●  SYMBOLOGY.ONLINE ANNOUNCEMENT

Faster pages,
open doors.

A season of work below the waterline: a rebuilt foundation, pages that travel light — and every synthesis on the site now open to read, no account required.

SEPTEMBER 13, 2026

Most of what changed this season you can't see. Symbology now runs on the same configuration-driven Kubernetes platform as the rest of Streetfortress — raised, rebuilt, and measured one pull request at a time — with a caching tier in front of the database and pages that ship a fraction of the bytes they used to. Every change was load-tested on a staging twin before it reached you, and a few taught us real lessons the hard way.

●  SAME SERVER, BEFORE & AFTER · SEPTEMBER 2026
Sustained requests / second
13 → 60
Concurrent readers at the limit
~960
Synthesis page, over the wire
143 KB → 10 KB
Errors under load, entire series
0
The full ladder — every configuration and release, knee by knee — is on the new engineering log.

The sign-up wall is gone. Guests used to hit a "create a free account" nudge after a few companies; we've removed it entirely. Every published synthesis is now open to anyone. Accounts remain what they should be: a way to watch companies and request coverage, with supporters keeping the deep-history perks.

The focus hasn't moved: synthesis quality, custodied from verifiable source to published prose. The parser work continues — a second round of upstream contributions to edgartools is in progress — and there are new features deep in the works that we're excited to show you when they're ready.

JULY 31, 2026

The initial dataset is complete.

New content is generally on pause. The companies we set out to cover are covered, and rather than racing to add more, we're turning back to look hard at what we've built. You can still put a new company in the queue: become a supporter, search for any ticker, and if we don't cover it yet its page will offer to synthesize it for you.

Why pause?

Because we found problems worth fixing first. In July we audited every document in the archive against current parser behavior. The sweep confirmed 277 findings across 189 filings — mostly sections stored far longer than their source should allow, plus a smaller set holding text from the wrong section of the filing entirely.

●  DOCUMENT AUDIT · JULY 24, 2026
Documents scanned
31,644
Candidates examined
1,089
Findings confirmed
277
Filings affected
189
Of the 277: 211 oversized sections, 66 wrong-section captures.

These oversized sections also drive most of our remaining external API use, so cleaning them up is both a correctness fix and a step further toward local-first synthesis. Faithfully custodying context — from verifiable source, through the LLM chain, into published prose — is the point of this project. Synthesis features stay paused until the sweep runs clean.

Along the way we've contributed parsing fixes upstream to edgartools, the open-source library that reads EDGAR for us: #906, #916, #921.

Expect styling updates as the site marches toward an ever more consistent design. And an open source cli so you can recreate similar synthesis at home. If symbology interests you: please show us your support!

In the meantime

If you're curious what else we do: