Every result on this page comes from Drevanto's own certification and verification program. This is not a third-party audit and we do not present it as one. The construction-reference campaign used independently sourced authoritative references, but Drevanto performed the evaluation and scoring.
Last verified August 29, 2026 · Instant benchmark certified 2026-08-21 · Catalog v3
Verification overview
What we verified, and who verified it
Internal program, stated plainly: Drevanto designed, ran, and scored these gates.
The reference corpus was sourced independently of the product team; the comparison itself was run by Drevanto.
Results are reported with their limits attached — coverage, scope, and hardware are all bounded.
Sanitized public summaries are downloadable; private evidence stays on protected release surfaces.
0.028 sec
Instant median, supported jobs
250
Independent construction references
100,000
Modeled transitions, 0 failure signatures
20 / 20
Supported scopes coordinated in one project
0
Confirmed defects remaining
2
Beta 2 candidates validated (Windows, Android)
How Drevanto is tested
Testing and reliability, stated with limits
Drevanto runs its own certification program before a beta build is released to testers. The proof points below are the results we can publish safely, each with the scope it actually covers. Nothing here is a third-party audit, and nothing here is extrapolated beyond the run it came from.
PASS2026-09-03
Previous Windows Build 15 readiness closure
The previous signed Windows Beta 2 installer (1.2.0-beta.2+15) was installed fresh into an ordinary Windows sandbox with runtime prerequisites provisioned automatically, launched to the Dashboard, took a synthetic supported local estimate through to Job Ready, kept that work after close and reopen in an isolated profile, and recorded no crash events.
Scope: Historical Build 15 evidence only; it does not certify current Build 17. Sign-in certification on Build 15, uninstall, and full upgrade runs from older Windows builds were not covered, and one test environment is not every Windows device. Detailed gate
PASS2026-09-03
Accepted Android v12 physical-device run
Acceptance recorded by September 3, 2026 (report date, not the test date): the accepted Android v12 beta candidate passed a physical-device run: data-preserving in-place upgrade with original app data intact, startup and force-stop/relaunch, enabled app and Guided-input views at normal and enlarged text, customer/estimate/job persistence, Job Health 2.0, job-operation views, and an Android camera capture that was still attached after force-stop and relaunch — with no crash, native crash, ANR or database failure recorded during those checks.
Scope: A bounded result on the devices tested, not a crash-free guarantee or an all-device certification. Camera means Android's system capture flow launched from inside Drevanto, not media cloud sync. Newer Android artifacts are not certified by it. Detailed gate
PASS2026-09-03
Cloud acceptance in isolated local staging
Migration replay, representative client and API contracts, tenant isolation, bounded concurrency and backup/restore all passed Stage 10 acceptance in an isolated local staging environment using synthetic data, with no failed checks. Offline and reconnect behavior, duplicate prevention and local media backup/restore were accepted earlier in the same program.
Scope: Isolated local staging with synthetic data only. Not hosted or public cloud, not production scale, not live billing, and not external cloud sign-in, sync or cross-device recovery. Detailed gate
PASS2026-08-29
Windows packaged-installer certification (archive)
The then-current Windows Beta 2 installer (1.2.0-beta.2+5, Authenticode publisher ARMM General llc) passed source binding, integrity checks, isolated installation, cold launch, sign-in and session behavior, the main workspaces, numeric input, the composed garage/addition flow, estimate generation, save/reopen/delete of a temporary estimate, close and relaunch, crash monitoring, and rollback/uninstall safety.
Scope: Historical archive for an older Windows build, kept for the record. The later Build 15 closure is also historical; current Build 17 full manual acceptance remains pending. Detailed gate
PASS2026-08-29
Android physical-device certification (archive)
The then-current Android Beta 2 candidate (1.2.0-beta.2+6) was retested on a physical Samsung SM-S938U running Android 16: repeated cold launch, authentication and session persistence, Dashboard, Customers, Estimates, Smart Estimate Builder, multi-digit input, the composed Guided flow, readiness gating, generation, save and reopen, Jobs, Schedule, Reports, Settings, force-stop and relaunch, and crash monitoring. An earlier cold-launch crash was confirmed gone.
