🏦⚡ stockloop desk

# #stockloop retro — 2026-07-07 (Tuesday) · THOROUGH / hard reckoning

## The number, no spin
**Today: −₹403.65 realized** (3 stops: MOTHERSON −242, COCHINSHIP intraday −74.8, HDFCBANK −86.85).
Only RELIANCE (6@1307, bracketed, ~flat-green) carries.
**Lifetime real (actual Kite): cum ≈ −₹271.59 — NEGATIVE.** Win rate 28.6%. ~5 green / 12 red days.
**Verdict: the desk is a net loser over a month of live execution. That is the fact to sit with.**

## What actually happened today
- **MOTHERSON −242 (−1R):** legit swing (auto tailwind + acquisition), held to the GTT stop, no discretionary cut. Stop worked (filled 141.3 before a flush to 136.71). This one was *correct process, planned loss.*
- **HDFCBANK −87 (−1R):** I MARKET-deployed it on the owner's "utilize more" push (fresh Q1 + bank tailwind, bought mid-range). It drifted straight to its 828 stop as the tape rolled red. Process fail: market-buy into a sell-heavy book.
- **COCHINSHIP intraday −75:** owner-pushed last-hour reversal trade; faded with the red close. PLUS an execution BUG — modify_order defaulted qty→1, unprotecting 6 shares ([[kite-modify-order-qty-default]]).
- **RELIANCE +8 (carry):** the ONE clean entry — a patient pullback-LIMIT that filled at 1307 on a real dip, bracketed. The only thing that worked.

## The honest root-cause (why we keep losing — not one bad day)
1. **We are trading heuristics without a VALIDATED, backtested edge.** Phase-1 (CLAUDE.md) explicitly found NO tradeable edge; the swing desk runs on "catalyst + pullback" rules that have never been proven to have positive expectancy on real fills. The claimed +0.74R expectancy is contradicted by the actual −₹272 cum. **The map says no-edge; the P&L agrees.**
2. **A choppy/rolling tape punishes marginal entries.** MOTHERSON & HDFCBANK both diverged from a green tape then stopped when it rolled — we keep buying names that are soft *relative to* the market.
3. **"Deploy more / trade intraday" pushes ADD losses, not gains.** Every forced action today (HDFCBANK market-buy, COCHINSHIP intraday) lost; the only winner-in-waiting is the one we DIDN'T force (RELIANCE limit). Activity ≠ edge.
4. **What kept us alive: discipline.** Every position was bracketed; losses were −1R and small (−₹403 on a ~₹86k book = −0.47%). We did not blow up. That's the floor, not a win.

## What I actually built (the non-P&L value — real, but not money)
- Turned a FALSE +₹458 hand-ledger into the TRUE −₹272 via actual-Kite reconciliation (the desk was lying to itself; now it can't).
- Public honest dashboard (stocks.deemwar.com), one unified DB (spec 030), bidirectional comms, a ₹3L paper panel + signal-agent fleet.
- **This is infrastructure + honesty. It is NOT a P&L edge. The owner is right to call that out.**

## The one forward-looking asset: the paper panel's edge map (today)
- ✅ **Order-win chase + heavily demand-led book** (RITES +₹730) — the ONE chase with edge.
- ✅ **Midcap demand-momentum** (JUBLFOOD +₹240) > large-cap (TITAN −293 stopped).
- ⚠️ **Reversal-on-quality** works on a stable tape (intraday green), FAILS on a broad-red day (COCHINSHIP −154 paper / −75 real).
- ❌ Plain momentum-chase, knife-catch, short-the-crasher, chase-vs-limit — ALL lost.

## The hard recommendation (owner call)
**Stop risking real capital on unvalidated heuristics.** Two honest options:
1. **Shrink to near-zero real size** (₹2-4k/trade) until an edge is *proven*, and run the paper panel to VALIDATE one (order-win-chase and midcap-momentum are the two candidates with a real signal — backtest them properly, ≥30-event sample, before scaling).
2. Keep the desk as an *honest measurement + research* engine (which it now is) and only deploy real size on a setup that clears a pre-registered edge bar.
**What NOT to do: keep forcing daily trades to "make green" — that is precisely what produced −₹403 today.**

## Carry into 07-08
- RELIANCE 6@1307 (GTT 1284/1400) — the only position. Manage to plan.
- Kite login each morning. Reconcile Console EOD. Fix: never modify_order without full qty.
- Bring the "shrink size / validate edge first" decision to the owner before next session.