Who gets to set
the pet fee?

The final terms — operator and acting operator — are the founder’s ruling of 2026-09-09, relayed by the coordinator; the ruling also fixes that the operator is an org-level role (a property points at exactly one operator org; no property-level operator entity) and that “owner-operator” is not a third role. The ruling is not yet in a design file; the design set on fable/arch-review still uses its earlier identifiers (FIRM_GRANTS, managing_firm, …), which this deck keeps verbatim inside code so its citations stay findable.

The three senses of “owner”: docs/planning/portfolio-architecture/how/AGENTS.md house rules and notes/want.md §1 — org_admin (src/lib/platform/auth/permissions.ts:52-57, level 1), property owner (landlord: ORG#{purpose:'owner_party'}, holds no rung), and the PMS credential holder (src/lib/data/types.ts:3371: “Property.ownerId is a Better Auth userId for PMS credentials rather than a landlord identity”). manage_roles and its ceiling: notes/decided-2026-09-08.md §6. The earlier acting-operator mechanism — per-value attribution, refusal of an unattributed write, confirm-or-re-author at handover — is design-final.md §20.82 FH1–FH5 and notes/want.md §10.63. D-0910-1 supersedes it as Yale’s answer: the dated mandate stays, but missing values use eligible PropFlow defaults until the operator overrides. Genuine transcription retains attribution; the stand-in may not invent an absent operator’s answer. Keep the honest limit from §10.61 OR3: the acting operator’s staff are the pen, so a rule the operator never stated can be typed and nothing refuses it. That hole produced the ruling: remove the authoring power rather than police it. Which keys may default remains open.

notes/stress-catalog.md: BENCH-A/B; ST-46 isolates all row families; ST-12b and ST-30 distinguish recipient REPORT rows from foreign operational rows.

design-final.md §20.5.1 requires authorization before fetching private data and preserves action/resource pairs, including lists, counts, search, exports and caches. §20.5.3 supplies explicitly entitled projections. “Wall” means no direct access to another customer's private data; approved disclosures are limited publications into the recipient’s own custody.

Today’s counts: notes/have.md §1d — 159 bare getProperties() callers versus 6 scoped at 8941d763fa; re-measured at 4f49cde3e9 as 156 versus 9 (notes/plan.md §10.11). src/lib/data/dynamo/property.ts:154-165 is the scan-then-filter. The “404 shape byte-identical to missing” rule is notes/want.md §4.

design-final.md §§2–3 and notes/stress-catalog.md BENCH-A-calB/ST-17 supply the a2 local calendar versus a1 company calendar example. Later calendar selection contracts govern actual endpoint choice, permissions and capacity.

notes/decided-2026-09-09.md D-0909-5 requires management scopes below a property and explicitly leaves re-siting of policy, chain and managing relationships unfinished. The diagram shows associations, not final attachment storage keys.

D-0910-1 (founder ruling, 2026-09-10): these fee defaults are the operator’s own values, not PropFlow defaults. Platform default eligibility for business/legal terms remains open.

notes/stress-catalog.md ST-161 supplies company 50 and a2 override 40. Its historic key spelling is knowledge.pets.fee; newer ST-900/902 use fees.pet. This deck uses “pet fee” in prose and the newer key where it shows shorthand.

The example is expressly qualified by notes/want.md §10.56 FG2–FG7: a current explicit operator grant is required for an active lower value. ST-451 states that ending an exception resumes the default. These are fixture amounts, not operating fee recommendations.

design-final.md §§3.2/3.4 and §20.3.2 define attachments, relationships, stable SLOT guards and immutable revisions. notes/want.md §10.56 supplies exact grant/target eligibility and provenance.

This table is a teaching projection, not a literal proposed database schema. Under D-0909-5 the exact keys for management scopes and the placement of operator-governed policy slots still need design. The $40 value comes from ST-161, qualified by current grants.

The numeric fixture is ST-161. The optional empty group is a teaching expansion of the permitted building → ordered chain groups → company walk, not an invented third fee. Source: design-final.md §20.4.1–3.

notes/want.md §10.56 requires explicit lower grants. D-0909-5 means split-property lookup must first identify the management scope; the drawing conveys intended authority routing, not settled SLOT/chain key placement. Ending a value resumes inheritance; typed suppression and an absent value are different (§20.4.2).

