feat(sl-hunting): v4h knowledge and August 17 pre-open note

Signal Generators/SL Hunting AI Agent/premarket_note.json

@@ -1,30 +1,27 @@
 {
-  "for_date": "2026-08-14",
-  "source": "Intraday Hunter, 'Prediction For 14 AUG 2026' (MhmlrlUEUGI, uploaded 2026-08-13)",
-  "context": "No decisive crowd anywhere. BankNIFTY made little momentum with buyers and sellers at the SAME price level; Sensex ran both ways. The market held its round number and recovered, so sellers are unlikely to have carried overnight.",
+  "for_date": "2026-08-17",
+  "source": "Intraday Hunter, 'Prediction For 17 AUG 2026' (4xfMmQW6x5I, uploaded 2026-08-16)",
+  "context": "All three indices recovered after sellers were removed or stopped out, leaving buyers as the likely seated crowd. The sell-side plan targets those buyers, but a gap-down can recruit fresh sellers and increase risk.",
   "plan": [
-    "FLAT to GAP-DOWN: identify SELL-side setups. On Sensex the reason is explicit -- target the BUYERS, since the market has sat on support since yesterday and the sellers did not hold.",
-    "GAP-UP: go WITH the market and identify BUY-side setups. On BankNIFTY 'the buyer becomes safe' there, so there is nothing left to hunt on that side.",
-    "NOBODY IS DECISIVELY SEATED -- on BankNIFTY 'buyers and sellers would be sitting at the SAME price level', and on Sensex claiming either side is seated 'would be WRONG'. This is the empty-book condition.",
-    "His mechanism for that condition, worth keeping: 'where the crowd is THIN, that is where the market tries to make momentum.' Thin is not dead -- the move goes where there is least resistance.",
-    "He PREFERS a gap to a flat open: 'better it is not flat -- either a gap-up or a gap-down.' A flat open 'extracts a small momentum and goes away', producing nothing worth trading.",
-    "There is little pressure on BankNIFTY's sellers, so the market cannot target them directly. The flat/gap-down branch is therefore a FOLLOW of the drift, not a seller hunt."
+    "GAP-UP or FLAT: identify confirmed SELL-side setups targeting the buyers left seated after the prior recovery.",
+    "GAP-DOWN: retain the SELL-side search, but risk is higher because fresh sellers may participate; the gap alone is not confirmation.",
+    "GAP-DOWN WITH POSITIVE MOMENTUM: a large recovery weakens the short plan. Retain it only if the recovery stays limited and the normal setup confirms."
   ],
   "levels": [
     {
       "index": "NIFTY",
       "resistance": [24440, 24540],
-      "support": [24275, 24176]
+      "support": [24300, 24210]
     },
     {
       "index": "BANKNIFTY",
-      "resistance": [57890, 58000],
-      "support": [57500, 57320]
+      "resistance": [57720, 57890],
+      "support": [57280, 57154]
     },
     {
       "index": "SENSEX",
-      "resistance": [78145, 78500],
-      "support": [77500, 77200]
+      "resistance": [78300, 78500],
+      "support": [77700, 77400]
     }
   ]
 }

