setup made examples visible at the moment of need.

Evidence table

Cohort	Participants	Completion	Median review	Readiness
Pilot A	42	84%	18.2 min	Monitor
Pilot B	56	88%	15.7 min	Ready
Partner demo	64	91%	13.9 min	Scale
Research review	26	79%	21.4 min	Revise

Participant voice
“Seeing a finished example beside the task made the first review feel obvious.”
Partner facilitator · Interview 08

Word Document1,246 words7,842 characters

v0.4.8

Workflows

Concrete workflows, not blank-page AI.

Each AI document workflow starts with the files already in a project folder and ends with an editable artifact you can inspect, revise, and keep across LaTeX, Markdown, DOCX, PPTX, and spreadsheet work.

Data to Document

Turn retrofit outcomes into a delivery review.

Start with regional grant delivery, installation, and savings data. Nebula traces the findings, then builds an editable DOCX review with the funding evidence still attached.

Open data workflow docs

Nebula WriterFileEditViewHelp

Saved

fx91%

	A	B	C	D	E	F	G	H	I
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

Spreadsheet24 populated rows216 populated cells

v0.4.8

Research to Deck

Turn a scientific paper into a presentation people can follow.

Start with LaTeX, a compiled paper, and references. Nebula traces the claims, methods, and results into an editable scientific PPTX presentation.

Open research workflow docs

Nebula WriterFileEditViewHelp

Saved

1\documentclass[11pt]{article}

3\usepackage[margin=1in]{geometry}

4\usepackage{hyperref}

5\usepackage{xcolor}

6\usepackage{graphicx,booktabs}

8\title{Human-Centered AI Systems}

9\author{Nebula Research Group}

10\date{May 2026}

12\begin{document}

13\maketitle

15\section*{Introduction}

16Research teams lose time when evidence moves manually

17from technical manuscripts into reports and slide decks.

19\section{Method}

20We combine qualitative coding with quantitative checks.

21y_i = X_i \beta + \gamma_t + \epsilon_i

23\section{Results}

24Median review time fell by 22\% across cohorts.

25Evidence links were retained for every summary claim.

27\begin{table}[h]

28\centering

29\begin{tabular}{lrr}

30Cohort & Completion & Review \\

31Pilot A & 84\% & 18.2 \\

32Partner demo & 91\% & 13.9 \\

33\end{tabular}

34\end{table}

36\begin{figure}[h]

37\caption{Completion rate by cohort}

38\label{fig:completion}

39\end{figure}

41\section{Limitations}

42Cohorts were not randomized; replication is required.

44\bibliographystyle{plain}

45\bibliography{references}

46\end{document}

Applied Human Systems · Vol. 12, No. 2 · May 2026

Human-Centered AI Systems in Local Research Workflows

Mira Chen, Jonah Patel, and the Nebula Research Group

Research Operations Lab · Chicago, Illinois

Abstract

We evaluate a local-first document agent across four research cohorts (n=188). Mixed-method evidence shows that source-linked drafting reduced median review time by 22% while preserving a verifiable route from every summary claim to its originating manuscript, table, or interview note.

Keywords: document systems, provenance, human–AI collaboration, research operations

1 Introduction

Research teams routinely reconstruct structure when evidence moves from technical manuscripts into reports and briefings. That handoff separates claims from the equations, citations, and observations needed to review them.

2 Study design

We compared four cohorts using completion logs, timed reviews, artifact audits, and 17 semi-structured interviews. The primary model controlled for cohort and week:

yᵢ = Xᵢβ + γₜ + εᵢ

Table 1. Study sample
Cohort	n	Weeks
Pilot A	42	6
Pilot B	56	6
Research review	26	4
Partner demo	64	5

3 Qualitative coding

Two researchers independently coded interview notes for setup friction, confidence, and evidence retrieval. Agreement reached κ=.82 before reconciliation.

4 Results

The partner cohort exceeded the 85% completion threshold and produced the fastest median review.

Figure 1. Completion by cohort

Partner91%

Pilot B88%

Pilot A84%

Review79%

Source: study outcomes log; n=188.

