Under the hood
A real operating system for one listing.
Code collects and normalizes the evidence. Models prepare reproducible guideposts. Greg investigates the complete board. Hard owner promises and write-safety gates control external action. Guesty confirms what actually happened.
Current public receipt: Aug 22, 1:11 PM ET
System map
Two wakeups. One verified operating loop.
The same production path runs at 7:05 AM and whenever a booking, cancellation, or alteration changes the calendar.
Wheelhouse, Guesty, Airbnb, history, market, panel
One privacy-safe date and stay contract
Clock, pace, comps, paths, geometry, cleaner
Complete evidence, conflicts, alternatives, final target
Owner rules, freshness, atomicity, concurrency, invariants
Wheelhouse prices, Guesty calendar rules
Fresh Guesty truth and bound receipts
Owner email, scoreboard, later outcome grade
Data contracts
What comes in, field by field.
Collectors are coded and repeatable. Greg receives a prepared evidence packet, but can still inspect the raw rationale and explore any source.
932reservation records
284,729posted-price records
10,956potential comps screened
4,896qualified substitutes
8,256panel member-date cells
12,960exact-date observations
Every run
Official RM API + OAuth MCPWheelhouse
ENDPOINTS / SURFACES
price_recommendations?attribution=true · price_calendar · neighborhood/pricing · neighborhood/occupancy · sets/{id}/time_series · sets/{id}/distribution · checkin_checkout
FIELDS CAPTURED
date · recommended price · seasonality · local demand · scarcity · occupancy pacing · historical anchoring · neighborhood range · minimum stay · posted price
USED FOR
The pricing prior, market pace, attribution, settings, and channel write surface.
Every run + event certainty poll
Official Open APIGuesty
ENDPOINTS / SURFACES
availability-pricing calendar · reservations · calendar write + readback
FIELDS CAPTURED
stay date · availability · booked or blocked · live price · check-in · check-out · booked at · stay length · guest count · channel · accommodation revenue
USED FOR
System of record for the real calendar, reservations, Sunday-cleaner rules, and independent confirmation.
Daily journal
Owner-authorized signed-in DOM captureAirbnb
ENDPOINTS / SURFACES
9 aggregate Performance pages · year-over-year + similar-listing modes
FIELDS CAPTURED
page views · wishlist additions · search impressions · conversion · lead time · nightly rate · occupancy · cancellations · length of stay
USED FOR
Shopper-funnel evidence that Wheelhouse and Guesty cannot see.
Rebuilt after new reservation history
Normalized Guesty corpus + price ledgerListing history
ENDPOINTS / SURFACES
8 years of reservations · 284,729 posted-price records
FIELDS CAPTURED
stay night · booked at · lead time · accepted rate · stay length · weekday · season · event class · cancellation
USED FOR
The listing-specific BookBy clock and historical booking curves. The COVID shutdown is excluded from demand inference.
Daily baseline + weekly scout review
Official Dynamic Set APIs + daily signed-in Calendar capture24-listing research panel
ENDPOINTS / SURFACES
sets/candidates · sets/{id}/listings · aggregated_metrics · time_series · distribution · Calendar export until per-member API ships
FIELDS CAPTURED
listing role · bedrooms · sleeps · distance · channel · daily ask · available · inferred booked · unavailable unknown · pickup · reopening
USED FOR
A maintained panel for near-term couples, long-lead families, and hyperlocal neighborhood demand.
The maintained comp panel
Not 24 random nearby listings.
The panel reflects how this unusual apartment competes. One-bedroom substitutes matter more for near-term couples. Capacity matters more for families booking farther out. Hyperlocal inventory captures neighborhood-specific need.
SCREEN10,956market candidates
→QUALIFY4,896plausible substitutes
→MAINTAIN24daily research panel
Stable core
A continuous time series. No more than two core changes per weekly review.
Research scouts
Rotating candidates used to test whether a better substitute belongs in the core.
Booking missions
Near-term couple, long-lead family, and neighborhood necessity.
LIVE + VERIFIED
What it can say: a member changed from available to inferred booked, the last observed asking rate, stay shape, lead time, and whether our listing was eligible.
What it cannot say: the transaction price, why the guest chose it, or whether Wheelhouse marked a night unavailable because it was booked or blocked.
Decision tools
Greg starts with six purpose-built instruments.
They make each run consistent and inspectable. They are guideposts, not blind constraints. Greg can challenge their assumptions and replace a soft recommendation, but the rationale must be recorded.
BookBy clock
Builds a separate historical booking window for each date class, including holidays, weekday or weekend, season, and event timing.