The two-question framing is the founder’s, relayed by the coordinator on 2026-09-09 (“Isn’t it really only like two directions, top down or bottom up? … all of it is defined from the top”); it is not in a design file. The slide agrees for POLICY/PARAMETER and shows where the premise breaks: IDENTITY does not walk (notes/want.md §3, “never inherited”), and COMPLIANCE_FLOOR is written by nobody and clamped last from the statute table (design-final.md §4.2, protected-classes.ts:54), with jurisdiction from the building else the org. The append wrinkle and the lock and home-unknown edges are the same §3 rows.

notes/want.md §3, the four class rows (lines 177–180 at the review’s HEAD): IDENTITY — property only, own node only, never inherited, “a known building never borrows a sibling’s identity (the real incident)”, with the one carve-out that while a caller’s home is unknown an org twin answers for timezone and jurisdiction; PARAMETER — org/group/property, nearest wins, replace, free-form knowledge append, the org lock masks nearer rows; POLICY — org only (plus an acquired chain group), a write below the root refused at write, always effectively locked; COMPLIANCE_FLOOR — nobody writes, clamp last, jurisdiction from the building else the org, stricter while unpinned.

design-final.md §4.2 key table: fees.* / pets.policy and knowledge.* as PARAMETER (structured inherits, free-form appends, “never merged”); fair_housing_text, consent_model, retention.days, clara.send_tiers as POLICY (“a building cannot widen who may send”); tcpa_contact_hours, state_notice_period, protected_classes as COMPLIANCE_FLOOR. §15: “The registry class list is the one irreversible decision (K12).” §17 #2: the founder signs the class list.

Examples: the $50/$40 pet fee is ST-161 (fixture amounts); the four Western Slope communities are Fede, Sep 8 02:12:51 (notes/transcripts/E-sep7-8-podcasts.md §2), historically modelled by notes/want.md §10.64 as the resident-facing community.quiet_hours key (PARAMETER, nearest-wins) beside contact.hours, which the TCPA floor clamps — the proposed two-key split (ST-474) is superseded by D-0910-2: ONE quiet-hours key means “may Clara speak FIRST”; agent-initiated messages are blocked during quiet hours, replies to a resident who wrote first are allowed. The phone number is the address claim of notes/want.md §2.2 (one head per physical number, attribute_not_exists), not a key class. The second axis is FG1 authorityDomain (notes/want.md §10.56) with the proposed agreement_term (§10.67) superseded by D-0910-4: the owner requests, the manager applies. No amounts or hours beyond the fixtures are invented.

16 September update. Historical delegation diagram. Launch uses role labels without ordinary privilege gates; explicit access scopes remain. Admin is Sean/Fede/Gera only. Current contract ↗

D-0910-3 (founder ruling, 2026-09-10): grants name buildings already acquired; no grant may cover future acquisitions. The proposed all_managed wildcard target is not adopted.

notes/want.md §10.56 FG2–FG6: exact parent grant lineage; bounded redelegation; no grant from staff title, missing lock, landlord projections, or an ordinary admin role. Active local policy requires the current operator grant and a valid value. Platform floors/property facts/account security retain independent authorities.

ST-903 supplies operator → regional PM → on-site staff delegation. ST-161 supplies the 25–75 bounds and 40/500 controls; this deck qualifies that historical case with the current FG contract.

notes/transcripts/A-aug19-20.md §2: Fede, 0820-0017 01:08:26 “all you have to do is Sean, Kenya, or the property team say, no, that’s not allowed. That’s it.” and 01:17:02–01:17:37 “it’s a staff reply… Sean decided this date, that’s the policy.” Gera’s intern hazard: E-sep7-8-podcasts.md §2, Sep 8 02:02:16.

notes/want.md §10.65 AUTHORSHIP_LADDER (AL1–AL4): role-derived operator authority by ROLE_HIERARCHY level inside scope (src/lib/platform/auth/permissions.ts:52-66: org_admin 1, regional 1.5, property_manager 2), a reply is policy at the author’s altitude, a candidate below it; the backfill attests today’s values; decision site_team_policy_authorship named with default 1.5. The names are the real Camellia site team from transcript A; the co-signer question is Fede’s own example (01:08). No amounts invented.