−22%median review time

100%claims source-linked

“The example and source arrived together, so I could review the reasoning—not rebuild the context.” —Interview 08

5 Limitations

Cohorts were not randomized and differed in facilitator experience. Results warrant a controlled replication rather than an unrestricted rollout.

6 Conclusion

Local-first generation can reduce handoff friction while retaining the provenance needed for credible review.

LaTeX Document184 words1,717 characters

v0.4.8

Deck to Decision Brief

Turn a launch review into a decision brief people can act on.

Start with the PowerPoint your launch team already uses. Nebula carries the recommendation, customer evidence, success thresholds, owners, and stop rules into an editable DOCX brief for Product Council.

Open launch workflow docs

Nebula WriterFileEditViewHelp

Saved

Presentation5 slides24 editable objects

v0.4.8

Useful applications

Start from the files already on your desk.

The useful moments are not blank prompts. They are half-finished folders, messy exports, and review packets that need to become a document someone can actually send.

Business operations · Leadership

Operating review to leadership update

Combine the quarterly narrative, budget variance, and customer themes into a concise leadership update with a decision log.

Example outputPPTX · 9 slides

Q2 operating reviewLeadership confidential · 26 Jun 2026

04 / Growth constraint

Onboarding is the fastest route back to plan.

Demand held. Activation friction delayed $420k of expansion and accounts for 63% of the quarter's revenue gap.

Time to value: −18%
At risk: $420k
Owner: Product Ops

Revenue variance bridge

Quarter vs. operating plan · $000s

−$420

DecisionFund the guided setup pod for 90 days; checkpoint activation on 30 September.

Sources: Summary!F18 · Packet p.12 · Customer theme 304 / 09

AnchorsSummary!F18Packet p.12Theme 3

Starts with

q2-operating-review.docx
budget-variance.xlsx
customer-themes.pdf

Ends with

leadership-update.pptx
decision-log.docx

Every recommendation remains linked to the operating evidence used to draft it.

Customer research · Product launch

Customer research to launch brief

Synthesize interviews, survey results, and the working brief into a launch plan the product and revenue teams can use.

Example outputDOCX · 14 pages

Research note 07June 2026

“Show me one good setup, then let me make it mine.”
Mara · Ops lead · Interview 08

Guided setup opportunity

From first click to first value

Seventeen customer conversations and 342 survey responses point to the same launch wedge: useful examples before configuration.

68%

asked for examples

12 / 17

activated with a guide

Day 30

launch checkpoint

Activation journeyMedian first session · 23 min

Choose a model

4 curated examples

Make it yours

Guided first edit

Share the win

Team-ready output

Interview 08 · Survey Q7 · Launch brief §47 / 14

AnchorsInterview 08Survey Q7Brief §4

Starts with

customer-interviews.docx
survey-results.csv
launch-brief.md

Ends with

launch-plan.docx
exec-readout.pptx

Themes, quotes, and launch claims stay connected through the final review.

Program evaluation · Decision memo

Assessment export to program memo

Reconcile missing post-tests, join attendance flags, and write a program memo with a transparent cleaning audit.

Example outputXLSX · audit + memo

Winter pilot · analysis workbook

Completion and assessment audit

AUDIT PASSED

188

valid records

91%

top completion

13.9m

median review

Program completion and assessment audit by cohort
Cohort	Raw	Valid	Lift	Status
North hub	48	47	+7.8 pp	Verified
East hub	44	41	+5.1 pp	Verified
South hub	51	49	+8.4 pp	Verified
Partner demo	45	44	+11.6 pp	Verified
Analysis set	188	181	+8.2 pp	Released

Cleaning disposition

n = 201

Method note

Post-test lift uses paired records only. Attendance under 60% remains in the sensitivity model, not the headline estimate.

Memo finding: Partner demo led completion at 91%; preserve the live walkthrough in the next cohort.

Cohorts!D14 · Audit row 22 · Methods p.2cleaning-audit.xlsx · v3

AnchorsCohorts!D14Audit row 22Methods p.2

Starts with

winter-pilot-scores.xlsx
attendance-export.csv
methodology-note.md

Ends with

pilot-evaluation-memo.docx
cleaning-audit.xlsx

Cleaned rows, formulas, and caveats remain attached to each memo claim.

