Where olab sits.
Two answers, and they point different ways. On what a product can do, olab now leads its segment. On what has been proven with real crews, it is last on this page. Both below, in that order.
Two answers, and they point different ways.
In September we ranked fifty-one things this segment's products can do and olab could not. All fifty-one are now in the build. That settles one question and leaves the other wide open.
The build half is done. The mission itself is not yet proven. On what the product can do, olab now leads its segment and holds a record-integrity position nobody else ships. On what has been shown to work for real people, it is last: no customer, no measured hour saved, no crew that has held the phone, no hosted region, no store listing, no certificate. Both are true at once — and only the second one can move from here.
Where olab leads today
- The second person. AI proposes with a confidence and a first-class “unclear”; the submitter can never be the confirmer; the human is your own named engineer, not a vendor's data-operations desk.
- The HV record. Safety documents with real earthing and lock semantics, switching programmes countersigned by three hands, ITRs typed to clause and unit, readings imported from the instrument, and a calibration register that reaches back and invalidates what a failed certificate touched.
- Money that cites records. Every pay-application line carries the ids of the records that earned it, and a claim above the verified quantity is refused by name.
- Engagement a works council can read. Self-facing rings, roster-aware streaks, crew rings with a floor of five, no leaderboards, no per-person scores, and a per-tenant kill switch. No competitor in this study ships any engagement mechanic at all.
Where olab is behind — plainly
- No customer. Zero. Every product on this page can point at a job that ran on it; olab cannot.
- No measured saving. Not one hour has been observed saved on a site. The calculator on this site is a model with cited inputs, and it says so.
- Nothing hosted, nothing in a store. Offline sync, push and the build profiles are written and green in tests; no region is running and nothing has been submitted.
- No certificate. The controls are built and mapped to SOC 2 and ISO 27001 control by control; the audits have not been bought.
Discipline is not a usability result and a test is not a customer. Read the left column as what the product is capable of, and the right column as everything a procurement checklist would ask for and not get.
Four ways sites run today.
Dimension by dimension against the three things olab actually replaces: the broad suites, the snag apps, and the spreadsheet-and-chat stack most packages really run on.
| Dimension | olab | Field management suites | Snag & punch-list apps | Spreadsheets, email & chat |
|---|---|---|---|---|
| Built around | Verified work records — and the instrument that measured them | A broad project and cost platform | Plans, tasks and tickets | Whatever each person kept |
| Where verification happens | In the capture moment, against that task's own checks | After capture, in a review step | After capture, by someone reading the photo later | In a conversation, if at all |
| AI posture | Labelled and named on the record, abstains as a verdict of its own, assesses work and never workers, switchable off per tenant | Platform-wide AI features applied to records | Typically search, drafting and summaries | None |
| Separation of duties | By construction — reviews, snags, receipts, releases, closes and even bulk actions each run the same refusal | Available through configurable workflow | Through statuses and assignment | By convention, and by memory |
| Test records | Typed to units, limits and clause references; a placeholder limit never passes; readings arrive from the instrument | A document attached to a task | Not what they set out to do | A scanned sheet in a folder |
| Per-person daily alignment | Rings — deterministic, self-facing, never ranked, switchable off | Dashboards, generally aimed at managers | A personal task list | A phone call at seven in the morning |
| How progress is claimed | A located quantity booked by one person, approved by another, complete only once QA closes | Percent complete entered against schedule or cost lines | Not what they set out to do | A percentage somebody types into a cell |
| How a claim is justified | Each pay-application line carries the record ids that earned it; above the verified quantity it is refused | Yes, and a substantial part of the price — but the line is a number, not a citation | Generally out of scope | A separate file, and a separate truth |
| Evidence at handover | A twelve-section turnover package the client accepts on the portal, with COBie and CMMS snapshots | Strong, once configured | Report exports per ticket set | Assembled by hand at the end |
| Pricing model | Free capture-only field seats, seats as the invoice unit, verified records as the meter — published, and recommended rather than decided | Annual contracts, often scaled to construction volume | Per-user tiers | No licence cost, and no audit trail |
| Deployment today | A preview build. Nothing hosted, nothing in a store, no certificate — the four blockers are written, not deployed | Running in production regions, with the certificates procurement asks for | Installed from the stores, offline in a basement | Already on every phone in the crew |
Categories rather than named products in this table, because capability varies widely between vendors inside each column — and the mature suites do a great deal that olab does not. The named products are marked one by one below.
Capability, product by product.
