Etf reit onboarding

ETF / REIT / InvIT Onboarding Pipeline (global-trend machinery)

The single source of truth for finding and grading global-trend funds, REITs and InvITs, and publishing them to the 🌍 Global & Passive Base tab. Follow the stages in order. This is the fund mirror of [[stock-onboarding-pipeline]]: the SAME board (binding Munger/Buffett CIO · 31 perspectives · 14 canon · 8 desks · WC) evaluates ETFs/REITs the same way it evaluates stocks — verdict → conviction → margin of safety. The design source of truth is /Users/Dhiraj/.claude/plans/build-a-plan-to-iterative-creek.md.

The retrieval plumbing is identical to the stock pipeline — all vault consultation goes through data/scripts/32_consult_brain.py (scoped, cited, structured), now carrying the four NEW fund models passive-etf | commodity-etf | reit | invit. ETF/REIT mechanical fund-quality reuses the MF methodology (extracted/mf/_mf_methodology.md + _mf_rubric.md); REIT/InvIT dossiers reuse the ALTIUS template (vault/research/dossiers/ALTIUS.md).

North star

The 🌍 tab is a forward BUY-LIST of durable global-trend exposures in search of a small set of CORE holds you keep for life — NOT a one-off cleanup of what's already held, and not a sprawling basket of every thematic ETF. The point is to buy good ETFs, keep good ETFs, and compound wealth via durable global megatrends — the fund-side parallel of the stock FOREVER/COMPOUND hunt. Bogle/Buffett's default (broad, lowest-cost index held for life) is the anchor; thematic trends are capped satellites you buy only when the underlying index is at a sane valuation. Finding that a trend is real but expensive right now → WATCH, don't buy is an acceptable, expected result. Patience on the entry is the edge.

The framework — evaluate ETFs / REITs LIKE stocks

Map the stock machinery's three sort keys onto funds so the board reasons identically:

Stock key	ETF analog	REIT / InvIT analog
Moat (durable advantage)	Trend durability (is the exposure a 10-yr structural tailwind, not a fad?) × wrapper quality (lowest TER · tightest tracking error · deepest AUM/liquidity · right structure · best tax bucket)	Asset/lease durability · tenant diversification · sponsor quality · contracted-revenue runway (WALE)
Conviction 1–5	trend-strength × wrapper-quality (a broad low-cost index defaults HIGH; a narrow thematic ETF needs higher proof to earn the same band)	adapted ALTIUS-style filters — InvIT/REIT conviction CAP 2 (yield instrument, not a compounder)
Margin of safety	index valuation vs its OWN history (CAPE / PE percentile) — don't pile into an expensive index; wait for the trend at a sane price	sustainable (return-ON-capital) distribution yield vs the ~9–10% required yield (ALTIUS method)

Verdict tiers (re-cut to mirror the stock 5-tier — so they read the same):

• CORE (≈ FOREVER): the broad, lowest-cost index you hold for life regardless of price (Bogle/Buffett default — broad-world, Nifty-500-style India core). The anchor. This is the only tier where price is not a gate.
• COMPOUND-TREND (≈ COMPOUND): a high-conviction durable global megatrend (AI/compute · US innovation · India structural growth · energy transition · EM …) via the best India-accessible wrapper, at a fair index valuation — buy and hold for years.
• WATCH: great trend + great wrapper, but the underlying index is expensive now → wait for the entry (the MoS gate is unmet).
• AVOID: a bad wrapper (high TER / poor structure / bad tax bucket / wide tracking error) OR a fad / fading trend.
• TOO_HARD: trend durability is genuinely unknowable, or there is no clean India-accessible wrapper for it.

Rank key = verdict → conviction → MoS (exactly the stock key). Here MoS = how far the index trades below its own fair/historical value (ETF) or how far the sustainable yield clears required (REIT/InvIT). This is what lets the board "look at ETFs the same way as stocks." TOO_HARD is quarantined OFF the buy-list.

