The Union (RS turn ∪ RS trend, stock-level) — Canonical Reference

🟡 STATUS — READ BEFORE ANY NUMBER ON THIS PAGE

This is a PRE-REGISTERED LEAD, not a deployed strategy, and NOT investment advice. Every number below is IN-SAMPLE — the Union was selected after seeing 2005–2026 across ~30 configurations, so its 17.5% CAGR is a research lead inflated by selection, not evidence. It is SEALED for forward testing (the union prereg notes, SHA-256 a9a14058f2140e22639b9504ab6d4af9c60fc76144de0f9f5e47f21b1b98d21c) and may not be deployed until it clears its frozen forward criteria. No live surface — there is deliberately no /dash page, because a page would imply tradeability the evidence does not yet support.

One-line definition: a long-only, stock-level book that holds every liquid Indian stock which is EITHER recovering from an oversold relative-strength dip (the turn) OR showing confirmed, persistent relative strength (the trend) — top 60 equal-weighted, idle capital parked in a Nifty Next 50 sleeve while the market is healthy, each name closed on a −20% trailing stop, rebalanced quarterly.

1. What it is — and what it is NOT

It is the answer to a question the desk pressed all through the 2026-07 RS arc: "we are taking the best of the best of the best stocks and still struggling — why?" The arc's central finding (ledger 15P) was that chasing the single strongest names buys peak volatility and loses to the variance toll. The Union is the opposite instinct made mechanical: hold strong-but-not-extreme names, caught either as they turn up or once their strength is confirmed, and let a stop and a sleeve carry the risk.

It is NOT the sector ladder (the sector rotation notes, which selects sector INDICES, not stocks). It is NOT deployed, and it is NOT a recommendation. It is a sealed lead awaiting its own forward test.

2. Our variation vs. the standard technique

Textbook momentum buys the top-ranked names and rebalances. The Union departs on three of the desk's axes: (a) two entry modes, not one — a mean-reversion turn signal (rare in momentum books) is unioned with a trend signal, because the arc proved they fire on different stocks at different times (11% overlap; the intersection is only 9% invested); (b) the middle of the strength band, not the peak — the persistence gate and the oversold-turn both deliberately avoid the extreme top decile that 15P showed is dominated by "merely good"; (c) idle capital is never dead — the un-invested fraction works the index sleeve (the V17 lesson) instead of sitting in cash or being force-concentrated into a handful of names.

3. How it works — THE COMPLETE RULESET (definitional)

A. Universe (checked each rebalance). NSE bhavcopy series EQ + BE + BZ, prices corporate-action adjusted (the adjust code), with the runtime split-ratio quarantine (the quarantine code, ~156 symbols whose recorded split ratios do not reconcile — excluded, nothing written to the DB). A stock is eligible only if its prior-month average daily traded value ≥ ₹5 crore (liquidity, no look-ahead). Dead names realise −50% on delisting; a name that simply doesn't print on a date is carried, not punished. Zero survivorship bias — every stock is present on exactly the dates it was tradeable.

B. Sector assignment (PIT). Each stock is mapped to one of the 16 NSE sector indices by which sector its trailing-500-day excess returns correlate with most. No membership table is used (reproduces NSE's own labels at ~85%). This is what lets both signals reference a stock's own sector.

C. The two signals — a stock qualifies if EITHER fires (the UNION):

• Signal A — TREND (RSI + persistence): price RSI(14) is above its own 50-day simple moving average (SMA beat EMA in testing), AND the stock beat its own sector index on ≥ 70% of the trailing quarter's trading days.
• Signal B — TURN ("6b"): the RSI(14) of the RS line (stock ÷ its own sector) was below 30 within the trailing ~60 days and has crossed back ≥ 30. No persistence gate — 6b fires early, before persistence can exist.
• The two are mutually exclusive by construction (a stock just turning up hasn't had a persistent quarter). Combined by OR only — the intersection (AND) collapses to ~9% invested and underperforms; do not use it.