The six products a specialty contractor will actually shortlist beside olab, marked capability by capability. These are capability marks, not grades: our own reading of each vendor's public documentation, they cut both ways, and four of olab's rows are honestly empty until someone opens an account, buys an audit or signs a customer.
| Capability | olab | Procore | Autodesk Build | Fieldwire | Novade | Dashpivot | Dalux |
|---|---|---|---|---|---|---|---|
| AI evidence with a disclosed abstention state | “unclear” is a verdict, not a silence | agents draft, a person approves; no confidence published | transparency cards; no abstention | threshold hidden — suppressed, not disclosed | “humans make the final call” | low-confidence fields flagged | |
| Submitter ≠ reviewer enforced by construction, not configured | configurable workflow | statuses and assignment | |||||
| Calibrated-instrument register that invalidates records retrospectively | ISO 9001 7.1.5, enforced at confirm | ||||||
| Test sheet bound to a standard's clause, with unit-aware limits | 17 sheets; licensed limits ship as placeholders | ||||||
| Instrument export ingested with its measured values (PowerDB, Fluke, OMICRON) | every mapping confirmed by a person | ||||||
| Electrical safety documents and switching programmes with real earthing semantics | PTW · sanction for test · limitation of access · EEWP | 14 permit types with credential gating; no switching | permit templates | ||||
| Commissioning levels L1–L5 with an integrated-systems-test record | gates released by a named party, never the starter — see the note below on the Cx specialists | ||||||
| Pay application whose lines cite the records that earned them | a claim above the verified quantity is refused | pay apps and change orders, uncited | Budget, lead-company only | ||||
| Engagement mechanics that a European works council can sign | self-facing, no leaderboards, kill switch, EN + DE pack | ||||||
| Public API, webhooks and an MCP server for agents | REST + 19 events + MCP; no marketplace listing | a marketplace of hundreds | |||||
| Offline capture that keeps working with no radio | outbox + media queue built and tested; hosted sync unopened | cache-only — its top field complaint | |||||
| Apps you install from the App Store and Google Play | build profiles ready; submission waits on the accounts | ||||||
| SOC 2 Type II, ISO 27001 and a hosted region under contract | controls built and mapped; audits not bought | EU zone scheduled | ISO 27001/27018, SOC 2 Type 2 | ||||
| Customers running jobs on it | none — a pilot has to earn this | ||||||
| A published price a twenty-person sub can read | published — and marked recommended, not decided | construction-volume pricing | free tier, then $39 | free Field tier |
Two honest caveats on the marks. The dedicated commissioning products are not in these columns — CxAlloy, Facility Grid and CxPlanner do own parts of the commissioning row, and CxPlanner's flat $5,160-a-year company plan is the price any company-wide plan of ours has to be argued against. And a mark reading not verified means our research could not confirm it either way, not that the vendor lacks it.
Everyone transcribes. Nobody verifies.
Voice-to-notes went mainstream in 2025 and 2026 — most of the products above will now turn a walked site and a spoken sentence into tidy text, and that is genuinely useful. Several will let an AI draft a record and a person approve it. What none of them publishes is the pair that makes a record hold up: a disclosed abstention state — the model allowed to say “unclear” — together with a rule that the person who submitted the work can never be the person who confirms it, the confirmer being your own named engineer rather than the vendor's data-operations desk. That is roughly a twelve-month position before “we show confidence” is table stakes, and it is the ground olab is built on.
And a second gap, wider and older: nobody owns the HV record.
Instrument software produces standards-assessed reports in a silo keyed to the asset. Project platforms own the WBS, the hold points and the handover but cannot ingest a measured value. Across the eleven commissioning products in the September study, three absences were universal: no unit-aware tolerance engine that evaluates a reading against an acceptance value, no calibrated-instrument register with expiry — section 8 of a standard test pack is calibration certificates and every product treats it as a file someone uploads — and no object binding a test step to a standard's clause. Meanwhile four standards that drive this segment's paperwork were rewritten in eighteen months, so every certificate library in the market has to be redone within a year. Those three absences are three of the things olab built first.
Findings from our own competitive study of forty-plus vendors across eight research streams, 3 September 2026, re-scored 5 September against the current build.
Five things you can buy from them today. Not from us.
- A hosted service. They run in production regions with a contract and an uptime page behind them. Olab runs on the machine you put it on: offline sync, the media queue and the region switch are written and tested, and no region is running. on request
- An app in the stores. Theirs install from the App Store and Google Play and work in a signal-dead basement. Olab's build profiles, permissions strings, deep links and push wiring are done and verified; submission waits on the developer accounts. with your accounts
- The certificates. SOC 2 Type II, ISO 27001, and in Novade's case Cyber Essentials Plus too. Olab's SSO, SCIM, session control, audit export and posture statement are built, and every control is mapped to its clause, its file and its test in the repository. The auditor turns that into a certificate; nobody has been hired. on request
- A marketplace. Procore's alone lists hundreds of certified connections. Olab ships a documented REST API, nineteen signed webhook events, an MCP server, and connectors for Procore, Autodesk ACC, BCF, P6/Asta, QuickBooks and Aconex — written against public documentation, with no partner listing behind them. with your accounts
- A track record. Years in production, named references, and a number for how much time they saved somebody. Olab has a seeded demo world, 1,105 tests and no customers. Every figure on this site is a model or a count of what the build does — never an observed result.
If your next job depends on any of those — a hosted region on Monday, a procurement checklist that wants a SOC 2 report, an app your crew installs from a store — a mature suite serves you better today, and we would rather say so here than in month two of a pilot. If the thing you cannot buy anywhere is a record that names the second person and proves the instrument was in calibration when it measured, that is the gap olab exists to fill.
A table is a claim.
A register is proof.
Run a pilot on one package — your own checklists, your own bill of quantities, one month's measurement. Print the register at the end and judge that instead of this page. Then measure the minutes: the first figure this site is allowed to quote is one you took on your own site. The price list is published, and marked recommended rather than decided.