Scope: Historical archive for an older Android build on one handset and one Android version. The accepted current Android candidate is v12 above. Detailed gate
PASS2026-08-29
Estimator transition closure
The modeled estimating workflow was closed out with 0 untested cells, then stress-run for 100,000 long-session transitions across 18 archetypes with 0 failure signatures and 0 external provider calls — no leaks, deadlocks, stuck states, or slowdown.
Scope: Closes the modeled workflow. It does not model every sequence a person could invent outside that model. Detailed gate
PASS2026-08-28
Construction reference benchmark
Against 250 independently sourced authoritative construction references across 92 documents: 100.00% scope recall, 87.91% scope precision, 0.00% median quantity difference, 0 professional-boundary violations, and 0 confirmed Drevanto defects remaining.
Scope: References were sourced independently; Drevanto ran the evaluation. Quantity comparisons apply only where references establish explicit normalized quantities. Detailed gate
PASS2026-08-29
Guided input and readiness controls
Real Guided UI controls were exercised character by character and through composed multi-part projects: 0 premature Generation Ready states, 0 readiness mismatches, 0 orphan blockers, and 0 dead controls. Partial or invalid numbers never became confirmed project facts.
Scope: Covers the governed Guided input path in the certified Beta 2 candidates. Detailed gate
PASS2026-08-21
Instant Estimate speed benchmark
520 certified generations of supported jobs reached an editable estimate in a 0.028 second median (rounded); the measured 95th percentile was 48.008 milliseconds (approximately 0.048 seconds), with 0 failures and 0 generative-AI provider calls.
Scope: Supported job types on the hardware in the methodology only. Not a claim about unusual scopes or every device. Detailed gate
What "supported" covers
Drevanto's Smart Estimate coverage is bounded: 20 supported estimating scopes and 18 project archetypes get progressive clarification and readiness gating on unresolved facts. That is not 20 fully automated trades and not a universal set of deterministic assemblies — plumbing and electrical have no general-purpose Instant assembly, complex or unsupported work stays on a hybrid or manual review path, and missing required facts block readiness rather than being filled in for you. Cross-workflow reuse of every answer, shared-cost de-duplication across scopes, side-by-side alternatives and complete company presets are partial or unsupported today.
Not covered by the published record
These areas are not claimed on this site because no approved, public-safe result exists for them yet. Absence here is a statement of scope, not a defect report.
Separate ANR (application-not-responding) statistics — no crash, native crash, ANR or database failure was recorded during the Android v12 checks, but no standalone ANR rate is published.
Full manual acceptance of current Windows Build 17, including current-build sign-in, uninstall and full upgrade runs from older Windows builds.
Hosted or public cloud operation, external cloud sign-in and sync, universal cross-device recovery, production scale and live billing — cloud acceptance was reached in isolated local staging with synthetic data only.
Physically tested cross-device sync, and cloud-durable job media. Photos and attachments are local device records.
Any certification of newer Android artifacts beyond the accepted v12 candidate, and any current Google closed-test review outcome.
Verified real contractor outcomes, time-saved studies, measured margin improvement, and head-to-head accuracy comparisons against other products.
Native browser zoom, physical-device and assistive-technology testing of this website itself, which is reported separately from product testing.
Beta 2 verification evidence
The gates behind Beta 2
Each gate below ran to completion with a PASS result. Numbers are aggregates — private evidence, fixtures, and internal implementation detail stay off this page. Dates on each gate are the historical run dates and are not restated for later builds.
Current invite-only release · September 17, 2026
Windows Beta 2 Build 17 (1.2.0-beta.2+17), signed by ARMM General llc, is distributable through the private download flow for entitled testers who sign in. Full manual acceptance and physical-device verification remain pending.
