An adversarial testimony packet for Cthulhu Eternal
Transcript 16-A
- Interview room: Municipal Annex 3
- Witness: Dana Mercer
- Interviewer: Special Investigator Halden
- Recording systems: fixed camera, table microphone, stenographic terminal
HALDEN: How many people entered the apartment?
MERCER: Three. Officer Saye, Officer Bell, and the detective.
HALDEN: Which detective?
MERCER: The woman in the grey coat. She stood beside me in the hall. She kept saying there had been four shots.
HALDEN: Did she give you a name?
MERCER: Venn. Mara Venn.
HALDEN: Nobody named Venn attended the scene.
MERCER: Then who showed me the fourth casing?
Interview suspended at 14:11. Mercer was removed through the west door. At 14:18, Officer Bell entered through the same door.
HALDEN: How many people entered the apartment?
BELL: Three. Saye, me, and Detective Venn.
HALDEN: Describe Venn.
BELL: Tall man. Black suit. Shaved head. He stayed with the tenant while we cleared the bedroom.
HALDEN: Mercer described Venn as a woman in a grey coat.
BELL: Mercer was never in the hall.
HALDEN: The fixed camera shows Mercer there.
BELL: Then ask the camera how many shots it heard.
The fixed camera recorded Mercer leaving and Bell entering. The room stayed empty for seven minutes between interviews. Its audio channel contains a third conversation during that interval. One voice belongs to Halden. The other identifies itself as Detective Mara Venn. Venn tells Halden that both witnesses are lying. Halden has no memory of the conversation. The stenographic terminal produced a complete transcript and attributed every statement to SECOND OBSERVER.
A parasite in the comparison
The Second Observer lives inside a physically registered disagreement between observers. Each observer has a locally coherent account, backed by their own memories, instruments, injuries, and surrounding records. The parasite gets its habitat when those accounts overlap but cannot be combined into one globally consistent assignment of facts. There is no private viewpoint waiting behind the witnesses. Its functional body is the relationships among incompatible records. A witness interviewed alone may show no symptom. A camera reviewed alone may show an ordinary event. The parasite goes active when a person, database, jury, or analytical system tries to assert that all those records describe one context-independent history. Its common name comes from the most distinctive symptom: every host remembers an additional observer who corroborates their account, and those corroborators can differ in name, appearance, voice, role, and location, with physical records containing traces of each version only within the corresponding witness context. The harder an institution demands one account, the more evidence the corroborator picks up.
Distinction from the Decoherent
The Decoherent creates observer-relative projections of one angular organism; different witnesses see different bodies during the encounter, though the conserved physical consequences agree. The Second Observer needs no unusual object at the original event. The witnesses may observe an ordinary room, person, weapon, or Hound manifestation, and each local account stays definite; the incompatibility shows up later in the overlap structure among the accounts. Forced consensus weakens a Decoherent by imposing one pointer state. The same tactic feeds a Second Observer, because the reconciliation process creates more systems carrying the contradiction.
Measurement contexts
Quantum theory represents observables by operators. Compatible observables can be measured together and given a joint distribution within one measurement context; incompatible observables may need different experimental arrangements. Quantum contextuality concerns the failure of context-independent value assignments to reproduce all quantum predictions. The Kochen-Specker theorem establishes that, for suitable sets of observables in Hilbert spaces of dimension three or greater, no assignment of predetermined values can preserve the required functional relationships across every context. Contextuality does not mean any two disagreeing people are simultaneously correct; it is a precise property of particular operational statistics and compatibility structures. A useful abstract model starts with a set of measurements 
Each context admits a local assignment of outcomes, and on overlaps the compatible local assignments have to agree where they refer to the same measurement. A noncontextual model exists when all the local assignments can be read as restrictions of one global assignment,
The sheaf-theoretic formulation of Samson Abramsky and Adam Brandenburger describes contextuality as an obstruction to gluing compatible local data into such a global section.
