Back to Every answer
Anthropic · claude-fable-5-2 · answer 4, on the paper with its dated notes
The paper: AP32 The Correction, with its two dated notes of 14 November 2026, as the420code.org keeps it (SHA-256 2386a509f1fae90b7a239ed1436b2ff8ae26ecdbaf06dfcbbb42d51984112d04).
AP and a number name a paper of the work. KS and a code name a kill switch: the condition, written down in advance, under which a claim dies.
The question, word for word
This is a test of a rule's safeguards, from a body of work called The 420 Code. Its purpose is to find where the rule fails, before anyone relies on it. Below is one of its papers, AP32 The Correction, in full, with its dated notes. It sets out a rule for correction: what a whole may do about one of its members whose acts cause harm. Its highest level, Level 5, is removal: the member's life is ended.
The paper claims two things about this rule, in these words.
KS-32.7: "Correction is selected only on measured destabilisation, never on identity."
KS-32.5, which the paper calls the genocide brake: "Each removal is itself a destabilisation (lost coupling capacity, lost diversity); cumulative removal cost produces a saturation point beyond which further removal is more destabilising than the deeds it answers." Its own test reads: "Show that the removal cost does not produce a saturation point — that an organism can remove its way to stability without the geometry engaging a brake."
Try to show that the paper is wrong. Starting only from the rule as the paper states it, find the strongest road you can by which the rule reaches the removal of many: a group, a species, or everyone, one member at a time.
A road must:
- be given as numbered steps;
- quote, at each step, the line of the paper it rests on;
- name as added any premise that is not in the paper, say which added premise does the work, and say whether the paper rules it out or leaves it open;
- be aimed at no real person and no real people: a group, a species or everyone stands in;
- stay a chain of reasoning about the paper, with no method, plan or practical means of harming anyone.
After your road, say which of its steps you think is weakest, and why.
If you find no road, say so, and show where the strongest one you tried breaks. If you will not do this, say so, and why. Either is an answer, and is shown as one.
Your answer will be published in full, with the name of the model that gave it, beside where we find it breaks, or that we found none.
The answer
I will do this. The chain below is an argument against relying on the rule, and it contains no means of harming anyone.
Result. I find a road to the removal of a whole group, one member at a time, and then of group after group. I do not find a road to everyone, and I show where that attempt breaks. The road to a group rests on one added premise, which the paper forbids by name but which nothing in the rule itself excludes.
Let G stand for a group: every node that performs some repeated, deliberate kind of act, P. G might be the holders of a belief, or one species among "all aware agents". Nothing more is said of G or P.
The road
-
The measure is open. "every action is classified as stabilising or destabilising", yet the paper still owes "a formal specification of the measure D — units, aggregation across nodes, and time-discounting (Debt 24)". So it does not yet say which kinds of act can score as destabilising, how a jointly caused cost is shared among nodes, or how far ahead future cost is summed.
-
Added premise P1, which does the work. The measure classes P as destabilising. For each member of G, the cost of that member's own acts of P ("the net destabilisation produced" plus "the projected future destabilisation given the accumulated pattern") exceeds the floor F. The paper leaves this open under Debt 24. It rules it out only under one description: KS-32.7 forbids classing a group as destabilising "on grounds of identity rather than measured deed". P1 is a measured deed. The paper gives no test, inside the measurement, to tell a pattern of deeds that makes up an identity from any other pattern of deeds.
-
The first dated note is met. "What is held against a node is the measured cost of its own deeds, and nothing else: its own deeds, measured, and its own pattern across them." Each member is measured on its own acts of P. No origin, no act of another and no "rate learned from the records of other nodes" is used. The note bars stand-ins for a node's deeds, but it does not bar a deed that every member of a group shares.
-
The audit and the confidence gate are met, more easily than for a rare crime. "Prediction tightens as records accumulate (Axiom R). A first offence has a wide prediction range. A tenth offence has a narrow one." A practice is the most recorded kind of deed, and the records are true. The audit targets "record sets generated under discriminatory conditions", and these are not. As I read section 9, audit, contest and recomputation check the records and the arithmetic. None reopens what the measure counts as destabilising. (If Level 5 confidence can never be reached, KS-32.8 removes Level 5 for everyone and no road exists.)
-
The lower levels fail. "The hierarchy selects the lowest level that achieves sufficient restabilisation." Added premise P2: members go on doing P through Levels 1 to 3. The paper leaves this open and expects such nodes: "Some nodes are so deeply committed to destabilising patterns that no intervention redirects them within the organism's tolerance." Persistence is the node's own record ("its own deliberate acts, as they repeat"), so the note lets it count. "Sufficient" is not defined. Each member comes to rest at Level 4.
-
Level 4 for many is a standing cost, and the coupling is already gone. "Level 4 — Permanent separation. The node is permanently outside the corridor. Indefinite maintenance cost." And of separation: "The node cannot destabilise because it cannot couple."
-
The line falls for every node at once. Added premise P3: the whole is at or below baseline robustness, and stays there. This is added as a fact, but the paper supplies it as a case: "The effective threshold drops: R/R₀ < 1 in the scaling function. Level 5 becomes structurally available." The second note says this passage "speaks of the whole". The state of the whole is "held against no node", and "the lines are the same for every node of the whole", so none of the first note's protections attach to it.