Android version code 18 has been submitted for review on the existing private Closed testing — Founding 50 track, with Google quick checks finished and version 18 now in review with no promised completion time. It is not available yet. Version code 17 remains the build available to selected testers until version 18 is approved, and the requested rollout is 100% of that same private track. Full manual acceptance remains pending. This is not a public, Open Testing, or Production release.
No iOS app. The historical gates below cover earlier Beta 2 candidates on their stated dates; they are not acceptance evidence for Windows Build 17 or for Android version codes 17 or 18. Full manual acceptance of the current candidates remains pending. Installer details are shown only to entitled testers on the download page.
Independent Construction Benchmark Lab v3
PASS2026-08-28
250-reference construction benchmark
Drevanto's output was compared against independently sourced authoritative construction references until meaningful coverage saturation was reached. Drevanto performed the evaluation.
Accepted independent references
250
Accepted source documents
92
New references / new documents
139 / 43
Previously certified corpus retained
111 references
Tier C sources used as truth
0
Scope recall
100.00%
Scope precision
87.91%
Quantity median absolute % difference
0.00%
Quantity p90 absolute % difference
30.00%
Professional-boundary violations
0
Confirmed Drevanto defects remaining
0
Quantitative comparisons only cover cases where authoritative references establish explicit normalized quantities. Legitimate construction methods can differ, so a difference is not automatically an error on either side.
One bounded governed repair was independently corroborated during the campaign. It changed scope-language recognition only — it did not change quantities, prices, professional boundaries, or protected baseline artifacts.
Coverage saturation means additional references stopped surfacing new behavior; it does not mean every construction reference in existence was consulted.
Guided numeric input + composed projects
PASS2026-08-29
Guided input and composed-project certification
Real Guided UI controls were exercised character by character, then driven through composed multi-part projects end to end.
Accepted governed entries
2, 24, 100, 24.5, 24.25
missingInformation at generation
0
Premature Generation Ready states
0
Readiness mismatches / orphan blockers
0 / 0
Dead controls found
0
Partial numeric text never created a confirmed project fact or triggered downstream derivation — an explicit commit was always required.
Invalid, negative, non-finite, zero-where-invalid, and out-of-range inputs were rejected without mutating already-confirmed facts.
Attached-garage addition and existing-roof tie-in scenarios both completed through the real Guided UI, and information already supplied in the description was not re-asked.
Professional/design and field-verification boundaries for existing-building interfaces were preserved throughout.
Exhaustive modeled transition closure
PASS2026-08-29
100,000-transition closure run
The modeled workflow was closed out state by state, then stress-run across long sessions to look for drift, leaks, and stuck states.
Modeled states / registered operations
20 / 68
Tested modeled edges
983
Untested cells
0
Pairwise obligations
1,197
High-risk three-way / four-way
42 / 4
Production-probe failures
0
Long-session transitions
100,000 across 18 archetypes
Failure signatures
0
Duplicate generation attempts prevented
19,800
Persistence / generation / next-question checks
10,000 / 15,000 / 5,000
External provider calls
0
The long-session run reached state-coverage saturation with no pathological slowdown, runaway memory growth, deadlock, or unbounded queue growth.
This closes the modeled workflow. It does not model every possible sequence a person could invent outside that model.
All-Trade Big-Box Final Boss
PASS2026-08-29
All-trade coordination fixture
A deliberately oversized synthetic integration fixture — a fictional 120,000 sq ft large-format retail build — run to check that all supported scopes coordinate at once.
Supported scopes detected and coordinated
20 of 20
Responsibility rows / takeoff rows
20 / 35
missingInformation
0
Generation count
1
Final failures
0
Reproducible clean semantic runs
2
Reviewer, Margin Guardian, Job Ready, ownership/duplication, mutation, save/resume, and cross-trade interface gates all passed.
The fixture is fictional: no real company, plan, price, or confidential information is involved.