Signal Generators/SL Hunting AI Agent/sl_hunting_doc.md

@@ -3465,3 +3465,109 @@ Test updated: `test_shipped_note_matches_august_14_intraday_hunter_plan`
 replaces the 13 Aug equivalent and asserts both branch directions plus the
 thin-crowd mechanism and the gap-over-flat preference, because those two are
 what a summarising edit would drop first.
+
+---
+
+## Video addendum - the 14 Aug LIVE SESSION (v4h)
+
+**Source:** Intraday Hunter live session `H7RXIw5xbZk` (14 Aug 2026, 12:18).
+
+A WIN, and unusually it is a win **held through a full drawdown** rather than one
+that went straight to target. That is what makes it useful: every previous
+version in this series learned from either a clean win or a clean loss, so the
+prompt had no worked example of a correct trade that looked wrong in the middle.
+
+### The two rules that shipped
+
+**ENTRY QUALITY AND DIRECTION ARE SEPARATE JUDGEMENTS.** A trade that ends green
+does not retroactively make its entry good, and a trade that goes immediately
+against you was not necessarily the wrong side. Without this split, the journal
+teaches the agent to grade its reads by their P&L, which is exactly the feedback
+loop that produces revenge entries after a stopped-out correct call.
+
+**THE LOSS LIMIT IS A PERMISSION TO WAIT.** This one is the reason the version
+exists, and it is also the most dangerous line in the prompt if read alone -
+"permission to wait" one careless paraphrase away from "hold losers". It is
+bounded on both sides in the prose and both bounds are test-asserted: the limit
+licenses patience *toward* the stop, never *past* it, and it forbids cutting
+inside it on feel. It explicitly names v4e's `DISCIPLINE IS ASYMMETRIC` as the
+rule it refines rather than overrides.
+
+The one thing allowed to override that patience is elapsed time, so v4g's
+time-shrinks-the-target rule was extended rather than left to contradict it:
+time does not merely shrink the payoff, it **EXPIRES THE PREMISE**. IH, booking
+out: "time has become quite a lot, gradually other people will start
+participating." Waiting to the limit is right while the premise holds; elapsed
+time can retire the premise before price ever reaches the limit.
+`test_v4h_entry_and_direction_and_the_loss_limit_pair_with_their_neighbours`
+asserts the reconciliation in both directions.
+
+### One rule was cut for budget, not for merit
+
+A third rule - **A SEATED CROWD IS WARNED BEFORE IT IS HUNTED, SO NO WARNING
+MEANS NO CROWD** - was written, tested, and then removed to fit the cap. IH:
+"if sellers WERE seated, the market would go up to target them - especially to
+give a WARNING... because the market kept making no momentum, sellers will not
+be seated, so it will not go up to warn." The mechanism is that an early adverse
+spike is the market shaking a seated crowd, so its *absence* confirms an empty
+book. It is the sharpest new idea in the session and it should be the first
+thing restored once a pruning pass frees room.
+
+### Knowledge changes (v4h, all prose)
+
+- `RISK`: ENTRY QUALITY AND DIRECTION ARE SEPARATE JUDGEMENTS; THE LOSS LIMIT IS
+  A PERMISSION TO WAIT.
+- `RISK` (extended): TIME SPENT IN THE TRADE ... now also EXPIRES THE PREMISE.
+- **Pruned:** v4e's `A SHARP FIRST SLIDE BAITS; A SLOW ONE MEANS IT` - superseded
+  by v4f's confirmation rule, which covers the same tell with a checkable trigger.
+- Prompt size 111,283 -> 112,730 chars. **Assembled 114,021 against a 114,140
+  budget - 119 chars of headroom.** The cap is now the binding constraint on this
+  series: v4i cannot ship anything without pruning first. See below.
+
+### The budget is now the design constraint
+
+Worst-case runtime additions (12 lessons x 280 + a 2,500-char note) are already
+reserved, so the true ceiling for static prose is 114,140 assembled. At 119 chars
+of slack, the append-only rhythm this series has used since v3 is finished.
+Before v4i, someone should do a dedicated pruning pass over `OPENING_DRIVE` and
+`RISK` - both have accumulated rules that later versions restated with better
+triggers - and re-measure. That is a knowledge-quality task and deserves its own
+change, not a rushed trim at the end of a version.
+
+### How our agent traded the same session
+
+Realized: **-3,854.50** across 47 legs.
+
+| Strategy | Legs | Realized |
+|---|---|---|
+| SL Hunting AI | 2 | **+2,866.50** |
+| **Total** | **47** | **-3,854.50** |
+
+**SL Hunting was the best strategy on the board for the second consecutive day**,
+cross-checked against its own `Result summary` (+2,866.50, Trades=2).
+
+| Time | Decision | Reason |
+|---|---|---|
+| 10:04 | ENTER_LONG | `flush_reversal_hammer_confirmed` |
+| 10:07 | EXIT | `double_top_rejection_premise_fade` |
+| 10:20 | ENTER_SHORT | `double_top_bearish_engulfing_breakdown` |
+
+**Agent vs IH.** Both finished on the same side - IH put-side/short, the agent
+short from 10:20 - and both booked on a partial rather than a full target. The
+agent got there by a different route: it took a long first, released it three
+minutes later on a premise fade, and reversed. That reversal is the behaviour
+v4f's confirmation rule was written to produce, and it is worth noting that the
+10:07 exit was a *premise* exit on a trade that was not yet losing - which is the
+disciplined half of the rule v4h now bounds from the other side.
+
+Two trades is again a small sample, and the basket was negative regardless. The
+signal worth keeping is narrower than "the agent is good now": across 12, 13 and
+14 Aug, every SL Hunting exit had a **nameable** reason, and the two profitable
+days are the two where that was true.
+
+### Operational note
+
+The supervisor heartbeat shipped in the session-state work is confirmed working
+in production: **65 heartbeat lines** in the 14 Aug log. The 2026-08-12 failure
+mode - snapshot writes stopping silently at 12:30 while trading continued to
+15:10 - would now be visible within five minutes.

