The review may be current. The product is not.

Type Veras into a search box this morning and the shelf looks oddly preserved. A 4.0 explainer sits beside a 4.5 upgrade video, both offered as if they describe the thing a practice could buy today.

They may be accurate. That is the problem. Accurate history can still become bad procurement advice.

What changed after the reviews in the search results?

The current Chaos Veras product page describes a product that spans sketches, images, prompts, 3D models, stills, animation and editable 3D scenes. It names GPT-6 Astra, Nano Banana and Opus 5.5 among more than 30 models. It also presents custom Nodes workflows, image-to-3D, image-to-video and integrations across Revit, SketchUp, Rhino, Archicad, Vectorworks, Forma, Allplan, Enscape, V-Ray and Corona.

That is materially broader than a review framed around one earlier release's rendering engine or prompt behavior. It does not make the older review false. It changes the object under review.

Chaos's June 2026 six-tool comparison itself lists Veras with Nano Banana 2 plus Stable Diffusion, seven BIM or CAD hosts and a starting annual price. The live product page now advertises a larger model roster and different plan details. Even two pages from the same company can describe different product moments.

A product name is no longer a sufficient citation

Cloud AI products change without a boxed release arriving on a desk. Engines rotate. Credit costs move. A plugin adds a host. A once-flat image workflow gains a graph and a path back into 3D. The logo stays polite and still.

Every consequential claim needs a coordinate. “Veras preserved geometry” is incomplete. Which engine, which control setting, which host, which input and when? “Veras costs $29 a month” is also incomplete. Which region, billing term, plan and checked date?

This is why our current-version retest rule treats model identity as part of the result. A tool review without it resembles a material specification that omits the product code because the manufacturer name felt close enough.

Which claims expire fastest?

Price ages in weeks. A model name can age between publication and the first search crawl. Interface criticism lasts longer, but only when the interface is still there. A review's durable value usually sits in its method: what input it used, what facts it protected and how it recognized failure.

How should a practice read the old review?

Start by splitting observations from conclusions. “The reviewer supplied a Revit viewport and compared four outputs” remains useful context. “This is the best engine for design development” expires when the selectable engines or controls change.

Then run the claim through four checks. Confirm the feature still exists. Confirm the named engine can still be selected. Confirm the host and version remain supported. Confirm the plan gives your users enough credits and rights for the intended work.

Do not quietly substitute today's product into yesterday's evidence. Our release-provenance check exists because features are often attributed to the wrong version after several announcements blur together. A citation should tell readers what shipped, when and where the claim came from.

Version the workflow, not merely the software

“Veras 5” would still be too coarse for an office standard. A repeatable workflow should record host application and version, Veras build, selected engine, control values, prompt, reference inputs, source view, output dimensions and date. If Nodes is involved, preserve the workflow version too.

The engine matters because the wrapper is no longer the whole product. A single Veras interface can route work through models with different instruction behavior, geometry retention, latency and credit consumption. Our Nano Banana 2 throughput audit separates a faster lane from an equivalent result. Speed is welcome. Equivalence must be shown.

Record the job the workflow was accepted to perform. Concept variation, material study, image polish, video and image-to-3D do not share one definition of success. An old review may remain excellent evidence for one job and useless for another that the product acquired later.

Search ranking is not recency control

Search engines rank pages for many reasons besides product currency. A clear 4.5 video can remain popular because it answers a crisp question. A reseller's 4.0 post can retain links. Neither page owes a future buyer a warning that Chaos has since changed the product.

The buyer therefore needs a two-tab habit. Keep the review open for method and observed behavior. Open the current vendor page or documentation beside it for present capability, compatibility and commercial terms. When the two disagree, label the difference. Do not average them into a fictional middle version.

A dated review can also reveal product direction. Comparing its boundary with the current page shows what the vendor decided to absorb: more engines, graph workflows, video and 3D generation. That shift matters culturally. Veras is being presented less as one renderer and more as an AI work surface inside several design hosts.

Give every review a five-line product stamp

Publish a small stamp near the verdict: tested or reviewed date, product build, selected engine, host application and commercial plan. Those five lines do more work than a paragraph of adjectives. They let a reader decide whether the evidence matches the product on offer.

The stamp should distinguish documentation review from hands-on testing. If the writer read vendor pages but did not run the tool, say so. If outputs were generated, name the input, settings and preserved artifacts. Our claim-pedigree method keeps vendor statements, third-party observations and documented tests in separate lanes.

Add a “checked current” date when an editor revisits the page without repeating the test. That date means the links, plan and feature names were checked. It does not renew an old performance verdict. A test result keeps its original clock.

For office procurement notes, add one more field: replacement trigger. Retest when the default engine changes, the credit unit changes, a required host gains or loses support, or the workflow acquires a new input route. A quarterly calendar reminder is useful, but a product event is more precise.

This modest stamp also protects good criticism. When a vendor fixes the exact failure a reviewer documented, the older article can remain online as history instead of being awkwardly rewritten into a permanent verdict. Readers see the boundary. The record stays honest.

ArchiGen's take: every verdict needs a product clock

Architecture practices already understand revision control. Nobody issues a detail because the search result looked authoritative while hiding which drawing set it came from. AI software deserves the same basic discipline.

Keep older reviews. They contain useful failures, good questions and evidence about how quickly the category moves. Strip them of present-tense authority until the current product confirms each material claim.

The product name is the folder. The version is the drawing.

Evidence note: checked 28 October 2026 against the live Chaos Veras product and pricing page and Chaos's six-tool comparison, last updated 18 June 2026. Today's ArchiGen sweep surfaced Veras 4.0 and 4.5 coverage alongside the current page. Product capabilities and prices cited here are vendor statements, not ArchiGen hands-on findings.