Experimental Strategies (ES-family) — DORMANT REGISTER
Why this page exists
An idea can be worth tracking without being worth trading. Before this page such an idea had two homes and both were wrong: the status matrix (which implies a validated lens) or nothing at all (which means the next session re-derives it and re-pays the cost). The ES-family is the third option — recorded, numbered, dormant, and permanently attached to its own evidence.
The binding rule for every ES entry: it may be watched, charted and discussed. It may never be promoted to a ranker, an alert, a position size or a published score without first beating its own recorded control numbers under a pre-registered harness. Softening an ES status is a blocking error, exactly as with the canonical strategies (README §"Status matrix").
Deliberately kept out of the status matrix so a dormant experiment can never be read as a deployed lens.
Register
| # | Name | Seeded from | Status | Signal verdict | Ledger |
|---|---|---|---|---|---|
| ES-1 | Wounded Compounder (deep-drawdown reversal) | KPIGREEN, 2026-09-05 | DORMANT — watch only | ❌ No positive expectancy at any threshold or quality level. (The first −19.2 pp headline was n=56 and did not survive widening — see the fence.) | § 2026-09-05 + correction |
ES-1 — Wounded Compounder
The thesis as stated (owner's framing, preserved): a stock beaten down hard but showing reversal character across every lens at once — pivots, CPR, delivery, fundamentals, technicals — is worth owning even while the market reads it as bearish.
Mechanical definition — 10 legs, match = passed / evaluable, signal at ≥ 90%:
| # | Leg | Test | Family |
|---|---|---|---|
| 1 | drawdown35 | close ≤ −35% from 252d high | price |
| 2 | near52wlow | close within +8% of 252d low | price |
| 3 | rare_air | ≤10% of last 800 closes are below today | price |
| 4 | deliv_spike | max delivery% of last 5d ≥ 1.25× own 250d median | tape |
| 5 | vol_dryup | today's turnover ≤ 35th pctile of own 251d | tape |
| 6 | below_mas | close < 50DMA and < 200DMA | price |
| 7 | value_ok | trailing PE ≤ 20 and PB ≤ 4 | fundamental |
| 8 | growth_real | 3y sales CAGR ≥ 25% | fundamental |
| 9 | lev_stress | borrowings +≥50% YoY or D/E ≥ 1.0 | fundamental |
| 10 | roce_decay | latest ROCE < mean of prior 3 years | fundamental |
Liquidity floor ≥ ₹1 cr/day 20-session ADV, NSE EQ/CM only. Fundamentals point-in-time via fundamentals_history.report_date. Cap buckets are SEBI-style rank (top 100 LARGE· 101–250 MID· 251+ SMALL) on close × NSE-filed share count.
The fence (read before using this list for anything)
1. The signal shows no positive expectancy — but the first headline was overstated. Backtested 2017-01 → 2026-09, 116,498 liquid deep-drawdown observations, survivorship-clean (delisted included — COFFEEDAY appears as a 2019 match, which is the point). At ≥90% the 12m median was −22.1% vs a −2.9% matched control (−19.2 pp) — but that cell is n=56. Widening to ≥80% (n=347) collapses the gap to roughly −1.4% to −1.9%. The −19.2 pp figure is small-sample and does not survive widening the threshold; it should never have been the headline. What survives both cuts: no threshold and no quality bucket produces a positive median. 2. The seed is not evidence. The 10 legs were written from one 2026 name, which therefore scores 10/10 by construction. Circular by design; treat its score as a label, not a result. 3. The governance legs are missing and cannot be added retroactively. Promoter-buying and pledge were part of the original thesis but are excluded: those feeds start ~Nov-2025 and pledge is a single snapshot per symbol (1,576 rows / 1,465 symbols; only 96 with >1 datapoint). Any pledge-aware ES-1 variant is a live forward overlay only — see the ownership arcs notes, the surface that already owns those feeds. 4. This family has a uniformly negative prior. Reversal-context (the reversal context notes), BOOK_YIELD, EARN_YIELD, PULLBACK, DELIV_MOM, Accumulation-as-alpha — all recorded failures in the ledger.
