CE-RC-01Core platform architectureAdditive — no existing module changedHumans remain decision-makers

Reality Kernel™

The central orchestration layer beneath every RealmaticOS module. Its single purpose is to continuously transform observations into understanding through an endless learning cycle. Every current module keeps working exactly as it does today; every future feature plugs in here rather than building a parallel architecture.

Kernel stages
8
Observation sources
12
Graph nodes
19
Graph edges
20
Reality Fitness
0.79
Confidence cap
0.92
Reality Computing Kernel™
Eight continuous stages · this cycle never ends
Observation Service
Ingest immutable observations from every source.
Proof of concept · Reality Model emits each canonical-entity version change into the Observation Service (obs-4414, obs-4417).
Service 01 · Observation Service
Immutable observations from every source
obs-4417government datacomputed
2026-08-05 14:12:00
NYC DOB posted a review-status change on ZRA-2026: routed to CPC calendar.
domain: regulatorystate: rs-midtownconfidence: 0.90entities: 3
evidence · DOB BIS #B00874392 (0.98)
obs-4416market datacomputed
2026-08-05 13:48:00
Mezzanine spreads on Midtown vertical assets tightened 38bps week-over-week.
domain: capitalstate: rs-midtownconfidence: 0.78entities: 2
evidence · Capital tape 08/05 (0.82)
obs-4415ai agentunderstood
2026-08-05 12:05:00
Permit Watch Agent detected a Con Ed feeder maintenance notice adjacent to Feeder-12.
domain: infrastructurestate: rs-midtownconfidence: 0.66entities: 2
evidence · Agent run 8841 (0.70)
obs-4414documentcomputed
2026-08-05 10:31:00
Envelope study v3 uploaded — TDR envelope revised to 84,000 sf (from 81,200 sf).
domain: operationalstate: rs-midtownconfidence: 0.87entities: 2
evidence · envelope-v3.pdf (0.93)
obs-4413calendarunderstood
2026-08-05 09:02:00
IC session moved to 08/09; TDR decision now precedes the CPC calendar date.
domain: executivestate: rs-midtownconfidence: 0.95entities: 2
evidence · Cal event IC-0809 (0.99)
obs-4412emailreceived
2026-08-04 17:40:00
Lender counsel flagged a covenant test tied to stabilized NOI timing.
domain: capitalstate: rs-midtownconfidence: 0.71entities: 2
evidence · Thread 'Covenant §4.2' (0.76)
obs-4411database eventcomputed
2026-08-04 15:18:00
Portfolio module wrote a revised hold-period assumption on the Midtown project.
domain: portfoliostate: rs-midtownconfidence: 0.83entities: 1
evidence · db:org_projects#midtown (0.90)
obs-4410userunderstood
2026-08-04 11:07:00
S. Chen annotated the LOI cap rationale in the deal room.
domain: executivestate: rs-midtownconfidence: 0.80entities: 2
evidence · Deal room note 771 (0.85)
Observations are append-only. History is never deleted; superseded observations are marked, not removed.
Service 02 · Reality State Manager
Continuously updating state, permanently accessible history
Midtown East — vertical programorg:realmatic-holdingsws:vertical-portfolioconfidence 0.84fitness 0.79
Entities
org:realmatic
asset:432park
project:midtown
asset:feeder12
doc:zra-2026
loc:midtown-east
Relationships
org:realmatic —owns→ asset:432park (0.94)
org:realmatic —owns→ project:midtown (0.90)
project:midtown —depends_on→ asset:432park (0.88)
asset:432park —depends_on→ asset:feeder12 (0.81)
org:coned —owns→ asset:feeder12 (0.97)
org:nyc-dob —regulates→ project:midtown (0.93)
doc:zra-2026 —influences→ project:midtown (0.86)
agent:permit-watch —uses→ doc:zra-2026 (0.80)
Objectives
Execute TDR transfer at ≤ $412M cap
Preserve 2029 stabilization window
Keep utility headroom non-binding
Constraints
CPC calendar timing
Feeder-12 headroom
Covenant §4.2 NOI timing
Opportunities
TDR transfer window opens ahead of CPC — +6–9% modeled IRR · p=0.61
Mezz refinance at tightened spreads — -38bps carry · p=0.54
Risks
Feeder binding under revised envelope · L=0.34 · high
Rate re-shock before close · L=0.22 · medium
Evidence
DOB BIS #B00874392 — CPC calendar status (0.98)
envelope-v3.pdf — 84,000 sf envelope (0.93)
Active Decisions
dec:loi-412
TDR transfer sequencing (IC 08/09)
v3 → v4
entities +2 · objectives +1 · constraints +1 · confidence +0.03 · fitness +0.05
Entities
Every node the state currently holds.
Relationships
Typed, confidence-weighted, evidence-backed edges.
Objectives
What this organization is trying to achieve.
Constraints
What limits progress right now.
Opportunities
Openings detected by the Reality Engine.
Risks
Exposures with likelihood and impact.
Confidence
Composite confidence, capped at 0.92.
Evidence
Provenance for every claim in the state.