HARD RULES (never violate)

1. The binding CIO is the arbiter. data/scripts/32_consult_brain.py --corpus binding (Munger/Buffett vault) gates trend-durability, conviction, verdict and MoS at Stages A/B/C. The new perspectives & canon (Bogle, Malkiel/Ellis, Damodaran-REITs, Dalio, Marks) are down-weighted advisory context surfaced via --corpus blended — they SHARPEN the call but NEVER reverse a --corpus binding verdict.
2. The vault is BINDING, not decoration. cites_principles may contain ONLY slugs the brain actually returns (principles[].slug / citation). No fabricated citations. If principles comes back empty/thin → ABSTAIN + flag, never pad. (29_vault_chat.py stays available for free-form cited Q&A / ad-hoc grilling.)
3. Never fabricate a trend, a TER, a tracking error, an index valuation, a DPU, or a yield. Every number is fetched (INDmoney / WebFetch / BSE filing) or it is flagged as unknown. A guessed number is worse than a missing one. (See [[feedback-paywall-enumeration-honesty]] / the never-fabricate rule in memory.)
4. Evaluate funds LIKE stocks — three sort keys, five verdict tiers, rank = verdict → conviction → MoS. A fund is not exempt from the discipline because it is "passive." A narrow thematic ETF must EARN its conviction band against trend-durability × wrapper-quality, exactly as a stock earns its moat grade.
5. Trend durability is the moat, and it is a HARD GATE before price. Order of analysis: is-the-trend-durable → is-the-wrapper-best-in-class → ONLY THEN index valuation. A fad trend, or a great trend wrapped in a costly/badly-structured/badly-taxed vehicle, cannot earn high conviction regardless of how cheap the index looks.
6. Conviction = JUDGMENT, not a formula (Munger: avoid false precision). Gates: fad/fading trend → AVOID; no clean wrapper → TOO_HARD; InvIT/REIT → conviction CAP 2 (a yield instrument cannot be a compounder, so its ceiling is WATCH — never COMPOUND-TREND or CORE). A broad lowest-cost index defaults to high conviction (low cost IS the durable edge); a narrow thematic needs higher proof for the same band.
7. Margin of safety is THE gate. ETF → index valuation vs its own history (CAPE/PE percentile); a durable trend at a stretched index → WATCH, not buy. REIT/InvIT → the sustainable (return-ON-capital) yield vs the ~9–10% required yield (ALTIUS method — strip return-OF-capital out of the headline DPU before judging). The ONLY tier exempt from the MoS gate is CORE (Bogle's "buy the haystack, hold for life regardless of price").
8. For passive ETFs, low cost IS the moat — moat-analysis does NOT apply. Do not run the business-moat taxonomy on a passive index fund; its edge is structural (cost + breadth + tracking), graded by the wrapper-quality read and the MF cost/overlap gates. moat-analysis applies ONLY to the trend-durability judgment (Stage A) and to REIT asset/tenancy durability — never to "the index fund's moat."
9. InvIT/REIT = yield instrument, dossier on the ALTIUS template. Full dossier at vault/research/dossiers/<CODE>.md (brain_model: reit or invit): 6-filter walk (Filter 2 = asset + contracted tenancy, Filter 3 = distribution coverage + return-OF vs return-ON split, Filter 5 = distribution-yield vs required, NOT a growth DCF), FFO/AFFO · LTV · occupancy · WALE, distribution-yield MoS, numeric buy-below + trim/sell triggers, kill criteria, 5-tier verdict. Live price via BSE code for InvITs (GOOGLEFINANCE NSE will not resolve — see ALTIUS).
10. NEVER edit the LOCKED Action Dashboard. The stock dashboard ([[feedback-sheet-is-locked]]) is canonical and untouched by this pipeline. ETFs/REITs publish ONLY to the 🌍 Global & Passive Base tab. Back up the 🌍 tab before any write; if a helper disagrees with the live tab, FIX THE HELPER, never the sheet.
11. Sell ₹ + Upside are written for every NON-CORE verdict — never blank them (mirrors the stock rule). WATCH / AVOID / TOO_HARD each carry a numeric Sell/trim target AND an Upside% (honestly negative when the price is above fair value). ONLY CORE / COMPOUND-TREND show "∞ hold". For REIT/InvIT the trim target is the price at which the yield compresses materially below required (ALTIUS: trim above ~₹185, sell above ~₹200).
12. 5-tier verdict exactly one per vehicle: CORE / COMPOUND-TREND / WATCH / AVOID / TOO_HARD.

