Timekeeping Parameter

Part of the cognition series. Builds on Temporal Compression and The Consolidation Codec.

Any system that keeps time under bounded memory converges on the same structure: store complete states occasionally, store differences the rest of the time. The ratio between the two is the keyframe rate. It is the survival parameter of temporal compression.

Why it converges

Three constraints force the structure:

  1. The buffer is finite. You can’t store every moment at full resolution.
  2. The sequence has temporal dependencies. Later states depend on earlier ones.
  3. Accumulated error grows with chain length. Each delta drifts from the original. Over enough steps, the reconstruction becomes unusable.

Given these constraints, any system that doesn’t periodically store a complete state will drift past recovery. The checkpoint isn’t an optimization. It’s a necessity. The only free parameter is how often. (This vocabulary has a formal algebraic counterpart — sheaf cohomology on directed graphs separates removable from irreducible error, and the checkpoint is the sub-complex where irreducible error vanishes.)

The evidence

The same five-part architecture appears independently across fifteen domains. Technical systems (top rows) are precise; institutional and cultural systems (lower rows) are less exact but structurally recognizable.

DomainCheckpoint (I-frame)Delta (P-frame)Cycle (GOP)Break (chain fragility)Reset
Video codecsIntra-coded framePredicted frameGOPReference lossKeyframe insertion
gitBase object / packfile snapshotDelta objectRepack cycleCorrupted objectRe-clone / fsck
DatabasesWAL checkpointWrite-ahead log entryCheckpoint intervalLog corruptionRecovery from checkpoint
ML trainingModel checkpointGradient updateCheckpoint intervalNaN / divergenceRollback to checkpoint
AccountingBalance sheetJournal entryMonth-end closeAudit discrepancyReconciliation
Oral poetryFormulaic epithetImprovised variationPerformance segmentAudience lostRe-invoke formula
Common lawLandmark caseSubsequent opinionEra of precedentOverturned rulingNew landmark
HistoriographyPeriod boundaryEvents betweenEraLost contextRevisionist re-encoding
ReligionFounding mythCultural practiceLiturgical cycleSchismReformation / revival
Nation-stateFounding documentLegislation / amendmentPolitical generationConstitutional crisisNew constitution
Scientific paradigmParadigm (Kuhn)Normal scienceResearch programAnomaly accumulationRevolution
Software architectureClean-sheet designFeature additionsMajor versionTechnical debt collapseRewrite

Twelve domains. Same architecture. The top five are precise: git’s packfile delta compression is I-frame/P-frame encoding on a DAG, and database WAL checkpointing is GOP structure. The lower ten are structural analogies, less exact but recognizable.

What kills systems: the evidence

Systems that get the keyframe rate wrong fail in predictable ways.

Too few keyframes (drift → death):

SystemWhat happenedChain lengthSource
Boeing 73757 years of inherited fuselage design, never a clean-sheet redesign. Engines mounted forward to fit, MCAS to mask pitch-up tendency. 346 dead.57 years, 4 generationsSeattle Times
KodakInvented digital camera in 1975, refused to reset founding assumption (photography = film). Bankruptcy 2012.37 yearsMIT Sloan
Qing Dynasty268 years without structural reform. Three failed partial resets (Self-Strengthening, Hundred Days, New Policies), each too late. Fell 1912.268 yearsBritannica
ZoroastrianismState religion of Persia → 200,000 worldwide. Orthodox refusal to accept converts. No mechanism to replenish.~1,400 yearsPluralism Project
Ariane 5 Flight 501Reused Ariane 4 inertial reference software without re-verifying against new flight profile. Integer overflow 37 seconds after launch. $370M lost.1 transplantMIT report
South Korea (TFR)Post-war I-frame: industrialize or die. Never decompressed after crisis ended. Reproduction was never in the checkpoint. TFR 0.75, population projected to halve by 2100.70 yearsStatistics Korea

Too many keyframes (overhead → death):

SystemWhat happenedReset frequencySource
France 1791–187014 constitutions in 80 years. No regime lasted long enough to build legitimacy.~1 every 6 yearsWikipedia
Mao’s Continuous RevolutionPermanent ideological reset destroyed institutions faster than they could function. 500K–2M dead.ContinuousStanford SPICE
Khmer Rouge Year ZeroTotal reset: all culture, institutions, professions destroyed and rebuilt from scratch. 2M dead out of 7M.Single totalWikipedia
Italy (postwar)68 governments in 76 years. No policy trajectory can accumulate momentum.~1 every 13 monthsEuronews
Digg v4100% rewrite discarded all accumulated community features. Lost 30% of audience in one month. Couldn’t roll back.Single totalSearchEngineLand

Got it right (calibrated keyframe rate → survival):

