SYS.01 · STRUCTURAL PATH-FINDING · STATUS // STANDING · POINTED AT ACTION
The third door.
Point it at a goal and it finds the non-obvious, workable way there. Not the front door
everyone competes at. Not the back door of known exceptions. The third — the door the drift leaves
open. The foundation's delta, pointed at action.
Every system has three ways in. The rulebook only admits to two.
The front door is where everyone competes.
The back door is the known exceptions. Both are priced accordingly.
The third exists because the real layout drifted from the
rulebook — and only one of them gets updated.
Read the rules against the live structure,
and it opens.No lock is picked. The map was wrong, in your favor.
Step through.On the far side: the machine that finds them.
00THE RUN — ONE LOOP, START TO FINISH
You bring a goal. The machine — a built harness that drives a model through each
expert's actual procedure — spreads into many expert views instead of collapsing to one, cuts on
your reactions, digs only where it fits, and chains the survivors into routes. Wide, cut, deep,
chain — then loop. Everything below is this loop, walked once, on one real target.
01THE TARGET
STAGE 00 · INTAKE & AIM
A run starts with one thing pinned and everything else deliberately loose. The governing
rule: constrain the destination, liberate the route. Lock what you actually want. Leave the how
wide open — including what you think you can't do.
OBJECTIVE // FIXED
"My HVAC company does four million a year. Double what it's worth, and sell in four
years."
The one running example on this page. A concrete business, a hard goal, a clock.
HELD LOOSE · CONSTRAINTSYou can't list them
cold. They surface when you react to a real option — "no, not that, because…". So they stay open,
and your kills fill them in.
HELD LOOSE · CAPABILITIESThe fastest way to
kill a great route is "I can't, so scrap it." The route gets found blind to what you can do,
and reconciled with your reach after.
!The trap it's built to dodge: quietly turning an assumption into a law. In one
real run the search forced every answer toward government programs because an early step assumed
"this has to be government-funded" — and every step after inherited it as if someone had set it.
Nobody did. Stripping out false walls like that is the single biggest source of new options.
02THROW ON EVERY LENS
STAGE 01 · WIDE CHEAP SCAN
The first pass is wide and cheap: throw on many experts' glasses at once, fast and shallow,
just to see which territories light up. A lens is one expert's way of seeing — their actual
procedure, encoded so the machine runs it instead of imitating it. Costume versus capability.
None of these experts ever saw your problem. That's the point.
The full rack in one figure, grouped by the question each lens asks. Open any
lens — the move it runs, and the shoulders it stands on. Where two lenses stand on the same
giant, a hairline joins them.
:: TAP A LENS — HAIRLINES = SHARED SHOULDERS
A
What's actually there — and what should be, and isn't?
structure and pattern · 05 LENS
Configuration / GestaltThe whole shape before any detail: is this situation type A or type B?+
What it searches for //
the situation's type first — then the few costly-to-fake features an adversary couldn't stage.
Successional readingWhat a sequence is about to become, and where it branches.+
What it searches for //
a unit's stage in a named trajectory — the next stage, the branch points, and whether the stage is being faked.
Extracted from
Industry life-cycle research (Klepper)Ecological successionOrganizational-lifecycle studies
swap the glasses.
03CUT WHAT'S DEAD, THEN DIVE
STAGES 02–04 · CUT → DIVE → VALIDATE
The wide scan hands you a menu. You react fast — kill what's dead — and each kill names a
constraint the machine couldn't have guessed. You prune by reaction, not recall: you can't
list your walls cold, but you know "no" the instant you see it. A killed idea isn't
waste — it locates a wall. The machine is echolocating the space, not pruning a tree.
TWO KILLS FROM THE TARGET'S RUN — EACH ONE NAMED A WALL
bring in a PE partner to fund the roll-up"no outside investor." — a wall, named
chase the build boom two metros over"won't relocate." — another
the dormant service-annuity route↳ survives — re-aimed, deeper
THEN THE SURVIVORS FIRE DEEP — the move that makes this more than a clever chatbot
01Inhabit the expertnot generic text in their vocabulary — their actual procedure.
→
02Name what the field stopped expectingthe load-bearing thing everyone quietly stopped checking.
→
03Run the oblique querythe search only that expert would think to run. Go get real facts.
→
04Reason again on what came backgrounded, dated, checked — not answered from memory.
ON THE TARGET · THE TEMPORAL LENS, DEEPIt doesn't just say "refrigerant rules are changing." It becomes the
refrigerant-regulation expert, realizes the durable value isn't the replacement boom everyone sees —
it's the service annuity on the equipment nobody's replacing — pulls the actual phase-down
schedule and supply data, and comes back with a grounded, dated, non-obvious play.
never what they do — always what they search.