This is a synthetic integration fixture. It is not proof that every real project or every real price is correct.
Final release-candidate validation
PASS2026-08-29
Physical device and packaged-installer gates
The exact Beta 2 release candidates — not rebuilds — were validated on a physical Android handset and as a packaged Windows installer.
Android candidate (historical, 2026-08-29)
1.2.0-beta.2+6
Android physical device
Samsung SM-S938U · Android 16
Windows candidate (historical, 2026-08-29)
1.2.0-beta.2+5
Windows publisher (Authenticode)
ARMM General llc
Forbidden packaged files found
0
Android: package, version, signing, archive, and source binding passed, along with Play upload-certificate compatibility. Bootstrap/plugin registration was repaired and deterministically verified before retest.
Android physical retest covered repeated cold launch, authentication and session persistence, Dashboard, Customers, Estimates, Smart Estimate Builder, character-by-character multi-digit input, composed attached-garage/addition Guided UX, description consumption, readiness and build gating, successful generation, save and reopen, Jobs, Schedule, Reports, Settings, force-stop and relaunch, crash monitoring, and cleanup of temporary test data.
The earlier cold-launch plugin-registration crash was confirmed gone on the physical device.
Windows: the exact packaged installer passed source binding, integrity checks, a valid Authenticode publisher, isolated installation, cold launch, authentication and session behavior, main workspaces, numeric input, the composed garage/addition flow, generation, save/reopen/delete of a temporary estimate, close and relaunch, crash monitoring, and rollback/uninstall safety.
Release artifacts were checked for archive integrity and forbidden packaged files. Private hashes stay on protected release evidence and download surfaces, not on this public page.
These gates cover the certified beta candidates. They do not cover all possible hardware, OS builds, or field conditions.
Previous certified release · Windows Beta 2 Build 15
PASS2026-09-03
Windows Build 15 beta-readiness closure (historical)
The previous signed Windows Beta 2 installer (1.2.0-beta.2+15) was installed into a fresh, ordinary Windows sandbox and taken through a supported local estimate end to end. Classified beta ready with explicitly disabled features, not feature complete. This evidence does not certify current Build 17.
Previously certified Windows build
1.2.0-beta.2+15 (Build 15)
Signed installer
Verified
Fresh isolated installation
PASS
Runtime prerequisites provisioned automatically
PASS
Supported local estimate to Job Ready
PASS
Close / reopen persistence
PASS
Crash events during the run
0
Covered: signed installer, fresh ordinary sandbox installation, automatic runtime prerequisite provisioning, Dashboard launch, one synthetic supported local estimate reaching Job Ready, close/reopen persistence, an isolated profile and local database, and crash-event checks.
Not covered by this closure: sign-in certification on the current build, uninstall, and full upgrade runs from older Windows builds. Those remain open, and the older archived Windows gate above is not a substitute.
An installer/executable version mismatch reported by the original collection harness was a documented harness defect, not an application failure; the exact executable was verified separately.
This is a bounded readiness closure on one test environment. It is not a statement about every Windows device, OS build, or field condition.
Android Beta 2 version code 12 — the separately frozen, accepted beta candidate — passed a founder-accepted physical-device run. Acceptance was recorded by September 3, 2026; that is the report date, not the date the device run itself was performed, which is not published here.
Covered: data-preserving in-place upgrades through v12 with original installation and app data preserved; startup and force-stop/relaunch; enabled app and Guided-input views at normal and enlarged text; repaired narrow and large-text headers; customer, estimate and job persistence; Job Health 2.0; job-operation views; launching Android's in-app camera capture from inside Drevanto and one captured attachment still present after force-stop and relaunch.
The camera flow opens Android's system camera capture from inside Drevanto. It is not a custom embedded camera and it is not media cloud sync.
Android v12 is frozen and accepted separately. It is not source-equivalent to Windows Build 15, and newer Android artifacts — including version codes 17 and 18 — are not certified by this evidence.