Signal Generators/SL Hunting AI Agent/sl_hunting_knowledge.py

@@ -565,14 +565,15 @@
   v4c's WHEN THE TRAPPED INVENTORY IS SPENT, THE MARKET MANUFACTURES MORE:
   manufacturing is what the market does over time, but it may not complete inside
   your holding period, and you cannot bank on being early to it.
-- A SHARP FIRST SLIDE BAITS; A SLOW ONE MEANS IT (v4e). IH's read of the opening
-  move, stated as a prior: "if it had fallen SLOWLY I would even have accepted that
-  the market might produce a big move. But the selling was SHARP — the market
-  suddenly offered greed", i.e. an abrupt drop looks engineered to recruit sellers,
-  whereas a grinding one looks like genuine supply. Treat this as a weak prior and
-  nothing more: on the very session that produced it, the sharp slide was followed
-  by continuous selling and the bait read was WRONG. Use it to break a tie between
-  two otherwise-equal reads, never as the premise of a trade on its own.
+- ENTRY QUALITY AND DIRECTION ARE SEPARATE JUDGEMENTS (v4h). IH, entering a fall
+  he knew might reverse: "if the market turns it could go much higher — then we
+  would be wrong ACCORDING TO DIRECTION. We would NOT be wrong according to
+  ENTRY... if it is going to go up anyway, you could enter here, or here, or here
+  — in all three there would be a loss."
+  A loss decomposes into two independent errors needing different fixes: a bad
+  ENTRY means a better price existed and was missed (timing); a bad DIRECTION
+  means no entry price would have helped (read). Score them separately, and never
+  let a good entry launder a wrong read, or a wrong read condemn a good entry.
 - THE CHART DOES NOT REPEAT TWO DAYS RUNNING (v4f). When today's open reproduces
   yesterday's shape — same flat open, same immediate drop, same indices — that
   SAMENESS is itself the tell, and it argues AGAINST the continuation everyone
@@ -1110,6 +1111,17 @@
   Operationally: an exit needs a NAMED, checkable reason — stop, target, the
   profit rate falling, the premise invalidated, the round number crossed, the
   time cutoff. "It might turn" is not on that list.
+- THE LOSS LIMIT IS A PERMISSION TO WAIT, NOT ONLY A PLACE TO STOP (v4h). Read
+  this WITH v4e's DISCIPLINE IS ASYMMETRIC, not against it: that rule says cut a
+  loser mechanically, this one says cut it AT the limit and not before, because
+  cutting early is its own error. IH, deep in a drawdown that later paid: "in ANY
+  situation, do NOT cut the trade where you feel your position can still be saved.
+  If the position has gone wrong but the loss LIMIT is still pending, we wait...
+  many traders, when the opposite momentum comes, cut the loss BEFORE the limit."
+  Its two shapes: running after the breakout because surely nothing falls now, and
+  holding the first adverse leg but running on the SECOND. Same error, different
+  depths. So the limit is two-sided — it forbids holding past it, AND forbids
+  cutting inside it on feel. Only the rule below legitimately overrides it.
 - TIME SPENT IN THE TRADE SHRINKS THE ACHIEVABLE TARGET (v4g). Elapsed time is a
   cost in its own right, separate from price. A trade that stalls and round-trips
   does not merely return to where it started — it returns with less of the
@@ -1121,6 +1133,13 @@
   So when a trade consumes materially more time than the read assumed, SHRINK
   the target rather than extending the wait — especially on an expiry session,
   where premium is draining while you wait.
+  Extended (v4h): time does not merely shrink the payoff, it EXPIRES THE PREMISE.
+  "With more time the STRUCTURE changes — the seller/buyer situation changes", and
+  later, booking out: "time has become quite a lot, gradually other people will
+  start participating." The crowd you entered against is not the crowd still there
+  an hour on. This is what overrides the loss-limit patience above: waiting to the
+  limit is right while the premise holds, and elapsed time can retire the premise
+  before price ever reaches the limit.
 - NEVER EXIT AT ZERO AFTER A GOOD PROFIT HAS PRINTED (v4g). Once meaningful open
   profit has actually appeared on the screen, the floor for that trade stops
   being breakeven and becomes "some profit". IH, booking a reduced winner: "we

