feat(sl-hunting): v5g, and the 18 SEP pre-open note

[DoRmAmMu1997]

Three commits: the knowledge rule for this session, the confirmed cause of its one open question, and tomorrow's pre-open note.

First, the part that worked

This was the first live session with the 17 Sep note, SLH-017 and SLH-018 all active — and SLH-017 did exactly what it was built for. Both entry rationales read the computed verdict off the tool and stated it:

09:39 — "Open classified FLAT on both NIFTY (-0.096%) and BankNIFTY (-0.295%), triggering the pre-open note's FLAT-open SELL branch"

10:18 — "Flat open (NIFTY -0.10%, BankNIFTY -0.30%, both below the 0.5% threshold) matches the pre-open note's SELL branch"

Yesterday the model computed 0.36% by hand, called it a gap-up, and took the wrong branch four times. Today it was handed FLAT, cited the threshold, and took the branch the note prescribed. SLH-018 never fired — no reasonless exit arrived, which is the right behaviour for a guard.

Then the part worth encoding

IH reached the same read and took the same side — his reasoning is the note's flat branch almost verbatim: "when the market moves UP and does not cross the round number, sellers gradually come there too. But the market keeps HOLDING in one place… even if someone made a trade, they do not hold a positional trade."

And he lost on it.

	approach	result
IH	entered at the open, held	loss limit exhausted, cut — "there is no need to apply the brain here, cut the trade and get out"
Book	two bounded windows, flat by 11:00	+₹11,355

#	entry	exit	basket
1	SHORT 23,282.75 @ 09:39	09:43, into the stall at support	+₹2,820
2	SHORT 23,302.90 @ 10:18	11:00 TIME_CUTOFF	+₹8,535

NIFTY chopped between roughly 23,217 and 23,305 all morning, broke that range near noon and ran to 23,358.70 by 12:38. The read was right for about ninety minutes and wrong afterwards.

What v5g encodes

The reflexive claim, and it's the missing half of v5b. v5b reads an eviction and licenses a follow rather than a hunt. What that follow inherits is that nobody is trapped — so nothing is forced to move. And the reasoning that clears the crowd —

"if someone made a put trade and the market gave no momentum for three or four hours, his premium decays and he will not hold"

— is a description of the conditions you are about to enter, on the same clock, paying the same decay. The eviction argument is evidence about the past; it promises nothing about the next hour.

Placed after the whole v5c block rather than beside v5b, because v5c opens "The twin of the rule above" and that reference points at v5b — inserting between them would have silently re-aimed it. Same class of mistake as splitting v4t from its v4u sub-bullet yesterday, caught this time before writing.

The honesty clause is part of the rule

Trade 2's exit was not a decision. The agent had been failing every bar since 10:42 — 19 agent_error rows, a canned SLHuntingAgentError per bar for the last eighteen minutes — and the position was closed by the 11:00 cutoff. What won was the box, not the judgement.

A mutation that removes that sentence is caught, because without it the day reads as the agent out-trading him, which is false and flattering.

Verification

Negative-tested 11 ways, all 11 caught, plus a control mutation of unasserted prose that correctly did not trip the test.

Gates: 606 master · 28 market-data-health · 1509 pytest (+1) · ruff 0.16.7 · mypy (80 files + master) · compileall · bandit · coverage 73.8%, policy pass.

Considered and not encoded

• His emotion / rule-discipline passage — "the more you think about things, the more your decisions will be wrong", "the easiest way to control emotion is to follow the rule". Third session running to make that point; already carried by v3y plus the RISK loss-limit rules.
• "If the loss limit is exhausted there is no need to apply the brain" — v5c's WHAT IT DOES NOT SAY clause stated from the other side.

The outage: confirmed, not open

The 18-minute run of agent_error rows from 10:42 was the Pro plan's 5-hour usage window expiring — confirmed by the operator. Nothing is broken and no code change is wanted.

What it adds is a number. The 11:00 cutoff is sized to that plan window, but the window does not end at 11:00 — today it ran out at about 10:42, roughly eighteen minutes early. So the agent can go dark before its own cutoff, and anything open in that gap is unmanaged: no premise check, no agent exit, only the host's mechanical risk and then the cutoff flatten.

Today that gap was worth +₹8,535, because the cutoff booked the second short about six points off the session's post-entry low, minutes before a reversal that ran 140 points the other way. That is luck, not design. The answer already in the book is v5g's PRACTICAL FORM — name the window before entering — which is worth more here than a new guard, because a position reasoned for a window ending at 11:00 is really operating on one that can end at 10:40.

Recorded too: the cause is not recoverable from the log. SLH-004 logs this class with canned, content-free messages on purpose, so a future outage of the same shape will again need the operator to say which it was.

The pre-open note for 18 Sep

From I7gORF7zm4o, uploaded 2026-09-17 21:25 IST.

Third consecutive note with the same branch shape — gap up → BUY, flat or gap down → SELL, for all three indices — so what the test protects is that the shape is not all there is. The reason moves each evening, and tonight it's whether a multi-day negative run has finally seated sellers.

Open	Side	His words
Flat / gap down	SELL	"sellers are not much seated here" — given on NIFTY as that branch's premise
Gap up	BUY	"because the market has been negative for many days, there COULD be sellers here, so in a gap-up case they can be targeted"

The conditional is the point: the sellers are hypothetical until the open shows them. An edit that asserts them rather than conditioning on them inverts the whole read, and the test catches exactly that mutation. Today's shape is why the question is open at all — "momentum turned positive but a rejection also formed".

NIFTY      R 23500 / 23360   S 23100 / 23000
BANKNIFTY  R 56650 / 56300   S 55810 / 55500
SENSEX     R 75200 / 74700   S 74000 / 73650

Only one value needed resolving tonight. BankNIFTY came through clean and is unchanged from the 17 Sep note — itself a useful check that nothing drifted. NIFTY was clean too, and note it is 23100 tonight, not the 23110 corrected yesterday, so the level genuinely moved rather than the ASR repeating itself. SENSEX's supports arrived as 7473650 again, byte-identical to the previous night's garble, and were confirmed to the same 74000 / 73650; its resistances were clean this time.

Negative-tested 17 ways, all 17 caught, plus a passing control — including guards that a reverted NIFTY 23110, a reordered BankNIFTY pair, or the SENSEX reading I offered and had rejected would each fail.

No doc section: notes haven't carried one (PR #168 precedent).

🤖 Generated with Claude Code