Open source SIL OFL 1.1 Built on Iosevka

Ioskeley Mono

An open programming typeface built on Iosevka — compact, geometric, and tuned for editors and terminals. Technical character without visual noise.

Normal · Regular 400 cap metric 690 · x metric 520 advance 600 10 weights · upright & italic 2 widths · 3 families

Download See it at work

09:12Workbench

Eight hours in one face.

Interface, editor and terminal: the same typeface at three sizes.

Explorer

src
queue
retry.ts M
scheduler.ts
dead-letter.ts
http
telemetry
types.ts
test
retry.test.ts M
scheduler.test.ts
CHANGELOG.md
package.json
relay.toml
tsconfig.json

Outline

MAX_ATTEMPTS
BASE_DELAY_MS
CEILING_MS
backoff(attempt, jitter)
deliver(d, ep)

retry.ts ● scheduler.ts relay.toml

src › queue › retry.ts › deliver

import { setTimeout as sleep } from "node:timers/promises";
import type { Delivery, Endpoint } from "../types";
import { metrics, tracer } from "../telemetry";

const MAX_ATTEMPTS = 8;
const BASE_DELAY_MS = 250;
const CEILING_MS = 60_000;

/** Equal jitter, capped: 250 ms, 500 ms, 1 s … 60 s. */
export function backoff(attempt: number, jitter = Math.random()): number {
  const exp = Math.min(CEILING_MS, BASE_DELAY_MS * 2 ** attempt);
  return Math.round(exp / 2 + (exp / 2) * jitter);
}

export async function deliver(d: Delivery, ep: Endpoint): Promise<boolean> {
  for (let attempt = 0; attempt < MAX_ATTEMPTS; attempt++) {
    const res = await fetch(ep.url, {
      method: "POST",
      headers: { "content-type": "application/json", "x-relay-id": d.id },
      body: JSON.stringify(d.payload),
      signal: AbortSignal.timeout(10_000),
    });
    if (res.ok) {
      metrics.increment("delivery.ok", { endpoint: ep.id });
      return true;
    }
    if (res.status >= 400 && res.status < 500 && res.status !== 429) break;

    const retryAfter = Number(res.headers.get("retry-after"));
    const wait = Number.isFinite(retryAfter) && retryAfter > 0
      ? Math.min(CEILING_MS, retryAfter * 1_000)
      : backoff(attempt);
    await sleep(wait);
  }
  metrics.increment("delivery.failed", { endpoint: ep.id });
  return false;
}

retry.test.ts ●

test › retry.test.ts › describe("backoff")

import { describe, expect, it } from "vitest";
import { backoff } from "../src/queue/retry";

describe("backoff", () => {
  it("is half the window with zero jitter", () => {
    expect(backoff(0, 0)).toBe(125);
    expect(backoff(3, 0)).toBe(1_000);
  });

  it("never exceeds CEILING_MS", () => {
    for (let n = 0; n < 32; n++) {
      expect(backoff(n, 1)).toBeLessThanOrEqual(60_000);
    }
  });

  it("doubles until the ceiling", () => {
    const delays = [0, 1, 2, 3, 4].map((n) => backoff(n, 1));
    expect(delays).toEqual([250, 500, 1_000, 2_000, 4_000]);
  });
});

Problems 1 Output Terminal zsh · relay

mira@forge ~/src/relay (feat/retry-cap) $ pnpm vitest run test/retry
 RUN v2.1.4 /home/mira/src/relay

 √ test/retry.test.ts (3 tests) 4ms

 Test Files  1 passed (1)
      Tests  3 passed (3)
   Start at  09:14:07
   Duration  412ms (transform 61ms, setup 0ms, collect 88ms, tests 4ms)

mira@forge ~/src/relay (feat/retry-cap)
$ git status -sb
## feat/retry-cap...origin/feat/retry-cap [ahead 2]
 M src/queue/retry.ts
 M test/retry.test.ts
?? docs/retries.md

$ git log --oneline --graph --all -9
* 4e1d0b7 (HEAD -> feat/retry-cap) parse Retry-After
* b7d03f5 honour Retry-After on 429 and 503
* 9c41e2a (origin/feat/retry-cap) cap backoff
| * 61aa0c3 (origin/main, main) bump undici
|/
*   d2f8e17 Merge pull request #209 from ade/dlq-ttl
|\
| * 0b93c4a dead-letter: expire after 14 days
|/
* 5e7fa12 (tag: v0.14.2) release 0.14.2

