Parsyra Conversion Examples,
Unedited

# Fieldnote 2.4 Release Overview

Product update memo — internal draft, April 2026

## Summary

Fieldnote 2.4 focuses on offline reliability. Sync conflicts dropped by roughly two thirds in the beta cohort after we replaced last-write-wins with a field-level merge. This memo covers what shipped, what we deferred, and the open questions for the 2.5 planning cycle.

## What Shipped

Field-level merge for notes edited on two devices while offline.

Background sync now retries with exponential backoff instead of a fixed 30-second timer.

Attachment uploads resume after a dropped connection rather than restarting.

A visible sync-state indicator in the note header, replacing the ambiguous spinner.

## Deferred to 2.5

Shared-notebook merge remains unsolved. When two collaborators restructure the same outline offline, the merge produces a valid but surprising order. We deferred this rather than ship a confusing resolution dialog.

Outline-aware merge for shared notebooks.

Selective sync for attachments over 50 MB.

## Open Questions

The beta cohort skews toward single-device users, so the conflict numbers above may understate multi-device behavior. We plan to re-run the analysis once the 2.4 rollout reaches half of the multi-device population.

# Northwind Tooling Co. — Q3 2026 Results

Unaudited condensed summary, prepared for the quarterly review

## Revenue by Segment

Segment

Q3 2026 Q2 2026 QoQ Hand tools $4,210k $3,955k +6.4%

Power tools

$2,875k $3,010k -4.5%

Accessories

$1,340k $1,180k +13.6% Service & parts $960k $915k +4.9%

Total

$9,385k $9,060k +3.6%

## Operating Costs

Line

Q3 2026

Share of revenue Cost of goods sold $5,210k

## 55.5%

Sales & marketing

$1,120k

## 11.9%

R&D

$860k

## 9.2%

General & administrative

$705k

## 7.5%

## Commentary

Accessories growth reflects the June launch of the modular case line. Power tools declined on channel inventory drawdown; sell-through remained flat. Gross margin held at 44.5% despite steel cost inflation, helped by the supplier consolidation completed in Q1.

Cash at quarter end was $6.2M with no drawn credit. The board approved a $1.5M capacity investment for the Tacoma line, expected online in Q2 2027.

# Northwind Tooling Co. — Q3 2026 Results

Unaudited condensed summary, prepared for the quarterly review

## Revenue by Segment

|  Segment | Q3 2026 | Q2 2026 | QoQ  |
| --- | --- | --- | --- |
|  Hand tools | $4,210k | $3,955k | +6.4%  |
|  Power tools | $2,875k | $3,010k | -4.5%  |
|  Accessories | $1,340k | $1,180k | +13.6%  |
|  Service & parts | $960k | $915k | +4.9%  |
|  **Total** | **$9,385k** | **$9,060k** | **+3.6%**  |

## Operating Costs

|  Line | Q3 2026 | Share of revenue  |
| --- | --- | --- |
|  Cost of goods sold | $5,210k | 55.5%  |
|  Sales & marketing | $1,120k | 11.9%  |
|  R&D | $860k | 9.2%  |
|  General & administrative | $705k | 7.5%  |

## Commentary

Accessories growth reflects the June launch of the modular case line. Power tools declined on channel inventory drawdown; sell-through remained flat. Gross margin held at 44.5% despite steel cost inflation, helped by the supplier consolidation completed in Q1.

Cash at quarter end was $6.2M with no drawn credit. The board approved a $1.5M capacity investment for the Tacoma line, expected online in Q2 2027.

# GREENFIELD NURSERY — DELIVERY NOTE

Date: 14 March 2026

Order no. 2026-0312

Deliver to: Harbor Community Garden, Pier Road allotments

|  Item | Qty | Unit | Total  |
| --- | --- | --- | --- |
|  Apple sapling (Discovery) | 6 | $18.50 | $111.00  |
|  Compost, 40 L bag | 10 | $7.25 | $72.50  |
|  Willow stakes, bundle | 4 | $9.00 | $36.00  |
|  Total due |  |  | $219.50  |

Payment terms: 30 days net. The variance of the germination estimate follows σ²/n as discussed with your volunteer coordinator.

Received in good condition — M. Aldous

# A Note on Weighted Moving Averages

Course handout — numerical methods seminar

## 1. Definition

Given a series x , …, x and weights w , the weighted moving average over a window of size k is

1 n i

k−1

k−1

m = ( ∑ w x ) / ( ∑ w )

t i=0 i t−i i=0 i

i

with the convention that the window is truncated at the start of the series. When w = λ for some 0 < λ <

i

1, the average discounts older observations geometrically.

## 2. Variance Under Independence

2

If the x are independent with common variance σ , the variance of the estimator is

i

2 2 2

Var(m ) = σ · ( ∑ w ) / ( ∑ w )

t i i

2

which is minimized by equal weights, giving the familiar σ /k. Geometric weights trade this efficiency for responsiveness: the effective sample size becomes (1+λ)/(1−λ) as k grows.

