Nine named
failure modes.
Each detection fires only when the failure mode is real. Each result has a citation chain back to ASHRAE, IPMVP, CORENET X, AACE TCM, or the relevant standard. Below: how each one works and which buyers use it.
5-Signal Claim Classifier.
The unsolved-field-wide gap, closed.
No production multi-agent system has automated claim-level verification for vertical agent fleets. We built one. Every claim entering our eight production squads passes a five-signal admission protocol before becoming part of any decision the OS makes downstream.
- Source authority — Tier-1 standards bodies (ASHRAE, IPMVP, CORENET) admit at full weight; Tier-2 trade press at 0.7; Reddit at 0.3 with required corroboration.
- Standards anchor — every numeric claim must cite a section/clause (e.g., ASHRAE G14 §5.3) or auto-downgrade.
- Numeric specificity — "around 14%" is rejected. "14.3% verified, ε=0.2 confidence interval" is admitted.
- Cross-source corroboration — single-source claims flagged for review unless source is Tier-1.
- Contradiction check — admitted claim is checked against the brain's existing knowledge; contradictions trigger reconciliation.
"14.3% verified savings · ASHRAE G14 §5.3" — full source authority + standards anchor + numeric specificity + 3-source corroboration + zero contradictions.
"AI-HVAC saves 30%" — failed numeric specificity (no anchor) + only Tier-3 source + Reddit-class authority.
Asset HK-12 · F&B retrofit triggers code-upgrade obligation. Sprinkler density required > existing capacity. Day 0 detection vs industry mean Day 47.
Retrofit Compliance Scan.
Catches code-triggered upgrades on Day 0.
When an owner touches an existing building, the local jurisdiction often imposes upgrade obligations the owner didn't know about. CRE-AD scans drawings + scope-of-work + jurisdiction-specific codes (CORENET X §8.1 in Singapore, IBC §3411 in the US, DOB BC §28 in NYC, Building Standards Law §86 in Japan) and flags the triggers before contracting starts.
- Drawing parser reads architectural + MEP scope-of-work and identifies trigger thresholds (area changes, occupancy classification shifts, structural interventions).
- Jurisdiction code-pack matching against the seven supported pacts; auto-downgrades confidence if jurisdiction is unsupported.
- Trigger surface generation — outputs the specific code clause + required upgrade with cost-impact estimate.
- Pre-contract gate — flags can be reviewed by a licensed architect before construction documents lock.
EVM Theater Detection.
Catches the gap between status reports and reality.
Earned-value reporting routinely tells owners the project is 78% done while procurement data shows 12% real progress. CRE-PM detects the divergence — across earned value (EV%), Last Planner percent-plan-complete (PPC%), and procurement spend — and fires when the spread breaches the threshold. 100% precision on a 100-project synthetic validation corpus.
- EV vs PPC vs procurement — three-axis divergence test; theater alert fires when the gap exceeds project-specific tolerance.
- Last Planner integration — pulls weekly PPC% from Last Planner system or equivalent agile-construction logs.
- Procurement reconciliation — checks against vendor invoicing, purchase orders, and committed-cost data.
- AACE TCM-anchored — every alert cites AACE TCM 7R-06 or the equivalent CPM Last Planner methodology.
Asset SG-04 reported 78% earned value, but procurement data shows 12% real progress. Pattern matches EVM-PPC divergence threshold.
Asset TW-08 · RFI velocity (28/wk · ↑ from 11) + minutes pattern (4 escalations) + schedule slip (12 days) → claim filing risk in 47 days.
Claims Early Warning.
Detects contractor patterns 30–60 days before filing.
By the time a contractor files a formal claim, the cost is already locked in. CRE-CON detects the pattern that always precedes a claim — RFI velocity spike, meeting-minute escalation language, schedule slip, and weather/scope-change correlation — and fires 30 to 60 days before the claim hits the project. Plus: real-time embodied-carbon tracking against the LCA budget, with auto-alerts when scope changes push embodied carbon past the contracted ceiling.
- RFI velocity model — weekly RFI count + classification (technical vs scope-change vs commercial) detects pattern shifts.
- Minutes escalation NLP — meeting minutes scanned for escalation language (delay, impact, defect, additional cost).
- Schedule slip + weather + scope correlation — three-way model triangulates root cause.
- Embodied carbon tracking — LCA-baseline + scope-change deltas + EPD lookups; alerts when ceiling approached.
AI-HVAC Feasibility Gate.
Blocks AI-HVAC pilots that won't pass IPMVP M&V.
95% of AI-HVAC pilots disappoint because no one ran the IPMVP feasibility check before signing. CRE-TS gates every AI-HVAC opportunity against the four IPMVP options (A retrofit isolation, B retrofit isolation with all parameters measured, C whole-facility, D calibrated simulation) and produces an audit-grade plan before any work starts.