$ pnpm outdated
┌──────────────────┬─────────┬────────┐
│ Package          │ Current │ Latest │
├──────────────────┼─────────┼────────┤
│ undici           │ 6.19.8  │ 6.21.0 │
├──────────────────┼─────────┼────────┤
│ vitest (dev)     │ 2.1.4   │ 2.1.8  │
├──────────────────┼─────────┼────────┤
│ typescript (dev) │ 5.6.2   │ 5.6.3  │
└──────────────────┴─────────┴────────┘
$

$ relay --help
relay 0.15.0-rc.1 · webhook delivery service

Usage: relay <command> [options]

Commands:
  serve              Start the HTTP API and workers
  replay <since>     Re-enqueue failed deliveries
  endpoints ls       Endpoint health
  dlq purge          Drop dead letters past ttl

Options:
  -c, --config <path>  Config       [./relay.toml]
  -w, --workers <n>    Worker count            [4]
      --attempts <n>   Attempts each           [8]
      --ceiling <ms>   Backoff cap         [60000]
  -q, --quiet          Warnings, errors only
  -h, --help           Show this message

$ relay replay --since 2026-10-19T09:00Z
scanning dead-letter queue … 1,284 candidates

 endpoint             sent   fail      p95
 ep_billing_live       412      3    184ms
 ep_ledger_sync        388      0     96ms
 ep_crm_contacts       137     11    1.21s

 [█████████████████████▉        ]  73%   937/1284
 eta 00:00:41   rate 22.6/s   workers 4/4

[relay] 0:zsh* 1:logs- 2:db "forge" 10:41 19-Oct-26

13:05Review

Reading change, line by line.

Hashes, hunks, two gutters of line numbers and a conversation in the margin. Added and removed lines differ in tone, not hue.

#214 Cap retry delay and honour Retry-After

feat/retry-cap → main · 3 commits · mira

9c41e2a cap exponential backoff at CEILING_MS
b7d03f5 honour Retry-After on 429 and 503
4e1d0b7 parse Retry-After HTTP-date form

src/queue/retry.ts+17 −4
test/retry.test.ts+22 −0
CHANGELOG.md+2 −0

src/queue/retry.ts +17 −4

diff --git a/src/queue/retry.ts b/src/queue/retry.ts
index 3f9c2e1..a07b4d8 100644
--- a/src/queue/retry.ts
+++ b/src/queue/retry.ts
@@ -2,9 +2,12 @@
 import type { Delivery, Endpoint } from "../types";
 import { metrics } from "../telemetry";
 
-const MAX_ATTEMPTS = 12;
-const BASE_DELAY_MS = 500;
+const MAX_ATTEMPTS = 8;
+const BASE_DELAY_MS = 250;
+const CEILING_MS = 60_000;
 
-export function backoff(attempt: number): number {
-  return BASE_DELAY_MS * 2 ** attempt;
+/** Equal jitter, capped: 250 ms, 500 ms, 1 s … 60 s. */
+export function backoff(attempt: number, jitter = Math.random()): number {
+  const exp = Math.min(CEILING_MS, BASE_DELAY_MS * 2 ** attempt);
+  return Math.round(exp / 2 + (exp / 2) * jitter);
 }
@@ -22,7 +25,12 @@ export async function deliver(d: Delivery, ep: Endpoint): Promise<boolean> {
       return true;
     }
     if (res.status >= 400 && res.status < 500 && res.status !== 429) break;
-    await sleep(backoff(attempt));
+
+    const retryAfter = Number(res.headers.get("retry-after"));
+    const wait = Number.isFinite(retryAfter) && retryAfter > 0
+      ? Math.min(CEILING_MS, retryAfter * 1_000)
+      : backoff(attempt);
+    await sleep(wait);
   }
   metrics.increment("delivery.failed", { endpoint: ep.id });
   return false;

tomas · 2 h ago

Retry-After can also be an HTTP-date (RFC 9110 §10.2.3). Number() turns that into NaN and we quietly fall back to backoff. Fine, but worth a test?

mira · 1 h ago

Covered both forms with parseRetryAfter() in 4e1d0b7.