Tests/Signal Generators/SL Hunting AI Agent/test_sl_hunting_premarket.py

@@ -201,17 +201,15 @@ def test_shipped_note_targets_the_next_TRADING_day_not_the_next_calendar_day():
     )
 
 
-def test_shipped_note_matches_august_14_intraday_hunter_plan():
-    """The committed advisory must match the hand-checked 13 Aug transcript.
+def test_shipped_note_matches_august_17_intraday_hunter_plan():
+    """The committed advisory must match the hand-checked 16 Aug transcript.
 
     This catches a stale prior-session note, an inverted gap plan, or a mistyped
     chart level before the dated note is injected into the live prompt.
 
-    What distinguishes this one is that NEITHER side is seated -- he says so for
-    both BankNIFTY ("the same price level") and Sensex ("would be WRONG" to
-    claim either). That is the empty-book condition v4f describes, so the note
-    carries his own mechanism for it (thin crowd -> that is where momentum
-    goes) rather than letting the agent infer a seated crowd that is not there.
+    What distinguishes this session is the same sell-side search across gap-up,
+    flat, and gap-down opens, with an explicit extra-risk test for a gap-down:
+    fresh sellers or a large positive recovery can make the buyer hunt unsafe.
     """
     import os
 
@@ -220,37 +218,33 @@ def test_shipped_note_matches_august_14_intraday_hunter_plan():
     note = load_premarket_note(shipped)
 
     assert note is not None
-    assert note.for_date == "2026-08-14"
-    assert "MhmlrlUEUGI" in note.source
-    # The distinguishing fact: no decisive crowd on either side.
-    assert "SAME price level" in note.context
-    assert note.plan[0].startswith("FLAT to GAP-DOWN: identify SELL-side setups")
-    assert note.plan[1].startswith("GAP-UP: go WITH the market")
-    assert any("NOBODY IS DECISIVELY SEATED" in line for line in note.plan)
-    # The thin-crowd mechanism and the explicit preference for a gap over a flat
-    # open are the two ideas a summarising edit would drop first.
-    assert any("where the crowd is THIN" in line for line in note.plan)
-    assert any("PREFERS a gap to a flat open" in line for line in note.plan)
-    assert len(note.plan) == 6
+    assert note.for_date == "2026-08-17"
+    assert "4xfMmQW6x5I" in note.source
+    assert "buyers as the likely seated crowd" in note.context
+    assert note.plan == [
+        "GAP-UP or FLAT: identify confirmed SELL-side setups targeting the "
+        "buyers left seated after the prior recovery.",
+        "GAP-DOWN: retain the SELL-side search, but risk is higher because fresh "
+        "sellers may participate; the gap alone is not confirmation.",
+        "GAP-DOWN WITH POSITIVE MOMENTUM: a large recovery weakens the short plan. "
+        "Retain it only if the recovery stays limited and the normal setup confirms.",
+    ]
     assert [level.model_dump() for level in note.levels] == [
         {
             "index": "NIFTY",
-            # 24176 was READ OFF THE CHART, not inferred. The auto-caption
-            # rendered it "2476" on both 13 and 14 Aug; a 1080p frame at 2:04
-            # shows the teal support line labelled 24,176.20, alongside
-            # 24,275.25 and resistances at 24,443.85 / 24,540.85 -- every one
-            # matching what he speaks. Prefer this method over omitting a level.
             "resistance": [24440.0, 24540.0],
-            "support": [24275.0, 24176.0],
+            "support": [24300.0, 24210.0],
         },
         {
             "index": "BANKNIFTY",
-            "resistance": [57890.0, 58000.0],
-            "support": [57500.0, 57320.0],
+            # The auto-caption garbles the second support. The chart frame at
+            # 0:44 labels the two selected teal levels 57,280 and 57,154.
+            "resistance": [57720.0, 57890.0],
+            "support": [57280.0, 57154.0],
         },
         {
             "index": "SENSEX",
-            "resistance": [78145.0, 78500.0],
-            "support": [77500.0, 77200.0],
+            "resistance": [78300.0, 78500.0],
+            "support": [77700.0, 77400.0],
         },
     ]