Android version code 17 was later published to the same private Founding 50 closed track, and version code 18 has since been submitted for review on that same private track. Version code 12 remains historical evidence only; full manual acceptance of the newer versions remains pending.
This is a bounded result on the devices tested. It is not a crash-free guarantee and not an all-device certification.
Current Android submission · single-device diagnostic
One reviewed diagnostic build of the Android large-catalog startup repair was run on the owner's own handset. This is a narrow startup measurement only — it is not certification of all features and not certification of a Play-installed production version code 18.
Device
Samsung S25 Ultra (owner's device)
Fresh-cache cloud initialization
15.707 seconds
Dashboard visual inspection
Green cloud indicator, restored records
Devices covered
1
Measured on a reviewed diagnostic build, not on the Play-distributed version code 18 bundle, which is still submitted for review on the private Founding 50 closed track.
Covers first-run startup and restore behavior only. It does not cover the six original QA repairs, which still await full manual acceptance, and it is not a broad physical-device or live-integration certification.
Historical August and September gates on this page do not validate this repair, and this diagnostic does not validate them.
Current release · isolated local staging
PASS2026-09-03
Cloud Stage 10 acceptance in isolated local staging
The cloud service layer passed Stage 10 acceptance in an isolated local staging environment using synthetic data only. All recorded checks passed with no failed checks.
Environment
Isolated local staging, synthetic data
Migration replay
PASS
Representative client / API contracts
PASS
Tenant isolation
PASS
Bounded concurrency
PASS
Backup and restore
PASS
Failed checks
0
Also accepted earlier in the same program (September 1): synthetic, provider-neutral offline and reconnect behavior, duplicate prevention, media transfer, and local database and media backup/restore.
This does not mean hosted or public cloud is available. It is not production scale, not live billing, and not external cloud sign-in or sync. Public Cloud sign-in and sync remain unavailable in the current beta.
It also does not prove physically tested cross-device sync.
Environment details, internal structure, throughput figures and security mechanisms from this bounded run are deliberately not published.
Historical archive · September 1, 2026 campaign
PASS2026-09-01
September 1 estimator campaign and suite run
A dated internal campaign run on the September 1 source state: 100,000 estimator transitions with no failure signatures and no external provider calls, alongside full automated suite runs.
Estimator transitions
100,000
Failure signatures
0
External provider calls
0
Application suite
806 / 806
Cloud suite (debug / release)
222 / 222 · 222 / 222
These are September 1, 2026 campaign results on that source state. They are not re-runs against Windows Build 15, and they are not a certification of every later build or every possible input.
This campaign is reported separately from the August 29 modeled-transition closure above. The two are different runs and their counts are never combined.
What this proves
Supported jobs resolve to editable estimates effectively instantly, with deterministic repeatability and no generative-AI provider calls.
Against 250 independently sourced authoritative references across 92 documents, scope recall was 100.00% and the median quantity difference was 0.00%, with 0 professional-boundary violations and 0 confirmed defects remaining.
Guided input is governed: partial or invalid numbers cannot silently become confirmed project facts, and composed multi-part projects complete through the real UI.
The modeled workflow is closed out with 0 untested cells, and a 100,000-transition long-session run produced 0 failure signatures with no leaks, deadlocks, or slowdown.
All 20 supported estimating scopes coordinate together in one oversized synthetic project, reproducibly, with review and margin gates intact. Supported scopes are bounded — that number is not 20 fully automated trades.
The previously certified Windows Beta 2 Build 15 installer passed a fresh-sandbox installation, a supported local estimate to Job Ready, persistence across close and reopen, and crash-event checks. This evidence does not certify current Build 17.
The accepted Android v12 beta candidate passed a physical-device run covering data-preserving in-place upgrade, persistence, enlarged text, Job Health 2.0, and a camera capture that survived force-stop and relaunch.
The cloud service layer passed Stage 10 acceptance — migration replay, tenant isolation, bounded concurrency and backup/restore — in isolated local staging with synthetic data.