Why it is here: how/CHANGELOG-fable.md rounds 2–3 — the round-2 critic found the design had defaulted to “only the org_admin,” which would have turned Kenya’s reply into a queue behind Sean and left the migrated customer unable to activate.

D-0910-2: quiet hours stay one key and gate direction, not all contact. Clara may reply to a resident who wrote first; the quiet-hours gate blocks Clara initiating.

notes/founders-custody-thread-2026-09-09.md §9 explicitly requires a per-key green indicator and, for a key the operator has NOT opened, a grayed-out disabled field carrying a route to message the operator — “if it doesn’t want to live there, it’s because the [operator] gave it permission to have an override.” A closed key never takes a value and never becomes a pending candidate.

notes/want.md §10.56 FG5 derives rungOverrideAllowed independently of the viewer and mayWrite from current grant, actor, placement and constraints. FG6 preserves locks/bounds. ST-448 exercises the distinction. §10.57 RQ5 makes the request action conditional on exact request authority and an available recipient. The UI is illustrative, not a claim of a shipped screen. Quiet hours deliberately has no invented interval.

notes/want.md §10.57 FIRM_REQUEST specifies the existing matter/conversation/task/notification workflow, current recipient, separate submission and intake rights, exact target command and committed-result-derived applied status. A stale appointment/value requires a refreshed proposal. ST-452–454 cover delivery, denied authority, stale requests and recovery.

notes/decided-2026-09-09.md D-0909-5: one property split between operators stays one PROP with management scopes beneath it. The record explicitly states the operator no longer sits on the building. Apartments and retail at 100 Main St is its own example.

~/astra-review/rounds/round-12.md, lines 27–39, uses operators F/G with disjoint residential/retail mandates and exposes the erroneous facility-wide constraint. This is the historical counterexample answered by the later ruling.

The counter-argument: notes/want.md §10.60 FACILITY_SCOPE (FS1–FS5) and design-final.md §20.78 — PROP# is the custody unit (wall, partition, roster stamp, line, PMS credential, the thing that transfers), so a scope beneath it would be a filter applied after the read; the founder’s “one prop” maps to the platform-private facility identity (MC2) and his “scope” to one PROP# per operator-managed slice. Three cold critics rated this the strongest reasoning in the set (how/CHANGELOG-fable.md rounds 1–5). The decision is open; nothing here is decided by the deck.

notes/decided-2026-09-09.md D-0909-5 explicitly lists unfinished re-siting of managingFirmOrgId/appointmentId, setting SLOTs, chain SLOTs and managed_by relationships. The table shows the required logical distinction only; it does not claim the MC facility guard or PROP custody schema has already been rewritten.

round-12.md lines 27–39 gives the independent-transfer consequence. Separate genuine co-located properties may use separate PROP rows; creating fake property identities to evade one operator's mandate is not the chosen split-property model.

notes/decided-2026-09-09.md D-0909-5, including the founder’s own third consequence: mixed-use routinely shares one street address, so “separate rows have separate addresses stops discriminating exactly where it is needed most.”

notes/want.md §10.60 / design-final.md §20.78: the argument that PROP# is the custody unit; FS1 (one slice per operator, facility_slice_overlap), FS2 (boundary, not address, is the discriminator), FS3 (the facility carries nothing resolvable), FS4 (never one line across two operators), FS5 (contact hours follow the resident’s slice). The “filter after the read” pattern is notes/want.md §4 and the Sep 7 leak in notes/transcripts/E-sep7-8-podcasts.md §6 (Fede, Sep 7 00:06:22: “I was able to see all the properties, all the units”).

Nothing on this slide is decided. Both columns are drawn from the sources named; the costs are the ones each side states, not the deck’s.

notes/stress-catalog.md BENCH-Y (line 411) and ST-146 L4 (line 416) explicitly define h1 as both run and held by org_st_hold, with the same resolution behavior as the ordinary managing case. WANT §10.55 preserves independent owned_by and managed_by roles. Placement of the appointment is updated conceptually to D-0909-5.