D. Ranking & sizing. Keep the first 60 qualifiers in the engine's qualifier order (symbol-load order — effectively an arbitrary, stable sample of the qualifying set), equal-weight (1/60 each). Correction (2026-07-16, ledger 16Z): this page previously said "rank by RSI strength" — the engine never ranked, and when ranking was actually tested, BOTH RSI-descending and RSI-ascending truncation LOST to the arbitrary order (15P physics: the top of the strength band buys the variance toll). The engine's behaviour, which produced every recorded number and runs the forward test, is definitional. Inverse-volatility sizing was tested (ledger 16X) and REJECTED as a wash — do not substitute it.

E. The residual sleeve (the V17 mechanic). The un-invested fraction (fewer than 60 names qualify, or stops have fired) is held in Nifty Next 50 while Nifty 500 ≥ its 200-day SMA, else in cash. The stock book is never touched by this switch. (Why sleeve-only: applied to the whole book, the same 200DMA kill destroys wealth — the standing V8→V17 lesson.)

F. Exit & cadence. A −20% trailing stop from each position's peak close, filled with 1% slippage. Quarterly rebalance. Costs 0.15% per side.

G. What was tested and must NOT be added (each rejected with numbers, ledger cited):

• Consistency gate on 6b — halves the return (6b + persistence are mutually exclusive). 16V.
• Market-stretch throttle on the invested fraction — worsened every metric. 16W.
• Inverse-volatility stock sizing — a wash, net slightly worse. 16X.
• BE-surveillance veto· fundamentals veto — one falsified, one inert. 16T.
• Sector-neutral name caps — starve the book; its sector concentration is load-bearing. 16Z.
• RSI-ranked truncation (either direction) — loses to the engine order (see D). 16Z.
• 6b threshold variants (25 / 35 / 25→30)· 6b∪6f· weekly-RS 6b· daily-AND-weekly confirmation — every one worse than the sealed rules. 16Z.
• Quality/valuation tilt at selection (Screener-table, G#8-disclosed) — dead; the 16T "fundamentals add nothing here" doctrine now covers tilts as well as vetoes. 16Z.
• ML re-ranking of the qualifiers — the pre-registered primary model failed its frozen bar (16AA); any new attempt requires a fresh pre-registration, never an in-sample score.

4. Status & candidate ladder

📋 The complete family record — every configuration in full (rulesets + all recorded numbers), the sealed validation (C1/C2/C3 significance· interim OOS· deflation· PBO), and the deflated forward expectations — lives in the **Union Ladder compendium**.