Active Decisions
Open decisions awaiting human judgment.
Service 03 · Reality Graph
Every entity automatically connected — confidence, timestamps, evidence on every edge
Realmatic HoldingsCon EdisonNYC DOBS. Chen — CSOJ. Reyes — IC ChairMidtown East Upzone432 Park EnvelopeFeeder-12 CapacityMidtown EastPermit Watch AgentCapital Scan AgentZRA-2026 FilingRate Re-shockFeeder BindingTDR Transfer WindowMezz RefinanceLOI @ $412M capRU · Utility-first underwritingRU · ZRA review lag ≈ 5.4 mo
Selected node
Midtown East Upzone
project:midtown · confidence 0.79
Relationships
Edge vocabulary · Owns · Depends On · Connected To · Influences · Reports To · Uses · Causes · Regulates · Learns From
Service 04 · Reality Engine
Reality changed — chg-9001 from obs-4417
What changed?
ZRA-2026 moved onto the CPC calendar — regulatory sequencing advanced by ~7 weeks.
Why?
DOB routing completed earlier than the modeled 5.4-month review lag (Reality Unit RU-zra-lag).
What else changes?
TDR transfer window now opens before the IC session on 08/09.
Envelope revision (84,000 sf) becomes decision-relevant this week, not next quarter.
Capital timeline compresses: mezz pricing must be locked earlier.
What risks appear?
Decision compression — IC reviews a revised envelope with 4 days of preparation.
Feeder-12 headroom re-tests against the larger envelope before capacity is re-validated.
What opportunities appear?
Sequence TDR transfer ahead of CPC to capture the tightened mezz spread.
Pull the covenant §4.2 renegotiation forward while lender appetite is elevated.
Which previous lessons apply?
Utility headroom must be re-validated whenever the envelope changes (RU · utility-first underwriting).
Calendar advances compress human review time more than they save schedule.
Similar historical Reality States
Apr 2026 — ZRA filed, optionality held74%
Hudson Yards air transfer, 202468%
Candidate Reality Paths
Transfer-first: execute TDR before CPC46%
horizon 2026-09-30 · Captures spread; requires expedited feeder re-test.
Validate-first: re-test Feeder-12, then transfer37%
horizon 2026-11-15 · Lower execution risk; spread likely widens.
Hold: let CPC land, reprice afterward17%
horizon 2027-01-31 · Maximum information; ceiling forfeited.
Service 05 · Recommendation Engine
Advisory only — humans remain the decision-makers
Re-test Feeder-12 headroom against the 84,000 sf envelope before the 08/09 IC
The envelope revision and the CPC advance landed in the same 48 hours. Utility headroom was validated against the prior 81,200 sf envelope, so the binding assessment underlying the LOI cap is now stale.
Assumptions
Con Ed can return a desktop re-test within 3 business days
Feeder maintenance notice does not affect Feeder-12 directly
Constraints
IC session 08/09
CPC calendar date fixed
Tradeoffs
Costs ~2 days of schedule
Reduces the chance of an IC decision on a stale capacity assumption
Expected consequences
Capacity assumption current at decision time
LOI cap defensible under §4.2 review
Supporting evidence
envelope-v3.pdf — 84,000 sf, sealed A-102 (0.93)
Con Ed capacity study 2025-11-09 — Validated at 81,200 sf (0.94)
The kernel never executes a recommendation. It prepares the decision; a human makes it.
Service 06 · Decision Attribution
Append-only, fully searchable — nothing is overwritten
Approve LOI at $412M cap2026-07-29
Decision-maker · S. Chen (CSO) — ratified by IC
AI contribution · Ranked 3 strategies; supplied confidence bands and the utility-headroom dependency.
Reason · Utility headroom validated and rate exposure bounded at the time of decision.
Alternatives · Hold + partial transfer | Assign optionality and monetize now
Expected · TDR executed by 2027-06; stabilized NOI window preserved.
confidence 0.83Con Ed capacity studyRate memo 07/22
File ZRA-20262026-04-11
Decision-maker · J. Reyes (IC Chair)
AI contribution · Modeled review-lag distribution; recommended filing ahead of the seasonal queue.
Reason · Filing early dominated all modeled alternatives on expected schedule.
Alternatives · Defer to Q3 filing window | File with reduced scope
Expected · CPC calendar placement within 5.4 months.
confidence 0.76Permit queue analysis 03/2026
Validate Feeder-12 before underwriting2025-11-09
Decision-maker · S. Chen (CSO)
AI contribution · Flagged capacity as the binding constraint with 0.5 likelihood.
Reason · Capacity was the highest-variance unknown in the model.
Alternatives · Underwrite with a capacity contingency
Expected · Capacity resolved before capital committed.
confidence 0.70Load curve Q4 2025
Service 07 · Consequence Tracking
Expected outcome versus actual outcome
csq-2201 · dec-43552026-01-15
Expected · Capacity resolved before capital committed.
Actual · Resolved; headroom non-binding at the then-current envelope.