## 3. Practical Note

For λ = 0.9 the effective sample size is 19, so a practitioner who wants the smoothing of a 19-point simple average with faster reaction to level shifts can use the geometric scheme without retuning the window length.

# A Note on Weighted Moving Averages

Course handout — numerical methods seminar

## 1. Definition

Given a series $x_1, \dots, x_n$ and weights $w_i$, the weighted moving average over a window of size $k$ is

$$m_t = (\sum_{i=0}^{k-1} w_i x_{t-i}) / (\sum_{i=0}^{k-1} w_i)$$

with the convention that the window is truncated at the start of the series. When $w_i = \lambda^i$ for some $0 < \lambda < 1$, the average discounts older observations geometrically.

## 2. Variance Under Independence

If the $x_i$ are independent with common variance $\sigma^2$, the variance of the estimator is

$$Var(m_t) = \sigma^2 \cdot (\sum w_i^2) / (\sum w_i)^2$$

which is minimized by equal weights, giving the familiar $\sigma^2/k$. Geometric weights trade this efficiency for responsiveness: the effective sample size becomes $(1+\lambda)/(1-\lambda)$ as $k$ grows.

## 3. Practical Note

For $\lambda = 0.9$ the effective sample size is 19, so a practitioner who wants the smoothing of a 19-point simple average with faster reaction to level shifts can use the geometric scheme without retuning the window length.

<article>
  <h1>How We Version Our Field Guides</h1>
  <p><em>Posted to the Fieldnote engineering blog, March 2026.</em></p>
  <p>Every field guide we publish is a set of Markdown files in a Git
  repository. Print editions are tagged releases. This post explains the
  branching model and the two rules that keep 40 contributors from
  stepping on each other.</p>
  <h2>The two rules</h2>
  <ol>
    <li>A chapter owner merges their own chapter. Nobody else does.</li>
    <li>Cross-chapter renames land in a dedicated <code>restructure/*</code>
    branch, never in a chapter branch.</li>
  </ol>
  <h2>Release cadence</h2>
  <table>
    <tr><th>Edition</th><th>Branch</th><th>Cadence</th></tr>
    <tr><td>Web</td><td><code>main</code></td><td>continuous</td></tr>
    <tr><td>Print</td><td><code>release/YYYY</code></td><td>yearly</td></tr>
    <tr><td>Translations</td><td><code>l10n/*</code></td><td>after print freeze</td></tr>
  </table>
  <blockquote>Versioning prose like code only works if the prose is
  actually plain text. That is the whole trick.</blockquote>
  ...

 How We Version Our Field Guides 

# How We Version Our Field Guides

_Posted to the Fieldnote engineering blog, March 2026._

Every field guide we publish is a set of Markdown files in a Git repository. Print editions are tagged releases. This post explains the branching model and the two rules that keep 40 contributors from stepping on each other.

## The two rules

1.  A chapter owner merges their own chapter. Nobody else does.
2.  Cross-chapter renames land in a dedicated `restructure/*` branch, never in a chapter branch.

## Release cadence

| Edition | Branch | Cadence |
| --- | --- | --- |
| Web | `main` | continuous |
| Print | `release/YYYY` | yearly |
| Translations | `l10n/*` | after print freeze |

The print freeze is the only synchronization point. Everything else flows continuously, and the [style guide](https://example.com/fieldnote-style-guide) resolves formatting arguments before they reach review.

> Versioning prose like code only works if the prose is actually plain text. That is the whole trick.

## What broke

Image-heavy chapters caused merge pain until we moved figures to per-chapter directories with numeric prefixes:

```
guides/
  coastal-birds/
    ch03-waders/
      fig-01-sandpiper.jpg
      text.md
```

Since that change, no figure conflict has required manual resolution.

## Slide 1

Community Workshop Program

Funding pitch — spring 2026

## Slide 2

Why now

Tool library membership doubled in 18 months

Waitlists on every repair workshop since October

Two volunteer instructors ready to lead new sessions

## Slide 3

Program economics

Line

Monthly

Covered by

Venue hire

$420

Council grant

Materials

$300 avg

Ticket sales

Instructor fees

$650 avg

This proposal

## Slide 4

What we are asking for

$7,800 to underwrite instructor fees for one year

Access to the annex room on weekday evenings

A decision before the summer schedule is printed

## Slide 5

Thank you

Questions — workshops@harborcommons.example

## Budget

| Q3 2026 Budget — Community Workshop |  |  |  |
| --- | --- | --- | --- |
| Category | Jul | Aug | Sep |
| Venue hire | 420 | 420 | 420 |
| Materials | 310 | 265 | 340 |
| Instructor fees | 600 | 600 | 750 |
| Insurance | 85 | 85 | 85 |
| Outreach & printing | 120 | 90 | 60 |
| Total | 1535 | 1460 | 1655 |

## Attendance

| Workshop | Seats | Attended | Waitlist |
| --- | --- | --- | --- |
| Intro woodworking | 12 | 12 | 5 |
| Bike repair basics | 10 | 8 | 0 |
| Sewing machine 101 | 12 | 11 | 3 |
| Home electrics safety | 16 | 14 | 2 |

# Sizing an A/B Test

Notes for the growth review. The formulas are written as TeX so they stay reviewable in version control.