- Baseline validity test — checks if the existing baseline is stable enough for measurement (no recent renovations, comparable weather, occupancy stable).
- Option recommendation — picks the IPMVP option that matches the project's measurement budget and rigor target.
- Savings projection with uncertainty — generates kW + kWh projections with explicit confidence intervals (not point estimates).
- Third-party M&V planning — flags the audit cadence + auditor type the project needs to accept.
Asset MY-02 · IPMVP Option C invalid — lighting renovation in baseline window. Switch to Option B (deemed savings) or wait until baseline stabilizes Q3.
Asset SG-04 · 14.3% verified savings · CV(RMSE) 8.2% ≤ 12.0% acceptance threshold.
Tue badge fusion · ε remaining 0.38 of 1.0 daily budget. k-anonymity floor (k=5) holds. PDPA + GDPR Art.9 + Colorado biometric consent all active.
Live People-Traffic HVAC + Privacy Broker.
15-min OCCUPANCY_BROADCAST that legal teams approve.
VergeSense, Density, OpenSensors all explicitly avoid badge-fused occupancy because of regulatory exposure. We don't. CRE-EN's privacy broker is a differential-privacy + k-anonymity gate that lets enterprise legal teams approve fused-occupancy data — and the OS uses that data to optimize HVAC at 15-min granularity, with verified savings tied to IPMVP M&V.
- Differential privacy with daily ε-budget — Laplace noise calibrated to the budget; no query crosses ε > 1.0 per day.
- k-anonymity floor — no occupancy bucket smaller than k=5 humans.
- 11-jurisdiction consent rule pack — GDPR Art.9, CCPA, Colorado biometric, PDPA, BIPA, EU AI Act, UK GDPR, SG PDPA, HK PDPO, JP APPI, AU Privacy Act.
- HVAC optimization downstream — privacy-gated occupancy feeds AI-HVAC at 15-min granularity with IPMVP-anchored savings claims.
KPI-Theater Alert.
When vendor green diverges from occupant red.
FM vendors report 96% green on every KPI. Occupants give the same building NPS −18. Both can be true and both can be measured — but only one is real. CRE-SS detects the divergence and surfaces it before contract renewal.
- Vendor KPI ingest — pulls scheduled cleaning, response time, work-order close from FM systems.
- Occupant signal — surveys, ticket sentiment, complaint patterns, badge dwell-time anomalies.
- Divergence model — vendor-green-occupant-red gap quantified; alert fires above threshold.
- Renewal decision support — outputs a "renew vs renegotiate vs replace" recommendation tied to the gap evidence.
Vendor "all green" / Occupant NPS −18. Cleaning KPI 96% on schedule; occupant complaints up 4.3× quarter-over-quarter on bathroom + lobby cleanliness.
HQ Tower 3. Tue forecast 1,180 occupants vs building capacity 940. Deficit of 240 — RTO mandate exceeds physical seats by 25.5%.
Policy-Space Mismatch.
RTO peak vs capacity gap, caught early.
When HR mandates "Tuesday RTO" and the building seats 940 but forecast attendance is 1,180, you have a policy-space mismatch. CRE-SP catches the gap before complaints, before reductions in collaboration quality, before quiet attrition. Alerts include scenario modeling for hot-desk allocation, satellite-office offload, or staggered-attendance fixes.
- Hybrid policy ingest — pulls "X days/week in office" + day-of-week weighting from HR systems.
- Capacity model — BOMA standard area + workpoint type + meeting-room conversion factor + neighborhood-based density.
- Forecast model — historical attendance + holiday + season + weather adjusted forecast vs capacity.
- Mismatch alert — fires when forecast peak exceeds 95% of capacity for any day.
Bayesian Deal Precedent Posterior.
Catches deals priced against pattern that doesn't match.
Most acquisitions are priced by reference to "similar deals." Most "similar" classifications are vibes. DIS computes a Bayesian posterior over the actual deal-precedent population — same market, same asset class, same buyer profile, same vintage — and surfaces when the proposed price implies a posterior the operator should reject.
- Precedent corpus — comparable transactions filtered on dimensional match (market, class, vintage, ownership-type).
- Bayesian posterior update — prior derived from cap-rate distribution, posterior updated as new comparables arrive.
- Pricing surface — outputs the implied cap-rate and asks "is this consistent with the posterior?"
- Decision flag — divergence from posterior > threshold triggers a "rebid or pass" recommendation.
Deal #001 · prior 0.12 → after 3 comparables: 0.74. Implied cap 5.4% vs posterior 5.9% — buyer should rebid.
Each detection runs on your portfolio in Beta Squad.
Six months. No NDA. No contract. Pick the squads that fit your work, give us fifteen minutes a week of structured feedback, and you decide whether the OS earns the relationship.