Pace controller
Compares booked inventory with the correct date-specific curve. Ahead creates upward pressure. Behind creates downward pressure. On pace supports a hold.
Comparable-market lens
Uses the broad market and the maintained 24-listing panel to size a move and test whether this apartment is still a strong eligible value.
Stay-path solver
Enumerates the two-to-fourteen-night stays still possible so related nights can be priced and protected as a complete product.
Protected Geometry
Outside D-30, uses arrival and departure restrictions to prevent one-night leftovers while preserving every clean stay. At D-30, the controls release automatically.
Cleaner controller
Creates and safely releases only the Sunday protection BookBy owns when a reservation edge would otherwise require Sunday cleaning.
ONE RECORD PER FUTURE DATE
current_pricetarget_pricesupported_rangeraise | hold | lowerpace_stateBookBy_phasemarket_anchorgeometry_rolecleaner_stateconfidencereasons[]execution_statereadbackWrite authority
Intelligence can explore. External action is tightly controlled.
The safety layer does not decide the price. It proves that the evidence is current, the action is complete, the owner's promises survive, and the channel accepted the exact result.
88/91current adversarial suite0 failed gates in this receipt
Stop before any qualified write.
Only one locked pricing authority may write.
Discard the plan, reread Guesty, and decide again.
Reject the whole batch before it reaches a channel.
Fail the geometry invariant.
Reject the write.
Fail the run and preserve the mismatch receipt.
Release only BookBy-owned protection after a fresh read proves it is safe.
signed target hash→single locked writer→Wheelhouse / Guesty→fresh Guesty readback→immutable receiptAccountability loop
Outcomes are graded without pretending correlation is causation.
A price change is not revenue. A later booking does not prove that price caused it. BookBy keeps an attribution ladder so verified operations, realized outcomes, and causal claims are never blended.
What was raised, held, lowered, or protected and whether the channel confirmed it.
What later booked, reopened, remained vacant, or preserved option value.
Accepted rate versus the frozen Wheelhouse baseline and prior displayed price.
A pattern becomes a proposed adjustment only after enough comparable evidence accumulates.
New model behavior is versioned, tested, and explained. There is no silent automatic retraining.
Production truth
No vapor. Every capability declares its boundary.
Green means external evidence verifies the behavior. Amber means it operates with a named gap. Gray means it can inform Greg but cannot write.
Can raise, hold, or lower an exact open night; booked and unavailable inventory stays immutable.
calendar=2026-08-22; rows=365Reviews normalized evidence and model intermediates, can explore dates with diagnostic tools, and may keep or replace any soft coded recommendation after recording evidence, counterevidence, alternatives, and uncertainty.
board=2026-08-22; judgment receipt=passed; applied=trueBoundary: The current reasoning receipt, exact final-board binding, override count, subsequent authority receipt, and Guesty readback do not all agree.
Forms its attack docket before seeing G.R.E.G., challenges the objective, commercial diagnosis, code, tools, and recovery paths, and may veto dates or block provider authority without mutating anything itself.
review completion=completed_validated; result=passed/surfaceAnalyzes orphan-producing stay paths without changing guest bookability.
candidate gaps=5; live controlled arrival/departure dates=0; previously observed BookBy dates=10Boundary: Candidate controls are analysis-only. BookBy has zero new Protected Geometry write authority until representative Airbnb shopper-path proof succeeds; previously applied controls remain a separately disclosed release obligation.
Blocks either side of a Sunday turnover and removes only obsolete BookBy-owned protection after a cancellation or alteration.
status=live_inventory_reconciled; protected edges=5Boundary: Cleaner-message delivery is a separate automation and is not proven by the inventory receipt.
Re-runs cleaner protection, geometry, and the full 365-date price decision after the board changes.
listener running=true; last exit=255Boundary: The listener and authoritative event operator are healthy, and the geometry step is installed and tested. The last real event predates the geometry receipt contract, so the next real booking, cancellation, or alteration must prove that combined path end to end.
Sends the event kind, dates/pace/Protected-Geometry result, cleaner result, price result, and whether attention is needed—without guest identity.
latest send=2026-08-12T01:05:47.084Z; provider status=200Boundary: Email delivery is proven, and the new geometry-aware notification passes privacy and idempotency tests. The next real booking, cancellation, or alteration must produce the first delivered version-2 geometry receipt.
Makes email the primary owner interface: Greg leads with whether the owner is needed, then reports verified pricing, pace, geometry, cleaner protection, and only the decisions Greg cannot make alone.
latest send=2026-08-20T17:05:47.401Z; provider=resend/200Boundary: The notifier is installed and tested, but today’s production delivery receipt is missing or incomplete.