D-0910-4 supersedes the landlord-written agreement-term proposal below. A spending ceiling is a scoped request to the managing company, not a binding owner write.

notes/stress-catalog.md ST-12b (line 137) has two owned_by rows with share 0.5 on a1. ST-30 says the recipient reads only its own published REPORT rows. notes/want.md §10.55 states there is no observer precedence or fall-through to another party's org. The accepted appointment is shown on a management scope under D-0909-5, not on PROP META.

The superseded row: notes/want.md §10.67 AGREEMENT_TERMS (AT1–AT3) / design-final.md §20.86 — a fourth authority domain held on the accepted managed_by agreement, written by the party the agreement names (the landlord for money terms), clamping the operator’s value like a floor. This is the review’s argued exception to D-0909-1; decision agreement_terms_domain is named and open.

D-0910-4 supersedes the AGREEMENT_TERMS proposal and ST-481 described below. The owner requests a spending ceiling for a named scope; the managing company applies it through the authorized request workflow. This is no longer an open choice of authority domain.

notes/transcripts/E-sep7-8-podcasts.md §2, Gera, Sep 8 01:54:08: “when we’re doing a maintenance request, we have, like, a threshold, like, oh, you’re only allowed to spend $300… Does the property define it? The org? The property manager?” The $300 is his own example figure, not a recommended default; the gauge deliberately shows “under” and “over” rather than inventing amounts.

notes/want.md §10.67 AGREEMENT_TERMS: authorityDomain:'agreement_term' on the accepted managed_by agreement’s authorityTermsRef; the clamp uses the existing LOCK operator; an approval above it escalates to the landlord’s named approver through the request workflow. ST-481 asserts it. Two cold critics judged the argument correct (how/CHANGELOG-fable.md round 4; round-5 critique). This was the argued exception to D-0909-1; D-0910-4 rejects it.

notes/stress-catalog.md BENCH-400 (line 74) supplies org_bigscatter, 400 PROP rows, one org line and town area tags. ST-44/ST-85 turn scattered street-address-as-UNIT records into property anchors while preserving actual unit records. Home icons are a schematic sample, not a literal count.

design-final.md §20.2 and OP1 distinguish setting-bearing chain groups from labels. Under D-0909-5 every managing appointment must be placed on the home's management scope; the table emphasizes the unchanged place/phone/town pattern without inventing scope keys.

notes/stress-catalog.md BENCH-A (line 62) and BENCH-B ×3 (line 71) provide shared versus per-building phone layouts. The notation is readable rather than literal; actual physical phone allocations are channel-qualified beneath the current address contract.

notes/want.md §10.58 FIRM_ADDRESS requires operator allocation authority, channel control and current appointments. §10.48 OUTBOUND_SELECTION / ST-413–416 require explicit cold-sender selectors. FA3/IS require original property evidence; a company/group route remains unbound even when only one building is currently reachable. These are within-operator layouts; cross-operator sharing for a split property remains open under D-0909-5.

notes/stress-catalog.md BENCH-A-set (line 64) gives g_ab as kind line_set, a1/a2 on +10004000101, and a3 on +10004000103. These are fixture numbers, not live contacts.

ST-413 (line 931) uses alloc_ab for a1/a2 and alloc_a3 for a3 via explicit cold selectors. WANT §10.58 retains operator-controlled native address commands. A policy delegation does not authorize a phone rebind; ST-455 requires a request followed by a qualified operator allocation command.

Founder rulings supplied for this revision, 2026-09-10: D-0910-1 retains shape C’s house and dated mandate, selected over A and B, but replaces its authoring half — a FOURTH shape. The founder: “Make a ConAm account and then set the items in there. But while we don’t have one, just let’s leave them blank — or maybe we can make a default that’s just like PropFlow defaults for all settings that we can configure somewhere. But anything will always just be a PropFlow default until the operator overrides it.” Transcription remains machinery for a value genuinely transcribed, not the mechanism making Yale answerable. D-0910-5 unifies onboarding and ordinary manager change; neither surviving name nor reconciled stress cases is claimed here.