## Confidence interval for a conversion rate

With $n$ visitors and $k$ conversions the point estimate is $\hat{p} = k/n$, and the normal-approximation interval at level $1-\alpha$ is

$$\hat{p} \pm z_{1-\alpha/2} \sqrt{\frac{\hat{p}(1-\hat{p})}{n}}$$

The approximation is reasonable once $n\hat{p} \geq 10$ and $n(1-\hat{p}) \geq 10$; below that use the Wilson interval instead.

## Required sample size

To detect an absolute lift of $\delta$ with power $1-\beta$, each arm needs roughly

$$n \approx \frac{2\bar{p}(1-\bar{p})\left(z_{1-\alpha/2} + z_{1-\beta}\right)^2}{\delta^2}$$

where $\bar{p}$ is the pooled rate. For $\bar{p} = 0.12$, $\delta = 0.02$, $\alpha = 0.05$ and $\beta = 0.2$ this gives about 4,150 visitors per arm.

## Why the test ran long

Traffic split unevenly for the first four days, so the effective $n$ was closer to the smaller arm than to the total. The variance term $\sqrt{\hat{p}(1-\hat{p})/n}$ shrinks with $\sqrt{n}$, so halving usable traffic costs roughly $\sqrt{2}$ in interval width.

# Sizing an A/B Test

Notes for the growth review. The formulas are written as TeX so they stay reviewable in version control.

## Confidence interval for a conversion rate

With $n$ visitors and $k$ conversions the point estimate is $\hat{p} = k/n$, and the normal-approximation interval at level $1-\alpha$ is

$$\hat{p} \pm z_{1-\alpha/2} \sqrt{\frac{\hat{p}(1-\hat{p})}{n}}$$

The approximation is reasonable once $n\hat{p} \geq 10$ and $n(1-\hat{p}) \geq 10$; below that use the Wilson interval instead.

## Required sample size

To detect an absolute lift of $\delta$ with power $1-\beta$, each arm needs roughly

$$n \approx \frac{2\bar{p}(1-\bar{p})\left(z_{1-\alpha/2} + z_{1-\beta}\right)^2}{\delta^2}$$

where $\bar{p}$ is the pooled rate. For $\bar{p} = 0.12$, $\delta = 0.02$, $\alpha = 0.05$ and $\beta = 0.2$ this gives about 4,150 visitors per arm.

## Why the test ran long

Traffic split unevenly for the first four days, so the effective $n$ was closer to the smaller arm than to the total. The variance term $\sqrt{\hat{p}(1-\hat{p})/n}$ shrinks with $\sqrt{n}$, so halving usable traffic costs roughly $\sqrt{2}$ in interval width.

<article>
  <h1>Sync Architecture Note</h1>
  <p>How a note travels from an offline edit to a merged document. The
  diagram below is kept in the repository as Mermaid source so it can be
  reviewed in a pull request like any other code.</p>

  <h2>Merge flow</h2>
  <pre><code class="language-mermaid">flowchart TD
    A[Offline edit] --&gt; B{Has remote changed?}
    B --&gt;|No| C[Fast-forward]
    B --&gt;|Yes| D[Field-level merge]
    D --&gt; E{Conflict?}
    E --&gt;|No| F[Write merged note]
    E --&gt;|Yes| G[Queue for review]
    C --&gt; F</code></pre>

  <h2>Retry timing</h2>
  <pre><code class="language-mermaid">sequenceDiagram
    participant App
    participant Queue
    participant Server
    App-&gt;&gt;Queue: enqueue(change)
    ...</code></pre>

  <h2>The same flow as a picture</h2>
  <figure>
    <img src="diagram.png" alt="Rendered flowchart of the merge path">
    <figcaption>Figure 1 — the merge path, exported as PNG.</figcaption>
  </figure>
  ...

 Sync Architecture Note 

# Sync Architecture Note

How a note travels from an offline edit to a merged document. The diagram below is kept in the repository as Mermaid source so it can be reviewed in a pull request like any other code.

## Merge flow

```mermaid
flowchart TD
    A[Offline edit] --> B{Has remote changed?}
    B -->|No| C[Fast-forward]
    B -->|Yes| D[Field-level merge]
    D --> E{Conflict?}
    E -->|No| F[Write merged note]
    E -->|Yes| G[Queue for review]
    C --> F
```

## Retry timing

```mermaid
sequenceDiagram
    participant App
    participant Queue
    participant Server
    App->>Queue: enqueue(change)
    Queue->>Server: POST /sync
    Server-->>Queue: 503
    Queue->>Queue: backoff x2
    Queue->>Server: POST /sync
    Server-->>App: 200 merged
```

## The same flow as a picture

![Rendered flowchart of the merge path](diagram.png)

Figure 1 — the merge path, exported as PNG for the slide deck.

## Notes

*   Mermaid source lives beside the prose, so diagram changes show up in review diffs.
*   The exported PNG is generated from the same source and is never edited by hand.