Forward-tracking roster — frozen 2026-09-04
Screen run as-of 2026-09-04 (NSE bhav EOD). Universe 3,130 priced symbols → 1,429 liquid and evaluable. 2 names clear ≥90%, both SMALL cap — no MID or LARGE name matched at any threshold, which is itself the finding: at this date the configuration is a small-cap phenomenon.
⚠ CORRECTED 2026-09-05. The first published roster listed five names at ≥90%. That run had a defect in the
rare_airleg. Re-measured directly: SHAKTIPUMP 17.8%, AURIONPRO 26.9%, INDOTHAI 16.5% of their last 800 closes sit below today's — all three FAIL a leg that requires ≤10%. Only two names clear 90%. The superseded roster is kept below so the error is legible.
| Symbol | Cap | Match | Close ₹ | Mcap ₹cr | PE | PB | Failed legs |
|---|---|---|---|---|---|---|---|
| KPIGREEN (seed) | SMALL | 10/10· 100% | 307.0 | 6,400 | 12.6 | 2.1 | none |
| TRANSRAILL | SMALL | 9/10· 90% | 429.3 | 5,764 | 14.3 | 2.5 | lev_stress |
| SHAKTIPUMP | SMALL | 8/10· 80% | 473.0 | 5,837 | 22.6 | 3.4 | rare_air, value_ok |
| AURIONPRO | SMALL | 8/10· 80% | 689.7 | 3,811 | 18.0 | 2.2 | rare_air, deliv_spike |
| INDOTHAI | SMALL | 8/10· 80% | 35.8 | 477 | 7.2 | 1.7 | rare_air, roce_decay |
TRANSRAILL carries a listing-history caveat. Its rare_air is computed over 420 bars, not 800 — it listed in Dec-2024. A recently-listed stock passes "rare air" almost by construction, because it has no old history to sit above. Its 0.5% is much weaker evidence than KPIGREEN's 0.5% over a full 800.
80–89% near-miss band (from the same defective run — treat as indicative only, not re-verified): ACC· DAMCAPITAL· HMAAGRO· IPL· IRCON· KITEX· KPEL· KPITTECH· NIITMTS· OSWALPUMPS· PATANJALI· RUSTOMJEE· SANDUMA· WAAREERTL· WIPRO
Does business quality rescue the pattern? (asked by the owner, 2026-09-05)
Widened to ≥80% matches, n=347 with a 12m forward — a far larger sample than the ≥90% cell — and split by business quality at signal date:
| Bucket | n | 12m median | Hit rate |
|---|---|---|---|
| Interest cover — worst third | 115 | −7.2% | 44.3% |
| Interest cover — middle third | 115 | +0.3% | 50.4% |
| Interest cover — best third | 116 | −0.9% | 47.4% |
| ROCE — worst third | 115 | −6.9% | 44.3% |
| ROCE — middle third | 115 | +0.3% | 50.4% |
| ROCE — best third | 117 | −0.8% | 47.0% |
| ROCE ≥15 AND interest cover ≥3 | 113 | −1.9% | 43.4% |
| everything else | 234 | −1.4% | 49.1% |
Read: low quality is genuinely the worst cell (−7%), so quality avoids the tail. But the explicit good-business combination does not beat the rest, and no bucket produces a positive median. Quality removes the disasters; it does not create an edge. The pattern is flat-to-slightly-negative whatever the business is.
Portfolio status
DECLINED — not created, and not an open task. The owner ruled on 2026-09-05 that no portfolio is needed. Nothing was written to stocks_in_play; there are no ES-1 rows in any book. A future session should read this as closed, not as work waiting to be picked up.
Adding a new ES entry
1. Number it sequentially (ES-2, ES-3…). Never renumber an existing one. 2. Give it a mechanical definition before looking at any outcome — pre-register the legs. 3. Run it against a matched control, not against the index. "Beat the market" is the wrong question for a conditional screen; "beat the names that share its precondition" is the right one. 4. Append the result to the strategy ledger — wins and failures both. 5. Add the row to the register above with its verdict visible in the row itself.
Related
README status matrix· the reversal context notes (the closest falsified neighbour)· the ownership arcs notes (owns the pledge/insider feeds ES-1 cannot backtest)· the patearn notes (the quality lens ES-1's value legs lean on)· the strategy ledger (all numbers).