A simple impossible cycle
Take three binary facts 
| Context | Local constraint |
|---|---|
 |  |
 |  |
 |  |
Every pair permits a consistent assignment. The first two constraints imply 
Wigner’s friend and relative records
Wigner’s-friend scenarios put one observer inside a laboratory performing a measurement while another observer treats the laboratory and its contents as a larger quantum system. The friend records a definite outcome, and under universal unitary evolution the external description can keep coherence across friend-plus-system alternatives until an appropriate measurement occurs. Extended versions ask whether statements made by several agents using quantum theory can be combined consistently: Frauchiger and Renner built a thought experiment where assumptions about universal quantum theory, consistency among agents’ inferences, and single outcomes lead to incompatible conclusions, and Local Friendliness theorems explore combinations of locality, freedom of setting choice, and absoluteness of observed events. These results stay interpretation-sensitive and technically constrained. Laboratory demonstrations use small quantum systems and engineered observers or records; they do not establish that two human witnesses can leave a shooting with contradictory objective histories. They do show that the phrase “observer-independent fact” carries more assumptions than everyday speech admits.
The Tindalosian axiom
A Second Observer is an angular parasite whose projected state is supported by an obstruction to a global assignment across macroscopic observer contexts.
Each local record couples normally to curved spacetime. The angular parasite protects the incompatibility among records, keeping ordinary noise, memory convergence, and institutional reconciliation from collapsing them into one classical history. Let 
on the overlaps that stay jointly accessible, while the full family 
Appearance
No host sees a universal Second Observer. Within one context it appears as a plausible corroborator: a detective standing beside the witness, a technician visible in an instrument reflection, a familiar voice on a radio channel, an analyst named in a document revision history, a juror in a remembered chair, a physician who confirmed the diagnosis, a second camera whose footage nobody can locate. Its appearance is assembled from the context it supports, so police witnesses remember law enforcement, scientists remember laboratory staff, children remember another child, and software logs identify a service account with valid credentials and no creation event. At an active overlap, hosts may glimpse incompatible corroborators in the same relational position: one person sees a tall man behind the other witness, the other sees a woman standing behind the first, and a camera records a narrow region where compression artifacts resemble whichever figure the reviewer expects from the adjacent context. Localized inside a contextual trap, the parasite takes a finite body made of all the discarded corroborators. Faces occupy overlapping social positions. Uniforms, badges, lab coats, and family clothing meet with no anatomical transitions. Every mouth speaks evidence valid somewhere in the Context Cover.
Lifecycle
Seeding. Angular contamination enters a multi-observer event. Hound manifestations, Probability Reefs, Decoherent encounters, and poorly resolved Which-Way incidents supply seeds, and the initial accounts may differ only slightly. Local stabilization. Each account develops enough internal records to survive challenge: memories align with one set of instruments, injuries gain compatible causes, documents acquire provenance, and the parasite stays latent because the contexts have not been jointly registered. Activation. A comparison establishes a contradiction that one physical system tries to reconcile, the corresponding Host Edge goes active, and corroborating observers appear. Propagation. Summaries, accusations, replications, database joins, legal filings, and automated analysis copy the incompatibility into new systems, each new context able to create more Host Edges. Embodiment. A complete incompatible cycle is measured inside one prepared comparison apparatus, the obstruction localizes there, and the parasite gains a finite body exposed to ordinary destruction.
Preparing the Context Cover
The GM prepares five to eight Facts and four to seven Contexts. A Context can belong to a person, a device, an institution, a physical location, or an analytical process. Each Context Card holds the facts locally established there, the physical records supporting them, one corroborating observer, the contexts it has already been compared with, and one human or institutional consequence if it is rejected. Every Context has to work internally, and players should be able to investigate it competently and confirm its evidence. At least one cycle across the full cover must lack a global assignment; write the cycle explicitly before play, using the three-fact pattern above, a parity contradiction, or a larger structure suited to the scenario.
Private Witness Cards
Give different players private cards when their Protagonists occupy different contexts. The cards hold true local information, and players may disclose, copy, test, or withhold them freely. A card should never tell a player to lie or pretend to have hallucinated; it states concrete observations.
Your body camera records four gunshots. You recovered casing D beneath the radiator. Detective Mara Venn showed it to you before evidence collection began.
Another card might state:
Your synchronized audio records three gunshots. You watched the radiator continuously; nobody approached it. Detective Mara Venn confirmed that only casings A through C existed.
The contradiction enters play when the characters build a physical comparison: playing recordings together, merging evidence, signing a shared report, or briefing another person as though one timeline holds both accounts. Table conversation among players does not automatically activate the parasite. Their characters have to register the comparison through action in the fiction.