Prediction accuracy
0.92
Confidence drift
+0.08
Reality fitness
0.81
Unexpected · Re-validation required on every envelope change
Lessons · Utility-first underwriting raises decision quality on vertical envelopes.
csq-2202 · dec-43902026-08-05
Expected · CPC calendar placement within 5.4 months.
Actual · Placement at 3.8 months — 7 weeks earlier than modeled.
Prediction accuracy
0.63
Confidence drift
-0.06
Reality fitness
0.72
Unexpected · Human review time compressed | Envelope revision collided with the calendar advance
Lessons · Review-lag model over-weights seasonal backlog. | Schedule gains transfer risk to decision preparation.
csq-2203 · dec-4412measurement open
Expected · TDR executed by 2027-06; stabilization window preserved.
Actual · In flight — measurement open.
Prediction accuracy
—
Confidence drift
+0.00
Reality fitness
0.79
Service 08 · Learning Engine
Reality Units™ · Consequence Chains™ · Reality Lessons™
Reality Units™
Utility headroom must be re-validated on every envelope revision.
infrastructure · confidence 0.78 · support 6 · contradictions 0
NYC ZRA review lag averages 5.4 months, high variance by season.
regulatory · confidence 0.63 · support 9 · contradictions 2
Mezz spread tightening on vertical assets persists ~5 weeks.
capital · confidence 0.57 · support 4 · contradictions 1
Schedule advances reduce decision preparation time proportionally.
executive · confidence 0.69 · support 5 · contradictions 0
Consequence Chains™
Envelope revised → Capacity assumption stale → LOI cap defensibility drops → IC decision quality falls
strength 0.74
CPC advance → Decision window compresses → Diligence shortened → Unmodeled risk enters the decision
strength 0.66
Spread tightening → Refi becomes accretive → Carry falls → Stabilization window widens
strength 0.58
Reality Lessons™
Re-validate every dependency the envelope touches, not just the envelope.
from csq-2201 · +0.08 confidence · applies to Acquisition, Construction, Infrastructure
Down-weight seasonal backlog in the ZRA review-lag model.
from csq-2202 · -0.06 confidence · applies to Regulatory, Forecast
When schedule accelerates, escalate preparation, not just execution.
from csq-2202 · +0.04 confidence · applies to Executive, Meeting Copilot
Confidence rises when evidence supports a conclusion and falls when reality disagrees. It never exceeds 0.92.
Service 09 · Evolution Engine
The platform becomes smarter through use
Reality Graph0.52 → 0.78
Edge coverage and evidence density across tracked entities.
Reality States0.58 → 0.84
State completeness across the nine slices.
Recommendation Models0.49 → 0.71
Acceptance-weighted usefulness of proposals.
Learning Models0.44 → 0.69
Lessons that later hold under new evidence.
Confidence Scores0.61 → 0.83
Calibration: stated confidence versus realized accuracy.
Reality Fitness0.55 → 0.79
How closely the model tracks the world it represents.
Confidence & Reality Fitness trend
ConfidenceReality Fitness2025-11 → 2026-08
Implementation guidance
Architecture first — one module wired as proof of concept, the rest integrate progressively
Module integration register
Reality Model™wired
emits · Canonical entity versions as observations
reads · Reality State entities & relationships
Signalsplanned
emits · Detected signals as observations
reads · Reality changes for prioritization
Executive Feed™planned
emits · Reader engagement events
reads · Recommendation queue
Decision Intelligence™planned
emits · Decisions
reads · Decision Attribution + Consequences
Forecast Engine™planned
emits · Predictions with horizons
reads · Consequence accuracy for calibration
Infrastructure Capacity™planned
emits · Capacity observations
reads · Constraint slice of Reality State
Construction Intelligence™planned
emits · Site progress observations
reads · Reality paths
AI Workforce™planned
emits · Agent findings as observations
reads · Objectives & constraints
Proof of concept · Reality Model™
01 Observe
Reality Model emits each canonical-entity version change into the Observation Service (obs-4414, obs-4417).
02 Understand
Reality State Manager folds those observations into rs-midtown, producing version 4.
03 Compute
Reality Engine derives chg-9001 with three candidate Reality Paths.
04 Recommend
Recommendation Engine returns rec-7701..7703, two behind approval gates.
05 Act
A human decides; Decision Attribution appends the record — nothing is overwritten.
06 Measure
Consequence Tracking opens csq-2203 and measures against the expected outcome.
07 Learn
Learning Engine mints Reality Units and Lessons; confidence moves with evidence.
08 Evolve
Evolution Engine lifts graph, state, and calibration metrics; Reality Fitness 0.55 → 0.79.
Kernel invariants
Observations are immutable — append only, never deleted.
Historical Reality States remain permanently accessible.
Every relationship carries confidence, timestamp, and evidence.
Recommendations are advisory; humans remain the decision-makers.
Decision Attribution is append-only and fully searchable.
Confidence never exceeds 0.92 and moves only with evidence.
Every future module integrates into the kernel — no parallel architectures.
Reality Kernel™ output is informational and advisory. It does not execute actions, contact third parties, or replace professional, legal, or financial judgment.