CONSULTING THE BRAIN (data/scripts/32_consult_brain.py — the ONE retrieval entrypoint)

set -a && source /Users/Dhiraj/dev/invest/.env && set +a && /Users/Dhiraj/dev/invest/.venv/bin/python \
  /Users/Dhiraj/dev/invest/data/scripts/32_consult_brain.py \
  --company "<trend or fund name>" \
  --model <passive-etf|commodity-etf|reit|invit> \
  --step <circle|conviction|intrinsic-value|margin-of-safety|opportunity-cost|verdict|inversion|capital-allocation|sizing> \
  [--corpus <binding|blended>] [--entities <peer-slugs,comma-sep>] [--k 8] \
  --json-out extracted/grilling/<CODE>_<step>.json

• NEW models (Phase 2, the only change to 32_consult_brain.py — 4 dicts: MODELS, MODEL_PHRASE, MOAT_HINT, MGMT_HINT):
  • passive-etf — broad/thematic index funds. MOAT_HINT = trend durability × wrapper quality (cost/tracking/liquidity/structure/tax); MGMT_HINT = tracking discipline + AMC/issuer reliability.
  • commodity-etf — gold/silver/commodity wrappers. MOAT_HINT = hedge/diversification role + storage/structure quality; valuation lens = real-asset / cycle, not earnings.
  • reit — real-estate trusts (e.g. Embassy). MOAT_HINT = asset + lease durability, tenant diversification, sponsor; MGMT_HINT = sponsor quality + capital allocation (dividend/M&A).
  • invit — infra trusts (e.g. ALTIUS/Embassy-style annuities). Same as reit but conviction CAP 2; MOAT_HINT = contracted-tenancy annuity + WALE; MGMT_HINT = sponsor + leverage discipline.
• What it does (unchanged): derives 2-3 model-aware retrieval questions, routes internally (specific model or entities → ScopedLanceDBBackend, 0 Azure calls; general + no entities → hybrid + GPT rerank — never stacked), RRF-merges + dedupes, hygiene-filters to mentors' atomics/raws + canon/perspectives + _synthesis (hubs/MOCs/_system/dossiers/positions excluded; every path verified on disk). Assets carry brain_model: <slug>.
• How to read the output: principles: [{claim, path, slug, citation, mentor, note_type, source_note, …}]. claim = the principle to apply; citation/slug = the ONLY tokens permitted in cites_principles; source_note = traceability to the raw transcript/letter/lecture. Empty principles → vault is thin here → ABSTAIN + flag, never pad.
• Corpus discipline: --corpus binding (Munger/Buffett ONLY) for every conviction/verdict/MoS/inversion gate — this is the arbiter. --corpus blended to ALSO surface the new world-canon (Bogle, Malkiel/Ellis, Damodaran-REITs, Dalio, Marks) + desk-synthesis atomics for trend-durability and wrapper context — advisory, down-weighted, never reverses a binding verdict.
• Step → lens map (binding): circle→circle (is this a trend/structure you understand?) · conviction→ moat(=trend-durability)+management(=wrapper/sponsor)+owner-earnings(=coverage for REITs) · intrinsic-value→ index-fair-value / REIT distributable-DCF · margin-of-safety→index-valuation-vs-history / yield-vs-required · opportunity-cost→opportunity-cost (vs the CORE index and the equity book) · verdict→trend+wrapper+opp-cost · inversion→fad-trend? + flawed-wrapper? + bubble-index? (disconfirming) · capital-allocation→REIT dividend/M&A discipline · sizing→kelly + conviction-cap + concentration (passive core disciplined, thematic capped).

