Blog

Engineering notes on how sf-intelligence works under the hood - the offline reasoning model, the trust model, and what changes in each release. Read-only, grounded, deterministic.

0.3.1 · 2026-08-22

0.3.1 — the tools audit themselves

A correctness release against tools that were already shipped: 39 verified defects across the permissions, schema and automation families, each reproduced against a real org - a reachability walk that called 85 of 186 Apex classes dead and certified itself complete, a stale retrieval that overwrote the current one so a removed permission still read as granted, and a response trimmer that had started shortening the list of things it had not checked. Plus the 18 regressions those fixes introduced, and the gates that now catch both.

› read the post

0.3.0 · 2026-08-10

0.3.0 — safer defaults, and claims that have to prove themselves

The release that narrows what the server exposes by default: a 19-tool core profile with sfi.run_analysis as the gateway, live consent as an org-bound expiring grant, network egress off, the refresh bug that could delete a vault and report success, and a generated manifest that turns capability drift into a build failure. Six breaking changes and the upgrade path.

› read the post

admin · 2026-07-29

What does this Flow do?

How to read a Salesforce Flow you did not build - identifying the trigger type, why entry conditions matter as much as the trigger, how decision outcomes actually evaluate, tracing every field it writes, the undefined-order problem with two record-triggered flows on one object, and what static reading genuinely cannot tell you.

› read the post

admin · 2026-07-28

Can I delete this Salesforce field?

Population tells you how much data is in a field; it never tells you what depends on it. A dependency-first cleanup method - what blocks a delete, what breaks silently afterwards, the four verdicts that make the output actionable, and the blind spots no static analysis closes.

› read the post

developer · 2026-07-21

Salesforce order of execution, explained

The full save-order sequence in plain English - where before-save flows, validation, triggers, workflow, and after-save flows fire, the classic bugs it causes, and how to see your own org's automation order and collisions offline.

› read the post

0.2.0 → 0.2.1 · 2026-07-20

How sf-intelligence reasons about your org

The deterministic, offline reasoning model behind sfi.interpret - a two-graph join of your org vault against a 143-concept, 195-rule org-independent Concept Model that returns cited, confidence-tiered structural claims. Plus what shipped in 0.2.0 and 0.2.1.

› read the post

› full capability mapget started