notes/transcripts/E-sep7-8-podcasts.md: Sean, Sep 8 01:53:47 “in Yale, we’re just the owner, not the property manager. ConAm is”; Sean 01:32:27 “ConAm’s gonna be a little slow… Yale is a guaranteed sale”; Fede 01:42:09 “that’s gonna take over a month.” Yale’s prod state: docs/planning/yale-integration/research/09-db-yale-jpco-state.md:39-44, :80 — a first-class row under org_jpco, 112 units, zero tenants.

notes/want.md §10.63 FIRM_HOUSE_BEFORE_SIGNING: three shapes argued (A: owner’s house then TRANSFER; B: operator’s house with PropFlow holding the keys; C, chosen: the operator’s house with the acting operator holding the login under a dated mandate), FH6 the pre-tenant shell re-home exception, FH8 the interval routing (the operator’s own staff author and route; the provisional admin transcribes and confirms only when no operator staff is present). Historical shape C was chosen over A and B on 2026-09-09; D-0910-1 retains its first half and supersedes its second; design-final.md §20.82 still carries firm_house_before_signing as a decision owed under its old identifier. The honest limit is §10.61 OR3 — the acting operator’s staff are the pen, so what the design guarantees is attribution, refusal of an unattributed write, and re-authoring at handover, not refusal of the write itself. That superseded authoring gap is the round-5 critique’s first finding (~/fable-review/scores.md, round 5) and led to D-0910-1; the older design contracts remain to be reconciled.

notes/transcripts/E-sep7-8-podcasts.md §1, Fede, Sep 8 02:21:22 (Hugo across two logins); §3, Gera, Sep 7 00:31:48 and Sep 8 02:20:17 (multi-org, multi-property dashboard).

notes/want.md §10.69 MULTI_ORG_VIEW: N memberships → N TenantContexts per request, union rendered client-side, cross_org_read_forbidden on any server-side join, membership_required drops an unheld org. The round-5 critique notes Hugo’s Grand Junction homes are Situs-owned and Western-Slope-managed, so that half of his view is ReportsContext over published rows (DEC §5), not a membership — stated on the slide. ST-487.

design-final.md §20.4.4/6 requires exact immutable evidence and current final authorization fences. OP2 and ST-902 require source displacement evidence even for an ordinary valid override.

notes/want.md §10.56 FG4–5 makes expired/revoked lower grants immediately inapplicable, separates current save authority from the indicator, and projects private grant/source evidence opaquely. Amounts are ST-161's teaching fixture.

~/astra-review/rounds/round-1.md through round-12.md, line 1 of each, supply the exact twelve scores and the 9.0 bar. The plot uses a linear vertical score scale; its baseline is not zero.

The review-loop explanation is the supplied brief's history; the reports directly support the deeper-follow-on pattern. R2 credits deterministic resolution and authorization fences; R7 credits constraint composition and effect settlement; R11 explicitly credits deterministic group/calendar/outbound selection. The reports alone do not independently certify every prior finding closed. Founder thread §11 puts correct direction ahead of score optimization.

Historical counterexamples from ~/astra-review/rounds/round-1.md lines 24–35; round-2.md lines 11–31; round-3.md lines 9–21. Scores from opening lines. The repair captions paraphrase the reports' proposed fixes; they do not assert deployed behavior or claim these historical gaps remain open today.

round-4.md lines 8–20 describes actor-authorized source reading plus recipient-bound output leaking internal material; lines 40–52 separately reject treating a legitimate pet-fee override as a same-position conflict.

round-5.md lines 25–45 shows STOP/START reviving prior grants; round-6.md lines 10–22 extends the flaw to a recycled phone number and the correct subject/contact binding. These are historical review findings, not legal advice or live incident claims.

round-7.md lines 9–19 describes restoring revoked authority/consumed effects. round-9.md lines 46–58 describes global recovery-journal contention between unrelated customers. round-11.md lines 8–26 describes rejected restrictive input while sealing, followed by reopening before retry. This progression directly illustrates deeper criticism of successive repairs.

round-12.md lines 10–18 goes deeper again: the newer pending guard can be spoofed before authentication. This deck does not label the recovery protocol fully proved.

round-8.md lines 29–35 requires complete travel-feasible itineraries despite non-overlapping bookings; the diagram intentionally invents no times or distances.

round-10.md lines 36–48 uses a web-form booking without an inbound address and a line-set tag outside policy inheritance. round-11.md line 4 credits the subsequent deterministic group/calendar/outbound selection rules.

round-12.md lines 27–39 gives the facility-level cardinality failure. notes/decided-2026-09-09.md D-0909-5 is newer and answers with scope-under-property. No later passing review is asserted.

notes/founders-custody-thread-2026-09-09.md §11 says direction matters more than getting a high score on the wrong plan. The deck paraphrases the instruction rather than inventing a quote.