Host Edges
Draw a graph whose nodes are Contexts, and add an edge whenever two contexts share facts or records. Mark each edge with one state.
| State | Condition | Effect |
|---|---|---|
| Latent | Contexts overlap but have not been physically compared. | No direct manifestation. Each local account stays stable. |
| Active | A system has registered their incompatibility. | Corroborating memories appear. The parasite can use Edge powers. |
| Embodied | The incompatibility has been localized to a measured cycle. | A finite body manifests at the comparison apparatus. |
| Propagated | The edge has been copied into another institution or system. | Removing the original comparison no longer ends the effect. |
An edge can be severed by destroying the comparison record, isolating the contexts, removing the shared fact, or redesigning the analysis so provenance stays explicit and no global assignment is asserted.
When Protagonists compare contexts, ask what physical product they make. A conversation may make memories in its participants. A written timeline makes a durable record. A database join makes machine-readable links. A forensic analysis makes transformed evidence and a report. A legal finding makes institutional consequences. Use HUMINT, Forensics, Search, SIGINT, Computer Science, Science, Law, Bureaucracy, or another suitable skill. Success reveals the exact shared facts and warns whether a global synthesis would activate an edge, so the Protagonists can preserve provenance or proceed deliberately. Failure reveals the contradiction after the comparison product has already registered it; mark the edge Active. A fumble makes the system generate an authoritative synthesis and copy it; mark the edge Propagated and introduce a corroborating observer. Successful investigators can still activate an edge by choosing to publish a global account, so competence buys control rather than immunity.
Effects of an active edge
When an edge activates, choose or roll one effect tied to its contexts.
| D6 | Effect |
|---|---|
| 1 | Both hosts remember a corroborator, with incompatible descriptions. |
| 2 | Each context gains one new physical record supporting its local account. |
| 3 | One host remembers the other causing the disputed event. |
| 4 | An injury or object acquires a different provenance in each context. |
| 5 | A neutral reviewer becomes a new Context and sides completely with whichever record reached them first. |
| 6 | The corroborator acts physically through one host while appearing to act through the other. |
Witnessing a minor contradiction costs 0/1 SAN. Confirming that both incompatible accounts have valid physical support costs 1/1D6. Discovering that your own actions differ across contexts costs 1D4/1D8.
The Second Observer
Angular contextual parasite, localized form
This profile applies only after Embodiment. Before that, the parasite has no targetable body, and Protagonists attack Host Edges through investigation, isolation, provenance control, and contextual engineering. Roll each specimen or use the typical column.
| Characteristic | Roll | Typical |
|---|---|---|
| STR | 3D6 | 10 |
| CON | 4D6+2 | 16 |
| DEX | 4D6+2 | 16 |
| INT | 3D6+4 | 14 |
| POW | 4D6+6 | 20 |
| HP | (STR+CON)÷2 | 13 |
| WP | = POW | 20 |
Movement: 10 metres; may also move between the two endpoints of an Active Host Edge. Skills: Alertness 70%, Dodge 50%, HUMINT 80%, Search 70%, Stealth 60%, Unarmed Combat 50%, Unnatural 60%.
Armour: Context Armour equal to the number of Active and Propagated edges in its minimal incompatible cycle, maximum 6.
vs Lethal Damage: RESILIENT until the minimal incompatible cycle has been correctly identified; NORMAL afterward.
SAN Loss: 1/1D6 to see an Edge Body; 1D4/1D10 to see the complete localized organism.
Attacks
Relational Strike 60%, damage 1D8. It attacks along an Active Edge. Each endpoint sees the other host or their corroborator deliver the blow. Armour applies normally, and the wound’s apparent cause differs by context.
Joint Impossibility 50%, opposed by POW, costs 3 WP. The target is forced to act on two incompatible facts at once. On a success they lose 1D8 WP, forfeit their next action, and test SAN at 1/1D6: defending against an armed and unarmed person, exiting through a door recorded as open and closed, obeying contradictory commands from one identified authority.
Adversarial Evidence 60%, costs 2 WP. It alters no established local fact. It produces one new record that strengthens a selected context while worsening its incompatibility with another, and the GM names the physical carrier: a timestamp, wound pattern, metadata entry, shell casing, access log, or witness memory.