FINANCE-DESK SPECIALISTS (additive advisory layer — they inform; the binding CIO at A/B/C decides)

Dispatch these sharpened desk skills to SHARPEN the analysis. They are an additive, --corpus blended-informed advisory layer (blended surfaces desk-synthesis atomics + the new world-canon ALONGSIDE binding); the perspectives they surface are down-weighted and NEVER reverse a --corpus binding verdict. The Munger/Buffett CIO stays THE arbiter. No new desk seats — the existing 8 desks gain ETF/REIT competence.

Skill	Attaches at	ETF role	REIT / InvIT role
moat-analysis	A (& B)	trend-durability judgment ONLY (NOT a passive ETF's "moat" — low cost is the moat, rule 8)	REIT asset / lease / tenancy durability
valuation-dcf-longrunway	B	n/a (index uses CAPE/PE-vs-history, not a DCF)	REIT distributable-DCF (FFO/AFFO over a long runway; the ALTIUS 12%-hurdle DCF)
bank-valuation	B	n/a	REIT leverage / justified-P-BV (LTV discipline, the levered-trust read)
forensic-accounting-redflags	B	ETF structure integrity (synthetic vs physical replication, securities-lending, issuer/counterparty risk) → CLEAN/CAUTION/RED	REIT sponsor + related-party + distribution-quality integrity (return-OF vs return-ON honesty)
capital-allocation-judge	B	n/a (passive — no capital allocation)	REIT dividend / M&A discipline (Outsiders scorecard on the sponsor; e.g. debt-reduction IPO)
portfolio-sizing	D	Max-size (passive core = disciplined anchor; thematic = capped satellite)	Max-size as an income sleeve, not a concentrated bet
decision-journal	D	thesis + kill-criteria + trend re-decision cadence	thesis + kill-criteria (yield-break, rate-shock, tenant impairment)
primary-research-sentiment	A & B	live trend sentiment (Reddit/X/news) — disconfirmation-seeking	tenant/sponsor/sector DD harvest

Also available: creating-financial-models (the DCF engine for REIT distributable-DCF and index fair-value), idea-researcher + deep-research (global-trend discovery & temporal "what's happening now" layering), grill-with-docs / grill-me (stress-test a trend conviction before it earns COMPOUND-TREND).

MODEL / EFFORT TIERING (max-effort, not ultracode)

• Stage A trend-durability scoring (the wealth engine): Sonnet workflow, vault-bound, one rigorous pass per trend; escalate a contested CORE/COMPOUND-TREND trend to Opus for the durability call.
• Stage B grading over the full set: Sonnet workflow (judgment, vault-bound, structured output) — the analyst-desk fan-out + MF-methodology's 8 fund subagents run in parallel.
• Stage C inversion (adversarial — fad? flawed wrapper? bubble index?): Opus/Fable (few agents, high stakes).
• Data fetch (TER/tracking/NAV/AUM/DPU/index-valuation) + 🌍 tab publish: mechanical (Bash / MCP) — no model reasoning.

STAGE A — Global-trend map (the discovery front-end / wealth engine)

Before grading any single fund, decide which durable trends are worth owning at all. This is the front-end that has no equivalent in a "cleanup" — it parallels the stock FOREVER/COMPOUND hunt.

1. Enumerate durable megatrends worth owning: broad-world · India structural growth · US innovation · AI / compute · energy transition · EM · gold/silver hedge sleeve. Write them to extracted/research/global_trend_map.md (the trend-map artifact).
2. Score each trend's DURABILITY with the binding CIO + the new perspectives: 32_consult_brain.py --company "<trend>" --model passive-etf --step circle (do I understand the structural driver?) then --step conviction (is this a 10-yr tailwind or a fad?). Pull Dalio/Marks/Bogle/megatrend voices via --corpus blended; cite ONLY returned slugs. A trend whose durability is unknowable → TOO_HARD, off the map. Optionally enrich with idea-researcher / deep-research for live, dated sentiment.
3. Map each KEPT trend to its best India-accessible wrapper: the lowest TER, tightest tracking error, deepest AUM/liquidity, best tax bucket vehicle (use mcp__indmoney__get_indian_stocks_details for the listed ETF + WebFetch mfdata.in/valueresearch for TER & tracking error; lookup_ind_keys first to resolve). If there is no clean India-accessible wrapper, the trend is still graded but flagged TOO_HARD-to-execute (the route may open later — US/LRS execution is kernel-deferred, out of scope to BUY, in scope to GRADE). Barrier: a ranked, cited trend map exists; every kept trend has a named best-wrapper candidate (or a TOO_HARD-to-execute flag).