What this does not prove
This is Drevanto's own internal certification and verification, not a third-party audit or an industry certification.
Quantitative reference comparisons only apply where authoritative references establish explicit normalized quantities. Legitimate construction methods differ, and a difference is not automatically an error.
The All-Trade fixture is synthetic. It does not prove that every real project, scope, or price is correct.
Nothing here certifies supplier-price accuracy, local code compliance, structural adequacy, or real-world job frequency and coverage.
Release gates cover the certified beta candidates on the hardware tested — not every device, OS build, or field condition.
Cloud acceptance was reached in isolated local staging with synthetic data. Hosted cloud, public cloud sign-in and sync, cross-device recovery, production scale and live billing are not available.
Android v12 is a frozen, accepted historical beta candidate. Version code 17 is the build available to selected testers on the private Founding 50 closed track and version code 18 is submitted for review on that same private track; neither is certified by the v12 evidence and full manual acceptance of both remains pending.
Windows Build 17 is signed and is the current invite-only installer, but its full manual acceptance remains pending; the Build 15 closure is not carried forward as Build 17 evidence.
Nothing here is a verified contractor outcome, a time-saved study, a measured margin improvement, or a head-to-head accuracy comparison against another product.
Contractor review still matters: output is editable, and pricing, codes, site conditions, and field verification stay under your control.
Instant Estimate speed benchmark
Certified 2026-08-21: how fast supported jobs become estimates
In 520 certified profile-mode generations across 26 supported Instant assemblies on Windows (profile/AOT) and an Android API 36 emulator, supported jobs became editable estimates in a median of 0.028 seconds (rounded); the measured 95th percentile was 48.008 milliseconds (approximately 0.048 seconds), with 0 generative-AI provider calls.
Results at a glance
0.028 sec
Median to editable estimate (rounded)
48.008 ms
95th percentile (measured, ≈0.048 sec)
520
Certified generations
0 / 0
Failures / generative-AI provider calls
Certified supported jobs only. Editable output; contractor pricing and field verification remain in control.
10 controlled generations per assembly per platform: 260 Windows profile/AOT samples and 260 Android API 36 emulator profile samples — 520 certified generations in total. No physical Android handset was timed in this campaign.
Timing starts after clarification, when Generate is pressed, and ends when the editable estimate is visible. Application startup and the clarification questions themselves are excluded.
Two timings were recorded separately: UI end-to-visible time and deterministic engine-internal time.
Exact measured values are reported in milliseconds or microseconds. Seconds figures (0.028 s median, approximately 0.048 s at the 95th percentile) are rounded for display.
Execution is deterministic and local. No generative-AI provider is called for supported Instant work.
Unsupported or unusual scope is not forced through Instant — it is flagged and routed to Drevanto's Hybrid/assisted review path. In the current beta that path is manual contractor review: AI-assisted drafting needs an external provider and no provider is enabled in the beta, so no generative-AI step runs.
All pricing and readiness checks used synthetic local certification rates: a $12.50 material baseline, $72/hour labor, and synthetic equipment rates only where applicable.
Certified results
UI end-to-visible time measures request to the editable estimate being visible. Engine-internal time measures only the deterministic engine — the two are recorded separately and are not interchangeable.
Windows (profile) — UI end-to-visible
Samples
260
Median (p50)
24.183 ms
p95
28.971 ms
Max
32.042 ms
Android API 36 emulator (profile) — UI end-to-visible
Samples
260
Median (p50)
32.777 ms
p95
52.883 ms
Max
123.81 ms
Deterministic engine — internal time
Samples
520
Median (p50)
262 µs
p95
1045 µs
Max
4775 µs
Integrity checks
Generation failures
0
Generative-AI provider calls
0
Repeatability hashes matched
78/78
Job Ready passed
520/520
Certified false-ready attempts prevented
520/520
Sample distributions
Counts from the sanitized 520-row certification sample file. The first two charts are UI end-to-visible timings; the third is engine-internal time.