• The UNION — the sealed lead. In-sample 2006–2026: the best full-period result of the RS arc. (Numbers live in the ledger, §16V, never restated here.)
• UNION-β14 — the pre-registered SIBLING lead (2026-07-16, ledger 16Y): the union plus exactly one rule — a qualifier whose trailing-250d beta vs Nifty 500 exceeds 1.4 is excluded at selection. In-sample it beats the union on return, drawdown, beta AND alpha together and flips the 2012–17 window positive, surviving four pre-declared kill checks (threshold plateau, beta-window, dead-cash decomposition, missing-data). Same epistemic class as the union — an in-sample-selected lead — so it is sealed beside it (the union beta14 prereg notes, SHA-256 08b46199…) with a frozen sibling- adjudication rule; the forward window judges both.
• UNION-C40RA — the THIRD pre-registered sibling (2026-07-16, ledger 16AB): β14 plus exactly two rules — rank the capped qualifiers by RISKADJ (6-month return ÷ 3-month volatility, the estate's best-of-32 factor) and hold the top 40. In-sample it is the family's best (all three windows alpha-positive) and it survives the 2%-slip and next-day-execution honesty passes; the worst honest case still clears every sibling. Third-generation in-sample selection — sealed (the union c40ra prereg notes, SHA-256 0715a0d9…) with the multiplicity risk disclosed inside and a frozen three-way adjudication. The family stops at three registrations.
• A2-COMPOSITE — a DEFERRED LEAD, not a registration (2026-07-16, ledger 16AE): the C40RA machinery with an ERA-RELATIVE liquidity floor (monthly ADV percentile calibrated to today's ₹5cr equivalent, clamped at ₹1cr) and the bear-state idle earning the documented risk-free rate. In-sample it is the first union-family book to clear the desk's 25% bar, surviving 2%/3%-slip and next-day-execution stress — but it CHANGES THE BOOK'S CHARACTER (small/mid-cap tilt in the early windows; median pick-ADV roughly a third of C40RA's; personal-scale execution only). The sibling family is closed at three, so this is recorded and HELD: registration awaits the 2026-10-03 forward verdict or the desk's explicit reopening (both queued decisions in the ledger entry).
• COMPOSITE-30 — THE CONFIRMED LEAD, the FOURTH sealed sibling (the desk 2026-07-16; ledger 16AF/16AH): the A2-composite plus concentration at top-30 (re-proven viable on the widened universe) and let-winners-run weight drift (retained names keep market-drifted weight, hard cap 5%). In-sample the family's best on return; the deeper drawdown was explicitly on the table when the desk confirmed it. The family was REOPENED by owner decision and this spec is SEALED (the union composite30 prereg notes, SHA-256 07ef2ef9…) with a four-way family adjudication. Fifth-generation in-sample selection, stated in the registration. Validated (ledger 16AL): the sealed ladder-validation protocol passed its frozen gate (the C40→COMPOSITE-30 increment is statistically real, p=0.014), survival 1.01 on the 2019+ interim hold-out, dead-name stress clean — with the era-floor rung flagged as the highest window-fit risk and the honest forward expectation deflated to ≈21.6% (the union ladder notes §9).
• A2-COMPOSITE — the lower-drawdown alternative, recorded IN FULL (the union ladder notes §5), unregistered, registration-ready on the desk's word.
• Signal A alone and Signal B alone — each beats the Next-50 bar in-sample but by less; the union beats both. Recorded reference points, not separate strategies.
• Rejected candidates (do not re-run): throttle (16W), inverse-vol (16X), the AND-intersection (16V), the twelve-candidate battery of 16Z (sector caps· RSI ranks· beta-rank-asc· 6b variants· 6b∪6f· weekly/MTF· quality tilt), the pre-registered ML rankers BOTH generations (16AA Ridge-primary· 16AG GBM-primary over the era-floor pool — the hand rule beat the machine both times; beta was the #1 learned feature all three runs), and the kills (16AC): trail widths other than 20· sleeve-index swaps· cap-floor refill· vs-bench consistency· monthly cadence (4th confirmation of the cadence law)· the cross-family LOWVOL_MOM blend (return-corr 0.83 — dilution). Concentration below top-30 and vol-based sizing stay dead (16AF narrowed the 16X sizing wall to vol-based only).

5. Known weakness (disclosed, not hidden)

Walk-forward, the Union is strongly positive in 2006–2011 and 2018–2026 but negative in 2012–2017 — a mid-cycle bull in which the book ran hot (high beta, near-fully invested) and still lagged. Two structurally-correct sizing fixes (exposure throttle 16W, inverse-vol 16X) both failed to close it, which established the weakness is selection (which stocks are picked in that regime), not sizing.

2026-07-16 update (ledger 16Y): the selection fix was found in-sample. A per-name trailing-beta cap at selection closes the window — and the effect survives a dead-cash decomposition, so it is the stock choice, not sleeve time. Because the fix was itself selected from a candidate battery on the same history, it earns only sibling pre-registration (see §4), not a claim. The forward window decides. The in-sample give-back is disclosed in the ledger: 2006–11 return is somewhat lower under the cap (fewer low-beta qualifiers existed then), at materially lower beta.