STAGE B — Grade (like stocks), branching by asset type at Stage 0

For each candidate wrapper, branch by asset type:

B-ETF — passive / thematic / commodity ETF

The mechanical fund-quality layer reuses the MF methodology (extracted/mf/_mf_methodology.md) and the 6-factor rubric (extracted/mf/_mf_rubric.md):

1. Fund facts (mechanical): TER, tracking error, AUM/liquidity, structure (physical vs synthetic), tax bucket, issuer. Source: INDmoney get_indian_stocks_details (nav/aum/ltp) + get_mf_funds_details where applicable + WebFetch (mfdata.in / valueresearch for TER & tracking error).
2. MF-methodology subagents (parallel): Book Recon · Fund Facts · Holdings Quality · Overlap Mapper · Category Benchmarker · Cost & SIP Auditor · Debt Sleeve · Global Sleeve — the same 8 fund subagents, producing the wrapper-quality evidence.
3. Apply the MF portfolio gates (the known structural checks): Overlap (effective overlap >40% → downgrade weaker one band — this reproduces the held-book overlaps: 4 commodity ETFs → consolidate to UTI Gold + ABSL Silver; Mirae FANG+ vs S&P-500-Top-50 overlap → retire the duplicate lot) · Concentration (single fund >25% of the passive sleeve) · Cap-bucket / sector-tilt (e.g. PSU-Bank-BeES) · Global (intl FOF + US ETFs >15% NW → consolidate to the lower-cost ETF).
4. Conviction (trend × wrapper): the binding 32_consult_brain.py --model passive-etf --step conviction read, applied as JUDGMENT. Broad lowest-cost index → high; narrow thematic → must prove durability for the same band. Bad wrapper (high TER / poor structure / bad tax / wide tracking) → caps conviction or → AVOID.
5. MoS (index valuation vs its OWN history): CAPE / PE percentile of the underlying index (WebFetch the index CAPE/PE; --step margin-of-safety). A durable trend at a stretched index → WATCH. Cheap-vs-history
   • durable + best wrapper → COMPOUND-TREND (or CORE if it is the broad anchor).
6. Verdict + Sell/Upside: one 5-tier verdict; numeric Sell/Upside for every non-CORE row (rule 11). forensic-accounting-redflags checks ETF structure integrity (synthetic/securities-lending/counterparty).

B-REIT / B-InvIT — full ALTIUS-template dossier

Write vault/research/dossiers/<CODE>.md to ALTIUS depth (read vault/research/dossiers/ALTIUS.md as the template; frontmatter note_type: dossier, brain_model: reit or invit):

1. 6-filter walk-through, yield-instrument-adapted: Filter 1 Circle · Filter 2 Moat = asset quality + contracted tenancy/escalators (WALE) · Filter 3 Owner-earnings = distribution coverage + return-OF vs return-ON-capital split (the heart — strip return-OF-capital out of the headline DPU) · Filter 4 Management = sponsor quality + capital allocation (capital-allocation-judge) · Filter 5 MoS = distribution-yield vs required (~9–10%), NOT a growth DCF · Filter 6 Opportunity cost.
2. The metrics: FFO / AFFO · LTV (vs SEBI ceiling) · occupancy · WALE · NAV premium/discount · DPU split (return-ON ₹ + return-OF ₹) · headline yield vs sustainable (income-only) yield vs required.
3. Valuation = distribution-yield MoS (+ a valuation-dcf-longrunway distributable-DCF cross-check at a strict hurdle, the ALTIUS 12% read; bank-valuation for leverage / justified-P-BV). Buy-below = the price where the SUSTAINABLE income-only yield clears required; trim/sell = where the yield compresses below required.
4. Conviction CAP 2 (rule 6 — yield instrument, ceiling = WATCH). Kill criteria (tenant/MSA impairment, rate shock, distribution-coverage break, leverage creep). forensic-accounting-redflags on sponsor + distribution-quality. 5-tier verdict (in practice WATCH or AVOID for a REIT/InvIT — never COMPOUND-TREND/CORE).
5. Data: BSE filings for DPU (live price via BSE code — GOOGLEFINANCE NSE will not resolve for InvITs). Barrier: every candidate has a verdict + conviction + MoS; REIT/InvIT dossiers mirror ALTIUS depth.