04READ THE PLAYER
STAGE 01 · FIRES WITH THE SCAN
The lenses ask what is here. Firing right alongside them, the exploit library asks
who is playing this, and how — each entry a recurring adversarial move stripped to its structural
logic. Every one names the exact reason its wall is passable: the lock. The catalog of locks is
small, and naming the lock is what makes it pickable.
FURTHER PATTERNS WITHHELD · IN ACTIVE USEthe edge you flash is the edge you lose
EVERY EXPLOIT CARRIES ITS INVERSION — THE ATTACK, TURNED INTO A BUILD
Reversal ✗"Stop using metrics. They just get gamed."
→
Inversion ✓Design a metric anchored to reality, and get
it adopted. Same lever, aimed at a real outcome.
In the library todaymarket-boundary arbitragechokepoint controleffort-asymmetryimpatience & liquidityinformation leakagefalse randomizationedge-sortingmispriced liabilitylayer migrationmetric colonizationtrust engineeringhost-machinery hijackwhy doors close
respect the engine. glitch the convention.
05SORT THE WALLS
STAGE 05 · SORT THE WALLS
Every finding gets sorted into one of three, and only one of them actually stops you.
A killed branch isn't a loss — it locates a wall and sharpens the map.
BEDROCKreal, fixed. Closes the branch —
which is information. Held as the current best read, challenged last, never gospel.
BUILTa rule people made. Gets the real
question: what specific reason is this one passable?
DECORATIONconvention with no teeth.
"That's not how it's done." Walk through it.
SORT THREE YOURSELF — FINDINGS FROM THE TARGET'S RUN (§01)
"the refrigerant phase-down date"
"you need an in-house crew to bid"
"HVAC sells at 4× — everyone knows"
AND WHEN THE WALL IS YOU — a capability, not a truth
THE BEST ROUTE · BLOCKED
ROUTE Aclearobvious · everyone's on it
ROUTE Bclearknown · priced accordingly
ROUTE CWALLruns straight to the goal · least contested · but you can't clear one hump
the reframe▸
RECLASSIFIED · A COST, NOT A WALL
ROUTE Astill just obvious
ROUTE Bstill just known
ROUTE CCOST · CLEARABLEbuy it · learn it · hire it · wait out its clock — now the best route is the live one
The fastest way to lose a great route is to call your own limit a wall. If Route C is
good enough, the one hump stops being "I can't" and becomes a smaller problem the machine solves on
itself — the cheapest, quietest way to clear it.
06CHAIN IT
STAGE 06 · CHAIN
One door rarely gets you all the way. Each opening is modeled as a state; the engine
links them so one step's result genuinely is the next step's entrance. "Do this, and also that"
is a portfolio. A route is a sequence you couldn't skip a step in. Borrowed most directly from
speedrunners, who chain single glitches by knowing exactly which state each trick leaves the game
in. And when a link's fix lives in another field entirely, the chain borrows it whole — most problems
wear a skeleton some other field already solved, under a different name. Below, the real shape from
the target's run.
BOOTSTRAPratchet a small position into a bigger one
INTERLEAVEtwo loops running, each feeding the other
CONSTRAINT-FORCEDthe blockage on the normal path is what surfaces the third one
TIMEDonly works inside a closing window · the step with a clock goes first
07THE DOORS IT HANDS BACK
STAGES 07–08 · RE-AIM ↻ → HANDBACK
You end with a short menu of real routes — each a sequenced plan carrying its evidence, the
exact lock it picks, its clock, its costs, and honest flags on whatever's still thin. Some are ordinary
and solid; at least one is usually something you'd never have thought to ask for. It is not a verdict
to obey — it's a set of doors you didn't know were there, and you choose.
THE SAME TEST // RUN 1 OF 3
08AGAINST THE AVERAGE MACHINE
A normal AI converges early and guesses confidently. This forces it to diverge
early and prove before it converges — and it moves the deciding out of the model, which only knows
the average, into reality: you, plus the evidence.
A GENERIC MODEL
Converges instantly. Hands you one answer.
Guesses from memory to fill the gaps.
The averaged middle of everything written.
You obey it.
THIRD DOOR ///
Diverges on purpose. A menu held open.
Goes and finds out, then checks what it found.
The oblique query only that expert would run.
You choose, sharper than when you started.
THE HONEST CONSTRAINTA full pass fans one problem across many lenses, fires them more than once, and searches
and validates each. That is genuinely expensive in compute and time, so the depth of a run is matched to
what the problem is worth. A run has to earn its expense.