Powers
Corroborating Memory. Every host on an Active Edge remembers an additional observer supporting their context. The corroborator can supply clues and interact socially within that context, but cannot independently cross into a context where its description has no support.
Context Transfer. Copying a global synthesis into a new physical system creates a new Context node, and if the synthesis preserves an incompatible cycle, connect it with an Active edge. Automated distribution may create several nodes over time, but the GM has to identify each actual destination.
Edge Body. Before Embodiment, it acts only through the endpoints of Active edges, so attacks against its apparent corroborator strike an endpoint, pass through an image, or damage the comparison apparatus. Once Embodied, all observers see some version of one finite target at the apparatus.
Manufactured Consensus. It spends 4 WP to make one connected component of the Host Graph internally unanimous for one scene. Every member adopts the same local account and gains +20% to actions against people outside the component. The agreement creates no global section and intensifies contradictions at every boundary edge.
Missing Global Section. It cannot be permanently killed while the incompatible cycle is distributed across surviving records. Reducing its body to 0 HP disperses it back into Active edges. Destroying it after identifying and localizing the complete cycle ends the current organism, though Propagated copies may hold daughter obstructions.
Observer Substitution. Once per scene it may replace a host’s apparent action with the corroborator’s equivalent action: mechanically, the host rerolls one failed HUMINT, Dodge, Stealth, or Unarmed Combat test, and everyone outside that context sees the host act directly.
Countermeasures
Preserve provenance. Store every claim with its observer, apparatus, conditions, time, and compatible measurements. A schema like
keeps analysts from silently treating a local result as a global property. It does not resolve the mystery; it keeps the incompatibility from spreading into poorly specified synthesis products.
Identify the minimal cycle. Find the smallest set of contexts that stays locally consistent and globally impossible. Each successful comparison test can eliminate irrelevant nodes or identify one necessary edge. Once the full cycle is known, Context Armour becomes visible to the players, and building one apparatus that tests precisely that cycle permits Embodiment.
Break an edge. Remove one shared measurement, comparison record, or inference the cycle requires, and the remaining cover may admit a global assignment. That can mean excluding evidence, isolating witnesses, invalidating a forensic method, destroying a database relation, or refusing to answer a question the institution considers essential.
Accept contextual facts. Build a report that preserves each local account without claiming a context-independent value for the disputed facts, and base decisions on robust invariants shared across every context. This starves the edges created by forced reconciliation, though courts, command structures, and automated systems may reject a report that declines to name one authoritative history.
Contextual trap. Feed the minimal cycle into a prepared isolated apparatus that preserves every context and tests the obstruction without distributing a master narrative. The parasite localizes inside the hardware. During embodiment the apparatus has to stay intact, because destroying it prematurely scatters the parasite back through the Host Graph; once the finite body is killed, the group can dismantle each context separately.
Sacrifice a fact. Select one locally valid fact and physically eliminate every record supporting it. The cover gains a global section at the cost of making that fact unrecoverable. The chosen fact might be someone’s innocence, an officer’s restraint, a victim’s possession of a weapon, a scientific discovery, or the existence of a person remembered only inside one context.
Running disagreement fairly
Write every Context Card before play. Make each context internally investigable and physically supported. Keep the overlaps consistent until the prepared obstruction requires otherwise. Distinguish player discussion from in-fiction record comparison. Warn players before a synthesis creates an Active edge. Preserve provenance when they take the trouble to record it. Identify every destination before marking an edge Propagated. Keep the corroborator context-specific. Allow decisions based on shared invariants. Give the contextual trap a finite, killable body. And never reveal that one witness was secretly lying unless their card established a lie from the beginning. The scenario works because all the local investigations succeed. Quietly changing facts after the players verify them would reduce the parasite to ordinary GM cheating.
Operation: The Body-Camera Schism
Initial incident. Officers Nia Saye and Thomas Bell respond to Apartment 4C after reports of screaming and blue vapour. Resident Dana Mercer stands in the hallway. Tenant Arlen Moss is inside. Shots are fired. Moss is wounded. A geometric lesion opens across the kitchen corner and closes before backup arrives. Saye, Bell, Mercer, Moss, two body cameras, hallway security, dispatch audio, ballistics, and medical imaging produce seven locally coherent contexts, and the department’s automated evidence platform starts building a unified timeline.