STAGE C — Inversion (Opus, adversarial)

For every CORE / COMPOUND-TREND candidate, a skeptic pulls disconfirming vault principles (32_consult_brain.py --company "<trend/fund>" --model <model> --step inversion) and argues the strongest case AGAINST: is the trend a fad / already fading? (the megatrend that everyone already owns) · is the wrapper structurally flawed? (synthetic replication, securities-lending counterparty, a tax trap, chronic tracking drift, thin AUM that could close) · is the index in a bubble? (CAPE in its top decile — a durable trend bought at a euphoric index is still a bad buy). Survive all three axes → keep the tier; else demote (default = demote). decision-journal records the pre-mortem + kill-criteria. Desk perspectives inform but never lift a fund into CORE/COMPOUND-TREND — the --corpus binding gate decides.

STAGE D — Buy-list + sizing (mechanical-ish)

1. portfolio-sizing sets Max-size (--step sizing --corpus blended): the passive base stays a disciplined CORE (the anchor, sized for life); thematic = capped satellites (Kelly-fraction / half-Kelly, conviction-scaled ceilings, concentration discipline — few bets, hold the good ones, don't over-trim winners). REIT/InvIT sized as an income sleeve, not a concentrated bet.
2. decision-journal records the thesis + kill-criteria per kept fund (what we believe, the bet size, what would change our mind, the trend re-decision cadence).
3. Rank the WHOLE set (verdict → conviction → MoS); TOO_HARD quarantined off the buy-list. Produce the ranked, cited buy-list (is there a CORE/COMPOUND-TREND worth buying NOW at a sane valuation, vs WATCH the expensive ones).

STAGE E — Publish to the 🌍 Global & Passive Base tab (mechanical, no rework)

⚠️ READ BEFORE TOUCHING THE SHEET — the LOCKED Action Dashboard is NEVER touched ([[feedback-sheet-is-locked]]). ETFs/REITs publish ONLY to the 🌍 tab. The 🌍 tab is RE-ARCHITECTED from scratch per the plan (the old CORE-PASSIVE/ACCUMULATE 11-col sketch is a data point, NOT a template). BACK UP the 🌍 tab first; same auto-coloring / locked-format discipline as the stock dashboard; if a helper disagrees with the live tab, FIX THE HELPER, never the sheet, and never bypass a helper with raw writes for new/reordered rows.

Re-architected 🌍 tab schema (reads like the stock dashboard — a decision cockpit): Rank · Vehicle · Trend/Exposure · Verdict · Conviction · Buy-zone (index-value) · Live NAV/price · Upside · TER · Tracking-error · Tax-bucket · Liquidity/AUM · Held? · Max-size · Why (1-line) · Sources with a REIT / InvIT block carrying: Yield · Sustainable-yield (income-only) · NAV premium/discount · LTV · Occupancy.