SystemKeyframe rateDurationWhy it worksSource
Catholic ecumenical councils~1 per 90 years (21 in 1,900 years)1,700 yearsSlow enough for continuity, frequent enough to prevent fatal drift. Trent answered the Reformation. Vatican II answered modernity.Catholic Answers
British constitutionIncremental partial keyframes (Magna Carta, Reform Acts, Parliament Acts)800 yearsNo total reset, ever. Gradual expansion within stable institutional frame.PolSci Institute
Vedic oral transmissionContinuous (11 redundant recitation modes)3,000+ yearsRedundant encoding as continuous error correction. Group verification — any mistake triggers restart.IJFMR
Aboriginal songlinesLandscape-anchored (physical terrain as external keyframe)7,000+ yearsExternal reference frame immune to cognitive drift. Multi-modal encoding (song, dance, painting).Scientific American
US Supreme Court~1 reversal per year (141 in 170 years)170 yearsStare decisis as default, with rare landmark reversals for constitutional drift.Constitution Center
Swiss ConfederationDistributed (26 cantons reset independently)734 yearsFederalism distributes keyframes across scales. Cantons can reset without destabilizing the whole.Liberty International
The MishnahEmergency keyframe (~200 CE)Chain savedOral Torah deliberately kept unwritten. Temple destruction threatened the chain. Emergency checkpoint to persistent storage.Chabad

What the tables show

1. Too few keyframes kills slowly. Boeing drifted for 57 years. The Qing drifted for 268. Korea is drifting now. Accumulated error becomes structural, and the system can no longer distinguish between what it is and what it was supposed to be. By the time someone calls for a reset, the cost of a keyframe has grown larger than the system can bear.

2. Too many keyframes kills fast. France cycled through 14 constitutions in 80 years. Mao’s Continuous Revolution destroyed institutions faster than they could function. The Khmer Rouge destroyed the substrate itself. Each reset discards accumulated P-frames that took years to build. Reset too often and nothing accumulates.

3. The survivors calibrate. The Catholic Church resets once per century. The British constitution patches continuously without ever fully resetting. The Vedas use redundant encoding as continuous error correction. Aboriginal songlines anchor to physical terrain that doesn’t drift. None found the rate by theory. All found it by selection: the ones that got it wrong aren’t here to compare.

Trauma compresses the keyframe

South Korea’s population crisis is a keyframe written under crisis that was never decompressed after the crisis ended.

The pre-war I-frame was Confucian: family continuity, ancestor worship, filial piety. Reproduction was in the checkpoint. The Korean War destroyed everything. Park Chung-hee’s developmental state wrote a new I-frame from rubble: industrialize or die. Intentionally monodimensional — a traumatized society rebuilding from nothing can’t afford a complex checkpoint. A simple I-frame compresses well, transmits easily, aligns the population toward a single objective.

It worked. One generation: agrarian poverty to OECD.

Then the crisis ended and the I-frame didn’t update. Family formation, communal obligation, generational continuity were never re-encoded into the new reference. They existed only as P-frame inheritance, increasingly distant from the active checkpoint. When accumulated economic and social pressure compressed the P-frames, reproduction was the first thing dropped. It was never in the I-frame. It was never protected from drift.

The standard explanations aren’t wrong. They’re downstream. Housing is expensive because the society optimized for one dimension and never re-encoded livability. Work culture is brutal because the I-frame says sacrifice, and every institutional delta since 1960 has been faithful to that reference. Education is a pressure cooker because the checkpoint rewards competition, not formation. These are consequences. The root cause is what the I-frame encodes and what it doesn’t.

“Economics” fails because Israel has comparable cost of living and a TFR of ~2.85. “Culture” fails because it doesn’t specify which parameter. The keyframe framing does: Israel’s I-frame encodes reproduction (“be fruitful and multiply” is in the Torah, which is the I-frame). Korea’s doesn’t. What’s in the checkpoint is protected from drift. What isn’t gets compressed away.

The generalization: the severity of the trauma that forces a new I-frame determines its dimensionality. The more severe, the simpler. Simple I-frames are adaptive in crisis. In peace, they become the chain that can’t encode what matters.

The parameter

The keyframe rate is not a design choice. It emerges from three quantities:

Every system in the tables above discovered this tradeoff independently. Codecs formalized it as rate-distortion optimization. Legal systems formalized it as stare decisis with rare reversals. Religions formalized it as liturgical cycles with occasional councils. The formalism differs. The parameter is the same.

The systems that persist are the ones that found, by trial or by selection, a keyframe rate matched to their constraints. The systems that died are the ones that didn’t. This is not a metaphor. It is the consequence of keeping time with a finite buffer.

How to prove this wrong

Score founding documents for dimensionality. Constitutional preambles, national curricula, liturgical texts, corporate mission statements — count how many distinct social functions each one encodes. Reproduction, education, defense, commerce, leisure, spiritual practice, environmental stewardship. Each is either in the checkpoint or it isn’t.

Then measure which outcomes survived a century of drift and which didn’t.

The prediction: outcomes encoded in the founding document persist under pressure. Outcomes absent from it collapse first when the system is stressed. If Korea’s founding documents encode development but not family formation, and Israel’s encode both, and the TFR tracks that difference after controlling for GDP, housing cost, and female labor participation — the parameter is real.

If it doesn’t track — if I-frame content has no predictive power over which outcomes survive drift — the framework is wrong. Not “needs refinement.” Wrong. Run the test.


Written via the double loop. For the technical crosswalk: Temporal Compression. For the formal reading section: june.kim/reading/temporal-compression.