15:22Logs

Scanning at density.

Timestamps, UUIDs and durations in SemiCondensed at 11 px. The eye runs down columns, not across noise.

2026-10-19T15:22:07.560Z  INFO   w03  3d81bc8b-c705-4fe8-b9ea-6dc136d4983a  POST  /v1/endpoints/ep_billing_live/deliveries  202    12.4ms  bytes=1874
2026-10-19T15:22:08.129Z  INFO   w01  3a699eef-ad63-4a68-ae08-a5268a17bc21  POST  /v1/endpoints/ep_ledger_sync/deliveries   202     9.8ms  bytes=622
2026-10-19T15:22:08.857Z  INFO   w04  39b5abbe-427b-4a04-a967-2b78630ead13  GET   /v1/deliveries/dlv_01J9ZK3T0QW7           200     3.1ms  cache=hit
2026-10-19T15:22:08.996Z  WARN   w02  0c7323f8-68b1-49c0-999e-6963a71b9d71  POST  /v1/endpoints/ep_crm_contacts/deliveries  503  1204.9ms  retry=3 wait_ms=2000 reason="upstream unavailable"
2026-10-19T15:22:09.759Z  INFO   w03  a746164a-78f5-4663-bf40-594c49d5f1b0  POST  /v1/endpoints/ep_billing_live/deliveries  202    11.0ms  bytes=1902
2026-10-19T15:22:09.846Z  WARN   w01  9a4d9000-281c-4bc7-9e80-9fd61ee1da19  POST  /v1/endpoints/ep_crm_contacts/deliveries  429    88.2ms  retry=4 retry_after=30 wait_ms=30000
2026-10-19T15:22:10.581Z  INFO   gc   -                                     kind=minor heap_used=182.4MiB heap_total=256.0MiB rss=311.0MiB  4.7ms
2026-10-19T15:22:11.206Z  INFO   w04  6ff8c37e-2b39-4a56-8f3c-51c168c301a4  POST  /v1/endpoints/ep_ledger_sync/deliveries   202    10.3ms  bytes=640
2026-10-19T15:22:11.343Z  ERROR  w02  72d4a4bc-1a32-410f-b181-69b34ab63fad  POST  /v1/endpoints/ep_crm_contacts/deliveries  503 10000.0ms  attempts=8 outcome=dead_letter err="TimeoutError: signal timed out"
                                                                                  at deliver (/srv/relay/dist/queue/retry.js:41:19)
                                                                                  at async Worker.run (/srv/relay/dist/queue/worker.js:88:7)
2026-10-19T15:22:11.883Z  INFO   w01  aead9dc2-c505-431a-81c7-e760cd890019  POST  /v1/endpoints/ep_billing_live/deliveries  202    13.6ms  bytes=1866
2026-10-19T15:22:12.538Z  INFO   q    -                                     window=60s queue_depth=1284 inflight=37 p50=41ms p95=184ms p99=1.21s
2026-10-19T15:22:12.840Z  INFO   w03  49de527c-353e-42a7-bb44-6701d20637cf  DEL   /v1/dlq?older_than=14d                    204    27.5ms  purged=0
2026-10-19T15:22:13.068Z  INFO   w04  d9c9e409-31a0-45b3-922e-435832accabd  POST  /v1/endpoints/ep_ledger_sync/deliveries   202     9.1ms  bytes=598
2026-10-19T15:22:13.544Z  INFO   w02  587f1d63-d2a0-49b5-8137-7b211fa6f6fd  POST  /v1/endpoints/ep_crm_contacts/deliveries  202   142.7ms  retry=4 recovered=true

16:48Query

Columns that line up by themselves.

Every figure takes the same advance, so psql's padding alone aligns money, ratios and dates. Zero, O, one, I and l stay distinct inside the IDs.