Executive leadership · Board review

Board packet to decision readout

Turn the Q3 packet, appendix metrics, and risk register into a concise readout with decisions, owners, and safeguards.

Example outputPPTX · board readout

Q3 board readout

Approve a staged rollout.

Two safeguards contain the material exposure while preserving the October window.

DECISION 03

Owner · COO

Decision conditions

2 material risks

01
Gate by service capacity.
Hold region two until first-response time is under four hours for ten days.
02
Cap commercial exposure.
Limit launch credits to $180k; CFO approval required above the cap.
03
Return with evidence.
Board checkpoint on 18 November with cohort retention and incident trend.

Risk heatmapResidual exposure

Low

Med

High

R-04

Med

R-07

Low

R-02

named owners

18 Nov

board checkpoint

Packet p.18 · Appendix B · Risk R-0403 / 08

AnchorsPacket p.18Appendix BRisk R-04

Starts with

q3-board-packet.docx
appendix-metrics.xlsx
risk-register.csv

Ends with

board-readout.pptx
decision-log.md

Each slide keeps a direct route back to the packet, appendix, or risk entry.

Built for serious work

Review agent work before it lands.

Sources, edits, and approval stay in one working view. Nebula never accepts a generated change for you.

Nebula WriterFileEditViewHelp

Saved

123456789

Northbridge Student SuccessEvidence dossier · Winter 2026

Tutoring renewal evidence: outcomes, attendance, and advising capacity

Prepared for the Academic Investment Committee · 10 pages

Evidence summary

The four-campus pilot improved gateway-course pass rates while keeping cost per completed tutoring sequence below the renewal threshold. Evening students benefited, but advisor capacity remains the principal implementation risk.

+8.4 ptscourse pass rate

74%sequence completion

$186cost / completer

Pass-rate lift by delivery mode

95% confidence interval

Day sections+8.4

Evening sections+7.9

Online support+5.6

Adjusted for prior GPA, credits attempted, campus, and gateway course.

Campus evidence table

Campus	Students	Completion	Pass-rate lift	Advisor load
North	214	78%	+9.1	Stable
Central	188	71%	+7.6	High
West	163	76%	+8.7	Stable
South	146	69%	+6.9	High

Finding 01 · Renewal case

Lift is consistent across delivery modes and campuses.

Finding 02 · Capacity caveat

Evening expansion requires 0.6 additional advisor FTE.

Word Document1,246 words7,842 characters

v0.4.8

Source-backed: Evidence stays attached to the claims it supports.
Review before save: Every insertion remains pending until you accept or discard it.
Ordinary project files: Your sources and final DOCX stay editable in the folder you control.

Get started

Start with the folder you already have.

Download Nebula Writer for macOS, Windows, or Linux and try a concrete workflow: leadership update, launch brief, program memo, or board readout.

Download free Read the docs

Nebula Writer is your document agent.

Partner onboarding pilot: outcomes and scale recommendation

Executive summary

Partner demo leads completion across cohorts

Evidence table

Concrete workflows, not blank-page AI.

Turn retrofit outcomes into a delivery review.

Turn a scientific paper into a presentation people can follow.

Human-Centered AI Systems in Local Research Workflows

Abstract

1 Introduction

2 Study design

3 Qualitative coding

4 Results

5 Limitations

6 Conclusion

Turn a launch review into a decision brief people can act on.

Make the first useful outcome happen before setup feels like work.

Confidence drops in the two minutes before the first useful result.

Four needs explain 81% of successful first sessions.

Prioritize relevance and control before adding more templates.

Ship one guided path, with evidence gates every seven days.

Make the first useful outcome happen before setup feels like work.

Start from the files already on your desk.

Operating review to leadership update

Onboarding is the fastest route back to plan.

Customer research to launch brief

From first click to first value

Assessment export to program memo

Completion and assessment audit

Board packet to decision readout

Review agent work before it lands.

Tutoring renewal evidence: outcomes, attendance, and advising capacity

Evidence summary

Campus evidence table

Start with the folder you already have.