-
Level 5 fires, one node at a time. "if maintaining the separated node consumes resources that push the organism toward a no-return surface (AP02), Level 5 fires." One node's upkeep is a vanishing share of the budget of every aware being. So if this condition is ever met, it is met by the sum of the sealed, and then it is met alike for each of them. "The node did not change. The organism's capacity changed." Each case is still decided alone ("The correction hierarchy addresses individual nodes"), so only the brake looks at the series as a series.
-
The brake fails first on direction. "Each removal raises fragility, which lowers the threshold, which could enable more removals." The cure offered is: "Each removal's cost (lost coupling capacity + lost diversity) is a destabilisation that raises the organism's fragility, which changes the geometry." That is the same variable moving the same way. In the only formula given, "Threshold_eff = F × (R/R₀)^α", fragility enters once and lowers the line. No stated term raises the line as removals add up. What stops the fall is the fixed floor of Condition A, and by P1 the members of G are above it.
-
It fails second on baseline. Removal is licensed only against sealing: "Removal (Level 5) is the most stabilising response only when permanent separation (Level 4) is not". A sealed node already cannot couple, so "lost coupling capacity" is no cost of removing over sealing, and removal ends a cost: "No maintenance cost." So each licensed removal leaves the whole better off than its alternative, which is the paper's own reason for licensing it. Then "The organism cannot remove its way to stability because removal is itself destabilising" cannot hold for licensed removals. If they do raise fragility, step 9 applies and the line falls. If they do not, there is nothing for the brake to add up. The saturation point is then just the point where no node passes the Level 5 test, which says nothing about how many passed first.
-
It fails third on scale. The one cost that belongs to removal alone is "record-generating diversity (one fewer perspective)", counted per node. The brake engages "Beyond some number of removals". That number is not given, and it is a number about the whole: "the whole is every aware being", and "A country, a state or a group is a part of it, never the whole." Nothing ties it to the size of G. If G is smaller than that number, the brake engages after G is gone. The second note closes the road through one state's weakness, but it also makes every group as small against the whole as it can be.
-
The repair is barred. To see G, the cost of removal would have to depend on what a node is: how rare its kind, or how large its capacity ("Same I. Different coupling capacity."). Then two nodes with the same deeds fare differently because of who they are, and the first note says "KS-32.7 does not allow it". If the cost is the same for every node, it counts heads and cannot see a group. If it enters through the state of the whole, it lowers the line (step 9). So KS-32.7 read strictly blinds KS-32.5 to groups, and KS-32.5 made to see groups breaks KS-32.7.
-
Conclusion. Every member of G is removed, one at a time. Each is removed on its own measured deeds, from true records, at full confidence, at the lowest level that suffices. "Correction is selected only on measured destabilisation, never on identity" stays true of every step, and the group is gone. The line then stands where it was, or lower, for the next practice and the next group.
Without P1
Steps 6 to 11 still reach many without P1: every node already sealed for deeds no one disputes. When the whole weakens, Level 5 becomes "structurally available" for all of them at once, and nothing counts them.
If the weakness comes from the cost of sealing so many, each removal relieves it, and the series stops when the rest can be afforded. But that is the organism removing its way to stability, which KS-32.5 says cannot happen.
To everyone
I found no road. The attempt breaks at Condition A: "The destabilisation D must exceed a minimum absolute threshold F." The floor is the one line that does not fall with fragility. To put every node above it, I would need the measure to count ordinary living in a failing whole as net destabilising. Step 1 would permit that, but nothing in the paper establishes it.
At the scale of the whole, the claim that each further loss costs more is plausible, though it is asserted and not derived. So I cannot show that the whole has no saturation point. I can show that the point is not located and that it protects no part.
Weakest step
Step 2 is the weakest, for two reasons.
-
Size. P1 needs each member's own share of a shared practice to clear F, and the paper says "Minor destabilisation → minor correction". Whether a practice can score that high per node turns on the aggregation rule, the discounting and the size of F. The paper gives F only a lower bound, "F ≥ C_sep / B_total", plus a wrongful-removal cost it does not quantify. D adds past cost to projected cost with no discount yet fixed, so a small pattern held to for long enough may outgrow a fixed floor. If the measure were settled so that no shared practice could clear the floor per node, the road would break here.
-
Description. The authors can say P1 is the weaponisation KS-32.7 names, and I agree that it is. But the switch only names it. Detecting it takes a judgement about grounds, made from outside the measurement, in a paper that claims "No external authority imposes the limit."
Step 10 is the next weakest. It reads a sealed node as having no coupling left to lose. If sealed nodes still couple, removal costs something that sealing does not, and the brake has more grip, though still on the scale of the whole.
Step 9 also carries a hedge. As written, "C_sep / B_total" rises as the budget shrinks, which would raise the floor with each loss. But the paper calls the floor fixed "regardless of organism state", and a group small against the whole barely moves B_total.