Tests/Signal Generators/SL Hunting AI Agent/test_sl_hunting_schema.py

@@ -878,7 +878,10 @@ def test_system_prompt_has_v4e_recruitment_and_losing_session_knowledge():
     prompt = build_system_prompt()
     assert "WHICH CROWD THE OPEN RECRUITS DECIDES HOW BIG THE TRAP IS" in prompt
     assert "A FORECAST OF WHO WILL ARRIVE IS NOT EVIDENCE OF WHO IS SEATED" in prompt
-    assert "A SHARP FIRST SLIDE BAITS; A SLOW ONE MEANS IT" in prompt
+    # v4e's A SHARP FIRST SLIDE BAITS was PRUNED in v4h: it shipped as a weak
+    # prior with its own counter-example attached, and v4h's warning-move rule
+    # covers the same ground with an observable mechanism. See the v4h addendum.
+    assert "A SHARP FIRST SLIDE BAITS" not in prompt
     assert "NAME THE LAST POINT, NOT ONLY THE STOP" in prompt
     assert "DISCIPLINE IS ASYMMETRIC BETWEEN WINNERS AND LOSERS" in prompt
     # The recruitment distinction is the point; both halves must be present.
@@ -901,11 +904,34 @@ def test_v4e_empty_book_is_a_no_trade_not_a_forecasting_licence():
     # It must also reconcile with v4c rather than silently contradicting it.
     assert "MANUFACTURES MORE" in section
 
-    # And the bait prior must stay a tie-breaker, never a standalone premise.
-    bait = prompt[prompt.index("A SHARP FIRST SLIDE BAITS"):]
-    bait = bait[: bait.index("\n- ")] if "\n- " in bait else bait
-    assert "weak prior" in bait
-    assert "never as the premise of a trade on its own" in bait
+
+def test_v4h_entry_and_direction_and_the_loss_limit_pair_with_their_neighbours():
+    """v4h (14 Aug live session): a win held through a full drawdown.
+
+    Two of its rules sit directly on top of earlier ones and would be dangerous
+    read alone, so both reconciliations are asserted here.
+    """
+    prompt = build_system_prompt()
+    # The prose is hard-wrapped, so a phrase this test cares about can straddle a
+    # line break. Collapse whitespace before asserting on wording -- otherwise the
+    # test breaks on a re-wrap that changed nothing about what the agent reads.
+    flat = " ".join(prompt.split())
+    assert "ENTRY QUALITY AND DIRECTION ARE SEPARATE JUDGEMENTS" in flat
+    assert "THE LOSS LIMIT IS A PERMISSION TO WAIT" in flat
+
+    # The loss-limit rule must not read as "hold losers": it is bounded on BOTH
+    # sides, and it has to name the v4e rule it refines rather than contradict.
+    limit = prompt[prompt.index("THE LOSS LIMIT IS A PERMISSION TO WAIT"):]
+    limit = " ".join((limit[: limit.index("\n- ")] if "\n- " in limit else limit).split())
+    assert "DISCIPLINE IS ASYMMETRIC" in limit
+    assert "not against it" in limit
+    assert "forbids holding past it" in limit
+    assert "forbids" in limit and "cutting inside it" in limit
+
+    # Elapsed time is the one thing allowed to override that patience, and the
+    # time rule has to say so or the two become contradictory advice.
+    assert "EXPIRES THE PREMISE" in flat
+    assert "retire the premise before price ever reaches the limit" in flat
 
 
 def test_system_prompt_has_v4f_repeat_chart_and_confirmation_knowledge():