• Rank key = lexicographic (verdict_tier [CORE < COMPOUND-TREND < WATCH < AVOID < TOO_HARD], −conviction, −MoS). No other tiebreakers. TOO_HARD quarantined off the buy-list rows.
• Values: conviction (capped 2 for REIT/InvIT), buy-zone = index-fair-value entry (ETF) or the sustainable-yield-clears-required price (REIT), verdict = the 5-tier. CORE / COMPOUND-TREND → "∞ hold"; WATCH / AVOID / TOO_HARD → numeric Sell/trim + Upside% (rule 11 — never blank a non-CORE Upside). Why = the one-line thesis from the dossier / fund brief (never an Obsidian [!info] callout).
• Publish runbook: (1) BACK UP the 🌍 tab (values + grid) to a revert point. (2) Build/extend the fund/REIT publish helper (parallels the stock multi_insert_* / sync_dashboard_extras.js helpers — reads the full grid, re-stamps Rank, carries any hyperlinks byte-for-byte, regenerates row-relative formulas, colors by verdict). Always --dry-run first. (3) Lock formatting if the layout changed.
• Reindex: new REIT/InvIT dossiers → .venv/bin/python data/scripts/26_build_lancedb_index.py --vault vault --db vault/.lancedb (auto-reloads the warm daemon; confirm /health). Remember the HEAD-WINDOW rule — a dossier correction must edit the head snapshot + one-line thesis, not just a tail append (only the ~4,000-char head vector is embedded).
• Update memory ([[reference-paise-se-paisa-sheet]] / the global-trend note) with the new 🌍 order + which trends are CORE/COMPOUND-TREND-worth-buying-now vs WATCH-the-expensive, and the held-book consolidations.

KEY REGISTRY

• Publish target: the 🌍 Global & Passive Base tab (re-architected) on the Paise se Paisa sheet (fileId 1N87younF990u-YGMOAiT8q6X-EtZ3jVovlWCF44orEY). The Action Dashboard is LOCKED and untouched.
• Brain models (Phase 2, data/scripts/32_consult_brain.py): passive-etf · commodity-etf · reit · invit — added to the 4 dicts (MODELS, MODEL_PHRASE, MOAT_HINT, MGMT_HINT); assets carry brain_model: <slug>. Retrieval auto-applies (no STEP/corpus/weight changes).
• Templates: REIT/InvIT dossier = vault/research/dossiers/ALTIUS.md (6-filter, DPU return-ON/return-OF split, distribution-yield MoS, buy-below/sell triggers, kill criteria, conviction cap 2). ETF fund-quality = extracted/mf/_mf_methodology.md (8 subagents + 4 gates + Buffett 5-filter) + extracted/mf/_mf_rubric.md (6-factor M1–M6 rubric + verdict bands + overlap rules).
• Scripts: brain consult data/scripts/32_consult_brain.py (the BINDING gate — scoped+cited); free-form vault RAG 29_vault_chat.py (ad-hoc only); index 26_build_lancedb_index.py; DCF engine via creating-financial-models / valuation-dcf-longrunway.
• Data sources: INDmoney (networth_holdings, get_indian_stocks_details nav/aum/ltp/sector, get_mf_funds_details, networth_allocation_breakdown, lookup_ind_keys) · WebSearch/WebFetch (TER & tracking error from mfdata.in / valueresearch; index CAPE/PE) · BSE filings (REIT/InvIT DPU; live price by BSE code) · optionally reddit / youtube MCP for live trend sentiment via idea-researcher.
• Artifacts: trend map extracted/research/global_trend_map.md; brain consults extracted/grilling/<CODE>_<step>.json; REIT/InvIT dossiers vault/research/dossiers/<CODE>.md.
• VERDICT TIERS: CORE (≈FOREVER, the broad lowest-cost anchor, price-exempt) · COMPOUND-TREND (≈COMPOUND, durable trend at a fair index valuation) · WATCH (great trend+wrapper, expensive index → wait) · AVOID (bad wrapper OR fad/fading trend) · TOO_HARD (durability unknowable OR no clean wrapper). Rank = verdict → conviction → MoS.
• CONVICTION (1–5, judgment within gates): broad lowest-cost index → high (low cost is the moat); narrow thematic → must prove trend-durability × wrapper-quality for the same band. Gates: fad/fading trend → AVOID; no clean wrapper → TOO_HARD; REIT/InvIT → CAP 2. MoS gate exempt only for CORE.
• OUT OF SCOPE: US-via-dollar/LRS execution (kernel-deferred — a trend may still be GRADED for when the route opens, just not BUYABLE); new desk SEATS (the 8 desks gain ETF/REIT competence, no new seats); active mutual-fund onboarding (a separate book); any trade; any edit to the LOCKED Action Dashboard.