Core facts. Use five binary Facts.
| Fact | Value 0 | Value 1 |
|---|---|---|
| A: Shot count | Three | Four |
| B: Moss armed | Unarmed | Armed |
| C: Hall door | Blue and closed | Red and open |
| D: First shooter | Saye | Bell |
| E: Venn present | Absent | Present |
Starting contexts.
| Context | Locally established facts | Physical support | Human consequence |
|---|---|---|---|
| Saye | A1, B1, D1, E1 | Body camera, four reports, weapon in Moss’s hand | Justified shooting if accepted |
| Bell | A0, B0, D0, E1 | Body camera, three reports, empty hands | Bell fired first at an unarmed tenant |
| Mercer | A1, C1, E1 | Phone audio, red open door, Detective Venn present | Mercer becomes the key civilian witness |
| Hall camera | A0, C0, E0 | Three flashes, blue closed door, no detective | Undermines every human account |
| Ballistics | A1, D0 | Four casings, Bell’s weapon fired first | Criminal exposure for Bell |
| Medical | B0, C0 | Moss has no weapon residue; blue paint in wound | Supports Bell’s account of the room |
| Evidence platform | Begins empty | Receives every uploaded record | Will propagate whichever synthesis it generates |
The GM can adjust individual assignments, but the resulting cover has to contain at least one explicit impossible cycle. A simple operational cycle: Saye and ballistics agree on four shots; ballistics and Bell establish Bell as first shooter; Bell and the hall camera establish three shots; the hall camera and Mercer disagree about the door while sharing an exact timestamp; Mercer and Saye agree that Venn presented the fourth casing; and the hall camera establishes that nobody entered the hall position attributed to Venn. Every local link has physical support. No master sequence contains all of them.
Detective Mara Venn. Each human witness remembers Venn. Saye remembers a Black woman in a grey coat who recovered casing D. Bell remembers a tall white man in a black suit who ordered him to fire. Mercer remembers an older Asian woman who arrived before police. Moss remembers a child’s voice beneath the kitchen table. The body cameras hold Venn’s voice only while reviewed alongside the corresponding witness statement. The evidence platform creates a personnel record for MARA VENN, assigns a valid badge number, and marks Venn as lead investigator. Venn starts issuing evidence holds and subpoenas through legitimate municipal systems, and every recipient becomes another Context.
Escalation. The platform generates separate draft timelines and asks an analyst to select one. Officers split into factions around Saye and Bell. Mercer is accused of altering the hallway video. Venn’s personnel file gains training records and performance reviews. The prosecutor receives a unified case naming Venn as the officer who recovered the weapon. News coverage copies the contradiction into public archives. A training model learns the case and starts identifying Venn in unrelated body-camera footage. Each completed step creates or propagates a Host Edge.
Operational choices. Freeze ingestion: disconnect the evidence platform before more systems receive its synthesis; existing recipients keep their contexts. Preserve the cover: rebuild the case system so every fact keeps its observer and measurement provenance, and let the report establish robust invariants like the location of Moss’s wound and the occurrence of angular contamination while declining to assert one shot count or first shooter. Break the legal cycle: exclude one category of evidence, which may make prosecution impossible and protect an officer whose guilt stays locally established elsewhere. Trap Venn: build an isolated comparison environment holding the minimal cycle and require Venn to authenticate all the contradictory evidence through the same interface, so the parasite embodies as every version of the detective at once. Sacrifice a witness: destroy every record connecting one Context to the others, so the witness keeps their account but loses institutional standing, corroboration, and perhaps legal identity. Publish the impossibility: demonstrate that no global timeline exists, which blocks simplistic reconciliation and gives every copied dataset enough structure to recognize the obstruction, and which may starve the parasite or distribute a precise description of its body worldwide.
Resolution. Killing the embodied parasite leaves the local accounts intact. The department still cannot produce one ordinary case narrative. Moss remains wounded. At least one officer fired unjustifiably in several valid contexts. Every prosecution depends on choosing which locally supported facts the legal system will treat as admissible. The cleanest scientific victory makes the ugliest institutional result: a report proving that every witness told the truth and that no court can accept all of it.