~/fable-review/scores.md, one line per round: 7.7, 7.7, 7.6, 8.3, 8.4, each with the critic’s top finding — reproduced on the axis labels. The plot reuses the earlier slide’s linear vertical scale (9.0 at the dashed line; baseline not zero) so the two series can be read together.

Per-round changes: docs/planning/portfolio-architecture/how/CHANGELOG-fable.md rounds 1–5 on branch fable/arch-review. The critics were fresh subagents each round, told no prior score, reading the same design set cold. These are scores of a written design, not a production pass rate; notes/founders-custody-thread-2026-09-09.md §11 puts direction ahead of the number.

notes/transcripts/E-sep7-8-podcasts.md §3, Fede, Sep 7 01:37:53: “the fastest path would be, like, is PropFlow the CMS [CRM], and then… We just send emails from their email box. And, like, use the number that we created to text people.” and 01:40:57: nobody types phone leads into the PMS.

The recommendation and its price: notes/plan.md §10.11 round 3/4 additions and notes/want.md §10.55 MC7’s cohort sentence (decision prewall_cohort_logins, both twins ST-475/ST-480, the digest row LD1/ST-483). The misread and its withdrawal: the round-5 critique’s third finding and the reviewer’s final report (~/fable-review/scores.md round 5). The design files were not edited after round 5, so the recommendation still reads “refuse” there; this slide is the correction of record until the next revision lands.

design-final.md §20.15: stronger reads, conditional writes, revisions, immutable pages and evidence, publication control, external service probes, and safe refusal/uncertainty are real costs. Earlier fixed read budgets and delivery dates are superseded. §20.77 moves the read boundary early and displaces UI/calendar convenience work; fresh delivery commitments require repricing.

The single-company comparison is an explicit tradeoff inference from those requirements, also requested in the brief. No new duration or cost estimate is invented. D-0909-5 leaves scope-related implementation changes owed.

Recorded from the founders’ decisions of 2026-09-09 and 2026-09-10. The work-order ruling is D-0909-3 in notes/decided-2026-09-09.md — the read-side isolation work “should be baked into the strategy from day one so we should probably put it in the early phases.” The other eight are D-0909-6 (order of work), 7 (hard cutover), 10 (verified disjoint boundary, “a shared street address is fine”), 11 (the channel org’s own on-call), 13 (the building’s own login wins), 14 (archiving — PropFlow staff only and retaining, with the address-release grace window explicitly left open), and 16(a)/(b)/(c) (two dashboard numbers, cap at eight, observe never on a phone number). Where the wording here is not the founder’s own, it is the relay’s, not a quotation.

The address ruling refines D-0909-5: two co-located properties stay separate rows, but the discriminator is the non-overlapping management scope, not the address label — the same correction the slide “The same ruling, two ways to store it” records as FS2.

notes/founders-custody-thread-2026-09-09.md §10 (lines 203–206) supplies higher single-family deposits, a contract percentage, first month/two months of rent, and 200 for one-bedroom / 300 for two-bedroom. These examples are not selected policy defaults.

notes/want.md §10.59 DEPOSIT_TERMS: proposed fixed, rent_multiple, contract_fraction formulas; per_scope cannot express within-building bedroom differences; target_schedule is a contingent recommendation. No approved representation, basis, rounding or jurisdiction contract is invented. ST-458–461 keep activation gated. Asset type is metadata, not an automatic policy rung.