Refreshes Guesty, Wheelhouse, settings, the daily Airbnb journal, and read-only model inputs before the 5:15 operator.
LaunchAgent loaded=true; runs=6; last exit=0Refreshes authority, reconciles constraints, prices every day, writes eligible changes, verifies Guesty, tests, and rebuilds receipts.
LaunchAgent loaded=true; runs=6; last exit=1Boundary: The scheduler may be healthy, but today's target/validation/authority/channel-readback chain is not completely proven in this copy.
Waits for an in-progress 5:15 run instead of racing it, retries one failed or missing run when the lock is free, and proves the matching success-or-failure email state.
LaunchAgent loaded=true; runs=6; last exit=1Boundary: The watchdog code or plist may exist, but it is not live until the explicit installer loads the LaunchAgent and a current terminal receipt proves the run.
Directly changes confidence, protects strong-demand inventory from premature cuts, and helps detect chronic underpricing.
exact-date neighborhood prices=719; occupancy dates=371Boundary: Wheelhouse aggregates are revision-prone, so BookBy uses relative movement and corroboration rather than calling them literal competitor bookings.
Turns an already-captured Wheelhouse alert surface into a privacy-safe secondary operator-health signal.
status=complete; source=2026-08-22T07:45:11.347ZBoundary: This is secondary alert evidence. Direct channel price readback remains authoritative, and BookBy never dismisses or changes Wheelhouse alerts automatically.
Separates expensive unsold asks from actual market clearance, so BookBy can protect tightening dates and refuse comp-driven raises when comparable supply remains plentiful.
exact dates=62; pickup window=7 daysBoundary: The current report exposes selected-set aggregates rather than individual shopper eligibility. It corroborates or vetoes direction but never sets a dollar by itself.
Turns the maintained 24-listing panel into anonymous exact-date current asks, clean stay availability, and longitudinal inferred pickup evidence that G.R.E.G. can inspect for a specific stay.
members=23; dates=344; observations=7912Boundary: Booked versus blocked is inferred from the supported Calendar display, and the last-observed asking rate is not a transaction price. This remains corroborating evidence, not independent price authority.
Learns which exact weeks for this listing tend to book early, late, high, or low and adjusts a market-first target only when corroborated.
dates=365; high-confidence=365; medium-confidence=0Boundary: The clock is descriptive and bounded; it cannot authorize a write by itself.
Adds visibility, wishlist, conversion, achieved-rate, lead-time, and cancellation weather that Wheelhouse does not have.
journal date=2026-08-22; status=fresh_capture_todayBoundary: The journal runs unattended, but a fresh Airbnb website capture still depends on an authorized signed-in browser session; credentials alone do not make browser login durable or safe. Funnel data changes broad diagnosis/confidence, not a date's exact dollar.
Challenges the market prior and exposes likely underpricing or an implausible cohort; it does not currently write prices.
candidate census=10022; endpoint complete=falseBoundary: The broad paid candidate census is analyzed in best-1/5/20/100/market rings, but individual exact-date eligibility and current restrictions are unavailable. It is deliberately not live price authority yet.
Maintains a channel-balanced panel for near-term couple, long-lead family, and hyperlocal demand. Sixteen core members preserve a continuous time series while eight scouts test whether better substitutes should be promoted.
qualified universe=4896; selected members=24Boundary: The selector and private activation plan exist, but the current paid-set membership is not independently proven as an exact match.
Research layer for valuing what a booking would permanently foreclose across the rest of a gap.
clean stay paths analyzed=1222Boundary: The joint solver currently fails its reconciliation gate and has zero live actions. The live price engine uses only the already-validated bounded geometry/option-value features.
Creates accountable evidence for future calibration without claiming a price caused a booking.
graded decision states=eligible:41499, observed-booked:2556, shadow:4619, shadow-booked:236Boundary: Decisions and later outcomes are recorded and graded, but the system does not yet automatically retrain or recalibrate its core parameters from those grades.
Separates verified work, observed outcomes, frozen counterfactuals, and causal claims so Greg can be credited and held accountable without inventing attribution.
verified channel changes=88; realized option-value nights=1Boundary: The outcome scorecard is missing or stale.
Audits available settings and can change supported controls under owner authority, but only after an exact preflight and readback contract exists.
latest audit=2026-08-22T17:08:41.937Z; controls audited=12Boundary: Airbnb host-fee translation, channel markup parity, and a cancellation-policy conflict require an explicit economic/policy decision. BookBy must not silently invent one.