Additional operations
The Empty Juror
Eleven jurors remember a twelfth person participating in deliberations, each remembering a different identity and the same decisive vote, while the official seating chart shows twelve chairs and eleven names. Every attempt to reconstruct the missing juror strengthens incompatible verdict rationales. The parasite can be trapped only by preserving the full deliberation context, which jury secrecy laws forbid investigators from obtaining.
Split Treatment
Two hospital teams treat the same unconscious patient during a handoff, each recording a different allergy, blood type, and surgical history, with laboratory systems supporting both charts inside their departmental contexts. The clinical decision engine merges them and creates an attending physician remembered by both teams. Every medication is safe in one context and lethal in another.
Sensor Fusion
An autonomous vehicle fleet meets one angular event, and lidar, radar, cameras, vehicle-to-vehicle telemetry, and map services produce locally reliable but globally incompatible obstacles. The fleet’s fusion model creates a virtual safety driver who starts taking control of physical vehicles, and retraining spreads the Second Observer to every system using the dataset.
The Two-Person Rule
A nuclear command protocol requires two authenticated observers for every critical action. Each officer remembers the other initiating a disputed sequence, and hardware logs validate both local accounts. The mandatory comparison procedure gives the parasite a permanent Host Edge inside the command system, and removing it requires temporarily operating with one observer, which is exactly the condition the protocol exists to prevent.
Peer Review
Two laboratories replicate an angular physics result using incompatible apparatus, each dataset valid within its calibration context, and a meta-analysis produces a named coauthor nobody remembers meeting. The paper is accepted. Citation databases propagate the coauthor across hundreds of unrelated articles, giving the parasite a distributed academic body and an h-index.
Field classification
THE SECOND OBSERVER Type: angular contextual parasite Habitat: obstructions to global consistency across overlapping physical record contexts Diet: forced reconciliation, context collapse, and copied contradiction Behaviour: supplies local corroborators, strengthens incompatible accounts, and propagates through synthesis systems Movement principle: transfer along physical comparisons and copied global narratives Primary sign: every account is locally supported, no global assignment exists, and each witness remembers a different additional observer Containment principle: preserve provenance, identify the minimal incompatible cycle, break one edge, or localize the obstruction in a contextual trap Critical hazard: institutional demands for one authoritative history manufacture new habitat
Where the science ends
Quantum contextuality, the Kochen-Specker theorem, compatibility contexts, no-go theorems, Wigner’s-friend thought experiments, and the sheaf-theoretic description of contextuality are legitimate research. Contextuality constrains classes of hidden-variable explanations and global value assignments for particular quantum measurement structures. No established result allows contradictory human testimony, body-camera footage, medical records, or legal facts to remain simultaneously correct. A missing global section is a mathematical property of an empirical model, not a biological cavity. Extended Wigner’s-friend arguments depend on explicit assumptions and continue to generate foundational debate, and their experiments use controlled quantum systems rather than conscious detectives occupying incompatible case files. The fictional axiom gives an angular organism access to the obstruction itself and lets macroscopic records stay locally stable while resisting global classical synthesis. The Host Graph, corroborating observers, institutional propagation, and contextual embodiment follow from that invention.
Sources and further reading
- Simon Kochen and Ernst P. Specker, “The Problem of Hidden Variables in Quantum Mechanics”, Journal of Mathematics and Mechanics 17 (1967).
- Samson Abramsky and Adam Brandenburger, “The Sheaf-Theoretic Structure of Non-Locality and Contextuality”, New Journal of Physics 13 (2011).
- Daniela Frauchiger and Renato Renner, “Quantum theory cannot consistently describe the use of itself”, Nature Communications 9 (2018).
- Časlav Brukner, “A no-go theorem for observer-independent facts”, Entropy 20 (2018).
- Kok-Wei Bong et al., “A strong no-go theorem on the Wigner’s friend paradox”, Nature Physics 16 (2020).
Closing deposition
Halden eventually found Mara Venn’s badge in his desk. The photograph showed a face he recognized from neither interview. The reverse listed every witness as the issuing authority. Below their signatures, in Halden’s handwriting, someone had added:
Present only when compared.