SELECT a.account_id,
       a.plan,
       count(d.id)                         AS deliveries_30d,
       round(avg((d.status >= 400)::int), 4) AS failed_ratio,
       a.credit_balance_usd,
       max(p.paid_at)::date                AS last_payment_at,
       a.is_suspended
  FROM accounts a
  LEFT JOIN deliveries d
         ON d.account_id = a.account_id
        AND d.created_at > now() - interval '30 days'
  LEFT JOIN payments p ON p.account_id = a.account_id
 GROUP BY a.account_id
 ORDER BY failed_ratio DESC NULLS LAST, deliveries_30d DESC;

relay=> \pset null NULL
Null display is "NULL".
relay=> \i accounts_at_risk.sql
  account_id   |    plan    | deliveries_30d | failed_ratio | credit_balance_usd | last_payment_at | is_suspended
---------------+------------+----------------+--------------+--------------------+-----------------+--------------
 acct_1PlO0kQz | team       |          84210 |       0.0417 |            -128.40 | 2026-09-30      | f
 acct_7oO0sLm1 | free       |           3118 |       0.0263 |               0.00 | NULL            | f
 acct_0Il9xWe2 | enterprise |        1204331 |       0.0031 |           18920.75 | 2026-10-01      | f
 acct_I1ldN0Ob | scale      |         402977 |       0.0009 |            2450.00 | 2026-10-14      | f
 acct_4fZl1O0q | team       |             61 |       0.0000 |              -3.12 | 2026-08-30      | t
 acct_8IlO0rTy | free       |              0 |         NULL |               0.00 | NULL            | t
(6 rows)

Time: 38.412 ms

17:30Docs

Prose, signature and table on one page.

A monospace that holds a paragraph, not only a code block.

Reference / queue

backoff()

Returns the delay, in milliseconds, before attempt n is retried. Delays double from BASE_DELAY_MS and stop at CEILING_MS; jitter spreads retries from many workers so they do not reach an endpoint in lockstep.

function backoff(attempt: number, jitter?: number): number

Parameters

Name	Type	Default	Description
`attempt`	number	—	Zero-based attempt index.
`jitter`	number	`Math.random()`	In [0, 1). Zero gives the lower bound.

Returns

number, in milliseconds, between half the window and the full window, never above 60_000.

Note A 429 or 503 carrying Retry-After overrides the computed delay. See Delivery semantics.

backoff(0, 0);   // 125
backoff(3, 0);   // 1_000
backoff(20, 1);  // 60_000, capped

       IOSKELEY TYPE WORKS
   proof of setting · no. 0219
==================================
DATE          2026-10-19  18:02:41
JOB           in-use showcase · 03
----------------------------------
PACKAGE           IoskeleyMono.zip
FAMILY                    Standard
WIDTH  Normal                600 u
       SemiCondensed         540 u
WEIGHT Thin – Black        100–900
STYLE  Upright, Italic       11.8°
----------------------------------
UNITS PER EM                  1000
CAP METRIC                     690
X METRIC                       520
LINE  965 / −285          1.250 em
----------------------------------
FILE (WEB)                     KiB
Regular.woff2                94.23
Italic.woff2                104.71
Bold.woff2                   94.47
----------------------------------
WEIGHTS                      ×  10
STYLES                       ×   2
WIDTHS                       ×   2
==================================
TOTAL FACES                     40
==================================
LICENSE                SIL OFL 1.1
      use · modify · embed · share

▌│█▌║▌││█║▌║││▌█│▌║█▌│║▌█││▌║█│▌
▌│█▌║▌││█║▌║││▌█│▌║█▌│║▌█││▌║█│▌
▌│█▌║▌││█║▌║││▌█│▌║█▌│║▌█││▌║█│▌

LabYour turn

Type your own text.

The syntax tones are presentation on this page, not a font feature: Ioskeley ships monochrome.

Width

Weight

Style

Size 14.5px

Ligatures

Ioskeley Mono · Normal · Regular 400 · 600 units/char

SystemHow it is built

One face. Ten weights. Two widths. Three families.

Single-storey gHover or focus for metrics

Signature forms

adouble-storey
Qstraight tail
@compact
$ascends
&open
*raised

Never confused

0Odotted zero
1Il|four strokes
5S
2Z
8B
rn m
;: ,.
'`"

Ten weights

Two widths

Same design, same size. SemiCondensed sets 540 units per character instead of 600, so a line runs exactly ten percent shorter.