D-0910-2 (2026-09-10) supersedes QUIET_HOURS_TWO_KEYS and its proposed monotonicity decision below. One key means “may Clara speak FIRST.” It does not block a reply to a resident who initiated. The shared-phone question remains open.

notes/decided-2026-09-09.md D-0909-5 consequences 1–3 explicitly leave phone sharing and the resident who can hear both scopes open. It also flags that address labels cannot reliably distinguish independent property identities; the discriminator needs design.

notes/want.md §10.56 FG6 and its four-community example distinguish permitted contact instants from forbidden quiet intervals. The founder's interval examples cannot be used as an unambiguous narrowing calculation until the semantics are specified.

Historical proposals (quiet-hours split superseded): notes/want.md §10.60 FS4 (a HEAD names one org; sharing across two customers refuses firm_address_authority_required, ST-466) and FS5 with §10.64 QUIET_HOURS_TWO_KEYS (contact.hours with DEC §2’s monotone; community.quiet_hours nearest-wins, no monotone; decision community_quiet_hours_monotone; ST-467, ST-474). Western Slope’s 3-of-4 communities (Fede, Sep 8 02:12:51) is the case that motivated the split.

The supplied founder rulings leave default eligibility and the surviving handover name explicitly open. D-0910-1’s product-versus-business/legal boundary is proposed, not selected. D-0910-5 requires one mechanism for both handover cases; eight stress cases still name TRANSFER, and ST-900’s “post-TRANSFER” precondition is stale after shape C removed that mechanism. The older round-5 findings below remain review debt, including ConAm never signing and then being replaced with tenants present.

The round-5 critique (recorded in ~/fable-review/scores.md and relayed in the reviewer’s final report): F1 notes/want.md §10.61 OR5 versus §10.63 FH1; F2 §10.63 FH5’s missing leg; F3 notes/plan.md §10.11 paraphrase; F4 design-final.md §1 #9, §7 #5, §15 and notes/want.md §10.9 versus FH6; F5 FG2 versus FT4 and two enum/attribute lists; F6 three cites (permissions.ts:66, :207; src/lib/platform/auth/helpers.ts:312-354); F7 custody split by function; F8 Hugo’s owned homes as ReportsContext; F9 the P0/P1 box sums to 19–29 working days.

The Yale PMS fact: docs/planning/yale-integration/research/09-db-yale-jpco-state.md:80 (rentRollSource=yardi, descriptive only) — Yale has no PMS write path in 2026 either way, which is the honest sentence the next reviser should add beside the fix.

Latest authority: founder rulings D-0910-1–5, supplied for this deck revision on 2026-09-10, supersede incompatible design passages below. Defaults replace stand-in authorship; quiet hours gate speaking first; grants name acquired buildings; owners request and managers apply; one handover mechanism has an undecided name. Default eligibility remains open.

Reading order for this lesson:

1. design-final.md: the record; later §20 contracts supersede incompatible older recipes.
2. notes/decided-2026-09-09.md: earlier founder rulings, especially D-0909-5; now qualified by D-0910-1–5.
3. notes/founders-custody-thread-2026-09-09.md §§6–11: operator rule book, green indicator, requests, deposits, direction over score.
4. notes/want.md §§10.55–10.59: proposed MANAGING_CUSTODY, FIRM_GRANTS, FIRM_REQUEST, FIRM_ADDRESS, DEPOSIT_TERMS; placement must now be revised for D-0909-5.
5. notes/want.md §11 (“the model after Sep 9, in nineteen lines”) then §§10.60–10.69: the Fable review’s rows — FACILITY_SCOPE, OPERATOR_OF_RECORD, FIRM_TOGGLE, FIRM_HOUSE_BEFORE_SIGNING, QUIET_HOURS_TWO_KEYS (superseded by D-0910-2), AUTHORSHIP_LADDER, the maintenance keys, AGREEMENT_TERMS (superseded by D-0910-4), the month-one rows, MULTI_ORG_VIEW — mirrored at design-final.md §21 and §§20.78–20.88 on branch fable/arch-review.
6. notes/stress-catalog.md: fixture stories and proposed acceptance cases.
7. ~/astra-review/rounds/round-1.md through round-12.md: historical adversarial findings.

The first five files are in the supplied portfolio-architecture/how/ design directory. All illustrations are inline SVG, all interactions run locally, and no external asset is required.

The deck teaches decided direction and proposed contracts; it does not claim implementation, completed scope re-siting, or a later passing review.