6. Data & provenance

NSE bhavcopy (primary source; corporate-action table VERIFIED complete vs NSE, ledger 15S) + NSE index closes (index_rows, adjusted). PIT throughout. Price-index benchmark (Nifty Next 50 price, not total-return) — disclosed wherever numbers are shown; the TRI re-cut is DONE (ledger 16AJ): the honest hurdle is Nifty Next 50 TRI 14.6%/yr and every family member's alpha survives it — TRI columns are reported beside every future judgment (the union ladder notes §8); the sealed criteria stay PR-vs-PR as frozen. The fundamentals_history veto experiment (16T) touched the Screener-sourced table read-only (Guardrail #8 disclosure) and is not part of the live ruleset.

7. Terminology canon

• The Union — hold a stock if the TURN signal OR the TREND signal fires. The binding name for this construct.
• Signal A / TREND — RSI(14) > its 50-day SMA, plus ≥70% same-quarter persistence vs the stock's own sector.
• Signal B / TURN / "6b" — RSI(14) of the RS line recovering from < 30 back above 30. ("6b" is its ID from the Dimension-6 reversal battery, ledger 16U, where it was the only one of eight indicators to survive.)
• The sleeve — idle capital in Nifty Next 50 above the 200DMA, else cash (the V17 mechanic).
• Distinguish RSI of price (Signal A's trigger) from RSI of the RS line (Signal B's trigger) — two different constructs.
• Do NOT confuse the Union (STOCK-level) with the sector ladder (the sector rotation notes, INDEX-level) or the descriptive RS suite (the relative strength notes).

8. Decision & session history

• 2026-07-15/16 (the RS arc) — the desk drove a full stock-level investigation after the sector ladder was found to select sectors not stocks. The arc produced 8 data/method retractions, killed ~9 ideas with numbers, and left exactly one survivor: the Union. Codex external review (ledger 15R) confirmed the honest framing. The two component signals trace directly to the desk's own theses (catch the turn; confirm persistence).

9. Open items

• The forward test is the ONLY thing that matters next — and it now adjudicates a THREE-SIBLING FAMILY. When a new quarter closes, run the Union (the union prereg notes), union-β14 (the union beta14 prereg notes) and union-C40RA (the union c40ra prereg notes) — touch no spec. Each is judged on its own frozen criteria; among passers, highest forward alpha graduates (the FOUR-way rule is frozen in the COMPOSITE-30 registration, which also records the desk's explicit reopening of the family). The scheduled 2026-10-03 task runs the union's engine; the same session runs ALL SIX ladder rows (the union ladder notes §8): the four sealed siblings judged on their own criteria, the two recorded composites for the record.
• ~~TR-benchmark re-cut~~ ✅ DONE (ledger 16AI/16AJ) — data + tool committed; the residue is the prod index_rows ingestion (feed-lane protocol), research needs are met from files.
• ~~The 2012–17 selection question~~ — answered in-sample 2026-07-16 (ledger 16Y, the beta cap) and moved into the sibling registration above. The remaining open questions are the forward evidence itself and the TR re-cut; the non-ML candidate space around this signal family is otherwise SPENT (16Z catalogues the kills). ML re-attempts require a fresh pre-registration (16AA).

10. Sources of truth

Ruleset + terminology: this page. **Every family configuration IN FULL (specs + all numbers, incl. the unregistered composites): the union ladder notes.** Every result number: the strategy ledger §§ 2026-07-16 → 2026-07-16 (the union arc), §§ 2026-07-16/Z/AA (the battery, the β14 sibling, the ML verdict) and §§ 2026-07-16/AC/AD (the battery, the C40RA sibling, the kills, the total-return measurement). Frozen specs + seals: the union prereg notes· the union beta14 prereg notes· the union c40ra prereg notes· the union ml prereg notes. Code: the cash blend research code, the cash 6b code, the dim6 code, the dim6g code, the cash throttle code, the cash ivol code, the union lab code, the union lab2 code, the union ml code, the union lab3 code, the union lab3b code, the blend u25 code.