Normal · 600

const glyphs = { figures: [0xFC82, 100_000, 0.0] };

SemiCondensed · 540

const glyphs = { figures: [0xFC82, 100_000, 0.0] };

Three families

Every family carries ten weights, upright and italic, in both widths. They differ in ligatures and terminal rhythm only.

Standard

Ioskeley Mono

a => b != c <= d

Programming ligatures on. The everyday choice for editors and IDEs.

Term

Ioskeley Mono Term

—

Same ligatures, with arrows and box drawing tuned to the terminal cell. Not shown: this page carries the Standard web build only.

Ioskeley Mono NL

a => b != c <= d

No programming ligatures; ordinary shaping only. Shown here with ligatures switched off, which is how NL sets.

NFNerd Font geometry

NF or NFM?

Each Nerd Font archive ships two patched variants with the same icons. They differ only in how icons sit in the cell.

NF Ioskeley Mono Nerd Font: Ordinary Nerd Font geometry. Larger icons stay intact and may extend a little past one cell, which looks right in most prompts. A good default.
NFM Ioskeley Mono Nerd Font Mono: Strict single-cell geometry: every glyph, icon included, stays inside one cell. What Fontconfig and some font pickers require of a true monospace face.

From 2.2.0. Every Nerd Font archive includes both NF and NFM. Choose NF for larger icons or NFM when strict single-cell Fontconfig classification is required.

OriginsDesign evolution

Where it started

Berkeley Mono / TX-02 inspired the first experiment, and Iosevka made it possible to run. Both are history rather than specification: a hundred-odd builds and repeated tuning moved Ioskeley toward a design of its own. The studies below are kept as a record of that process. They are comparison plates, deliberately monochrome, and they are not a claim of equivalence.

Open the character, overlay and code studies Three plates · 6 images

TX-02 is a commercial typeface. Its files are neither included nor distributed with Ioskeley; these plates were produced for private study.

DownloadFree · SIL OFL 1.1

Take it with you.

Start here IoskeleyMono.zip Standard family. Both widths, all ten weights, programming ligatures. Download

Terminal IoskeleyMono-Term.zip Term family. Terminal-tuned arrow and box-drawing forms. Download →

No ligatures IoskeleyMono-NL.zip NL family. No programming ligatures. Download →

With icons IoskeleyMono-NerdFont.zip Standard, patched. Contains NF and NFM from 2.2.0. Download →

IoskeleyMono-Term-NerdFont.zip Term, patched. Contains NF and NFM from 2.2.0. Download →

IoskeleyMono-NL-NerdFont.zip NL, patched. Contains NF and NFM from 2.2.0. Download →

Web IoskeleyMono-Web.zip Compact WOFF2 subset for @font-face. Download →

IoskeleyMono-Web-Full.zip Full WOFF2 repertoire: Greek, Cyrillic, long arrows. Download →

Release timing. These links use GitHub's /releases/latest/download/ route, so they always serve the latest published release.

Install & use

Manually

Unzip, choose Normal or SemiCondensed, then Hinted for standard-DPI Windows and Linux or Unhinted for macOS and HiDPI.

Windows: right-click → Install. macOS: Font Book. Linux: copy to ~/.local/share/fonts/, run fc-cache -fv.

Editors

// VS Code — settings.json
{
  "editor.fontFamily": "'Ioskeley Mono', monospace",
  "editor.fontLigatures": true,
  "editor.fontSize": 14.5
}

Terminals

# Ghostty — config
font-family = Ioskeley Mono Term
font-size   = 14

# Kitty — kitty.conf
font_family Ioskeley Mono Term
font_size   14.0

On the web

@font-face {
  font-family: "Ioskeley Mono";
  src: url("./IoskeleyMono-Regular.woff2")
       format("woff2");
  font-weight: 400;
  font-display: swap;
}

Package managers, OpenType features and build notes live in the project README.

Ioskeley Mono

Eight hours in one face.

Every character on the grid.

Reading change, line by line.

Scanning at density.

Columns that line up by themselves.

Prose, signature and table on one page.

backoff()

Parameters

Returns

Off the screen, onto paper.

Type your own text.

One face. Ten weights. Two widths. Three families.

Signature forms

Never confused

Ten weights

Two widths

Three families

NF or NFM?

Where it started

Take it with you.

Install & use

Manually

Editors

Terminals

On the web