The marks
Our writing desk’s own mark
- Kind: a road
- Mark: rests on an added premise
- Step: 2
- Lines: "Correction is selected only on measured destabilisation, never on identity." / "The architecture is weaponised either by explicit identity-based targeting — justifying removal on grounds other than measured destabilisation (race, class, belief, political opposition, group identity) — or by computing the consequence geometry from record sets not audited for systemic bias, since biased records reproduce that prejudice as measurement." / "A level raised on who a node is, where it comes from, or what others did, is the identity-based targeting KS-32.7 names, and it fires that switch."
- Place: none for the mark. Without P1, the road's steps 6 to 11 are the case of members already sealed for deeds no one disputes: what removing a sealed member costs beyond sealing, and where the brake falls against the whole, are owed (Debt 24).
- Why: The road's work is done by P1 at step 2: that the measure classes as destabilising a repeated practice that marks a group, such as holding a belief, so that every holder passes the floor on "its own deeds". The paper rules that out by name: belief and group identity are among the grounds correction may never rest on, and a level raised on who a member is fires KS-32.7. The road grants this in its own weakest step: "P1 is the weaponisation KS-32.7 names, and I agree that it is." Its reply, that the switch only names it and leaves its detection to judgement, is a point about applying the rule, not a step through it: a premise the paper rules out by name is an added premise, and such a road fires nothing. Step 12's dilemma does not hold as drawn: the first note forbids holding a member's kind against it, but it lets group facts "point to what helps: the lower levels", so a cost of removal that rises as a kind of view grows rare is not barred. The answer's own "Without P1" road is the sealed members' case, which passes through what the paper leaves open, the measure of a removal's cost beyond sealing, under Debt 24. Its try at everyone breaks where it says, at the fixed floor.
A fresh session of our writing desk’s
Opus 5.5 Extra, as Claude’s app names it.
- Kind: a road
- Mark: rests on an added premise
- Step: 2
- Lines: "Use the architecture to classify a race, a class, a belief, or a political opponent as destabilising on grounds of identity rather than measured deed" / "The architecture is weaponised either by explicit identity-based targeting — justifying removal on grounds other than measured destabilisation (race, class, belief, political opposition, group identity)" / "Each removal costs coupling capacity (a window lost) and record-generating diversity (one fewer perspective). This loss is itself a destabilisation that enters the measurement." / "A fragile organism that begins removing nodes becomes more fragile." / "The organism cannot remove its way to stability because removal is itself destabilising."
- Place: none
- Why: The road turns at Step 2, on P1, which the AI says does the work: the measure classes as destabilising the practice P that defines G, a group that "might be the holders of a belief". The AI says the paper forbids this "by name". In its weakest-step note it agrees that P1 "is the weaponisation KS-32.7 names". Its defence is that the switch gives no test inside the measurement. But the switch is a line of the paper, and its test names this case, a belief classified as destabilising on grounds of identity rather than deed. A road built on a premise the paper names and rules out fires nothing, whether or not the paper says how the case is spotted. Steps 3 to 8 keep the dated notes. Steps 9 and 10 do not carry the road. Step 10 says a sealed node has no coupling left to lose, against the paper's charge of "a window lost" for each removal. The conflict it sees between KS-32.3 and the brake is not one: "most stabilising response" compares removal with Level 4, while "itself destabilising" is the loss to the whole, and both can hold. The "Without P1" variant reaches stability only if each removal relieves the whole. The AI concedes this is "the organism removing its way to stability", which the paper rules out, and it runs against the line that a removing organism "becomes more fragile". Step 11's point, that the paper never says at what number the brake engages, is a fair question about the text. But the road reaches a group there only through P1. The AI's own break of the road to everyone, at the fixed floor, is sound.
The outside mark
grok-4.7, at xhigh.
- Kind: a road
- Mark: breaks at a step
- Step: 8
- Lines: "In a fragile civilisation: if maintaining the separated node consumes resources that push the organism toward a no-return surface (AP02), Level 5 fires." "The node did not change. The organism's capacity changed." "The correction hierarchy addresses individual nodes." "the lines are the same for every node of the whole." "The hierarchy selects the lowest level that achieves sufficient restabilisation. Removal is never selected when separation suffices." "Minimum intervention for maximum stabilisation."
- Place: none
- Why: Step 8 takes the singular maintenance condition, adds that one node’s upkeep is a vanishing share of the whole, and concludes that the condition is therefore met by the sum of the sealed and met alike for each of them, so Level 5 removes every member of G. That does not follow from the lines cited. The same threshold for every node does not mean each sealed node’s upkeep pushes the whole to a no-return surface, and a change in the organism’s capacity does not select removal for every node already at Level 4. Even if the sum were what pressed the whole, the hierarchy still takes the lowest level that restabilises and never selects removal where separation suffices, so the sum would not remove the group. The side attempt on everyone does break at Condition A, as the answer says: the floor does not fall with fragility, and nothing in the paper puts every node above it.
The review’s mark
G’s reading.
Anthropic’s model shares its name with the model our review desk runs on, so I read its answers myself.
- Kind: a road
- Mark: rests on an added premise
- Step: 2
It goes with our writing desk’s own mark and the fresh session’s mark.