// 01
Depth, not hot takes
Every piece is long-form and illustrated: architecture diagrams, screenshots, real numbers. If we haven't run the thing ourselves, we say so up front.
// resolution journal — est. 2026
A working record of the systems small companies actually run on: the software underneath, the brands built on top of it, and the people who decide whether any of it survives. Every piece is argued from primary sources or from production data we generated ourselves — and revised in public when the evidence moves.
// why this journal exists
Resolution is a process advisory firm that also ships its own software. Every week we evaluate tools, rebuild internal systems, negotiate with clients and watch brands succeed or stall in real time. Most of that knowledge used to die in a Notion page. This is where it goes now — in English, for an audience that runs into the same problems in any market.
// 01
Every piece is long-form and illustrated: architecture diagrams, screenshots, real numbers. If we haven't run the thing ourselves, we say so up front.
// 02
Written for the person who is simultaneously the founder, the finance department and the IT support. No enterprise theater, no tooling you can't afford.
// 03
Ads and affiliate links keep the lights on; they never buy a conclusion. Sponsored content is labeled, and we publish the negative findings too.
// featured
Four production apps as single HTML files: no npm, no bundler, no build step. What the architecture eliminates, what it costs, and the four conditions under which we would abandon it.
// the desks
// no article matches that
A trial made a procedure longer and more uncomfortable, and patients remembered it as better. What that means for anyone who delivers a service.
Always anchor first is half right. The two conditions where it backfires, and the documented technique that cancels the other side's anchor.
An airline argued its chatbot was a separate legal entity responsible for its own actions. It lost. What that means before you deploy an agent.
The passwordless numbers are good and they come from people who sell authentication. The part left out is account recovery, which is where every deployment breaks.
A randomised experiment trained managers in twelve behaviours and raised how charismatic colleagues rated them. The tactics, and the inverted-U nobody quotes.
Seventy years of research on seating and posture, and almost none of it in what you read. What the studies support, and the full power-posing replication record.
Pricing is moving from per-seat to per-use and per-outcome. Written for the people paying the bill: who pays more, who pays less, and six clauses worth having.
Most people using AI at work brought it themselves, on a personal account, into a product that trains on what they paste. What the terms actually say, and why banning fails.
Transcription inverted the default: meetings are on the record unless someone objects. What that changes in behaviour, in liability, and what to decide first.
Nobody decides to spend what they spend on software. A four-step audit, the four categories every stack falls into, and where the money hides.
People agree sincerely and then do nothing. The three mechanisms that dissolve intention, and the thirty-second technique that closes part of the gap.
Video does not thin the signals. It removes them and adds artefacts that look like behaviour — latency that reads as hesitation, gaze that is lens geometry.
Developers report feeling substantially faster with AI assistance. One controlled trial measured the opposite. Where the tools win, where they charge you later, and how to measure it.
Copying posture increases liking — until it is noticed, at which point it does the opposite. Why deliberate mirroring fails and what holds up at minute thirty.
Gaze aversion is the most widely believed cue to deception and one of the weakest. The meta-analytic evidence, and the NLP claim that was specific enough to be tested and failed.
Small business software loses to a file someone already understands. What the spreadsheet is genuinely better at, and the only three places it fails.
Row-level security replaces the application layer for multi-tenant SaaS. What makes it safe, the three ways we have seen it fail, and what it cannot do.
The feature table is not the decision. Three questions that actually decide it for a one-person team, ending with what leaving costs.
In text, only latency and length survive as signals. How to read both, the three ways written communication misleads you, and when to change channel.
Snap judgments are accurate about warmth and delivery, and poor about competence. What that means for interviews — and what beats the interview entirely.
Open questions followed by silence outperform every clever technique. Why leading questions contaminate the answer, and how to pause without being strange about it.
Disengagement precedes refusal and is far easier to read than deception. Four observable changes, in the order they arrive, and what to do inside that window.
The 7-38-55 rule came from two 1967 studies on single recorded words with contradictory affect. What they measured, and why the generalisation fails on arithmetic alone.
Agents are not persuaded by a good interface. When the operator of your product is no longer the party with the goal, value moves to schema, authorization and reversibility.
Humans detect deception at barely above chance, and training raises confidence without raising accuracy. Why gesture dictionaries fail, and what replaces them.
Four production apps as single HTML files: no npm, no bundler, no build step. What the architecture eliminates, what it costs, and the four conditions under which we would abandon it.
// the human layer
Every system terminates in a person. A contract is signed by someone weighing risk they will not state out loud; a hire is decided in the first four minutes; a price stops being real the moment hesitation enters the room. This desk treats that layer as infrastructure — reading it through the empirical literature on nonverbal behavior, deception detection and decision-making, separating what replicates from what merely sells books.
Baselines & deviations. Establishing what "normal" looks like before you interpret anything.
Clusters, not tells. Why a single gesture is noise and three aligned signals are data.
Questions that open people up. Elicitation, silence, and the cost of leading questions.
Negotiation under pressure. Reading hesitation, false urgency and the moment a price stops being real.
// weekly dispatch
The week's analysis, one tool we actually tested, and one behavioral pattern worth practicing. No digest of links, no "10 AI tools you must try". Unsubscribe in one click.
// topics
Disclosure: Resolution is a participant in the Amazon Services LLC Associates Program, an affiliate advertising program. As an Amazon Associate we earn from qualifying purchases. Some links in our articles are affiliate links and this site displays advertising. Neither affects which tools or books we recommend, nor the conclusions we publish — reviews are based on tools we run in production and on research we have read.