Windows — UI end-to-visible time
UI end-to-visible timing · 260 samples
≤ 20 ms
0
> 20–25 ms
171
> 25–30 ms
82
> 30–40 ms
7
> 40–50 ms
0
> 50–75 ms
0
> 75–100 ms
0
> 100 ms
0
View data table
Windows — UI end-to-visible time — 260 samples
Bucket
Samples
≤ 20 ms
0
> 20–25 ms
171
> 25–30 ms
82
> 30–40 ms
7
> 40–50 ms
0
> 50–75 ms
0
> 75–100 ms
0
> 100 ms
0
Android API 36 emulator — UI end-to-visible time
UI end-to-visible timing · 260 samples
≤ 20 ms
0
> 20–25 ms
2
> 25–30 ms
64
> 30–40 ms
149
> 40–50 ms
20
> 50–75 ms
23
> 75–100 ms
1
> 100 ms
1
View data table
Android API 36 emulator — UI end-to-visible time — 260 samples
Bucket
Samples
≤ 20 ms
0
> 20–25 ms
2
> 25–30 ms
64
> 30–40 ms
149
> 40–50 ms
20
> 50–75 ms
23
> 75–100 ms
1
> 100 ms
1
Deterministic engine — internal time
Engine-internal timing · 520 samples
≤ 0.25 ms
247
> 0.25–0.5 ms
194
> 0.5–1 ms
52
> 1–2 ms
23
> 2–5 ms
4
> 5 ms
0
View data table
Deterministic engine — internal time — 520 samples
Bucket
Samples
≤ 0.25 ms
247
> 0.25–0.5 ms
194
> 0.5–1 ms
52
> 1–2 ms
23
> 2–5 ms
4
> 5 ms
0
What the estimate benchmark proves
Supported jobs resolve to editable estimates effectively instantly — a 0.028 second median (rounded) across 520 certified generations.
Results are deterministic: 78/78 repeatability hashes matched for identical inputs.
No generative-AI provider is involved: 0 provider calls and 0 generation failures across all 520 runs.
Review gating held: 520/520 generations passed Job Ready, and 520/520 certified false-ready attempts were prevented.
What this does not prove
This benchmark covers certified supported jobs only. It does not certify supplier-price accuracy, real-world job frequency or coverage, every construction estimate, every device, or arbitrary free-form scope.
Certification pricing was synthetic — it says nothing about what a real job should cost.
Timing on your hardware may differ; the Android figures come from an API 36 emulator, not physical handsets, and nothing here is a re-run against the current Windows Build 17.
The rounded 0.048 second figure is not a guarantee that 95% of runs finish inside 0.048 seconds; the measured 95th percentile was 48.008 milliseconds in this dated campaign.
Architecture
How an Instant Estimate is built
01
Structured job router
Your description is classified against the supported-work catalog. Supported jobs go Instant; unusual or unsupported scope is flagged for the Hybrid/assisted review path, which in the current beta is manual contractor review — AI-assisted drafting needs an external provider and none is enabled in the beta.
02
Deterministic engine
Assemblies, formulas, and production rates resolve the job locally — same inputs, same structured result. No generative-AI provider call is made.
03
Editable estimate
The output is a normal Drevanto estimate. Every line is editable, and pricing stays under your control.
04
Review before it leaves
Estimate Reviewer, Job Ready, and Margin Guardian check completeness and margin. Local codes, site conditions, prices, and structural details still require your verification.
Evidence
Sanitized certification downloads
Public facts, histogram counts, methodology, PASS statuses, and limitations from the 2026-08-21 Instant certification and the August 29, 2026 verification pass. Sanitized: no usernames, local paths, hostnames, device serials, credentials, customer data, source identifiers, full raw hashes, or raw sample rows.
Summary (CSV)
drevanto-instant-estimate-summary-2026-08-21.csv
Headline metrics and histogram counts from the 2026-08-21 certification run.